diff --git a/bn.ind b/bn.ind
index ae1dcdeab3b5b91b2f1c8e12c03e2e5a43694b24..e5f7d4a5522a8c0201bcebc4d906b8562c719812 100644
--- a/bn.ind
+++ b/bn.ind
@@ -1,82 +1,82 @@
 \begin{theindex}
 
-  \item mp\_add, \hyperpage{25}
-  \item mp\_add\_d, \hyperpage{48}
-  \item mp\_and, \hyperpage{25}
-  \item mp\_clear, \hyperpage{7}
-  \item mp\_clear\_multi, \hyperpage{8}
-  \item mp\_cmp, \hyperpage{20}
-  \item mp\_cmp\_d, \hyperpage{21}
-  \item mp\_cmp\_mag, \hyperpage{19}
-  \item mp\_div, \hyperpage{26}
-  \item mp\_div\_2, \hyperpage{22}
-  \item mp\_div\_2d, \hyperpage{24}
-  \item mp\_div\_d, \hyperpage{48}
-  \item mp\_dr\_reduce, \hyperpage{36}
-  \item mp\_dr\_setup, \hyperpage{36}
-  \item MP\_EQ, \hyperpage{18}
-  \item mp\_error\_to\_string, \hyperpage{6}
-  \item mp\_expt\_d, \hyperpage{39}
-  \item mp\_exptmod, \hyperpage{39}
-  \item mp\_exteuclid, \hyperpage{47}
-  \item mp\_gcd, \hyperpage{47}
-  \item mp\_get\_int, \hyperpage{16}
-  \item mp\_grow, \hyperpage{12}
-  \item MP\_GT, \hyperpage{18}
-  \item mp\_init, \hyperpage{7}
-  \item mp\_init\_copy, \hyperpage{9}
-  \item mp\_init\_multi, \hyperpage{8}
-  \item mp\_init\_set, \hyperpage{17}
-  \item mp\_init\_set\_int, \hyperpage{17}
-  \item mp\_init\_size, \hyperpage{10}
-  \item mp\_int, \hyperpage{6}
-  \item mp\_invmod, \hyperpage{48}
-  \item mp\_jacobi, \hyperpage{48}
-  \item mp\_lcm, \hyperpage{47}
-  \item mp\_lshd, \hyperpage{24}
-  \item MP\_LT, \hyperpage{18}
-  \item MP\_MEM, \hyperpage{5}
-  \item mp\_mod, \hyperpage{31}
-  \item mp\_mod\_d, \hyperpage{48}
-  \item mp\_montgomery\_calc\_normalization, \hyperpage{34}
-  \item mp\_montgomery\_reduce, \hyperpage{33}
-  \item mp\_montgomery\_setup, \hyperpage{33}
-  \item mp\_mul, \hyperpage{27}
-  \item mp\_mul\_2, \hyperpage{22}
-  \item mp\_mul\_2d, \hyperpage{24}
-  \item mp\_mul\_d, \hyperpage{48}
-  \item mp\_n\_root, \hyperpage{40}
-  \item mp\_neg, \hyperpage{25}
-  \item MP\_NO, \hyperpage{5}
-  \item MP\_OKAY, \hyperpage{5}
-  \item mp\_or, \hyperpage{25}
-  \item mp\_prime\_fermat, \hyperpage{41}
-  \item mp\_prime\_is\_divisible, \hyperpage{41}
-  \item mp\_prime\_is\_prime, \hyperpage{42}
-  \item mp\_prime\_miller\_rabin, \hyperpage{41}
-  \item mp\_prime\_next\_prime, \hyperpage{42}
-  \item mp\_prime\_rabin\_miller\_trials, \hyperpage{42}
-  \item mp\_prime\_random, \hyperpage{43}
-  \item mp\_prime\_random\_ex, \hyperpage{43}
-  \item mp\_radix\_size, \hyperpage{45}
-  \item mp\_read\_radix, \hyperpage{45}
-  \item mp\_read\_unsigned\_bin, \hyperpage{46}
-  \item mp\_reduce, \hyperpage{32}
-  \item mp\_reduce\_2k, \hyperpage{37}
-  \item mp\_reduce\_2k\_setup, \hyperpage{37}
-  \item mp\_reduce\_setup, \hyperpage{32}
-  \item mp\_rshd, \hyperpage{24}
-  \item mp\_set, \hyperpage{15}
-  \item mp\_set\_int, \hyperpage{16}
-  \item mp\_shrink, \hyperpage{11}
-  \item mp\_sqr, \hyperpage{29}
-  \item mp\_sub, \hyperpage{25}
-  \item mp\_sub\_d, \hyperpage{48}
-  \item mp\_to\_unsigned\_bin, \hyperpage{46}
-  \item mp\_toradix, \hyperpage{45}
-  \item mp\_unsigned\_bin\_size, \hyperpage{46}
-  \item MP\_VAL, \hyperpage{5}
-  \item mp\_xor, \hyperpage{25}
-  \item MP\_YES, \hyperpage{5}
+  \item mp\_add, \hyperpage{29}
+  \item mp\_add\_d, \hyperpage{52}
+  \item mp\_and, \hyperpage{29}
+  \item mp\_clear, \hyperpage{11}
+  \item mp\_clear\_multi, \hyperpage{12}
+  \item mp\_cmp, \hyperpage{24}
+  \item mp\_cmp\_d, \hyperpage{25}
+  \item mp\_cmp\_mag, \hyperpage{23}
+  \item mp\_div, \hyperpage{30}
+  \item mp\_div\_2, \hyperpage{26}
+  \item mp\_div\_2d, \hyperpage{28}
+  \item mp\_div\_d, \hyperpage{52}
+  \item mp\_dr\_reduce, \hyperpage{40}
+  \item mp\_dr\_setup, \hyperpage{40}
+  \item MP\_EQ, \hyperpage{22}
+  \item mp\_error\_to\_string, \hyperpage{10}
+  \item mp\_expt\_d, \hyperpage{43}
+  \item mp\_exptmod, \hyperpage{43}
+  \item mp\_exteuclid, \hyperpage{51}
+  \item mp\_gcd, \hyperpage{51}
+  \item mp\_get\_int, \hyperpage{20}
+  \item mp\_grow, \hyperpage{16}
+  \item MP\_GT, \hyperpage{22}
+  \item mp\_init, \hyperpage{11}
+  \item mp\_init\_copy, \hyperpage{13}
+  \item mp\_init\_multi, \hyperpage{12}
+  \item mp\_init\_set, \hyperpage{21}
+  \item mp\_init\_set\_int, \hyperpage{21}
+  \item mp\_init\_size, \hyperpage{14}
+  \item mp\_int, \hyperpage{10}
+  \item mp\_invmod, \hyperpage{52}
+  \item mp\_jacobi, \hyperpage{52}
+  \item mp\_lcm, \hyperpage{51}
+  \item mp\_lshd, \hyperpage{28}
+  \item MP\_LT, \hyperpage{22}
+  \item MP\_MEM, \hyperpage{9}
+  \item mp\_mod, \hyperpage{35}
+  \item mp\_mod\_d, \hyperpage{52}
+  \item mp\_montgomery\_calc\_normalization, \hyperpage{38}
+  \item mp\_montgomery\_reduce, \hyperpage{37}
+  \item mp\_montgomery\_setup, \hyperpage{37}
+  \item mp\_mul, \hyperpage{31}
+  \item mp\_mul\_2, \hyperpage{26}
+  \item mp\_mul\_2d, \hyperpage{28}
+  \item mp\_mul\_d, \hyperpage{52}
+  \item mp\_n\_root, \hyperpage{44}
+  \item mp\_neg, \hyperpage{29}
+  \item MP\_NO, \hyperpage{9}
+  \item MP\_OKAY, \hyperpage{9}
+  \item mp\_or, \hyperpage{29}
+  \item mp\_prime\_fermat, \hyperpage{45}
+  \item mp\_prime\_is\_divisible, \hyperpage{45}
+  \item mp\_prime\_is\_prime, \hyperpage{46}
+  \item mp\_prime\_miller\_rabin, \hyperpage{45}
+  \item mp\_prime\_next\_prime, \hyperpage{46}
+  \item mp\_prime\_rabin\_miller\_trials, \hyperpage{46}
+  \item mp\_prime\_random, \hyperpage{47}
+  \item mp\_prime\_random\_ex, \hyperpage{47}
+  \item mp\_radix\_size, \hyperpage{49}
+  \item mp\_read\_radix, \hyperpage{49}
+  \item mp\_read\_unsigned\_bin, \hyperpage{50}
+  \item mp\_reduce, \hyperpage{36}
+  \item mp\_reduce\_2k, \hyperpage{41}
+  \item mp\_reduce\_2k\_setup, \hyperpage{41}
+  \item mp\_reduce\_setup, \hyperpage{36}
+  \item mp\_rshd, \hyperpage{28}
+  \item mp\_set, \hyperpage{19}
+  \item mp\_set\_int, \hyperpage{20}
+  \item mp\_shrink, \hyperpage{15}
+  \item mp\_sqr, \hyperpage{33}
+  \item mp\_sub, \hyperpage{29}
+  \item mp\_sub\_d, \hyperpage{52}
+  \item mp\_to\_unsigned\_bin, \hyperpage{50}
+  \item mp\_toradix, \hyperpage{49}
+  \item mp\_unsigned\_bin\_size, \hyperpage{50}
+  \item MP\_VAL, \hyperpage{9}
+  \item mp\_xor, \hyperpage{29}
+  \item MP\_YES, \hyperpage{9}
 
 \end{theindex}
diff --git a/bn.pdf b/bn.pdf
index 52421fd5981b12f134d6f2352d6c7d978f6ffb32..fbd5b2adc106eacf9ae64849d42aa15100c676c4 100644
Binary files a/bn.pdf and b/bn.pdf differ
diff --git a/bn.tex b/bn.tex
index a2545862568c61fac98beb3c3e5028d769a42bf9..74a4f017bc688919884ede09af708bae84477f8b 100644
--- a/bn.tex
+++ b/bn.tex
@@ -49,7 +49,7 @@
 \begin{document}
 \frontmatter
 \pagestyle{empty}
-\title{LibTomMath User Manual \\ v0.30}
+\title{LibTomMath User Manual \\ v0.32}
 \author{Tom St Denis \\ tomstdenis@iahu.ca}
 \maketitle
 This text, the library and the accompanying textbook are all hereby placed in the public domain.  This book has been 
@@ -96,27 +96,34 @@ LibTomMath is meant to be very ``GCC friendly'' as it comes with a makefile well
 also build in MSVC, Borland C out of the box.  For any other ISO C compiler a makefile will have to be made by the end
 developer.  
 
-To build the library for GCC simply issue the 
-
+\subsection{Static Libraries}
+To build as a static library for GCC issue the following
 \begin{alltt}
 make
 \end{alltt}
 
-command.  This will build the library and archive the object files in ``libtommath.a''.  Now you simply link against that
-and include ``tommath.h'' within your programs.  
-
-Alternatively to build with MSVC type
-
+command.  This will build the library and archive the object files in ``libtommath.a''.  Now you link against 
+that and include ``tommath.h'' within your programs.  Alternatively to build with MSVC issue the following
 \begin{alltt}
 nmake -f makefile.msvc
 \end{alltt}
 
-This will build the library and archive the object files in ``tommath.lib''.  This has been tested with MSVC version 6.00
-with service pack 5.  
+This will build the library and archive the object files in ``tommath.lib''.  This has been tested with MSVC 
+version 6.00 with service pack 5.  
+
+\subsection{Shared Libraries}
+To build as a shared library for GCC issue the following
+\begin{alltt}
+make -f makefile.shared
+\end{alltt}
+This requires the ``libtool'' package (common on most Linux/BSD systems).  It will build LibTomMath as both shared
+and static then install (by default) into /usr/lib as well as install the header files in /usr/include.  The shared 
+library (resource) will be called ``libtommath.la'' while the static library called ``libtommath.a''.  Generally 
+you use libtool to link your application against the shared object.  
 
-There is limited support for making a ``DLL'' in windows via the ``makefile.cygwin\_dll'' makefile.  It requires Cygwin
-to work with since it requires the auto-export/import functionality.  The resulting DLL and imprt library ``libtomcrypt.dll.a''
-can be used to link LibTomMath dynamically to any Windows program using Cygwin.
+There is limited support for making a ``DLL'' in windows via the ``makefile.cygwin\_dll'' makefile.  It requires 
+Cygwin to work with since it requires the auto-export/import functionality.  The resulting DLL and import library 
+``libtommath.dll.a'' can be used to link LibTomMath dynamically to any Windows program using Cygwin.
 
 \subsection{Testing}
 To build the library and the test harness type
@@ -144,6 +151,96 @@ This will output a row of numbers that are increasing.  Each column is a differe
 that is being performed.  The numbers represent how many times the test was invoked.  If an error is detected the program
 will exit with a dump of the relevent numbers it was working with.
 
+\section{Build Configuration}
+LibTomMath can configured at build time in three phases we shall call ``depends'', ``tweaks'' and ``trims''.  
+Each phase changes how the library is built and they are applied one after another respectively.  
+
+To make the system more powerful you can tweak the build process.  Classes are defined in the file
+``tommath\_superclass.h''.  By default, the symbol ``LTM\_ALL'' shall be defined which simply 
+instructs the system to build all of the functions.  This is how LibTomMath used to be packaged.  This will give you 
+access to every function LibTomMath offers.
+
+However, there are cases where such a build is not optional.  For instance, you want to perform RSA operations.  You 
+don't need the vast majority of the library to perform these operations.  Aside from LTM\_ALL there is 
+another pre--defined class ``SC\_RSA\_1'' which works in conjunction with the RSA from LibTomCrypt.  Additional 
+classes can be defined base on the need of the user.
+
+\subsection{Build Depends}
+In the file tommath\_class.h you will see a large list of C ``defines'' followed by a series of ``ifdefs''
+which further define symbols.  All of the symbols (technically they're macros $\ldots$) represent a given C source
+file.  For instance, BN\_MP\_ADD\_C represents the file ``bn\_mp\_add.c''.  When a define has been enabled the
+function in the respective file will be compiled and linked into the library.  Accordingly when the define
+is absent the file will not be compiled and not contribute any size to the library.
+
+You will also note that the header tommath\_class.h is actually recursively included (it includes itself twice).  
+This is to help resolve as many dependencies as possible.  In the last pass the symbol LTM\_LAST will be defined.  
+This is useful for ``trims''.
+
+\subsection{Build Tweaks}
+A tweak is an algorithm ``alternative''.  For example, to provide tradeoffs (usually between size and space).
+They can be enabled at any pass of the configuration phase.
+
+\begin{small}
+\begin{center}
+\begin{tabular}{|l|l|}
+\hline \textbf{Define} & \textbf{Purpose} \\
+\hline BN\_MP\_DIV\_SMALL & Enables a slower, smaller and equally \\
+                          & functional mp\_div() function \\
+\hline
+\end{tabular}
+\end{center}
+\end{small}
+
+\subsection{Build Trims}
+A trim is a manner of removing functionality from a function that is not required.  For instance, to perform
+RSA cryptography you only require exponentiation with odd moduli so even moduli support can be safely removed.  
+Build trims are meant to be defined on the last pass of the configuration which means they are to be defined
+only if LTM\_LAST has been defined.
+
+\subsubsection{Moduli Related}
+\begin{small}
+\begin{center}
+\begin{tabular}{|l|l|}
+\hline \textbf{Restriction} & \textbf{Undefine} \\
+\hline Exponentiation with odd moduli only & BN\_S\_MP\_EXPTMOD\_C \\
+                                           & BN\_MP\_REDUCE\_C \\
+                                           & BN\_MP\_REDUCE\_SETUP\_C \\
+                                           & BN\_S\_MP\_MUL\_HIGH\_DIGS\_C \\
+                                           & BN\_FAST\_S\_MP\_MUL\_HIGH\_DIGS\_C \\
+\hline Exponentiation with random odd moduli & (The above plus the following) \\
+                                           & BN\_MP\_REDUCE\_2K\_C \\
+                                           & BN\_MP\_REDUCE\_2K\_SETUP\_C \\
+                                           & BN\_MP\_REDUCE\_IS\_2K\_C \\
+                                           & BN\_MP\_DR\_IS\_MODULUS\_C \\
+                                           & BN\_MP\_DR\_REDUCE\_C \\
+                                           & BN\_MP\_DR\_SETUP\_C \\
+\hline Modular inverse odd moduli only     & BN\_MP\_INVMOD\_SLOW\_C \\
+\hline Modular inverse (both, smaller/slower) & BN\_FAST\_MP\_INVMOD\_C \\
+\hline
+\end{tabular}
+\end{center}
+\end{small}
+
+\subsubsection{Operand Size Related}
+\begin{small}
+\begin{center}
+\begin{tabular}{|l|l|}
+\hline \textbf{Restriction} & \textbf{Undefine} \\
+\hline Moduli $\le 2560$ bits              & BN\_MP\_MONTGOMERY\_REDUCE\_C \\
+                                           & BN\_S\_MP\_MUL\_DIGS\_C \\
+                                           & BN\_S\_MP\_MUL\_HIGH\_DIGS\_C \\
+                                           & BN\_S\_MP\_SQR\_C \\
+\hline Polynomial Schmolynomial            & BN\_MP\_KARATSUBA\_MUL\_C \\
+                                           & BN\_MP\_KARATSUBA\_SQR\_C \\
+                                           & BN\_MP\_TOOM\_MUL\_C \\ 
+                                           & BN\_MP\_TOOM\_SQR\_C \\
+
+\hline
+\end{tabular}
+\end{center}
+\end{small}
+
+
 \section{Purpose of LibTomMath}
 Unlike  GNU MP (GMP) Library, LIP, OpenSSL or various other commercial kits (Miracl), LibTomMath was not written with 
 bleeding edge performance in mind.  First and foremost LibTomMath was written to be entirely open.  Not only is the 
diff --git a/bn_error.c b/bn_error.c
index 7cac6497aa4c3f6389b2c3be16612a28693182e2..1546784167bd61ea83add9c568125bde29178fa2 100644
--- a/bn_error.c
+++ b/bn_error.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_ERROR_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 static const struct {
      int code;
@@ -39,3 +40,4 @@ char *mp_error_to_string(int code)
    return "Invalid error code";
 }
 
+#endif
diff --git a/bn_fast_mp_invmod.c b/bn_fast_mp_invmod.c
index c9a945982649aa3100b299452cd6ce192330c854..492a3f1f6dc099f1b42a4fd1beed8bb828b7ba06 100644
--- a/bn_fast_mp_invmod.c
+++ b/bn_fast_mp_invmod.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_FAST_MP_INVMOD_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,12 +14,11 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* computes the modular inverse via binary extended euclidean algorithm, 
  * that is c = 1/a mod b 
  *
- * Based on mp_invmod except this is optimized for the case where b is 
+ * Based on slow invmod except this is optimized for the case where b is 
  * odd as per HAC Note 14.64 on pp. 610
  */
 int
@@ -141,3 +142,4 @@ top:
 __ERR:mp_clear_multi (&x, &y, &u, &v, &B, &D, NULL);
   return res;
 }
+#endif
diff --git a/bn_fast_mp_montgomery_reduce.c b/bn_fast_mp_montgomery_reduce.c
index c7a3413e2523841f620b57c2387a797fb2ada19a..7373ae6b88bd57fbf10fb705c1e5c2f3d2372aef 100644
--- a/bn_fast_mp_montgomery_reduce.c
+++ b/bn_fast_mp_montgomery_reduce.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_FAST_MP_MONTGOMERY_REDUCE_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,11 +14,10 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* computes xR**-1 == x (mod N) via Montgomery Reduction
  *
- * This is an optimized implementation of mp_montgomery_reduce
+ * This is an optimized implementation of montgomery_reduce
  * which uses the comba method to quickly calculate the columns of the
  * reduction.
  *
@@ -165,3 +166,4 @@ fast_mp_montgomery_reduce (mp_int * x, mp_int * n, mp_digit rho)
   }
   return MP_OKAY;
 }
+#endif
diff --git a/bn_fast_s_mp_mul_digs.c b/bn_fast_s_mp_mul_digs.c
index 75fa706e6e8e998a0b704b92c64c1a05da82b831..92b50bb7201436f726fafb354868dbf5218b7a63 100644
--- a/bn_fast_s_mp_mul_digs.c
+++ b/bn_fast_s_mp_mul_digs.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_FAST_S_MP_MUL_DIGS_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* Fast (comba) multiplier
  *
@@ -33,8 +34,9 @@
 int
 fast_s_mp_mul_digs (mp_int * a, mp_int * b, mp_int * c, int digs)
 {
-  int     olduse, res, pa, ix;
-  mp_word W[MP_WARRAY];
+  int     olduse, res, pa, ix, iz;
+  mp_digit W[MP_WARRAY];
+  register mp_word  _W;
 
   /* grow the destination as required */
   if (c->alloc < digs) {
@@ -43,82 +45,52 @@ fast_s_mp_mul_digs (mp_int * a, mp_int * b, mp_int * c, int digs)
     }
   }
 
-  /* clear temp buf (the columns) */
-  memset (W, 0, sizeof (mp_word) * digs);
+  /* number of output digits to produce */
+  pa = MIN(digs, a->used + b->used);
 
-  /* calculate the columns */
-  pa = a->used;
-  for (ix = 0; ix < pa; ix++) {
-    /* this multiplier has been modified to allow you to 
-     * control how many digits of output are produced.  
-     * So at most we want to make upto "digs" digits of output.
-     *
-     * this adds products to distinct columns (at ix+iy) of W
-     * note that each step through the loop is not dependent on
-     * the previous which means the compiler can easily unroll
-     * the loop without scheduling problems
-     */
-    {
-      register mp_digit tmpx, *tmpy;
-      register mp_word *_W;
-      register int iy, pb;
+  /* clear the carry */
+  _W = 0;
+  for (ix = 0; ix <= pa; ix++) { 
+      int      tx, ty;
+      int      iy;
+      mp_digit *tmpx, *tmpy;
 
-      /* alias for the the word on the left e.g. A[ix] * A[iy] */
-      tmpx = a->dp[ix];
+      /* get offsets into the two bignums */
+      ty = MIN(b->used-1, ix);
+      tx = ix - ty;
 
-      /* alias for the right side */
-      tmpy = b->dp;
-
-      /* alias for the columns, each step through the loop adds a new
-         term to each column
-       */
-      _W = W + ix;
+      /* setup temp aliases */
+      tmpx = a->dp + tx;
+      tmpy = b->dp + ty;
 
-      /* the number of digits is limited by their placement.  E.g.
-         we avoid multiplying digits that will end up above the # of
-         digits of precision requested
+      /* this is the number of times the loop will iterrate, essentially its 
+         while (tx++ < a->used && ty-- >= 0) { ... }
        */
-      pb = MIN (b->used, digs - ix);
+      iy = MIN(a->used-tx, ty+1);
 
-      for (iy = 0; iy < pb; iy++) {
-        *_W++ += ((mp_word)tmpx) * ((mp_word)*tmpy++);
+      /* execute loop */
+      for (iz = 0; iz < iy; ++iz) {
+         _W += ((mp_word)*tmpx++)*((mp_word)*tmpy--);
       }
-    }
 
+      /* store term */
+      W[ix] = ((mp_digit)_W) & MP_MASK;
+
+      /* make next carry */
+      _W = _W >> ((mp_word)DIGIT_BIT);
   }
 
   /* setup dest */
-  olduse = c->used;
+  olduse  = c->used;
   c->used = digs;
 
   {
     register mp_digit *tmpc;
-
-    /* At this point W[] contains the sums of each column.  To get the
-     * correct result we must take the extra bits from each column and
-     * carry them down
-     *
-     * Note that while this adds extra code to the multiplier it 
-     * saves time since the carry propagation is removed from the 
-     * above nested loop.This has the effect of reducing the work 
-     * from N*(N+N*c)==N**2 + c*N**2 to N**2 + N*c where c is the 
-     * cost of the shifting.  On very small numbers this is slower 
-     * but on most cryptographic size numbers it is faster.
-     *
-     * In this particular implementation we feed the carries from
-     * behind which means when the loop terminates we still have one
-     * last digit to copy
-     */
     tmpc = c->dp;
-    for (ix = 1; ix < digs; ix++) {
-      /* forward the carry from the previous temp */
-      W[ix] += (W[ix - 1] >> ((mp_word) DIGIT_BIT));
-
+    for (ix = 0; ix < digs; ix++) {
       /* now extract the previous digit [below the carry] */
-      *tmpc++ = (mp_digit) (W[ix - 1] & ((mp_word) MP_MASK));
+      *tmpc++ = W[ix];
     }
-    /* fetch the last digit */
-    *tmpc++ = (mp_digit) (W[digs - 1] & ((mp_word) MP_MASK));
 
     /* clear unused digits [that existed in the old copy of c] */
     for (; ix < olduse; ix++) {
@@ -128,3 +100,4 @@ fast_s_mp_mul_digs (mp_int * a, mp_int * b, mp_int * c, int digs)
   mp_clamp (c);
   return MP_OKAY;
 }
+#endif
diff --git a/bn_fast_s_mp_mul_high_digs.c b/bn_fast_s_mp_mul_high_digs.c
index 779aa140066c5d2aaa9c2411d8170b9192ec360e..9e0cf55aa18ea5b2634b71219aebf8e6dd8092eb 100644
--- a/bn_fast_s_mp_mul_high_digs.c
+++ b/bn_fast_s_mp_mul_high_digs.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_FAST_S_MP_MUL_HIGH_DIGS_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,10 +14,9 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
- #include <tommath.h>
 
-/* this is a modified version of fast_s_mp_mul_digs that only produces
- * output digits *above* digs.  See the comments for fast_s_mp_mul_digs
+/* this is a modified version of fast_s_mul_digs that only produces
+ * output digits *above* digs.  See the comments for fast_s_mul_digs
  * to see how it works.
  *
  * This is used in the Barrett reduction since for one of the multiplications
@@ -26,73 +27,69 @@
 int
 fast_s_mp_mul_high_digs (mp_int * a, mp_int * b, mp_int * c, int digs)
 {
-  int     oldused, newused, res, pa, pb, ix;
-  mp_word W[MP_WARRAY];
+  int     olduse, res, pa, ix, iz;
+  mp_digit W[MP_WARRAY];
+  mp_word  _W;
 
-  /* calculate size of product and allocate more space if required */
-  newused = a->used + b->used + 1;
-  if (c->alloc < newused) {
-    if ((res = mp_grow (c, newused)) != MP_OKAY) {
+  /* grow the destination as required */
+  pa = a->used + b->used;
+  if (c->alloc < pa) {
+    if ((res = mp_grow (c, pa)) != MP_OKAY) {
       return res;
     }
   }
 
-  /* like the other comba method we compute the columns first */
-  pa = a->used;
-  pb = b->used;
-  memset (W + digs, 0, (pa + pb + 1 - digs) * sizeof (mp_word));
-  for (ix = 0; ix < pa; ix++) {
-    {
-      register mp_digit tmpx, *tmpy;
-      register int iy;
-      register mp_word *_W;
+  /* number of output digits to produce */
+  pa = a->used + b->used;
+  _W = 0;
+  for (ix = digs; ix <= pa; ix++) { 
+      int      tx, ty, iy;
+      mp_digit *tmpx, *tmpy;
 
-      /* work todo, that is we only calculate digits that are at "digs" or above  */
-      iy = digs - ix;
+      /* get offsets into the two bignums */
+      ty = MIN(b->used-1, ix);
+      tx = ix - ty;
 
-      /* copy of word on the left of A[ix] * B[iy] */
-      tmpx = a->dp[ix];
+      /* setup temp aliases */
+      tmpx = a->dp + tx;
+      tmpy = b->dp + ty;
 
-      /* alias for right side */
-      tmpy = b->dp + iy;
-     
-      /* alias for the columns of output.  Offset to be equal to or above the 
-       * smallest digit place requested 
+      /* this is the number of times the loop will iterrate, essentially its 
+         while (tx++ < a->used && ty-- >= 0) { ... }
        */
-      _W = W + digs;     
-      
-      /* skip cases below zero where ix > digs */
-      if (iy < 0) {
-         iy    = abs(iy);
-         tmpy += iy;
-         _W   += iy;
-         iy    = 0;
-      }
+      iy = MIN(a->used-tx, ty+1);
 
-      /* compute column products for digits above the minimum */
-      for (; iy < pb; iy++) {
-         *_W++ += ((mp_word) tmpx) * ((mp_word)*tmpy++);
+      /* execute loop */
+      for (iz = 0; iz < iy; iz++) {
+         _W += ((mp_word)*tmpx++)*((mp_word)*tmpy--);
       }
-    }
+
+      /* store term */
+      W[ix] = ((mp_digit)_W) & MP_MASK;
+
+      /* make next carry */
+      _W = _W >> ((mp_word)DIGIT_BIT);
   }
 
   /* setup dest */
-  oldused = c->used;
-  c->used = newused;
+  olduse  = c->used;
+  c->used = pa;
 
-  /* now convert the array W downto what we need
-   *
-   * See comments in bn_fast_s_mp_mul_digs.c
-   */
-  for (ix = digs + 1; ix < newused; ix++) {
-    W[ix] += (W[ix - 1] >> ((mp_word) DIGIT_BIT));
-    c->dp[ix - 1] = (mp_digit) (W[ix - 1] & ((mp_word) MP_MASK));
-  }
-  c->dp[newused - 1] = (mp_digit) (W[newused - 1] & ((mp_word) MP_MASK));
+  {
+    register mp_digit *tmpc;
+
+    tmpc = c->dp + digs;
+    for (ix = digs; ix <= pa; ix++) {
+      /* now extract the previous digit [below the carry] */
+      *tmpc++ = W[ix];
+    }
 
-  for (; ix < oldused; ix++) {
-    c->dp[ix] = 0;
+    /* clear unused digits [that existed in the old copy of c] */
+    for (; ix < olduse; ix++) {
+      *tmpc++ = 0;
+    }
   }
   mp_clamp (c);
   return MP_OKAY;
 }
+#endif
diff --git a/bn_fast_s_mp_sqr.c b/bn_fast_s_mp_sqr.c
index f62ae54e380f07fe2a739b436869e03bfa55d5d4..9f6962dce5d93c058bfc16332378699da41119fc 100644
--- a/bn_fast_s_mp_sqr.c
+++ b/bn_fast_s_mp_sqr.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_FAST_S_MP_SQR_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* fast squaring
  *
@@ -31,109 +32,98 @@
  * Based on Algorithm 14.16 on pp.597 of HAC.
  *
  */
+/* the jist of squaring...
+
+you do like mult except the offset of the tmpx [one that starts closer to zero]
+can't equal the offset of tmpy.  So basically you set up iy like before then you min it with
+(ty-tx) so that it never happens.  You double all those you add in the inner loop
+
+After that loop you do the squares and add them in.
+
+Remove W2 and don't memset W
+
+*/
+
 int fast_s_mp_sqr (mp_int * a, mp_int * b)
 {
-  int     olduse, newused, res, ix, pa;
-  mp_word W2[MP_WARRAY], W[MP_WARRAY];
-
-  /* calculate size of product and allocate as required */
-  pa = a->used;
-  newused = pa + pa + 1;
-  if (b->alloc < newused) {
-    if ((res = mp_grow (b, newused)) != MP_OKAY) {
+  int       olduse, res, pa, ix, iz;
+  mp_digit   W[MP_WARRAY], *tmpx;
+  mp_word   W1;
+
+  /* grow the destination as required */
+  pa = a->used + a->used;
+  if (b->alloc < pa) {
+    if ((res = mp_grow (b, pa)) != MP_OKAY) {
       return res;
     }
   }
 
-  /* zero temp buffer (columns)
-   * Note that there are two buffers.  Since squaring requires
-   * a outer and inner product and the inner product requires
-   * computing a product and doubling it (a relatively expensive
-   * op to perform n**2 times if you don't have to) the inner and
-   * outer products are computed in different buffers.  This way
-   * the inner product can be doubled using n doublings instead of
-   * n**2
-   */
-  memset (W,  0, newused * sizeof (mp_word));
-  memset (W2, 0, newused * sizeof (mp_word));
-
-  /* This computes the inner product.  To simplify the inner N**2 loop
-   * the multiplication by two is done afterwards in the N loop.
-   */
-  for (ix = 0; ix < pa; ix++) {
-    /* compute the outer product
-     *
-     * Note that every outer product is computed
-     * for a particular column only once which means that
-     * there is no need todo a double precision addition
-     * into the W2[] array.
-     */
-    W2[ix + ix] = ((mp_word)a->dp[ix]) * ((mp_word)a->dp[ix]);
-
-    {
-      register mp_digit tmpx, *tmpy;
-      register mp_word *_W;
-      register int iy;
-
-      /* copy of left side */
-      tmpx = a->dp[ix];
-
-      /* alias for right side */
-      tmpy = a->dp + (ix + 1);
-
-      /* the column to store the result in */
-      _W = W + (ix + ix + 1);
-
-      /* inner products */
-      for (iy = ix + 1; iy < pa; iy++) {
-          *_W++ += ((mp_word)tmpx) * ((mp_word)*tmpy++);
+  /* number of output digits to produce */
+  W1 = 0;
+  for (ix = 0; ix <= pa; ix++) { 
+      int      tx, ty, iy;
+      mp_word  _W;
+      mp_digit *tmpy;
+
+      /* clear counter */
+      _W = 0;
+
+      /* get offsets into the two bignums */
+      ty = MIN(a->used-1, ix);
+      tx = ix - ty;
+
+      /* setup temp aliases */
+      tmpx = a->dp + tx;
+      tmpy = a->dp + ty;
+
+      /* this is the number of times the loop will iterrate, essentially its 
+         while (tx++ < a->used && ty-- >= 0) { ... }
+       */
+      iy = MIN(a->used-tx, ty+1);
+
+      /* now for squaring tx can never equal ty 
+       * we halve the distance since they approach at a rate of 2x
+       * and we have to round because odd cases need to be executed
+       */
+      iy = MIN(iy, (ty-tx+1)>>1);
+
+      /* execute loop */
+      for (iz = 0; iz < iy; iz++) {
+         _W += ((mp_word)*tmpx++)*((mp_word)*tmpy--);
       }
-    }
+
+      /* double the inner product and add carry */
+      _W = _W + _W + W1;
+
+      /* even columns have the square term in them */
+      if ((ix&1) == 0) {
+         _W += ((mp_word)a->dp[ix>>1])*((mp_word)a->dp[ix>>1]);
+      }
+
+      /* store it */
+      W[ix] = _W;
+
+      /* make next carry */
+      W1 = _W >> ((mp_word)DIGIT_BIT);
   }
 
   /* setup dest */
   olduse  = b->used;
-  b->used = newused;
-
-  /* now compute digits
-   *
-   * We have to double the inner product sums, add in the
-   * outer product sums, propagate carries and convert
-   * to single precision.
-   */
-  {
-    register mp_digit *tmpb;
-
-    /* double first value, since the inner products are
-     * half of what they should be
-     */
-    W[0] += W[0] + W2[0];
+  b->used = a->used+a->used;
 
+  {
+    mp_digit *tmpb;
     tmpb = b->dp;
-    for (ix = 1; ix < newused; ix++) {
-      /* double/add next digit */
-      W[ix] += W[ix] + W2[ix];
-
-      /* propagate carry forwards [from the previous digit] */
-      W[ix] = W[ix] + (W[ix - 1] >> ((mp_word) DIGIT_BIT));
-
-      /* store the current digit now that the carry isn't
-       * needed
-       */
-      *tmpb++ = (mp_digit) (W[ix - 1] & ((mp_word) MP_MASK));
+    for (ix = 0; ix < pa; ix++) {
+      *tmpb++ = W[ix] & MP_MASK;
     }
-    /* set the last value.  Note even if the carry is zero
-     * this is required since the next step will not zero
-     * it if b originally had a value at b->dp[2*a.used]
-     */
-    *tmpb++ = (mp_digit) (W[(newused) - 1] & ((mp_word) MP_MASK));
 
-    /* clear high digits of b if there were any originally */
+    /* clear unused digits [that existed in the old copy of c] */
     for (; ix < olduse; ix++) {
       *tmpb++ = 0;
     }
   }
-
   mp_clamp (b);
   return MP_OKAY;
 }
+#endif
diff --git a/bn_mp_2expt.c b/bn_mp_2expt.c
index 5c4e25649681da0328930d2979bfe860033e4e42..45a68184ae3d82b34a94e14e76cb4142eacd44a3 100644
--- a/bn_mp_2expt.c
+++ b/bn_mp_2expt.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_2EXPT_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* computes a = 2**b 
  *
@@ -36,7 +37,8 @@ mp_2expt (mp_int * a, int b)
   a->used = b / DIGIT_BIT + 1;
 
   /* put the single bit in its place */
-  a->dp[b / DIGIT_BIT] = 1 << (b % DIGIT_BIT);
+  a->dp[b / DIGIT_BIT] = ((mp_digit)1) << (b % DIGIT_BIT);
 
   return MP_OKAY;
 }
+#endif
diff --git a/bn_mp_abs.c b/bn_mp_abs.c
index a5e9ab02af57807af18ac1a080d91589d821d839..34f810fdae524d7d4dc8d0d451df2bf4a5abdc8d 100644
--- a/bn_mp_abs.c
+++ b/bn_mp_abs.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_ABS_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* b = |a| 
  *
@@ -35,3 +36,4 @@ mp_abs (mp_int * a, mp_int * b)
 
   return MP_OKAY;
 }
+#endif
diff --git a/bn_mp_add.c b/bn_mp_add.c
index f625195376696022bdd721af238a3a14dac0caa2..554b7f7467dd3f5ef52d7ce0dca31d2ba420bf55 100644
--- a/bn_mp_add.c
+++ b/bn_mp_add.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_ADD_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* high level addition (handles signs) */
 int mp_add (mp_int * a, mp_int * b, mp_int * c)
@@ -45,3 +46,4 @@ int mp_add (mp_int * a, mp_int * b, mp_int * c)
   return res;
 }
 
+#endif
diff --git a/bn_mp_add_d.c b/bn_mp_add_d.c
index edc93c1fa582b875e6ccdc156294ec4559c3103b..bdd02809857f41a50d2232a9f9fbb7d94c5a589d 100644
--- a/bn_mp_add_d.c
+++ b/bn_mp_add_d.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_ADD_D_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* single digit addition */
 int
@@ -101,3 +102,4 @@ mp_add_d (mp_int * a, mp_digit b, mp_int * c)
   return MP_OKAY;
 }
 
+#endif
diff --git a/bn_mp_addmod.c b/bn_mp_addmod.c
index 691dc5088cfdb9efd0db460cc3318f2c0068fea4..13eb33f572e4238001da8313b63632e5904b31fb 100644
--- a/bn_mp_addmod.c
+++ b/bn_mp_addmod.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_ADDMOD_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* d = a + b (mod c) */
 int
@@ -33,3 +34,4 @@ mp_addmod (mp_int * a, mp_int * b, mp_int * c, mp_int * d)
   mp_clear (&t);
   return res;
 }
+#endif
diff --git a/bn_mp_and.c b/bn_mp_and.c
index 9abb6184049bd40e3842f4e9fe09735a7976cd76..61dc386863171cc9492afa8af476a82a8c69ca46 100644
--- a/bn_mp_and.c
+++ b/bn_mp_and.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_AND_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* AND two ints together */
 int
@@ -49,3 +50,4 @@ mp_and (mp_int * a, mp_int * b, mp_int * c)
   mp_clear (&t);
   return MP_OKAY;
 }
+#endif
diff --git a/bn_mp_clamp.c b/bn_mp_clamp.c
index c6285ceade392280359faf879db0c4161c29ab27..c17261118566b515150c943cffd1327b5dcdded8 100644
--- a/bn_mp_clamp.c
+++ b/bn_mp_clamp.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_CLAMP_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* trim unused digits 
  *
@@ -36,3 +37,4 @@ mp_clamp (mp_int * a)
     a->sign = MP_ZPOS;
   }
 }
+#endif
diff --git a/bn_mp_clear.c b/bn_mp_clear.c
index afe9b26342d7b441150d8aeb132d0c7a32dd1250..534264804fb0f2d877a4453eb54369ffcbf04f57 100644
--- a/bn_mp_clear.c
+++ b/bn_mp_clear.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_CLEAR_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,16 +14,19 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* clear one (frees)  */
 void
 mp_clear (mp_int * a)
 {
+  int i;
+
   /* only do anything if a hasn't been freed previously */
   if (a->dp != NULL) {
     /* first zero the digits */
-    memset (a->dp, 0, sizeof (mp_digit) * a->used);
+    for (i = 0; i < a->used; i++) {
+        a->dp[i] = 0;
+    }
 
     /* free ram */
     XFREE(a->dp);
@@ -32,3 +37,4 @@ mp_clear (mp_int * a)
     a->sign  = MP_ZPOS;
   }
 }
+#endif
diff --git a/bn_mp_clear_multi.c b/bn_mp_clear_multi.c
index d262b2ff095d7da71bcf6ab33017e98fe9ed9754..24cbe7329cca3029710a20a579a9c2bc7bcd6559 100644
--- a/bn_mp_clear_multi.c
+++ b/bn_mp_clear_multi.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_CLEAR_MULTI_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 #include <stdarg.h>
 
 void mp_clear_multi(mp_int *mp, ...) 
@@ -26,3 +27,4 @@ void mp_clear_multi(mp_int *mp, ...)
     }
     va_end(args);
 }
+#endif
diff --git a/bn_mp_cmp.c b/bn_mp_cmp.c
index ed27d21fb61cf096cfe099e66258c791f0e57e8a..583b5f866d744819314c393bf86bd04f16f93268 100644
--- a/bn_mp_cmp.c
+++ b/bn_mp_cmp.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_CMP_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* compare two ints (signed)*/
 int
@@ -35,3 +36,4 @@ mp_cmp (mp_int * a, mp_int * b)
      return mp_cmp_mag(a, b);
   }
 }
+#endif
diff --git a/bn_mp_cmp_d.c b/bn_mp_cmp_d.c
index 6baa172cd8d7d69195e0849a5003775bb0299277..882b1c93db34cbf129f74430f183ae827717adbb 100644
--- a/bn_mp_cmp_d.c
+++ b/bn_mp_cmp_d.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_CMP_D_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* compare a digit */
 int mp_cmp_d(mp_int * a, mp_digit b)
@@ -36,3 +37,4 @@ int mp_cmp_d(mp_int * a, mp_digit b)
     return MP_EQ;
   }
 }
+#endif
diff --git a/bn_mp_cmp_mag.c b/bn_mp_cmp_mag.c
index b50f769dbc4ec64ff4a1dae73d2fe6e5a0d39a14..a0f351c76bf39eee9a0c68a615b6b815665ea8fc 100644
--- a/bn_mp_cmp_mag.c
+++ b/bn_mp_cmp_mag.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_CMP_MAG_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* compare maginitude of two ints (unsigned) */
 int mp_cmp_mag (mp_int * a, mp_int * b)
@@ -47,3 +48,4 @@ int mp_cmp_mag (mp_int * a, mp_int * b)
   }
   return MP_EQ;
 }
+#endif
diff --git a/bn_mp_cnt_lsb.c b/bn_mp_cnt_lsb.c
index 07cb709ca1df560e53ff6d7876f6f63b2f6d329e..571f03f9b21e67f1db0e57365281fd1d0f58e605 100644
--- a/bn_mp_cnt_lsb.c
+++ b/bn_mp_cnt_lsb.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_CNT_LSB_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 static const int lnz[16] = { 
    4, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0
@@ -45,3 +46,4 @@ int mp_cnt_lsb(mp_int *a)
    return x;
 }
 
+#endif
diff --git a/bn_mp_copy.c b/bn_mp_copy.c
index 1c45f3e03889671f0b3f3390260f19f9d17899de..183ec9b6de8d58601ec678fb98ea56635b976542 100644
--- a/bn_mp_copy.c
+++ b/bn_mp_copy.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_COPY_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* copy, b = a */
 int
@@ -60,3 +61,4 @@ mp_copy (mp_int * a, mp_int * b)
   b->sign = a->sign;
   return MP_OKAY;
 }
+#endif
diff --git a/bn_mp_count_bits.c b/bn_mp_count_bits.c
index 33b971bdc0e64af29eead1efdde15ef7f3eb9dd6..f3f85ac60efa28511d45d76f3a911264418b5349 100644
--- a/bn_mp_count_bits.c
+++ b/bn_mp_count_bits.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_COUNT_BITS_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* returns the number of bits in an int */
 int
@@ -37,3 +38,4 @@ mp_count_bits (mp_int * a)
   }
   return r;
 }
+#endif
diff --git a/bn_mp_div.c b/bn_mp_div.c
index 652a0944f06a8c7178dd276424b28967c3be08f5..39d921a2ddf59a5c87a3f315135197a64ae98c62 100644
--- a/bn_mp_div.c
+++ b/bn_mp_div.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_DIV_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,78 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
+
+#ifdef BN_MP_DIV_SMALL
+
+/* slower bit-bang division... also smaller */
+int mp_div(mp_int * a, mp_int * b, mp_int * c, mp_int * d)
+{
+   mp_int ta, tb, tq, q;
+   int    res, n, n2;
+
+  /* is divisor zero ? */
+  if (mp_iszero (b) == 1) {
+    return MP_VAL;
+  }
+
+  /* if a < b then q=0, r = a */
+  if (mp_cmp_mag (a, b) == MP_LT) {
+    if (d != NULL) {
+      res = mp_copy (a, d);
+    } else {
+      res = MP_OKAY;
+    }
+    if (c != NULL) {
+      mp_zero (c);
+    }
+    return res;
+  }
+	
+  /* init our temps */
+  if ((res = mp_init_multi(&ta, &tb, &tq, &q, NULL) != MP_OKAY)) {
+     return res;
+  }
+
+
+  mp_set(&tq, 1);
+  n = mp_count_bits(a) - mp_count_bits(b);
+  if (((res = mp_copy(a, &ta)) != MP_OKAY) ||
+      ((res = mp_copy(b, &tb)) != MP_OKAY) || 
+      ((res = mp_mul_2d(&tb, n, &tb)) != MP_OKAY) ||
+      ((res = mp_mul_2d(&tq, n, &tq)) != MP_OKAY)) {
+      goto __ERR;
+  }
+
+  while (n-- >= 0) {
+     if (mp_cmp(&tb, &ta) != MP_GT) {
+        if (((res = mp_sub(&ta, &tb, &ta)) != MP_OKAY) ||
+            ((res = mp_add(&q, &tq, &q)) != MP_OKAY)) {
+           goto __ERR;
+        }
+     }
+     if (((res = mp_div_2d(&tb, 1, &tb, NULL)) != MP_OKAY) ||
+         ((res = mp_div_2d(&tq, 1, &tq, NULL)) != MP_OKAY)) {
+           goto __ERR;
+     }
+  }
+
+  /* now q == quotient and ta == remainder */
+  n  = a->sign;
+  n2 = (a->sign == b->sign ? MP_ZPOS : MP_NEG);
+  if (c != NULL) {
+     mp_exch(c, &q);
+     c->sign  = n2;
+  }
+  if (d != NULL) {
+     mp_exch(d, &ta);
+     d->sign = n;
+  }
+__ERR:
+   mp_clear_multi(&ta, &tb, &tq, &q, NULL);
+   return res;
+}
+
+#else
 
 /* integer signed division. 
  * c*b + d == a [e.g. a/b, c=quotient, d=remainder]
@@ -187,7 +260,7 @@ int mp_div (mp_int * a, mp_int * b, mp_int * c, mp_int * d)
    */
   
   /* get sign before writing to c */
-  x.sign = a->sign;
+  x.sign = x.used == 0 ? MP_ZPOS : a->sign;
 
   if (c != NULL) {
     mp_clamp (&q);
@@ -209,3 +282,7 @@ __T1:mp_clear (&t1);
 __Q:mp_clear (&q);
   return res;
 }
+
+#endif
+
+#endif
diff --git a/bn_mp_div_2.c b/bn_mp_div_2.c
index ebad7ae471e50197ec314b2fdc9195080bd3e587..577799717764d4526799dda0d10a30ad71885505 100644
--- a/bn_mp_div_2.c
+++ b/bn_mp_div_2.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_DIV_2_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* b = a/2 */
 int mp_div_2(mp_int * a, mp_int * b)
@@ -60,3 +61,4 @@ int mp_div_2(mp_int * a, mp_int * b)
   mp_clamp (b);
   return MP_OKAY;
 }
+#endif
diff --git a/bn_mp_div_2d.c b/bn_mp_div_2d.c
index 75f090ce9b5be1b3d33fc8108166ecfb467d72cd..cf103f2ba3f2b364b2d87ad940f6497e87044fe7 100644
--- a/bn_mp_div_2d.c
+++ b/bn_mp_div_2d.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_DIV_2D_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* shift right by a certain bit count (store quotient in c, optional remainder in d) */
 int mp_div_2d (mp_int * a, int b, mp_int * c, mp_int * d)
@@ -89,3 +90,4 @@ int mp_div_2d (mp_int * a, int b, mp_int * c, mp_int * d)
   mp_clear (&t);
   return MP_OKAY;
 }
+#endif
diff --git a/bn_mp_div_3.c b/bn_mp_div_3.c
index 3bc2bd2f7fb7cf8a8f721ecbd7528653c839c049..7cbafc145f4cc780576b22416fff9572db21ebb0 100644
--- a/bn_mp_div_3.c
+++ b/bn_mp_div_3.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_DIV_3_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* divide by three (based on routine from MPI and the GMP manual) */
 int
@@ -71,3 +72,4 @@ mp_div_3 (mp_int * a, mp_int *c, mp_digit * d)
   return res;
 }
 
+#endif
diff --git a/bn_mp_div_d.c b/bn_mp_div_d.c
index f8caf83627e85358b07b88d39156a4e9b2987854..9b58aa667c12c11fca6b4aa5be25683cd1d2acd0 100644
--- a/bn_mp_div_d.c
+++ b/bn_mp_div_d.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_DIV_D_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 static int s_is_power_of_two(mp_digit b, int *p)
 {
@@ -54,7 +55,7 @@ int mp_div_d (mp_int * a, mp_digit b, mp_int * c, mp_digit * d)
   /* power of two ? */
   if (s_is_power_of_two(b, &ix) == 1) {
      if (d != NULL) {
-        *d = a->dp[0] & ((1<<ix) - 1);
+        *d = a->dp[0] & ((((mp_digit)1)<<ix) - 1);
      }
      if (c != NULL) {
         return mp_div_2d(a, ix, c, NULL);
@@ -62,10 +63,12 @@ int mp_div_d (mp_int * a, mp_digit b, mp_int * c, mp_digit * d)
      return MP_OKAY;
   }
 
+#ifdef BN_MP_DIV_3_C
   /* three? */
   if (b == 3) {
      return mp_div_3(a, c, d);
   }
+#endif
 
   /* no easy answer [c'est la vie].  Just division */
   if ((res = mp_init_size(&q, a->used)) != MP_OKAY) {
@@ -100,3 +103,4 @@ int mp_div_d (mp_int * a, mp_digit b, mp_int * c, mp_digit * d)
   return res;
 }
 
+#endif
diff --git a/bn_mp_dr_is_modulus.c b/bn_mp_dr_is_modulus.c
index 0206e0ed237de8ff70bd217fdfd9b6da94b47931..5ef78a39734f8b817da0f6222ae87bab8962a89a 100644
--- a/bn_mp_dr_is_modulus.c
+++ b/bn_mp_dr_is_modulus.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_DR_IS_MODULUS_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* determines if a number is a valid DR modulus */
 int mp_dr_is_modulus(mp_int *a)
@@ -35,3 +36,4 @@ int mp_dr_is_modulus(mp_int *a)
    return 1;
 }
 
+#endif
diff --git a/bn_mp_dr_reduce.c b/bn_mp_dr_reduce.c
index bad240aa6c125dbb759b89a5654d34c415b40b6a..308b80a1ce4355068656bef85734e2175f36707a 100644
--- a/bn_mp_dr_reduce.c
+++ b/bn_mp_dr_reduce.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_DR_REDUCE_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* reduce "x" in place modulo "n" using the Diminished Radix algorithm.
  *
@@ -86,3 +87,4 @@ top:
   }
   return MP_OKAY;
 }
+#endif
diff --git a/bn_mp_dr_setup.c b/bn_mp_dr_setup.c
index f269436c18295a49b5e2a9b3030009a6b634011a..029d310879a7d071e1b9d1323f60f0eba347b371 100644
--- a/bn_mp_dr_setup.c
+++ b/bn_mp_dr_setup.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_DR_SETUP_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* determines the setup value */
 void mp_dr_setup(mp_int *a, mp_digit *d)
@@ -24,3 +25,4 @@ void mp_dr_setup(mp_int *a, mp_digit *d)
         ((mp_word)a->dp[0]));
 }
 
+#endif
diff --git a/bn_mp_exch.c b/bn_mp_exch.c
index b0f1cd000cd98e303bd2fd541fb6010accea50b5..0ef485adb6ae5d67ed402c74a27e2c1066650efb 100644
--- a/bn_mp_exch.c
+++ b/bn_mp_exch.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_EXCH_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* swap the elements of two integers, for cases where you can't simply swap the 
  * mp_int pointers around
@@ -26,3 +27,4 @@ mp_exch (mp_int * a, mp_int * b)
   *a = *b;
   *b = t;
 }
+#endif
diff --git a/bn_mp_expt_d.c b/bn_mp_expt_d.c
index 6ec12a494c0be976f890a6e68e64673739263bdd..fdb8bd91450f655d899926b882e88720a550f61e 100644
--- a/bn_mp_expt_d.c
+++ b/bn_mp_expt_d.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_EXPT_D_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* calculate c = a**b  using a square-multiply algorithm */
 int mp_expt_d (mp_int * a, mp_digit b, mp_int * c)
@@ -49,3 +50,4 @@ int mp_expt_d (mp_int * a, mp_digit b, mp_int * c)
   mp_clear (&g);
   return MP_OKAY;
 }
+#endif
diff --git a/bn_mp_exptmod.c b/bn_mp_exptmod.c
index 21d0cc50c2a435d6d18a7887f6c8b4c6e6a19f3a..da88fecaa7d43cd7707d7cd1aa8ccc7b82e4e098 100644
--- a/bn_mp_exptmod.c
+++ b/bn_mp_exptmod.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_EXPTMOD_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 
 /* this is a shell function that calls either the normal or Montgomery
@@ -31,6 +32,7 @@ int mp_exptmod (mp_int * G, mp_int * X, mp_int * P, mp_int * Y)
 
   /* if exponent X is negative we have to recurse */
   if (X->sign == MP_NEG) {
+#ifdef BN_MP_INVMOD_C
      mp_int tmpG, tmpX;
      int err;
 
@@ -57,22 +59,42 @@ int mp_exptmod (mp_int * G, mp_int * X, mp_int * P, mp_int * Y)
      err = mp_exptmod(&tmpG, &tmpX, P, Y);
      mp_clear_multi(&tmpG, &tmpX, NULL);
      return err;
+#else 
+     /* no invmod */
+     return MP_VAL
+#endif
   }
 
+#ifdef BN_MP_DR_IS_MODULUS_C
   /* is it a DR modulus? */
   dr = mp_dr_is_modulus(P);
+#else
+  dr = 0;
+#endif
 
+#ifdef BN_MP_REDUCE_IS_2K_C
   /* if not, is it a uDR modulus? */
   if (dr == 0) {
      dr = mp_reduce_is_2k(P) << 1;
   }
+#endif
     
   /* if the modulus is odd or dr != 0 use the fast method */
+#ifdef BN_MP_EXPTMOD_FAST_C
   if (mp_isodd (P) == 1 || dr !=  0) {
     return mp_exptmod_fast (G, X, P, Y, dr);
   } else {
+#endif
+#ifdef BN_S_MP_EXPTMOD_C
     /* otherwise use the generic Barrett reduction technique */
     return s_mp_exptmod (G, X, P, Y);
+#else
+    /* no exptmod for evens */
+    return MP_VAL;
+#endif
+#ifdef BN_MP_EXPTMOD_FAST_C
   }
+#endif
 }
 
+#endif
diff --git a/bn_mp_exptmod_fast.c b/bn_mp_exptmod_fast.c
index c281733971ee30067af4a89d2abecfda2d7f0d8f..4351e60f9b4367eae3450753afa6ec1d25f03665 100644
--- a/bn_mp_exptmod_fast.c
+++ b/bn_mp_exptmod_fast.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_EXPTMOD_FAST_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* computes Y == G**X mod P, HAC pp.616, Algorithm 14.85
  *
@@ -84,29 +85,52 @@ mp_exptmod_fast (mp_int * G, mp_int * X, mp_int * P, mp_int * Y, int redmode)
 
   /* determine and setup reduction code */
   if (redmode == 0) {
+#ifdef BN_MP_MONTGOMERY_SETUP_C     
      /* now setup montgomery  */
      if ((err = mp_montgomery_setup (P, &mp)) != MP_OKAY) {
         goto __M;
      }
+#else
+     err = MP_VAL;
+     goto __M;
+#endif
 
      /* automatically pick the comba one if available (saves quite a few calls/ifs) */
+#ifdef BN_FAST_MP_MONTGOMERY_REDUCE_C
      if (((P->used * 2 + 1) < MP_WARRAY) &&
           P->used < (1 << ((CHAR_BIT * sizeof (mp_word)) - (2 * DIGIT_BIT)))) {
         redux = fast_mp_montgomery_reduce;
-     } else {
+     } else 
+#endif
+     {
+#ifdef BN_MP_MONTGOMERY_REDUCE_C
         /* use slower baseline Montgomery method */
         redux = mp_montgomery_reduce;
+#else
+        err = MP_VAL;
+        goto __M;
+#endif
      }
   } else if (redmode == 1) {
+#if defined(BN_MP_DR_SETUP_C) && defined(BN_MP_DR_REDUCE_C)
      /* setup DR reduction for moduli of the form B**k - b */
      mp_dr_setup(P, &mp);
      redux = mp_dr_reduce;
+#else
+     err = MP_VAL;
+     goto __M;
+#endif
   } else {
+#if defined(BN_MP_REDUCE_2K_SETUP_C) && defined(BN_MP_REDUCE_2K_C)
      /* setup DR reduction for moduli of the form 2**k - b */
      if ((err = mp_reduce_2k_setup(P, &mp)) != MP_OKAY) {
         goto __M;
      }
      redux = mp_reduce_2k;
+#else
+     err = MP_VAL;
+     goto __M;
+#endif
   }
 
   /* setup result */
@@ -116,16 +140,21 @@ mp_exptmod_fast (mp_int * G, mp_int * X, mp_int * P, mp_int * Y, int redmode)
 
   /* create M table
    *
-   * The M table contains powers of the input base, e.g. M[x] = G^x mod P
+
    *
    * The first half of the table is not computed though accept for M[0] and M[1]
    */
 
   if (redmode == 0) {
+#ifdef BN_MP_MONTGOMERY_CALC_NORMALIZATION_C
      /* now we need R mod m */
      if ((err = mp_montgomery_calc_normalization (&res, P)) != MP_OKAY) {
        goto __RES;
      }
+#else 
+     err = MP_VAL;
+     goto __RES;
+#endif
 
      /* now set M[1] to G * R mod m */
      if ((err = mp_mulmod (G, &res, P, &M[1])) != MP_OKAY) {
@@ -269,7 +298,7 @@ mp_exptmod_fast (mp_int * G, mp_int * X, mp_int * P, mp_int * Y, int redmode)
       * to reduce one more time to cancel out the factor
       * of R.
       */
-     if ((err = mp_montgomery_reduce (&res, P, mp)) != MP_OKAY) {
+     if ((err = redux(&res, P, mp)) != MP_OKAY) {
        goto __RES;
      }
   }
@@ -285,3 +314,5 @@ __M:
   }
   return err;
 }
+#endif
+
diff --git a/bn_mp_exteuclid.c b/bn_mp_exteuclid.c
index cb3f78726e6181382d72d0591c84cbda2bb87d88..545450bb8807d04a75a4e268919ca3cee79171c8 100644
--- a/bn_mp_exteuclid.c
+++ b/bn_mp_exteuclid.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_EXTEUCLID_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* Extended euclidean algorithm of (a, b) produces 
    a*u1 + b*u2 = u3
@@ -67,3 +68,4 @@ int mp_exteuclid(mp_int *a, mp_int *b, mp_int *U1, mp_int *U2, mp_int *U3)
 _ERR: mp_clear_multi(&u1, &u2, &u3, &v1, &v2, &v3, &t1, &t2, &t3, &q, &tmp, NULL);
    return err;
 }
+#endif
diff --git a/bn_mp_fread.c b/bn_mp_fread.c
index ef665e5f8059b9926a7be2c98fa0c906b764f8ff..293df3fa6776596a0ab33f9bf6d2f8e55a9a513b 100644
--- a/bn_mp_fread.c
+++ b/bn_mp_fread.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_FREAD_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* read a bigint from a file stream in ASCII */
 int mp_fread(mp_int *a, int radix, FILE *stream)
@@ -59,3 +60,4 @@ int mp_fread(mp_int *a, int radix, FILE *stream)
    return MP_OKAY;
 }
 
+#endif
diff --git a/bn_mp_fwrite.c b/bn_mp_fwrite.c
index 2853ec1b58e653f68a960ada3a3b39a398fea243..8fa3129bebe034f2c934d5367d56e1481a490deb 100644
--- a/bn_mp_fwrite.c
+++ b/bn_mp_fwrite.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_FWRITE_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 int mp_fwrite(mp_int *a, int radix, FILE *stream)
 {
@@ -44,3 +45,4 @@ int mp_fwrite(mp_int *a, int radix, FILE *stream)
    return MP_OKAY;
 }
 
+#endif
diff --git a/bn_mp_gcd.c b/bn_mp_gcd.c
index 78ddda225605e2d9e7bf441c7137a6d9b1e06526..1cd21fcd67ba8a09068d522f50a733bccaf5627b 100644
--- a/bn_mp_gcd.c
+++ b/bn_mp_gcd.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_GCD_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* Greatest Common Divisor using the binary method */
 int mp_gcd (mp_int * a, mp_int * b, mp_int * c)
@@ -105,3 +106,4 @@ __V:mp_clear (&u);
 __U:mp_clear (&v);
   return res;
 }
+#endif
diff --git a/bn_mp_get_int.c b/bn_mp_get_int.c
index 41df6e116d91ccd5c23d784d355b4f482900c05d..034467b7468bf5225e0cfe6c62748f78daa0ccba 100644
--- a/bn_mp_get_int.c
+++ b/bn_mp_get_int.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_GET_INT_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* get the lower 32-bits of an mp_int */
 unsigned long mp_get_int(mp_int * a) 
@@ -37,3 +38,4 @@ unsigned long mp_get_int(mp_int * a)
   /* force result to 32-bits always so it is consistent on non 32-bit platforms */
   return res & 0xFFFFFFFFUL;
 }
+#endif
diff --git a/bn_mp_grow.c b/bn_mp_grow.c
index 43a30284b45db90f3c8928f596d1f0ef0cf88fdd..12a78a86c485ade0891abde0f809cf81dab3ed8f 100644
--- a/bn_mp_grow.c
+++ b/bn_mp_grow.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_GROW_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* grow as required */
 int mp_grow (mp_int * a, int size)
@@ -49,3 +50,4 @@ int mp_grow (mp_int * a, int size)
   }
   return MP_OKAY;
 }
+#endif
diff --git a/bn_mp_init.c b/bn_mp_init.c
index 5c5c1adf81cf86bd79871d111ae0e4ce7ea0dba6..9d705541ba5636b5ec1a194dc53609138f74c459 100644
--- a/bn_mp_init.c
+++ b/bn_mp_init.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_INIT_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,17 +14,23 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
-/* init a new bigint */
+/* init a new mp_int */
 int mp_init (mp_int * a)
 {
+  int i;
+
   /* allocate memory required and clear it */
-  a->dp = OPT_CAST(mp_digit) XCALLOC (sizeof (mp_digit), MP_PREC);
+  a->dp = OPT_CAST(mp_digit) XMALLOC (sizeof (mp_digit) * MP_PREC);
   if (a->dp == NULL) {
     return MP_MEM;
   }
 
+  /* set the digits to zero */
+  for (i = 0; i < MP_PREC; i++) {
+      a->dp[i] = 0;
+  }
+
   /* set the used to zero, allocated digits to the default precision
    * and sign to positive */
   a->used  = 0;
@@ -31,3 +39,4 @@ int mp_init (mp_int * a)
 
   return MP_OKAY;
 }
+#endif
diff --git a/bn_mp_init_copy.c b/bn_mp_init_copy.c
index 1c5da4d131218fabfb5aac596e95d78cee5b88b8..b1b0fa2aa669312051cc0abd644c48626361cb7c 100644
--- a/bn_mp_init_copy.c
+++ b/bn_mp_init_copy.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_INIT_COPY_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* creates "a" then copies b into it */
 int mp_init_copy (mp_int * a, mp_int * b)
@@ -24,3 +25,4 @@ int mp_init_copy (mp_int * a, mp_int * b)
   }
   return mp_copy (b, a);
 }
+#endif
diff --git a/bn_mp_init_multi.c b/bn_mp_init_multi.c
index 99ce331086107d221c1d2b6b006a6a933c508895..8cb123a900a1a11ec2040f9c2f592eb4ff3522d3 100644
--- a/bn_mp_init_multi.c
+++ b/bn_mp_init_multi.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_INIT_MULTI_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 #include <stdarg.h>
 
 int mp_init_multi(mp_int *mp, ...) 
@@ -51,3 +52,4 @@ int mp_init_multi(mp_int *mp, ...)
     return res;                /* Assumed ok, if error flagged above. */
 }
 
+#endif
diff --git a/bn_mp_init_set.c b/bn_mp_init_set.c
index c8d8bf825231283bdcc116faf51d63f35bc52b66..0251e611063bf3d694cb2b45de6aa7d3fafed3f0 100644
--- a/bn_mp_init_set.c
+++ b/bn_mp_init_set.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_INIT_SET_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* initialize and set a digit */
 int mp_init_set (mp_int * a, mp_digit b)
@@ -24,3 +25,4 @@ int mp_init_set (mp_int * a, mp_digit b)
   mp_set(a, b);
   return err;
 }
+#endif
diff --git a/bn_mp_init_set_int.c b/bn_mp_init_set_int.c
index 2d6628d547587b5a874bd17cbaf4e7d0a81c2493..f59fd190176a9d071cf00910699c82c882fbf4e2 100644
--- a/bn_mp_init_set_int.c
+++ b/bn_mp_init_set_int.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_INIT_SET_INT_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* initialize and set a digit */
 int mp_init_set_int (mp_int * a, unsigned long b)
@@ -23,3 +24,4 @@ int mp_init_set_int (mp_int * a, unsigned long b)
   }
   return mp_set_int(a, b);
 }
+#endif
diff --git a/bn_mp_init_size.c b/bn_mp_init_size.c
index c763ee0f2a961a1ba3ea99db2c2c48e06099bb96..845ce2c0484cd75b819efca26320cd3ac65e1b3c 100644
--- a/bn_mp_init_size.c
+++ b/bn_mp_init_size.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_INIT_SIZE_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,22 +14,31 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* init an mp_init for a given size */
 int mp_init_size (mp_int * a, int size)
 {
+  int x;
+
   /* pad size so there are always extra digits */
   size += (MP_PREC * 2) - (size % MP_PREC);	
   
   /* alloc mem */
-  a->dp = OPT_CAST(mp_digit) XCALLOC (sizeof (mp_digit), size);
+  a->dp = OPT_CAST(mp_digit) XMALLOC (sizeof (mp_digit) * size);
   if (a->dp == NULL) {
     return MP_MEM;
   }
+
+  /* set the members */
   a->used  = 0;
   a->alloc = size;
   a->sign  = MP_ZPOS;
 
+  /* zero the digits */
+  for (x = 0; x < size; x++) {
+      a->dp[x] = 0;
+  }
+
   return MP_OKAY;
 }
+#endif
diff --git a/bn_mp_invmod.c b/bn_mp_invmod.c
index 4ac5b2f67642254fb68025a3e35b0402cf78a223..46118ad88182ae24dc2bbebfaf6b2eefdaa69bc4 100644
--- a/bn_mp_invmod.c
+++ b/bn_mp_invmod.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_INVMOD_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,163 +14,26 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* hac 14.61, pp608 */
 int mp_invmod (mp_int * a, mp_int * b, mp_int * c)
 {
-  mp_int  x, y, u, v, A, B, C, D;
-  int     res;
-
   /* b cannot be negative */
   if (b->sign == MP_NEG || mp_iszero(b) == 1) {
     return MP_VAL;
   }
 
+#ifdef BN_FAST_MP_INVMOD_C
   /* if the modulus is odd we can use a faster routine instead */
   if (mp_isodd (b) == 1) {
     return fast_mp_invmod (a, b, c);
   }
-  
-  /* init temps */
-  if ((res = mp_init_multi(&x, &y, &u, &v, 
-                           &A, &B, &C, &D, NULL)) != MP_OKAY) {
-     return res;
-  }
-
-  /* x = a, y = b */
-  if ((res = mp_copy (a, &x)) != MP_OKAY) {
-    goto __ERR;
-  }
-  if ((res = mp_copy (b, &y)) != MP_OKAY) {
-    goto __ERR;
-  }
+#endif
 
-  /* 2. [modified] if x,y are both even then return an error! */
-  if (mp_iseven (&x) == 1 && mp_iseven (&y) == 1) {
-    res = MP_VAL;
-    goto __ERR;
-  }
-
-  /* 3. u=x, v=y, A=1, B=0, C=0,D=1 */
-  if ((res = mp_copy (&x, &u)) != MP_OKAY) {
-    goto __ERR;
-  }
-  if ((res = mp_copy (&y, &v)) != MP_OKAY) {
-    goto __ERR;
-  }
-  mp_set (&A, 1);
-  mp_set (&D, 1);
+#ifdef BN_MP_INVMOD_SLOW_C
+  return mp_invmod_slow(a, b, c);
+#endif
 
-top:
-  /* 4.  while u is even do */
-  while (mp_iseven (&u) == 1) {
-    /* 4.1 u = u/2 */
-    if ((res = mp_div_2 (&u, &u)) != MP_OKAY) {
-      goto __ERR;
-    }
-    /* 4.2 if A or B is odd then */
-    if (mp_isodd (&A) == 1 || mp_isodd (&B) == 1) {
-      /* A = (A+y)/2, B = (B-x)/2 */
-      if ((res = mp_add (&A, &y, &A)) != MP_OKAY) {
-         goto __ERR;
-      }
-      if ((res = mp_sub (&B, &x, &B)) != MP_OKAY) {
-         goto __ERR;
-      }
-    }
-    /* A = A/2, B = B/2 */
-    if ((res = mp_div_2 (&A, &A)) != MP_OKAY) {
-      goto __ERR;
-    }
-    if ((res = mp_div_2 (&B, &B)) != MP_OKAY) {
-      goto __ERR;
-    }
-  }
-
-  /* 5.  while v is even do */
-  while (mp_iseven (&v) == 1) {
-    /* 5.1 v = v/2 */
-    if ((res = mp_div_2 (&v, &v)) != MP_OKAY) {
-      goto __ERR;
-    }
-    /* 5.2 if C or D is odd then */
-    if (mp_isodd (&C) == 1 || mp_isodd (&D) == 1) {
-      /* C = (C+y)/2, D = (D-x)/2 */
-      if ((res = mp_add (&C, &y, &C)) != MP_OKAY) {
-         goto __ERR;
-      }
-      if ((res = mp_sub (&D, &x, &D)) != MP_OKAY) {
-         goto __ERR;
-      }
-    }
-    /* C = C/2, D = D/2 */
-    if ((res = mp_div_2 (&C, &C)) != MP_OKAY) {
-      goto __ERR;
-    }
-    if ((res = mp_div_2 (&D, &D)) != MP_OKAY) {
-      goto __ERR;
-    }
-  }
-
-  /* 6.  if u >= v then */
-  if (mp_cmp (&u, &v) != MP_LT) {
-    /* u = u - v, A = A - C, B = B - D */
-    if ((res = mp_sub (&u, &v, &u)) != MP_OKAY) {
-      goto __ERR;
-    }
-
-    if ((res = mp_sub (&A, &C, &A)) != MP_OKAY) {
-      goto __ERR;
-    }
-
-    if ((res = mp_sub (&B, &D, &B)) != MP_OKAY) {
-      goto __ERR;
-    }
-  } else {
-    /* v - v - u, C = C - A, D = D - B */
-    if ((res = mp_sub (&v, &u, &v)) != MP_OKAY) {
-      goto __ERR;
-    }
-
-    if ((res = mp_sub (&C, &A, &C)) != MP_OKAY) {
-      goto __ERR;
-    }
-
-    if ((res = mp_sub (&D, &B, &D)) != MP_OKAY) {
-      goto __ERR;
-    }
-  }
-
-  /* if not zero goto step 4 */
-  if (mp_iszero (&u) == 0)
-    goto top;
-
-  /* now a = C, b = D, gcd == g*v */
-
-  /* if v != 1 then there is no inverse */
-  if (mp_cmp_d (&v, 1) != MP_EQ) {
-    res = MP_VAL;
-    goto __ERR;
-  }
-
-  /* if its too low */
-  while (mp_cmp_d(&C, 0) == MP_LT) {
-      if ((res = mp_add(&C, b, &C)) != MP_OKAY) {
-         goto __ERR;
-      }
-  }
-  
-  /* too big */
-  while (mp_cmp_mag(&C, b) != MP_LT) {
-      if ((res = mp_sub(&C, b, &C)) != MP_OKAY) {
-         goto __ERR;
-      }
-  }
-  
-  /* C is now the inverse */
-  mp_exch (&C, c);
-  res = MP_OKAY;
-__ERR:mp_clear_multi (&x, &y, &u, &v, &A, &B, &C, &D, NULL);
-  return res;
+  return MP_VAL;
 }
+#endif
diff --git a/bn_mp_invmod_slow.c b/bn_mp_invmod_slow.c
new file mode 100644
index 0000000000000000000000000000000000000000..8ecb009488bcb3214b24f13047c6a616a510df0a
--- /dev/null
+++ b/bn_mp_invmod_slow.c
@@ -0,0 +1,171 @@
+#include <tommath.h>
+#ifdef BN_MP_INVMOD_SLOW_C
+/* LibTomMath, multiple-precision integer library -- Tom St Denis
+ *
+ * LibTomMath is a library that provides multiple-precision
+ * integer arithmetic as well as number theoretic functionality.
+ *
+ * The library was designed directly after the MPI library by
+ * Michael Fromberger but has been written from scratch with
+ * additional optimizations in place.
+ *
+ * The library is free for all purposes without any express
+ * guarantee it works.
+ *
+ * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
+ */
+
+/* hac 14.61, pp608 */
+int mp_invmod_slow (mp_int * a, mp_int * b, mp_int * c)
+{
+  mp_int  x, y, u, v, A, B, C, D;
+  int     res;
+
+  /* b cannot be negative */
+  if (b->sign == MP_NEG || mp_iszero(b) == 1) {
+    return MP_VAL;
+  }
+
+  /* init temps */
+  if ((res = mp_init_multi(&x, &y, &u, &v, 
+                           &A, &B, &C, &D, NULL)) != MP_OKAY) {
+     return res;
+  }
+
+  /* x = a, y = b */
+  if ((res = mp_copy (a, &x)) != MP_OKAY) {
+    goto __ERR;
+  }
+  if ((res = mp_copy (b, &y)) != MP_OKAY) {
+    goto __ERR;
+  }
+
+  /* 2. [modified] if x,y are both even then return an error! */
+  if (mp_iseven (&x) == 1 && mp_iseven (&y) == 1) {
+    res = MP_VAL;
+    goto __ERR;
+  }
+
+  /* 3. u=x, v=y, A=1, B=0, C=0,D=1 */
+  if ((res = mp_copy (&x, &u)) != MP_OKAY) {
+    goto __ERR;
+  }
+  if ((res = mp_copy (&y, &v)) != MP_OKAY) {
+    goto __ERR;
+  }
+  mp_set (&A, 1);
+  mp_set (&D, 1);
+
+top:
+  /* 4.  while u is even do */
+  while (mp_iseven (&u) == 1) {
+    /* 4.1 u = u/2 */
+    if ((res = mp_div_2 (&u, &u)) != MP_OKAY) {
+      goto __ERR;
+    }
+    /* 4.2 if A or B is odd then */
+    if (mp_isodd (&A) == 1 || mp_isodd (&B) == 1) {
+      /* A = (A+y)/2, B = (B-x)/2 */
+      if ((res = mp_add (&A, &y, &A)) != MP_OKAY) {
+         goto __ERR;
+      }
+      if ((res = mp_sub (&B, &x, &B)) != MP_OKAY) {
+         goto __ERR;
+      }
+    }
+    /* A = A/2, B = B/2 */
+    if ((res = mp_div_2 (&A, &A)) != MP_OKAY) {
+      goto __ERR;
+    }
+    if ((res = mp_div_2 (&B, &B)) != MP_OKAY) {
+      goto __ERR;
+    }
+  }
+
+  /* 5.  while v is even do */
+  while (mp_iseven (&v) == 1) {
+    /* 5.1 v = v/2 */
+    if ((res = mp_div_2 (&v, &v)) != MP_OKAY) {
+      goto __ERR;
+    }
+    /* 5.2 if C or D is odd then */
+    if (mp_isodd (&C) == 1 || mp_isodd (&D) == 1) {
+      /* C = (C+y)/2, D = (D-x)/2 */
+      if ((res = mp_add (&C, &y, &C)) != MP_OKAY) {
+         goto __ERR;
+      }
+      if ((res = mp_sub (&D, &x, &D)) != MP_OKAY) {
+         goto __ERR;
+      }
+    }
+    /* C = C/2, D = D/2 */
+    if ((res = mp_div_2 (&C, &C)) != MP_OKAY) {
+      goto __ERR;
+    }
+    if ((res = mp_div_2 (&D, &D)) != MP_OKAY) {
+      goto __ERR;
+    }
+  }
+
+  /* 6.  if u >= v then */
+  if (mp_cmp (&u, &v) != MP_LT) {
+    /* u = u - v, A = A - C, B = B - D */
+    if ((res = mp_sub (&u, &v, &u)) != MP_OKAY) {
+      goto __ERR;
+    }
+
+    if ((res = mp_sub (&A, &C, &A)) != MP_OKAY) {
+      goto __ERR;
+    }
+
+    if ((res = mp_sub (&B, &D, &B)) != MP_OKAY) {
+      goto __ERR;
+    }
+  } else {
+    /* v - v - u, C = C - A, D = D - B */
+    if ((res = mp_sub (&v, &u, &v)) != MP_OKAY) {
+      goto __ERR;
+    }
+
+    if ((res = mp_sub (&C, &A, &C)) != MP_OKAY) {
+      goto __ERR;
+    }
+
+    if ((res = mp_sub (&D, &B, &D)) != MP_OKAY) {
+      goto __ERR;
+    }
+  }
+
+  /* if not zero goto step 4 */
+  if (mp_iszero (&u) == 0)
+    goto top;
+
+  /* now a = C, b = D, gcd == g*v */
+
+  /* if v != 1 then there is no inverse */
+  if (mp_cmp_d (&v, 1) != MP_EQ) {
+    res = MP_VAL;
+    goto __ERR;
+  }
+
+  /* if its too low */
+  while (mp_cmp_d(&C, 0) == MP_LT) {
+      if ((res = mp_add(&C, b, &C)) != MP_OKAY) {
+         goto __ERR;
+      }
+  }
+  
+  /* too big */
+  while (mp_cmp_mag(&C, b) != MP_LT) {
+      if ((res = mp_sub(&C, b, &C)) != MP_OKAY) {
+         goto __ERR;
+      }
+  }
+  
+  /* C is now the inverse */
+  mp_exch (&C, c);
+  res = MP_OKAY;
+__ERR:mp_clear_multi (&x, &y, &u, &v, &A, &B, &C, &D, NULL);
+  return res;
+}
+#endif
diff --git a/bn_mp_is_square.c b/bn_mp_is_square.c
index 1f01bca51e808b0b152e31503a851e82c625c292..969d2377d1f13fccc57057cf0e0bef33b265bf13 100644
--- a/bn_mp_is_square.c
+++ b/bn_mp_is_square.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_IS_SQUARE_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* Check if remainders are possible squares - fast exclude non-squares */
 static const char rem_128[128] = {
@@ -69,7 +70,7 @@ int mp_is_square(mp_int *arg,int *ret)
      return MP_OKAY;
   }
 
-  /* product of primes less than 2^31 */
+
   if ((res = mp_init_set_int(&t,11L*13L*17L*19L*23L*29L*31L)) != MP_OKAY) {
      return res;
   }
@@ -101,3 +102,4 @@ int mp_is_square(mp_int *arg,int *ret)
 ERR:mp_clear(&t);
   return res;
 }
+#endif
diff --git a/bn_mp_jacobi.c b/bn_mp_jacobi.c
index 634370f289236e57347274ee08df9bf37c9a2dc8..1c69cfde920b58d2d59e564a7b741e7f31b7c9ba 100644
--- a/bn_mp_jacobi.c
+++ b/bn_mp_jacobi.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_JACOBI_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* computes the jacobi c = (a | n) (or Legendre if n is prime)
  * HAC pp. 73 Algorithm 2.149
@@ -97,3 +98,4 @@ __P1:mp_clear (&p1);
 __A1:mp_clear (&a1);
   return res;
 }
+#endif
diff --git a/bn_mp_karatsuba_mul.c b/bn_mp_karatsuba_mul.c
index 169dacf90536a1d3d35755dedb6217a7fb9a1dd2..daa78c7158849df7cbe01a5b405b7841cbe9ce49 100644
--- a/bn_mp_karatsuba_mul.c
+++ b/bn_mp_karatsuba_mul.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_KARATSUBA_MUL_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* c = |a| * |b| using Karatsuba Multiplication using 
  * three half size multiplications
@@ -76,9 +77,6 @@ int mp_karatsuba_mul (mp_int * a, mp_int * b, mp_int * c)
     goto X0Y0;
 
   /* now shift the digits */
-  x0.sign = x1.sign = a->sign;
-  y0.sign = y1.sign = b->sign;
-
   x0.used = y0.used = B;
   x1.used = a->used - B;
   y1.used = b->used - B;
@@ -162,3 +160,4 @@ X0:mp_clear (&x0);
 ERR:
   return err;
 }
+#endif
diff --git a/bn_mp_karatsuba_sqr.c b/bn_mp_karatsuba_sqr.c
index c3356135927bd3c0dc9b68d1b9777fc8a081e298..315ceabfae224e3de5ef054b18bbd3a89963231c 100644
--- a/bn_mp_karatsuba_sqr.c
+++ b/bn_mp_karatsuba_sqr.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_KARATSUBA_SQR_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,12 +14,11 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* Karatsuba squaring, computes b = a*a using three 
  * half size squarings
  *
- * See comments of mp_karatsuba_mul for details.  It 
+ * See comments of karatsuba_mul for details.  It 
  * is essentially the same algorithm but merely 
  * tuned to perform recursive squarings.
  */
@@ -113,3 +114,4 @@ X0:mp_clear (&x0);
 ERR:
   return err;
 }
+#endif
diff --git a/bn_mp_lcm.c b/bn_mp_lcm.c
index 4aa5749084ff905f240296b99e94583bb5852b26..340d757b7c0f214f12d309bb3eef82b2b6c1ff1c 100644
--- a/bn_mp_lcm.c
+++ b/bn_mp_lcm.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_LCM_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* computes least common multiple as |a*b|/(a, b) */
 int mp_lcm (mp_int * a, mp_int * b, mp_int * c)
@@ -52,3 +53,4 @@ __T:
   mp_clear_multi (&t1, &t2, NULL);
   return res;
 }
+#endif
diff --git a/bn_mp_lshd.c b/bn_mp_lshd.c
index f1702376c7adcbf95789602295a129b9009a24b5..398b6482280a959ade5ad51494797bb3ce367ffb 100644
--- a/bn_mp_lshd.c
+++ b/bn_mp_lshd.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_LSHD_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* shift left a certain amount of digits */
 int mp_lshd (mp_int * a, int b)
@@ -59,3 +60,4 @@ int mp_lshd (mp_int * a, int b)
   }
   return MP_OKAY;
 }
+#endif
diff --git a/bn_mp_mod.c b/bn_mp_mod.c
index ad963a9761fde2e5525f1311d8f5e1c58032d0a5..75779bb0cfbba1fea657494bbd532b8eb611bf9a 100644
--- a/bn_mp_mod.c
+++ b/bn_mp_mod.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_MOD_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* c = a mod b, 0 <= c < b */
 int
@@ -40,3 +41,4 @@ mp_mod (mp_int * a, mp_int * b, mp_int * c)
   mp_clear (&t);
   return res;
 }
+#endif
diff --git a/bn_mp_mod_2d.c b/bn_mp_mod_2d.c
index 89e9081c5beab89cdb0a3fbcefdec16f7b2a2ea1..f81a0d41de94e4a66a78b139a9896e0318d869d0 100644
--- a/bn_mp_mod_2d.c
+++ b/bn_mp_mod_2d.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_MOD_2D_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* calc a value mod 2**b */
 int
@@ -47,3 +48,4 @@ mp_mod_2d (mp_int * a, int b, mp_int * c)
   mp_clamp (c);
   return MP_OKAY;
 }
+#endif
diff --git a/bn_mp_mod_d.c b/bn_mp_mod_d.c
index c5dbe46a019d2cc5278b9ec3295855305d8added..8a2ad242450d75d4070414c53d12f282085a807f 100644
--- a/bn_mp_mod_d.c
+++ b/bn_mp_mod_d.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_MOD_D_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,10 +14,10 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 int
 mp_mod_d (mp_int * a, mp_digit b, mp_digit * c)
 {
   return mp_div_d(a, b, NULL, c);
 }
+#endif
diff --git a/bn_mp_montgomery_calc_normalization.c b/bn_mp_montgomery_calc_normalization.c
index a38555cec4c58cde5cc1fc25965400539061e7cf..0a760cfe1c08b6b8fb820140671067721d91845c 100644
--- a/bn_mp_montgomery_calc_normalization.c
+++ b/bn_mp_montgomery_calc_normalization.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_MONTGOMERY_CALC_NORMALIZATION_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,31 +14,31 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
-/* calculates a = B^n mod b for Montgomery reduction
- * Where B is the base [e.g. 2^DIGIT_BIT].
- * B^n mod b is computed by first computing
- * A = B^(n-1) which doesn't require a reduction but a simple OR.
- * then C = A * B = B^n is computed by performing upto DIGIT_BIT
+/*
  * shifts with subtractions when the result is greater than b.
  *
  * The method is slightly modified to shift B unconditionally upto just under
  * the leading bit of b.  This saves alot of multiple precision shifting.
  */
-int
-mp_montgomery_calc_normalization (mp_int * a, mp_int * b)
+int mp_montgomery_calc_normalization (mp_int * a, mp_int * b)
 {
   int     x, bits, res;
 
   /* how many bits of last digit does b use */
   bits = mp_count_bits (b) % DIGIT_BIT;
 
-  /* compute A = B^(n-1) * 2^(bits-1) */
-  if ((res = mp_2expt (a, (b->used - 1) * DIGIT_BIT + bits - 1)) != MP_OKAY) {
-    return res;
+
+  if (b->used > 1) {
+     if ((res = mp_2expt (a, (b->used - 1) * DIGIT_BIT + bits - 1)) != MP_OKAY) {
+        return res;
+     }
+  } else {
+     mp_set(a, 1);
+     bits = 1;
   }
 
+
   /* now compute C = A * B mod b */
   for (x = bits - 1; x < (int)DIGIT_BIT; x++) {
     if ((res = mp_mul_2 (a, a)) != MP_OKAY) {
@@ -51,3 +53,4 @@ mp_montgomery_calc_normalization (mp_int * a, mp_int * b)
 
   return MP_OKAY;
 }
+#endif
diff --git a/bn_mp_montgomery_reduce.c b/bn_mp_montgomery_reduce.c
index 52b98cdb0e1305c42b8ede0820955e9324bd7427..3095fa76b530ac74c6194fc2b82907ec99523ef0 100644
--- a/bn_mp_montgomery_reduce.c
+++ b/bn_mp_montgomery_reduce.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_MONTGOMERY_REDUCE_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* computes xR**-1 == x (mod N) via Montgomery Reduction */
 int
@@ -23,7 +24,7 @@ mp_montgomery_reduce (mp_int * x, mp_int * n, mp_digit rho)
 
   /* can the fast reduction [comba] method be used?
    *
-   * Note that unlike in mp_mul you're safely allowed *less*
+   * Note that unlike in mul you're safely allowed *less*
    * than the available columns [255 per default] since carries
    * are fixed up in the inner loop.
    */
@@ -46,7 +47,7 @@ mp_montgomery_reduce (mp_int * x, mp_int * n, mp_digit rho)
     /* mu = ai * rho mod b
      *
      * The value of rho must be precalculated via
-     * bn_mp_montgomery_setup() such that
+     * montgomery_setup() such that
      * it equals -1/n0 mod b this allows the
      * following inner loop to reduce the
      * input one digit at a time
@@ -110,3 +111,4 @@ mp_montgomery_reduce (mp_int * x, mp_int * n, mp_digit rho)
 
   return MP_OKAY;
 }
+#endif
diff --git a/bn_mp_montgomery_setup.c b/bn_mp_montgomery_setup.c
index 36253afb736650b283be32c31f7bdf4f1cd26b05..9dfc0878a951be1214a7e46570a40f29dedb0749 100644
--- a/bn_mp_montgomery_setup.c
+++ b/bn_mp_montgomery_setup.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_MONTGOMERY_SETUP_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* setups the montgomery reduction stuff */
 int
@@ -47,7 +48,8 @@ mp_montgomery_setup (mp_int * n, mp_digit * rho)
 #endif
 
   /* rho = -1/m mod b */
-  *rho = (((mp_digit) 1 << ((mp_digit) DIGIT_BIT)) - x) & MP_MASK;
+  *rho = (((mp_word)1 << ((mp_word) DIGIT_BIT)) - x) & MP_MASK;
 
   return MP_OKAY;
 }
+#endif
diff --git a/bn_mp_mul.c b/bn_mp_mul.c
index 6f3c4911d2b13f7ffb1b3230e4a5de1c6f9c9953..f9cfa09e7628478931f92aaddc6dad42116733b8 100644
--- a/bn_mp_mul.c
+++ b/bn_mp_mul.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_MUL_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* high level multiplication (handles sign) */
 int mp_mul (mp_int * a, mp_int * b, mp_int * c)
@@ -21,12 +22,18 @@ int mp_mul (mp_int * a, mp_int * b, mp_int * c)
   neg = (a->sign == b->sign) ? MP_ZPOS : MP_NEG;
 
   /* use Toom-Cook? */
+#ifdef BN_MP_TOOM_MUL_C
   if (MIN (a->used, b->used) >= TOOM_MUL_CUTOFF) {
     res = mp_toom_mul(a, b, c);
+  } else 
+#endif
+#ifdef BN_MP_KARATSUBA_MUL_C
   /* use Karatsuba? */
-  } else if (MIN (a->used, b->used) >= KARATSUBA_MUL_CUTOFF) {
+  if (MIN (a->used, b->used) >= KARATSUBA_MUL_CUTOFF) {
     res = mp_karatsuba_mul (a, b, c);
-  } else {
+  } else 
+#endif
+  {
     /* can we use the fast multiplier?
      *
      * The fast multiplier can be used if the output will 
@@ -35,14 +42,21 @@ int mp_mul (mp_int * a, mp_int * b, mp_int * c)
      */
     int     digs = a->used + b->used + 1;
 
+#ifdef BN_FAST_S_MP_MUL_DIGS_C
     if ((digs < MP_WARRAY) &&
         MIN(a->used, b->used) <= 
         (1 << ((CHAR_BIT * sizeof (mp_word)) - (2 * DIGIT_BIT)))) {
       res = fast_s_mp_mul_digs (a, b, c, digs);
-    } else {
-      res = s_mp_mul (a, b, c);
-    }
+    } else 
+#endif
+#ifdef BN_S_MP_MUL_DIGS_C
+      res = s_mp_mul (a, b, c); /* uses s_mp_mul_digs */
+#else
+      res = MP_VAL;
+#endif
+
   }
-  c->sign = neg;
+  c->sign = (c->used > 0) ? neg : MP_ZPOS;
   return res;
 }
+#endif
diff --git a/bn_mp_mul_2.c b/bn_mp_mul_2.c
index f947d754e42ed5bc578d05d8c9947f1d41ccb219..6936681f2446fa095bae604ffd9f62ee103cdcc9 100644
--- a/bn_mp_mul_2.c
+++ b/bn_mp_mul_2.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_MUL_2_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* b = a*2 */
 int mp_mul_2(mp_int * a, mp_int * b)
@@ -74,3 +75,4 @@ int mp_mul_2(mp_int * a, mp_int * b)
   b->sign = a->sign;
   return MP_OKAY;
 }
+#endif
diff --git a/bn_mp_mul_2d.c b/bn_mp_mul_2d.c
index f5fa1d072cff0a4eef2a7c3c0939ef4370c468e5..04cb8dd9e301bd0967f876f20c82af6e2f62712d 100644
--- a/bn_mp_mul_2d.c
+++ b/bn_mp_mul_2d.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_MUL_2D_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* shift left by a certain bit count */
 int mp_mul_2d (mp_int * a, int b, mp_int * c)
@@ -77,3 +78,4 @@ int mp_mul_2d (mp_int * a, int b, mp_int * c)
   mp_clamp (c);
   return MP_OKAY;
 }
+#endif
diff --git a/bn_mp_mul_d.c b/bn_mp_mul_d.c
index 658fe01ce5294eb2293457a3603f93ed690a32b4..f9363619104d5d6c308e8971f1447a7e4f9acc3a 100644
--- a/bn_mp_mul_d.c
+++ b/bn_mp_mul_d.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_MUL_D_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* multiply by a digit */
 int
@@ -70,3 +71,4 @@ mp_mul_d (mp_int * a, mp_digit b, mp_int * c)
 
   return MP_OKAY;
 }
+#endif
diff --git a/bn_mp_mulmod.c b/bn_mp_mulmod.c
index 77f8459210a4da96d8174ba9f5e002cd22b377eb..d34e90ac319827b8eaf88ce0cb120cd0e8fbb2ce 100644
--- a/bn_mp_mulmod.c
+++ b/bn_mp_mulmod.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_MULMOD_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* d = a * b (mod c) */
 int
@@ -33,3 +34,4 @@ mp_mulmod (mp_int * a, mp_int * b, mp_int * c, mp_int * d)
   mp_clear (&t);
   return res;
 }
+#endif
diff --git a/bn_mp_n_root.c b/bn_mp_n_root.c
index a79af174e3e7d73cf7d9b5aac63eede7ee373d83..94899037ef7ef6e9e6b21103a0280ad417adc3a9 100644
--- a/bn_mp_n_root.c
+++ b/bn_mp_n_root.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_N_ROOT_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* find the n'th root of an integer 
  *
@@ -124,3 +125,4 @@ __T2:mp_clear (&t2);
 __T1:mp_clear (&t1);
   return res;
 }
+#endif
diff --git a/bn_mp_neg.c b/bn_mp_neg.c
index f9de6e4348fa807d052ac13af26bf5aa531bf2d3..3a991db5aa80cf1556d525a404a3f3054591db27 100644
--- a/bn_mp_neg.c
+++ b/bn_mp_neg.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_NEG_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* b = -a */
 int mp_neg (mp_int * a, mp_int * b)
@@ -26,3 +27,4 @@ int mp_neg (mp_int * a, mp_int * b)
   }
   return MP_OKAY;
 }
+#endif
diff --git a/bn_mp_or.c b/bn_mp_or.c
index 793e8e9a701ac48298f81f06f3af8e761b5a9b4b..dccee7e1b9b97ae12de60b78d892fd2e1cef79a4 100644
--- a/bn_mp_or.c
+++ b/bn_mp_or.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_OR_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* OR two ints together */
 int mp_or (mp_int * a, mp_int * b, mp_int * c)
@@ -42,3 +43,4 @@ int mp_or (mp_int * a, mp_int * b, mp_int * c)
   mp_clear (&t);
   return MP_OKAY;
 }
+#endif
diff --git a/bn_mp_prime_fermat.c b/bn_mp_prime_fermat.c
index 7d58785642373a907f863ae2931a86c342318197..fe17aaab1a3dbd478147e1b3d5d1e9a85094001d 100644
--- a/bn_mp_prime_fermat.c
+++ b/bn_mp_prime_fermat.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_PRIME_FERMAT_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* performs one Fermat test.
  * 
@@ -54,3 +55,4 @@ int mp_prime_fermat (mp_int * a, mp_int * b, int *result)
 __T:mp_clear (&t);
   return err;
 }
+#endif
diff --git a/bn_mp_prime_is_divisible.c b/bn_mp_prime_is_divisible.c
index 86392d17f04f91eb58becd64f9032eb3a8f77c9b..22ec1aeacdf571b4136bfa9f389f54e7c32d9c3c 100644
--- a/bn_mp_prime_is_divisible.c
+++ b/bn_mp_prime_is_divisible.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_PRIME_IS_DIVISIBLE_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* determines if an integers is divisible by one 
  * of the first PRIME_SIZE primes or not
@@ -42,3 +43,4 @@ int mp_prime_is_divisible (mp_int * a, int *result)
 
   return MP_OKAY;
 }
+#endif
diff --git a/bn_mp_prime_is_prime.c b/bn_mp_prime_is_prime.c
index 88a777b41b082ed8dd13c6b4580f7fe187b9a119..c2354d2f656f015a2ae6cbc92016c1b8c7c54eb9 100644
--- a/bn_mp_prime_is_prime.c
+++ b/bn_mp_prime_is_prime.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_PRIME_IS_PRIME_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,12 +14,11 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* performs a variable number of rounds of Miller-Rabin
  *
  * Probability of error after t rounds is no more than
- * (1/4)^t when 1 <= t <= PRIME_SIZE
+
  *
  * Sets result to 1 if probably prime, 0 otherwise
  */
@@ -75,3 +76,4 @@ int mp_prime_is_prime (mp_int * a, int t, int *result)
 __B:mp_clear (&b);
   return err;
 }
+#endif
diff --git a/bn_mp_prime_miller_rabin.c b/bn_mp_prime_miller_rabin.c
index 0c55d9c7750d6afe275f93fbbdaec7adfe5ff28c..22dec2f2e7441660ad7029d298f497522996c3a9 100644
--- a/bn_mp_prime_miller_rabin.c
+++ b/bn_mp_prime_miller_rabin.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_PRIME_MILLER_RABIN_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* Miller-Rabin test of "a" to the base of "b" as described in 
  * HAC pp. 139 Algorithm 4.24
@@ -95,3 +96,4 @@ __R:mp_clear (&r);
 __N1:mp_clear (&n1);
   return err;
 }
+#endif
diff --git a/bn_mp_prime_next_prime.c b/bn_mp_prime_next_prime.c
index f45af81eff9f10c7eec1c4ccdd1d3d94d4cd618f..c478ce5a1a1866cfb521fc9446e07a3c9e4e455b 100644
--- a/bn_mp_prime_next_prime.c
+++ b/bn_mp_prime_next_prime.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_PRIME_NEXT_PRIME_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* finds the next prime after the number "a" using "t" trials
  * of Miller-Rabin.
@@ -162,3 +163,4 @@ __ERR:
    return err;
 }
 
+#endif
diff --git a/bn_prime_sizes_tab.c b/bn_mp_prime_rabin_miller_trials.c
similarity index 82%
rename from bn_prime_sizes_tab.c
rename to bn_mp_prime_rabin_miller_trials.c
index c6e1dadfb3dcb47cf6c40811917cc4e6c0130dce..d1d0867196fd9b7da9913d731df845811885d38f 100644
--- a/bn_prime_sizes_tab.c
+++ b/bn_mp_prime_rabin_miller_trials.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_PRIME_RABIN_MILLER_TRIALS_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,9 +14,8 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
-/* this table gives the # of rabin miller trials for a prob of failure lower than 2^-96 */
+
 static const struct {
    int k, t;
 } sizes[] = {
@@ -25,13 +26,8 @@ static const struct {
 {   640,     6 },
 {   768,     5 },
 {   896,     4 },
-{  1024,     4 },
-{  1152,     3 },
-{  1280,     3 },
-{  1408,     3 },
-{  1536,     3 },
-{  1664,     3 },
-{  1792,     2 } };
+{  1024,     4 }
+};
 
 /* returns # of RM trials required for a given bit size */
 int mp_prime_rabin_miller_trials(int size)
@@ -45,7 +41,8 @@ int mp_prime_rabin_miller_trials(int size)
           return (x == 0) ? sizes[0].t : sizes[x - 1].t;
        }
    }
-   return 1;
+   return sizes[x-1].t + 1;
 }
 
 
+#endif
diff --git a/bn_mp_prime_random_ex.c b/bn_mp_prime_random_ex.c
index 147721bda7e5db226c693daa69e85f1b395003ca..2c4f4f07a1ea8ab0179686053da3a212b15c2e55 100644
--- a/bn_mp_prime_random_ex.c
+++ b/bn_mp_prime_random_ex.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_PRIME_RANDOM_EX_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* makes a truly random prime of a given size (bits),
  *
@@ -92,6 +93,9 @@ int mp_prime_random_ex(mp_int *a, int t, int size, int flags, ltm_prime_callback
 
       /* is it prime? */
       if ((err = mp_prime_is_prime(a, t, &res)) != MP_OKAY)           { goto error; }
+      if (res == MP_NO) {  
+         continue;
+      }
 
       if (flags & LTM_PRIME_SAFE) {
          /* see if (a-1)/2 is prime */
@@ -116,3 +120,4 @@ error:
 }
 
 
+#endif
diff --git a/bn_mp_radix_size.c b/bn_mp_radix_size.c
index ae2d09a705a9113d31a01e19b87e7671472eca75..30b78d9927e64aaa6f9fbb0898ffdf660064a825 100644
--- a/bn_mp_radix_size.c
+++ b/bn_mp_radix_size.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_RADIX_SIZE_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* returns size of ASCII reprensentation */
 int mp_radix_size (mp_int * a, int radix, int *size)
@@ -63,3 +64,4 @@ int mp_radix_size (mp_int * a, int radix, int *size)
   return MP_OKAY;
 }
 
+#endif
diff --git a/bn_mp_radix_smap.c b/bn_mp_radix_smap.c
index 2c4d96633c50545e9b056caf5319ab1dbb7aed83..bc7517dac46c6ade3975a509318d33e0afd4e262 100644
--- a/bn_mp_radix_smap.c
+++ b/bn_mp_radix_smap.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_RADIX_SMAP_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,7 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* chars used in radix conversions */
 const char *mp_s_rmap = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz+/";
+#endif
diff --git a/bn_mp_rand.c b/bn_mp_rand.c
index c7449947b7fb3a6b44281d52cbfa2096cc0ffaa3..1cc47f19694d46eb4cd3b566cc19965015795fe1 100644
--- a/bn_mp_rand.c
+++ b/bn_mp_rand.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_RAND_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* makes a pseudo-random int of a given size */
 int
@@ -47,3 +48,4 @@ mp_rand (mp_int * a, int digits)
 
   return MP_OKAY;
 }
+#endif
diff --git a/bn_mp_read_radix.c b/bn_mp_read_radix.c
index 653a891c14697c2b332dd5adf931bf860c684cfc..704bd0fed68a053cee6c047f20c88d35b99a5a0e 100644
--- a/bn_mp_read_radix.c
+++ b/bn_mp_read_radix.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_READ_RADIX_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* read a string [ASCII] in a given radix */
 int mp_read_radix (mp_int * a, char *str, int radix)
@@ -74,3 +75,4 @@ int mp_read_radix (mp_int * a, char *str, int radix)
   }
   return MP_OKAY;
 }
+#endif
diff --git a/bn_mp_read_signed_bin.c b/bn_mp_read_signed_bin.c
index 39d861832839e07920c073d4956ff00a19285307..814d6c12f33e562c3b94d288cad133babb3d8acc 100644
--- a/bn_mp_read_signed_bin.c
+++ b/bn_mp_read_signed_bin.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_READ_SIGNED_BIN_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* read signed bin, big endian, first byte is 0==positive or 1==negative */
 int
@@ -34,3 +35,4 @@ mp_read_signed_bin (mp_int * a, unsigned char *b, int c)
 
   return MP_OKAY;
 }
+#endif
diff --git a/bn_mp_read_unsigned_bin.c b/bn_mp_read_unsigned_bin.c
index cb11d873316a17751f6f5be7567de266122b967e..946457de27197be0f0667573c7f2382e1c02c07e 100644
--- a/bn_mp_read_unsigned_bin.c
+++ b/bn_mp_read_unsigned_bin.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_READ_UNSIGNED_BIN_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* reads a unsigned char array, assumes the msb is stored first [big endian] */
 int
@@ -48,3 +49,4 @@ mp_read_unsigned_bin (mp_int * a, unsigned char *b, int c)
   mp_clamp (a);
   return MP_OKAY;
 }
+#endif
diff --git a/bn_mp_reduce.c b/bn_mp_reduce.c
index d4c5b1076889812ea2450d87e0f791a8acda305f..cfcb55a7ef3088a67a8f09ef5469b8703d9985da 100644
--- a/bn_mp_reduce.c
+++ b/bn_mp_reduce.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_REDUCE_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* reduces x mod m, assumes 0 < x < m**2, mu is 
  * precomputed via mp_reduce_setup.
@@ -38,9 +39,20 @@ mp_reduce (mp_int * x, mp_int * m, mp_int * mu)
       goto CLEANUP;
     }
   } else {
+#ifdef BN_S_MP_MUL_HIGH_DIGS_C
     if ((res = s_mp_mul_high_digs (&q, mu, &q, um - 1)) != MP_OKAY) {
       goto CLEANUP;
     }
+#elif defined(BN_FAST_S_MP_MUL_HIGH_DIGS_C)
+    if ((res = fast_s_mp_mul_high_digs (&q, mu, &q, um - 1)) != MP_OKAY) {
+      goto CLEANUP;
+    }
+#else 
+    { 
+      res = MP_VAL;
+      goto CLEANUP;
+    }
+#endif
   }
 
   /* q3 = q2 / b**(k+1) */
@@ -82,3 +94,4 @@ CLEANUP:
 
   return res;
 }
+#endif
diff --git a/bn_mp_reduce_2k.c b/bn_mp_reduce_2k.c
index f6b40cf1808fc93dce243a320357a6002197454a..a5a9c7436f506326f920a3033bfb39d44d4e2264 100644
--- a/bn_mp_reduce_2k.c
+++ b/bn_mp_reduce_2k.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_REDUCE_2K_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* reduces a modulo n where n is of the form 2**p - d */
 int
@@ -54,3 +55,4 @@ ERR:
    return res;
 }
 
+#endif
diff --git a/bn_mp_reduce_2k_setup.c b/bn_mp_reduce_2k_setup.c
index 6bdfe7f26a9587ba0f32596f2d9b1535891f3f53..5e1fb6e845ee9c124cf5dd403dfcb7104b120d1f 100644
--- a/bn_mp_reduce_2k_setup.c
+++ b/bn_mp_reduce_2k_setup.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_REDUCE_2K_SETUP_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* determines the setup value */
 int 
@@ -40,3 +41,4 @@ mp_reduce_2k_setup(mp_int *a, mp_digit *d)
    mp_clear(&tmp);
    return MP_OKAY;
 }
+#endif
diff --git a/bn_mp_reduce_is_2k.c b/bn_mp_reduce_is_2k.c
index d43b9ffb2c20a9cee31ecb628c62e6ddf70de094..fc81397be11b5c0c4c44475672a57b2de3cf8a00 100644
--- a/bn_mp_reduce_is_2k.c
+++ b/bn_mp_reduce_is_2k.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_REDUCE_IS_2K_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,12 +14,12 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* determines if mp_reduce_2k can be used */
 int mp_reduce_is_2k(mp_int *a)
 {
-   int ix, iy, iz, iw;
+   int ix, iy, iw;
+   mp_digit iz;
    
    if (a->used == 0) {
       return 0;
@@ -34,7 +36,7 @@ int mp_reduce_is_2k(mp_int *a)
              return 0;
           }
           iz <<= 1;
-          if (iz > (int)MP_MASK) {
+          if (iz > (mp_digit)MP_MASK) {
              ++iw;
              iz = 1;
           }
@@ -43,3 +45,4 @@ int mp_reduce_is_2k(mp_int *a)
    return 1;
 }
 
+#endif
diff --git a/bn_mp_reduce_setup.c b/bn_mp_reduce_setup.c
index 1ba69e9648bfca7933b38773db74fb5e108b956a..99f158a57ce7f08321910295026636fbc0b8fbc2 100644
--- a/bn_mp_reduce_setup.c
+++ b/bn_mp_reduce_setup.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_REDUCE_SETUP_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,13 +14,11 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* pre-calculate the value required for Barrett reduction
  * For a given modulus "b" it calulates the value required in "a"
  */
-int
-mp_reduce_setup (mp_int * a, mp_int * b)
+int mp_reduce_setup (mp_int * a, mp_int * b)
 {
   int     res;
   
@@ -27,3 +27,4 @@ mp_reduce_setup (mp_int * a, mp_int * b)
   }
   return mp_div (a, b, a, NULL);
 }
+#endif
diff --git a/bn_mp_rshd.c b/bn_mp_rshd.c
index 87d6e3b0012b0824f41910bf620422af611b3052..913dda6ded301cebf6650e810be8995667b331f7 100644
--- a/bn_mp_rshd.c
+++ b/bn_mp_rshd.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_RSHD_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* shift right a certain amount of digits */
 void mp_rshd (mp_int * a, int b)
@@ -64,3 +65,4 @@ void mp_rshd (mp_int * a, int b)
   /* remove excess digits */
   a->used -= b;
 }
+#endif
diff --git a/bn_mp_set.c b/bn_mp_set.c
index 25ccf64e9cdd884661c0719deda7c5d00267af36..078fd5fbbe8cd3cf9daf2c3e06da0322efaa2b17 100644
--- a/bn_mp_set.c
+++ b/bn_mp_set.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_SET_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* set to a digit */
 void mp_set (mp_int * a, mp_digit b)
@@ -21,3 +22,4 @@ void mp_set (mp_int * a, mp_digit b)
   a->dp[0] = b & MP_MASK;
   a->used  = (a->dp[0] != 0) ? 1 : 0;
 }
+#endif
diff --git a/bn_mp_set_int.c b/bn_mp_set_int.c
index 64f1aed5bfbb86d3aaabfeaf7210c060c12c6ac4..bd47136c34be9681c67ed0ae935e46f4d8ad21a9 100644
--- a/bn_mp_set_int.c
+++ b/bn_mp_set_int.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_SET_INT_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* set a 32-bit const */
 int mp_set_int (mp_int * a, unsigned long b)
@@ -40,3 +41,4 @@ int mp_set_int (mp_int * a, unsigned long b)
   mp_clamp (a);
   return MP_OKAY;
 }
+#endif
diff --git a/bn_mp_shrink.c b/bn_mp_shrink.c
index daefed2b4432a4d104084d499b7b37ba2f1b543f..b31f9d22950f02103cd1508e8d7ac55212bdb03c 100644
--- a/bn_mp_shrink.c
+++ b/bn_mp_shrink.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_SHRINK_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* shrink a bignum */
 int mp_shrink (mp_int * a)
@@ -27,3 +28,4 @@ int mp_shrink (mp_int * a)
   }
   return MP_OKAY;
 }
+#endif
diff --git a/bn_mp_signed_bin_size.c b/bn_mp_signed_bin_size.c
index 76d085cb441b8abb15c16bb160797aa82c7f048b..30048cb8496885f849a5bfc3f590d0a1b25ad817 100644
--- a/bn_mp_signed_bin_size.c
+++ b/bn_mp_signed_bin_size.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_SIGNED_BIN_SIZE_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,10 +14,10 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* get the size for an signed equivalent */
 int mp_signed_bin_size (mp_int * a)
 {
   return 1 + mp_unsigned_bin_size (a);
 }
+#endif
diff --git a/bn_mp_sqr.c b/bn_mp_sqr.c
index 4857994d08d347fd4fec4ed96f0d550c7484c185..b1fdb57c93a86ce5c2d5c19608e856d34f4d4d8f 100644
--- a/bn_mp_sqr.c
+++ b/bn_mp_sqr.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_SQR_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* computes b = a*a */
 int
@@ -20,22 +21,34 @@ mp_sqr (mp_int * a, mp_int * b)
 {
   int     res;
 
+#ifdef BN_MP_TOOM_SQR_C
   /* use Toom-Cook? */
   if (a->used >= TOOM_SQR_CUTOFF) {
     res = mp_toom_sqr(a, b);
   /* Karatsuba? */
-  } else if (a->used >= KARATSUBA_SQR_CUTOFF) {
+  } else 
+#endif
+#ifdef BN_MP_KARATSUBA_SQR_C
+if (a->used >= KARATSUBA_SQR_CUTOFF) {
     res = mp_karatsuba_sqr (a, b);
-  } else {
+  } else 
+#endif
+  {
+#ifdef BN_FAST_S_MP_SQR_C
     /* can we use the fast comba multiplier? */
     if ((a->used * 2 + 1) < MP_WARRAY && 
          a->used < 
          (1 << (sizeof(mp_word) * CHAR_BIT - 2*DIGIT_BIT - 1))) {
       res = fast_s_mp_sqr (a, b);
-    } else {
+    } else
+#endif
+#ifdef BN_S_MP_SQR_C
       res = s_mp_sqr (a, b);
-    }
+#else
+      res = MP_VAL;
+#endif
   }
   b->sign = MP_ZPOS;
   return res;
 }
+#endif
diff --git a/bn_mp_sqrmod.c b/bn_mp_sqrmod.c
index 778b68e84abcc7cb903dd5279fd4d43703c53d6e..1923be43cd7e787c704ad9c962effdb232d61ad4 100644
--- a/bn_mp_sqrmod.c
+++ b/bn_mp_sqrmod.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_SQRMOD_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* c = a * a (mod b) */
 int
@@ -33,3 +34,4 @@ mp_sqrmod (mp_int * a, mp_int * b, mp_int * c)
   mp_clear (&t);
   return res;
 }
+#endif
diff --git a/bn_mp_sqrt.c b/bn_mp_sqrt.c
index ec9d1026cff6bf60660759ef3d9b6f3dcbe76d90..76cec878edaaadc16274438beaefa23610c4d120 100644
--- a/bn_mp_sqrt.c
+++ b/bn_mp_sqrt.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_SQRT_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* this function is less generic than mp_n_root, simpler and faster */
 int mp_sqrt(mp_int *arg, mp_int *ret) 
@@ -73,3 +74,4 @@ E2: mp_clear(&t1);
   return res;
 }
 
+#endif
diff --git a/bn_mp_sub.c b/bn_mp_sub.c
index eb99583f7dd118db896d8571d10ac63117a6434a..97495f42f585a21ec8bd40dbb792238cb1b779b8 100644
--- a/bn_mp_sub.c
+++ b/bn_mp_sub.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_SUB_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* high level subtraction (handles signs) */
 int
@@ -51,3 +52,4 @@ mp_sub (mp_int * a, mp_int * b, mp_int * c)
   return res;
 }
 
+#endif
diff --git a/bn_mp_sub_d.c b/bn_mp_sub_d.c
index 6368970f67d010f90e595d4a9978d37b9d5300a1..4923ddeb96c6154fe472fbbc850c4cffb27d07c2 100644
--- a/bn_mp_sub_d.c
+++ b/bn_mp_sub_d.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_SUB_D_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* single digit subtraction */
 int
@@ -81,3 +82,4 @@ mp_sub_d (mp_int * a, mp_digit b, mp_int * c)
   return MP_OKAY;
 }
 
+#endif
diff --git a/bn_mp_submod.c b/bn_mp_submod.c
index 868dd2f44ca466565b14a8072871dbc0c690f917..b999c85459d813e35b14dae38a45321129344040 100644
--- a/bn_mp_submod.c
+++ b/bn_mp_submod.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_SUBMOD_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* d = a - b (mod c) */
 int
@@ -34,3 +35,4 @@ mp_submod (mp_int * a, mp_int * b, mp_int * c, mp_int * d)
   mp_clear (&t);
   return res;
 }
+#endif
diff --git a/bn_mp_to_signed_bin.c b/bn_mp_to_signed_bin.c
index 1e21667e28f3537f229b9a6d18cfa76b21dd03be..0e40d0fa34b801f13126aad42c7a3f1acf6305e1 100644
--- a/bn_mp_to_signed_bin.c
+++ b/bn_mp_to_signed_bin.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_TO_SIGNED_BIN_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* store in signed [big endian] format */
 int
@@ -26,3 +27,4 @@ mp_to_signed_bin (mp_int * a, unsigned char *b)
   b[0] = (unsigned char) ((a->sign == MP_ZPOS) ? 0 : 1);
   return MP_OKAY;
 }
+#endif
diff --git a/bn_mp_to_unsigned_bin.c b/bn_mp_to_unsigned_bin.c
index f7cae6910de7c5c6f6d46c1128e41f8bc7b0b75b..763e34615dad467a33fc594c835da06dc0736675 100644
--- a/bn_mp_to_unsigned_bin.c
+++ b/bn_mp_to_unsigned_bin.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_TO_UNSIGNED_BIN_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* store in unsigned [big endian] format */
 int
@@ -41,3 +42,4 @@ mp_to_unsigned_bin (mp_int * a, unsigned char *b)
   mp_clear (&t);
   return MP_OKAY;
 }
+#endif
diff --git a/bn_mp_toom_mul.c b/bn_mp_toom_mul.c
index 50660ffda80001f8420982da469cc2a840f7c8d6..2d66779c9702144f80658e84d0e5cdf441feb281 100644
--- a/bn_mp_toom_mul.c
+++ b/bn_mp_toom_mul.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_TOOM_MUL_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,9 +14,13 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
-/* multiplication using the Toom-Cook 3-way algorithm */
+/* multiplication using the Toom-Cook 3-way algorithm 
+ *
+ * Much more complicated than Karatsuba but has a lower asymptotic running time of 
+ * O(N**1.464).  This algorithm is only particularly useful on VERY large
+ * inputs (we're talking 1000s of digits here...).
+*/
 int mp_toom_mul(mp_int *a, mp_int *b, mp_int *c)
 {
     mp_int w0, w1, w2, w3, w4, tmp1, tmp2, a0, a1, a2, b0, b1, b2;
@@ -270,3 +276,4 @@ ERR:
      return res;
 }     
      
+#endif
diff --git a/bn_mp_toom_sqr.c b/bn_mp_toom_sqr.c
index de3094aa898b55ead3a425c1298bab750674ec3e..8c46fead8db6f7a16049515671a047629bd28652 100644
--- a/bn_mp_toom_sqr.c
+++ b/bn_mp_toom_sqr.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_TOOM_SQR_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* squaring using Toom-Cook 3-way algorithm */
 int
@@ -218,3 +219,4 @@ ERR:
      return res;
 }
 
+#endif
diff --git a/bn_mp_toradix.c b/bn_mp_toradix.c
index c605867eacbe79831ce371a37dcba1d54aebd8ef..a206d5eee00a36bf04f0a367893c646aaf7a4369 100644
--- a/bn_mp_toradix.c
+++ b/bn_mp_toradix.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_TORADIX_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* stores a bignum as a ASCII string in a given radix (2..64) */
 int mp_toradix (mp_int * a, char *str, int radix)
@@ -67,3 +68,4 @@ int mp_toradix (mp_int * a, char *str, int radix)
   return MP_OKAY;
 }
 
+#endif
diff --git a/bn_mp_toradix_n.c b/bn_mp_toradix_n.c
index d2f6ec20c75b55210cc782b7276c315f24ab1a79..7d435587d0f8caf2e52ffd08cbf736ef1a8e2af3 100644
--- a/bn_mp_toradix_n.c
+++ b/bn_mp_toradix_n.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_TORADIX_N_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* stores a bignum as a ASCII string in a given radix (2..64) 
  *
@@ -81,3 +82,4 @@ int mp_toradix_n(mp_int * a, char *str, int radix, int maxlen)
   return MP_OKAY;
 }
 
+#endif
diff --git a/bn_mp_unsigned_bin_size.c b/bn_mp_unsigned_bin_size.c
index 47bbc648892d5172649b9de473be7fc9b1406ce1..80da415a8dbf7b2b7900a4e60b28a1a4052b0f96 100644
--- a/bn_mp_unsigned_bin_size.c
+++ b/bn_mp_unsigned_bin_size.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_UNSIGNED_BIN_SIZE_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* get the size for an unsigned equivalent */
 int
@@ -21,3 +22,4 @@ mp_unsigned_bin_size (mp_int * a)
   int     size = mp_count_bits (a);
   return (size / 8 + ((size & 7) != 0 ? 1 : 0));
 }
+#endif
diff --git a/bn_mp_xor.c b/bn_mp_xor.c
index 36e6733e67cc0b7eb83cb2452ac831378816549d..192aacc4e8bf2ae846f1b648adf6ee22e50d468f 100644
--- a/bn_mp_xor.c
+++ b/bn_mp_xor.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_XOR_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* XOR two ints together */
 int
@@ -36,10 +37,11 @@ mp_xor (mp_int * a, mp_int * b, mp_int * c)
   }
 
   for (ix = 0; ix < px; ix++) {
-    t.dp[ix] ^= x->dp[ix];
+
   }
   mp_clamp (&t);
   mp_exch (c, &t);
   mp_clear (&t);
   return MP_OKAY;
 }
+#endif
diff --git a/bn_mp_zero.c b/bn_mp_zero.c
index a888589f1bd3dcfc9cbe57d9314668a61c093e00..0097598caceff783121434b8b61df12f6985b45f 100644
--- a/bn_mp_zero.c
+++ b/bn_mp_zero.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_MP_ZERO_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* set to zero */
 void
@@ -22,3 +23,4 @@ mp_zero (mp_int * a)
   a->used = 0;
   memset (a->dp, 0, sizeof (mp_digit) * a->alloc);
 }
+#endif
diff --git a/bn_prime_tab.c b/bn_prime_tab.c
index 7dfdc8ce3f52f50e3b3dcf6d9df828158586e5d6..18ecc47aeebefcc7742abfd61ba32628cb580e94 100644
--- a/bn_prime_tab.c
+++ b/bn_prime_tab.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_PRIME_TAB_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 const mp_digit __prime_tab[] = {
   0x0002, 0x0003, 0x0005, 0x0007, 0x000B, 0x000D, 0x0011, 0x0013,
   0x0017, 0x001D, 0x001F, 0x0025, 0x0029, 0x002B, 0x002F, 0x0035,
@@ -53,3 +54,4 @@ const mp_digit __prime_tab[] = {
   0x062B, 0x062F, 0x063D, 0x0641, 0x0647, 0x0649, 0x064D, 0x0653
 #endif
 };
+#endif
diff --git a/bn_reverse.c b/bn_reverse.c
index 89cbbb7c555800b1e69f9a12fad84786920eea61..851a6e8b6b7c7ed226c867f2b115ae4fad33a7f0 100644
--- a/bn_reverse.c
+++ b/bn_reverse.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_REVERSE_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* reverse an array, used for radix code */
 void
@@ -31,3 +32,4 @@ bn_reverse (unsigned char *s, int len)
     --iy;
   }
 }
+#endif
diff --git a/bn_s_mp_add.c b/bn_s_mp_add.c
index e07931ab6d601fe8e33fe0cdabeec39bb2336a83..2b378ae641c576de2176a6c2cdd60e0e153d139a 100644
--- a/bn_s_mp_add.c
+++ b/bn_s_mp_add.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_S_MP_ADD_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* low level addition, based on HAC pp.594, Algorithm 14.7 */
 int
@@ -101,3 +102,4 @@ s_mp_add (mp_int * a, mp_int * b, mp_int * c)
   mp_clamp (c);
   return MP_OKAY;
 }
+#endif
diff --git a/bn_s_mp_exptmod.c b/bn_s_mp_exptmod.c
index 28c14fd9f0222e3f8565143f181dac20e0bb1318..4f1032a94b6dc32d41f0b90717641dba8aeb5f08 100644
--- a/bn_s_mp_exptmod.c
+++ b/bn_s_mp_exptmod.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_S_MP_EXPTMOD_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 #ifdef MP_LOW_MEM
    #define TAB_SIZE 32
@@ -232,3 +233,4 @@ __M:
   }
   return err;
 }
+#endif
diff --git a/bn_s_mp_mul_digs.c b/bn_s_mp_mul_digs.c
index a5b10675df9f422ad3e7f032326f685cb23eb4a0..d9f0a56eb15214ae8729af8ac404b04cf3d1e31e 100644
--- a/bn_s_mp_mul_digs.c
+++ b/bn_s_mp_mul_digs.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_S_MP_MUL_DIGS_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* multiplies |a| * |b| and only computes upto digs digits of result
  * HAC pp. 595, Algorithm 14.12  Modified so you can control how 
@@ -83,3 +84,4 @@ s_mp_mul_digs (mp_int * a, mp_int * b, mp_int * c, int digs)
   mp_clear (&t);
   return MP_OKAY;
 }
+#endif
diff --git a/bn_s_mp_mul_high_digs.c b/bn_s_mp_mul_high_digs.c
index 84cc7d0a132e1c0eb651a1f38821946200a1b558..a060248bf77addebdcb8c3f05e217967cf5121ef 100644
--- a/bn_s_mp_mul_high_digs.c
+++ b/bn_s_mp_mul_high_digs.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_S_MP_MUL_HIGH_DIGS_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* multiplies |a| * |b| and does not compute the lower digs digits
  * [meant to get the higher part of the product]
@@ -27,10 +28,12 @@ s_mp_mul_high_digs (mp_int * a, mp_int * b, mp_int * c, int digs)
   mp_digit tmpx, *tmpt, *tmpy;
 
   /* can we use the fast multiplier? */
+#ifdef BN_FAST_S_MP_MUL_HIGH_DIGS_C
   if (((a->used + b->used + 1) < MP_WARRAY)
       && MIN (a->used, b->used) < (1 << ((CHAR_BIT * sizeof (mp_word)) - (2 * DIGIT_BIT)))) {
     return fast_s_mp_mul_high_digs (a, b, c, digs);
   }
+#endif
 
   if ((res = mp_init_size (&t, a->used + b->used + 1)) != MP_OKAY) {
     return res;
@@ -71,3 +74,4 @@ s_mp_mul_high_digs (mp_int * a, mp_int * b, mp_int * c, int digs)
   mp_clear (&t);
   return MP_OKAY;
 }
+#endif
diff --git a/bn_s_mp_sqr.c b/bn_s_mp_sqr.c
index 3a00a4ed6f52beb667e83d6fff63076813ae2ec2..4d128047cd0eebc6e4fe64472c8adcd8c9b0428c 100644
--- a/bn_s_mp_sqr.c
+++ b/bn_s_mp_sqr.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_S_MP_SQR_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* low level squaring, b = a*a, HAC pp.596-597, Algorithm 14.16 */
 int
@@ -77,3 +78,4 @@ s_mp_sqr (mp_int * a, mp_int * b)
   mp_clear (&t);
   return MP_OKAY;
 }
+#endif
diff --git a/bn_s_mp_sub.c b/bn_s_mp_sub.c
index 4381af892e1551cac9c69f97641faf590ee4c9a2..5b7aef9d69cb4e43885b38cd733dde8c583d4210 100644
--- a/bn_s_mp_sub.c
+++ b/bn_s_mp_sub.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BN_S_MP_SUB_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,7 +14,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* low level subtraction (assumes |a| > |b|), HAC pp.595 Algorithm 14.9 */
 int
@@ -81,3 +82,4 @@ s_mp_sub (mp_int * a, mp_int * b, mp_int * c)
   return MP_OKAY;
 }
 
+#endif
diff --git a/bncore.c b/bncore.c
index acb78a0ab02ff2a6b1c4e2fedd51067a17288c05..cf8a15a136cddd90ecaee7c796434c5920d41173 100644
--- a/bncore.c
+++ b/bncore.c
@@ -1,3 +1,5 @@
+#include <tommath.h>
+#ifdef BNCORE_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -12,20 +14,18 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* Known optimal configurations
 
  CPU                    /Compiler     /MUL CUTOFF/SQR CUTOFF
 -------------------------------------------------------------
- Intel P4               /GCC v3.2     /        70/       108
- AMD Athlon XP          /GCC v3.2     /       109/       127
-
+ Intel P4 Northwood     /GCC v3.4.1   /        88/       128/LTM 0.32 ;-)
+ 
 */
 
-/* configured for a AMD XP Thoroughbred core with etc/tune.c */
-int     KARATSUBA_MUL_CUTOFF = 109,      /* Min. number of digits before Karatsuba multiplication is used. */
-        KARATSUBA_SQR_CUTOFF = 127,      /* Min. number of digits before Karatsuba squaring is used. */
+int     KARATSUBA_MUL_CUTOFF = 88,      /* Min. number of digits before Karatsuba multiplication is used. */
+        KARATSUBA_SQR_CUTOFF = 128,     /* Min. number of digits before Karatsuba squaring is used. */
         
         TOOM_MUL_CUTOFF      = 350,      /* no optimal values of these are known yet so set em high */
         TOOM_SQR_CUTOFF      = 400; 
+#endif
diff --git a/booker.pl b/booker.pl
index 4f4231f216cccf0632dc0f1b7f5faa6669efc90d..5c77e538613ac3494e7738b434b85eacea5d68bc 100644
--- a/booker.pl
+++ b/booker.pl
@@ -82,8 +82,9 @@ while (<IN>) {
          # scan till next end of comment, e.g. skip license 
          while (<SRC>) {
             $text[$line++] = $_;
-            last if ($_ =~ /tommath\.h/);
+            last if ($_ =~ /math\.libtomcrypt\.org/);
          }
+         <SRC>;   
       }
       
       $inline = 0;
diff --git a/callgraph.txt b/callgraph.txt
new file mode 100644
index 0000000000000000000000000000000000000000..56d4f8bbb6d9d399092dbc14b6ef71cf146f36bc
--- /dev/null
+++ b/callgraph.txt
@@ -0,0 +1,10168 @@
+BN_PRIME_TAB_C
+
+
+BN_MP_SQRT_C
++--->BN_MP_N_ROOT_C
+|   +--->BN_MP_INIT_C
+|   +--->BN_MP_SET_C
+|   |   +--->BN_MP_ZERO_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_EXPT_D_C
+|   |   +--->BN_MP_INIT_COPY_C
+|   |   +--->BN_MP_SQR_C
+|   |   |   +--->BN_MP_TOOM_SQR_C
+|   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_MP_KARATSUBA_SQR_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_FAST_S_MP_SQR_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_SQR_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_MUL_C
+|   |   |   +--->BN_MP_TOOM_MUL_C
+|   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   +--->BN_MP_KARATSUBA_MUL_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_MUL_DIGS_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   +--->BN_MP_MUL_C
+|   |   +--->BN_MP_TOOM_MUL_C
+|   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_SUB_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_KARATSUBA_MUL_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_SUB_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_S_MP_MUL_DIGS_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_SUB_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_MUL_D_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_DIV_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_COUNT_BITS_C
+|   |   +--->BN_MP_MUL_2D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_2D_C
+|   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_INIT_COPY_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_CMP_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   +--->BN_MP_SUB_D_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ADD_D_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_EXCH_C
+|   +--->BN_MP_CLEAR_C
++--->BN_MP_ZERO_C
++--->BN_MP_INIT_COPY_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
++--->BN_MP_RSHD_C
++--->BN_MP_DIV_C
+|   +--->BN_MP_CMP_MAG_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_INIT_MULTI_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_SET_C
+|   +--->BN_MP_COUNT_BITS_C
+|   +--->BN_MP_MUL_2D_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_LSHD_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CMP_C
+|   +--->BN_MP_SUB_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_ADD_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_DIV_2D_C
+|   |   +--->BN_MP_MOD_2D_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   +--->BN_MP_EXCH_C
+|   +--->BN_MP_CLEAR_MULTI_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_INIT_SIZE_C
+|   +--->BN_MP_LSHD_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_MUL_D_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CLEAR_C
++--->BN_MP_ADD_C
+|   +--->BN_S_MP_ADD_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CMP_MAG_C
+|   +--->BN_S_MP_SUB_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
++--->BN_MP_DIV_2_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_CMP_MAG_C
++--->BN_MP_EXCH_C
++--->BN_MP_CLEAR_C
+
+
+BN_MP_CMP_D_C
+
+
+BN_MP_EXCH_C
+
+
+BN_MP_IS_SQUARE_C
++--->BN_MP_MOD_D_C
+|   +--->BN_MP_DIV_D_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_DIV_2D_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_INIT_C
+|   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_DIV_3_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_INIT_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_INIT_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_C
++--->BN_MP_INIT_SET_INT_C
+|   +--->BN_MP_INIT_C
+|   +--->BN_MP_SET_INT_C
+|   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_MUL_2D_C
+|   |   |   +--->BN_MP_COPY_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CLAMP_C
++--->BN_MP_MOD_C
+|   +--->BN_MP_INIT_C
+|   +--->BN_MP_DIV_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_SET_C
+|   |   +--->BN_MP_COUNT_BITS_C
+|   |   +--->BN_MP_MUL_2D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_2D_C
+|   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_INIT_COPY_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_MUL_D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_ADD_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_EXCH_C
++--->BN_MP_GET_INT_C
++--->BN_MP_SQRT_C
+|   +--->BN_MP_N_ROOT_C
+|   |   +--->BN_MP_INIT_C
+|   |   +--->BN_MP_SET_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_EXPT_D_C
+|   |   |   +--->BN_MP_INIT_COPY_C
+|   |   |   +--->BN_MP_SQR_C
+|   |   |   |   +--->BN_MP_TOOM_SQR_C
+|   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   |   +--->BN_MP_KARATSUBA_SQR_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   |   +--->BN_FAST_S_MP_SQR_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_S_MP_SQR_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_MP_MUL_C
+|   |   |   |   +--->BN_MP_TOOM_MUL_C
+|   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   |   +--->BN_MP_KARATSUBA_MUL_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_S_MP_MUL_DIGS_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_MUL_C
+|   |   |   +--->BN_MP_TOOM_MUL_C
+|   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_MP_KARATSUBA_MUL_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_MUL_DIGS_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MUL_D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_MP_COUNT_BITS_C
+|   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_DIV_2D_C
+|   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_INIT_COPY_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_CMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_MP_SUB_D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_ADD_D_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_ZERO_C
+|   +--->BN_MP_INIT_COPY_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_RSHD_C
+|   +--->BN_MP_DIV_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_SET_C
+|   |   +--->BN_MP_COUNT_BITS_C
+|   |   +--->BN_MP_MUL_2D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_2D_C
+|   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_MUL_D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_ADD_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_DIV_2_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CMP_MAG_C
+|   +--->BN_MP_EXCH_C
+|   +--->BN_MP_CLEAR_C
++--->BN_MP_SQR_C
+|   +--->BN_MP_TOOM_SQR_C
+|   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   +--->BN_MP_INIT_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_MOD_2D_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_COPY_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_MUL_2_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_2_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MUL_2D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MUL_D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_3_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_INIT_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_KARATSUBA_SQR_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_INIT_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_FAST_S_MP_SQR_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_S_MP_SQR_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_INIT_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_C
++--->BN_MP_CMP_MAG_C
++--->BN_MP_CLEAR_C
+
+
+BN_MP_NEG_C
++--->BN_MP_COPY_C
+|   +--->BN_MP_GROW_C
+
+
+BN_MP_EXPTMOD_C
++--->BN_MP_INIT_C
++--->BN_MP_INVMOD_C
+|   +--->BN_FAST_MP_INVMOD_C
+|   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ABS_C
+|   |   +--->BN_MP_SET_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_DIV_2_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_MP_CMP_D_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_INVMOD_SLOW_C
+|   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_SET_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_DIV_2_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_MP_CMP_D_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   +--->BN_MP_CLEAR_C
++--->BN_MP_CLEAR_C
++--->BN_MP_ABS_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
++--->BN_MP_CLEAR_MULTI_C
++--->BN_MP_DR_IS_MODULUS_C
++--->BN_MP_REDUCE_IS_2K_C
+|   +--->BN_MP_REDUCE_2K_C
+|   |   +--->BN_MP_COUNT_BITS_C
+|   |   +--->BN_MP_DIV_2D_C
+|   |   |   +--->BN_MP_COPY_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_MUL_D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_COUNT_BITS_C
++--->BN_MP_EXPTMOD_FAST_C
+|   +--->BN_MP_COUNT_BITS_C
+|   +--->BN_MP_MONTGOMERY_SETUP_C
+|   +--->BN_FAST_MP_MONTGOMERY_REDUCE_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_S_MP_SUB_C
+|   +--->BN_MP_MONTGOMERY_REDUCE_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_S_MP_SUB_C
+|   +--->BN_MP_DR_SETUP_C
+|   +--->BN_MP_DR_REDUCE_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_S_MP_SUB_C
+|   +--->BN_MP_REDUCE_2K_SETUP_C
+|   |   +--->BN_MP_2EXPT_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_REDUCE_2K_C
+|   |   +--->BN_MP_DIV_2D_C
+|   |   |   +--->BN_MP_COPY_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_MUL_D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_MONTGOMERY_CALC_NORMALIZATION_C
+|   |   +--->BN_MP_2EXPT_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_SET_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_MUL_2_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_MULMOD_C
+|   |   +--->BN_MP_MUL_C
+|   |   |   +--->BN_MP_TOOM_MUL_C
+|   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_KARATSUBA_MUL_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_MUL_DIGS_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_MOD_C
+|   |   |   +--->BN_MP_DIV_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   +--->BN_MP_SET_C
+|   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_DIV_2D_C
+|   |   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_INIT_COPY_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   +--->BN_MP_SET_C
+|   |   +--->BN_MP_ZERO_C
+|   +--->BN_MP_MOD_C
+|   |   +--->BN_MP_DIV_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_MP_COPY_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_C
+|   |   |   +--->BN_MP_SUB_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_DIV_2D_C
+|   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_INIT_COPY_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_SQR_C
+|   |   +--->BN_MP_TOOM_SQR_C
+|   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_SUB_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_KARATSUBA_SQR_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_SUB_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_FAST_S_MP_SQR_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_S_MP_SQR_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   +--->BN_MP_MUL_C
+|   |   +--->BN_MP_TOOM_MUL_C
+|   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_SUB_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_KARATSUBA_MUL_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_SUB_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_S_MP_MUL_DIGS_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   +--->BN_MP_EXCH_C
++--->BN_S_MP_EXPTMOD_C
+|   +--->BN_MP_COUNT_BITS_C
+|   +--->BN_MP_REDUCE_SETUP_C
+|   |   +--->BN_MP_2EXPT_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_DIV_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_MP_COPY_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   +--->BN_MP_SET_C
+|   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_C
+|   |   |   +--->BN_MP_SUB_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_DIV_2D_C
+|   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_INIT_COPY_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_MOD_C
+|   |   +--->BN_MP_DIV_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_MP_COPY_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   +--->BN_MP_SET_C
+|   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_C
+|   |   |   +--->BN_MP_SUB_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_DIV_2D_C
+|   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_INIT_COPY_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_SQR_C
+|   |   +--->BN_MP_TOOM_SQR_C
+|   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_SUB_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_KARATSUBA_SQR_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_SUB_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_FAST_S_MP_SQR_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_S_MP_SQR_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   +--->BN_MP_REDUCE_C
+|   |   +--->BN_MP_INIT_COPY_C
+|   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_MUL_C
+|   |   |   +--->BN_MP_TOOM_MUL_C
+|   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_KARATSUBA_MUL_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_MUL_DIGS_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_S_MP_MUL_HIGH_DIGS_C
+|   |   |   +--->BN_FAST_S_MP_MUL_HIGH_DIGS_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_FAST_S_MP_MUL_HIGH_DIGS_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MOD_2D_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_S_MP_MUL_DIGS_C
+|   |   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_D_C
+|   |   +--->BN_MP_SET_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_MUL_C
+|   |   +--->BN_MP_TOOM_MUL_C
+|   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_SUB_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_KARATSUBA_MUL_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_SUB_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_S_MP_MUL_DIGS_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   +--->BN_MP_SET_C
+|   |   +--->BN_MP_ZERO_C
+|   +--->BN_MP_EXCH_C
+
+
+BN_MP_OR_C
++--->BN_MP_INIT_COPY_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
++--->BN_MP_CLAMP_C
++--->BN_MP_EXCH_C
++--->BN_MP_CLEAR_C
+
+
+BN_MP_ZERO_C
+
+
+BN_MP_GROW_C
+
+
+BN_MP_COUNT_BITS_C
+
+
+BN_MP_PRIME_FERMAT_C
++--->BN_MP_CMP_D_C
++--->BN_MP_INIT_C
++--->BN_MP_EXPTMOD_C
+|   +--->BN_MP_INVMOD_C
+|   |   +--->BN_FAST_MP_INVMOD_C
+|   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_MP_COPY_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_ABS_C
+|   |   |   +--->BN_MP_SET_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_SUB_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_INVMOD_SLOW_C
+|   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_MP_COPY_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_SET_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_SUB_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_ABS_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLEAR_MULTI_C
+|   +--->BN_MP_DR_IS_MODULUS_C
+|   +--->BN_MP_REDUCE_IS_2K_C
+|   |   +--->BN_MP_REDUCE_2K_C
+|   |   |   +--->BN_MP_COUNT_BITS_C
+|   |   |   +--->BN_MP_DIV_2D_C
+|   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_COUNT_BITS_C
+|   +--->BN_MP_EXPTMOD_FAST_C
+|   |   +--->BN_MP_COUNT_BITS_C
+|   |   +--->BN_MP_MONTGOMERY_SETUP_C
+|   |   +--->BN_FAST_MP_MONTGOMERY_REDUCE_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   +--->BN_MP_MONTGOMERY_REDUCE_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   +--->BN_MP_DR_SETUP_C
+|   |   +--->BN_MP_DR_REDUCE_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   +--->BN_MP_REDUCE_2K_SETUP_C
+|   |   |   +--->BN_MP_2EXPT_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_REDUCE_2K_C
+|   |   |   +--->BN_MP_DIV_2D_C
+|   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MONTGOMERY_CALC_NORMALIZATION_C
+|   |   |   +--->BN_MP_2EXPT_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_SET_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MULMOD_C
+|   |   |   +--->BN_MP_MUL_C
+|   |   |   |   +--->BN_MP_TOOM_MUL_C
+|   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_KARATSUBA_MUL_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_S_MP_MUL_DIGS_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_MOD_C
+|   |   |   |   +--->BN_MP_DIV_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   +--->BN_MP_SET_C
+|   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_DIV_2D_C
+|   |   |   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_INIT_COPY_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_SET_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_MOD_C
+|   |   |   +--->BN_MP_DIV_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_DIV_2D_C
+|   |   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_INIT_COPY_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_SQR_C
+|   |   |   +--->BN_MP_TOOM_SQR_C
+|   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_KARATSUBA_SQR_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_FAST_S_MP_SQR_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_SQR_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_MUL_C
+|   |   |   +--->BN_MP_TOOM_MUL_C
+|   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_KARATSUBA_MUL_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_MUL_DIGS_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_EXCH_C
+|   +--->BN_S_MP_EXPTMOD_C
+|   |   +--->BN_MP_COUNT_BITS_C
+|   |   +--->BN_MP_REDUCE_SETUP_C
+|   |   |   +--->BN_MP_2EXPT_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_DIV_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   +--->BN_MP_SET_C
+|   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_DIV_2D_C
+|   |   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_INIT_COPY_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MOD_C
+|   |   |   +--->BN_MP_DIV_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   +--->BN_MP_SET_C
+|   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_DIV_2D_C
+|   |   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_INIT_COPY_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_SQR_C
+|   |   |   +--->BN_MP_TOOM_SQR_C
+|   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_KARATSUBA_SQR_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_FAST_S_MP_SQR_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_SQR_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_REDUCE_C
+|   |   |   +--->BN_MP_INIT_COPY_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_MUL_C
+|   |   |   |   +--->BN_MP_TOOM_MUL_C
+|   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_KARATSUBA_MUL_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_S_MP_MUL_DIGS_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_S_MP_MUL_HIGH_DIGS_C
+|   |   |   |   +--->BN_FAST_S_MP_MUL_HIGH_DIGS_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_FAST_S_MP_MUL_HIGH_DIGS_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_MUL_DIGS_C
+|   |   |   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_SUB_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_SET_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MUL_C
+|   |   |   +--->BN_MP_TOOM_MUL_C
+|   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_KARATSUBA_MUL_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_MUL_DIGS_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_SET_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_EXCH_C
++--->BN_MP_CMP_C
+|   +--->BN_MP_CMP_MAG_C
++--->BN_MP_CLEAR_C
+
+
+BN_MP_SUBMOD_C
++--->BN_MP_INIT_C
++--->BN_MP_SUB_C
+|   +--->BN_S_MP_ADD_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CMP_MAG_C
+|   +--->BN_S_MP_SUB_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
++--->BN_MP_CLEAR_C
++--->BN_MP_MOD_C
+|   +--->BN_MP_DIV_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_INIT_MULTI_C
+|   |   +--->BN_MP_SET_C
+|   |   +--->BN_MP_COUNT_BITS_C
+|   |   +--->BN_MP_MUL_2D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_2D_C
+|   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_INIT_COPY_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_MUL_D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_ADD_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_EXCH_C
+
+
+BN_MP_MOD_2D_C
++--->BN_MP_ZERO_C
++--->BN_MP_COPY_C
+|   +--->BN_MP_GROW_C
++--->BN_MP_CLAMP_C
+
+
+BN_MP_TORADIX_N_C
++--->BN_MP_INIT_COPY_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
++--->BN_MP_DIV_D_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_DIV_2D_C
+|   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_MOD_2D_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   +--->BN_MP_DIV_3_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_INIT_SIZE_C
+|   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_EXCH_C
+|   +--->BN_MP_CLEAR_C
++--->BN_MP_CLEAR_C
+
+
+BN_MP_CMP_C
++--->BN_MP_CMP_MAG_C
+
+
+BNCORE_C
+
+
+BN_MP_TORADIX_C
++--->BN_MP_INIT_COPY_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
++--->BN_MP_DIV_D_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_DIV_2D_C
+|   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_MOD_2D_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   +--->BN_MP_DIV_3_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_INIT_SIZE_C
+|   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_EXCH_C
+|   +--->BN_MP_CLEAR_C
++--->BN_MP_CLEAR_C
+
+
+BN_MP_ADD_D_C
++--->BN_MP_GROW_C
++--->BN_MP_SUB_D_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_CLAMP_C
+
+
+BN_MP_DIV_3_C
++--->BN_MP_INIT_SIZE_C
+|   +--->BN_MP_INIT_C
++--->BN_MP_CLAMP_C
++--->BN_MP_EXCH_C
++--->BN_MP_CLEAR_C
+
+
+BN_FAST_S_MP_MUL_DIGS_C
++--->BN_MP_GROW_C
++--->BN_MP_CLAMP_C
+
+
+BN_MP_SQRMOD_C
++--->BN_MP_INIT_C
++--->BN_MP_SQR_C
+|   +--->BN_MP_TOOM_SQR_C
+|   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_MOD_2D_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_COPY_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_MUL_2_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_2_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MUL_2D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MUL_D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_3_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_KARATSUBA_SQR_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_FAST_S_MP_SQR_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_S_MP_SQR_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_C
++--->BN_MP_CLEAR_C
++--->BN_MP_MOD_C
+|   +--->BN_MP_DIV_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_INIT_MULTI_C
+|   |   +--->BN_MP_SET_C
+|   |   +--->BN_MP_COUNT_BITS_C
+|   |   +--->BN_MP_MUL_2D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_2D_C
+|   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_INIT_COPY_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_MUL_D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_ADD_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_EXCH_C
+
+
+BN_MP_INVMOD_C
++--->BN_FAST_MP_INVMOD_C
+|   +--->BN_MP_INIT_MULTI_C
+|   |   +--->BN_MP_INIT_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_ABS_C
+|   +--->BN_MP_SET_C
+|   |   +--->BN_MP_ZERO_C
+|   +--->BN_MP_DIV_2_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_SUB_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CMP_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   +--->BN_MP_CMP_D_C
+|   +--->BN_MP_ADD_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_EXCH_C
+|   +--->BN_MP_CLEAR_MULTI_C
+|   |   +--->BN_MP_CLEAR_C
++--->BN_MP_INVMOD_SLOW_C
+|   +--->BN_MP_INIT_MULTI_C
+|   |   +--->BN_MP_INIT_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_SET_C
+|   |   +--->BN_MP_ZERO_C
+|   +--->BN_MP_DIV_2_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_ADD_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_SUB_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CMP_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   +--->BN_MP_CMP_D_C
+|   +--->BN_MP_CMP_MAG_C
+|   +--->BN_MP_EXCH_C
+|   +--->BN_MP_CLEAR_MULTI_C
+|   |   +--->BN_MP_CLEAR_C
+
+
+BN_MP_AND_C
++--->BN_MP_INIT_COPY_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
++--->BN_MP_CLAMP_C
++--->BN_MP_EXCH_C
++--->BN_MP_CLEAR_C
+
+
+BN_MP_MUL_D_C
++--->BN_MP_GROW_C
++--->BN_MP_CLAMP_C
+
+
+BN_FAST_MP_INVMOD_C
++--->BN_MP_INIT_MULTI_C
+|   +--->BN_MP_INIT_C
+|   +--->BN_MP_CLEAR_C
++--->BN_MP_COPY_C
+|   +--->BN_MP_GROW_C
++--->BN_MP_ABS_C
++--->BN_MP_SET_C
+|   +--->BN_MP_ZERO_C
++--->BN_MP_DIV_2_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_SUB_C
+|   +--->BN_S_MP_ADD_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CMP_MAG_C
+|   +--->BN_S_MP_SUB_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
++--->BN_MP_CMP_C
+|   +--->BN_MP_CMP_MAG_C
++--->BN_MP_CMP_D_C
++--->BN_MP_ADD_C
+|   +--->BN_S_MP_ADD_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CMP_MAG_C
+|   +--->BN_S_MP_SUB_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
++--->BN_MP_EXCH_C
++--->BN_MP_CLEAR_MULTI_C
+|   +--->BN_MP_CLEAR_C
+
+
+BN_MP_FWRITE_C
++--->BN_MP_RADIX_SIZE_C
+|   +--->BN_MP_COUNT_BITS_C
+|   +--->BN_MP_INIT_COPY_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_DIV_D_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_DIV_2D_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_DIV_3_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_CLEAR_C
++--->BN_MP_TORADIX_C
+|   +--->BN_MP_INIT_COPY_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_DIV_D_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_DIV_2D_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_DIV_3_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_CLEAR_C
+
+
+BN_S_MP_SQR_C
++--->BN_MP_INIT_SIZE_C
+|   +--->BN_MP_INIT_C
++--->BN_MP_CLAMP_C
++--->BN_MP_EXCH_C
++--->BN_MP_CLEAR_C
+
+
+BN_MP_N_ROOT_C
++--->BN_MP_INIT_C
++--->BN_MP_SET_C
+|   +--->BN_MP_ZERO_C
++--->BN_MP_COPY_C
+|   +--->BN_MP_GROW_C
++--->BN_MP_EXPT_D_C
+|   +--->BN_MP_INIT_COPY_C
+|   +--->BN_MP_SQR_C
+|   |   +--->BN_MP_TOOM_SQR_C
+|   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_SUB_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_KARATSUBA_SQR_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_SUB_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_FAST_S_MP_SQR_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_S_MP_SQR_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_MUL_C
+|   |   +--->BN_MP_TOOM_MUL_C
+|   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_SUB_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   +--->BN_MP_KARATSUBA_MUL_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_SUB_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_S_MP_MUL_DIGS_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
++--->BN_MP_MUL_C
+|   +--->BN_MP_TOOM_MUL_C
+|   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_MOD_2D_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_MUL_2_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_2_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MUL_2D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MUL_D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_3_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_KARATSUBA_MUL_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_S_MP_MUL_DIGS_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_C
++--->BN_MP_SUB_C
+|   +--->BN_S_MP_ADD_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CMP_MAG_C
+|   +--->BN_S_MP_SUB_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
++--->BN_MP_MUL_D_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_DIV_C
+|   +--->BN_MP_CMP_MAG_C
+|   +--->BN_MP_ZERO_C
+|   +--->BN_MP_INIT_MULTI_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_COUNT_BITS_C
+|   +--->BN_MP_MUL_2D_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CMP_C
+|   +--->BN_MP_ADD_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_DIV_2D_C
+|   |   +--->BN_MP_MOD_2D_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   +--->BN_MP_EXCH_C
+|   +--->BN_MP_CLEAR_MULTI_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_INIT_SIZE_C
+|   +--->BN_MP_INIT_COPY_C
+|   +--->BN_MP_LSHD_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_RSHD_C
+|   +--->BN_MP_RSHD_C
+|   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CLEAR_C
++--->BN_MP_CMP_C
+|   +--->BN_MP_CMP_MAG_C
++--->BN_MP_SUB_D_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_ADD_D_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_EXCH_C
++--->BN_MP_CLEAR_C
+
+
+BN_MP_PRIME_RABIN_MILLER_TRIALS_C
+
+
+BN_MP_RADIX_SIZE_C
++--->BN_MP_COUNT_BITS_C
++--->BN_MP_INIT_COPY_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
++--->BN_MP_DIV_D_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_DIV_2D_C
+|   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_MOD_2D_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   +--->BN_MP_DIV_3_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_INIT_SIZE_C
+|   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_EXCH_C
+|   +--->BN_MP_CLEAR_C
++--->BN_MP_CLEAR_C
+
+
+BN_MP_READ_SIGNED_BIN_C
++--->BN_MP_READ_UNSIGNED_BIN_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_ZERO_C
+|   +--->BN_MP_MUL_2D_C
+|   |   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CLAMP_C
+
+
+BN_MP_PRIME_RANDOM_EX_C
++--->BN_MP_READ_UNSIGNED_BIN_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_ZERO_C
+|   +--->BN_MP_MUL_2D_C
+|   |   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_PRIME_IS_PRIME_C
+|   +--->BN_MP_CMP_D_C
+|   +--->BN_MP_PRIME_IS_DIVISIBLE_C
+|   |   +--->BN_MP_MOD_D_C
+|   |   |   +--->BN_MP_DIV_D_C
+|   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_DIV_2D_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_INIT_C
+|   |   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_INIT_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_INIT_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_INIT_C
+|   +--->BN_MP_SET_C
+|   |   +--->BN_MP_ZERO_C
+|   +--->BN_MP_PRIME_MILLER_RABIN_C
+|   |   +--->BN_MP_INIT_COPY_C
+|   |   |   +--->BN_MP_COPY_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_SUB_D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_ADD_D_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CNT_LSB_C
+|   |   +--->BN_MP_DIV_2D_C
+|   |   |   +--->BN_MP_COPY_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_EXPTMOD_C
+|   |   |   +--->BN_MP_INVMOD_C
+|   |   |   |   +--->BN_FAST_MP_INVMOD_C
+|   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_ABS_C
+|   |   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   |   +--->BN_MP_INVMOD_SLOW_C
+|   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_MP_ABS_C
+|   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   +--->BN_MP_DR_IS_MODULUS_C
+|   |   |   +--->BN_MP_REDUCE_IS_2K_C
+|   |   |   |   +--->BN_MP_REDUCE_2K_C
+|   |   |   |   |   +--->BN_MP_COUNT_BITS_C
+|   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_COUNT_BITS_C
+|   |   |   +--->BN_MP_EXPTMOD_FAST_C
+|   |   |   |   +--->BN_MP_COUNT_BITS_C
+|   |   |   |   +--->BN_MP_MONTGOMERY_SETUP_C
+|   |   |   |   +--->BN_FAST_MP_MONTGOMERY_REDUCE_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_MONTGOMERY_REDUCE_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_DR_SETUP_C
+|   |   |   |   +--->BN_MP_DR_REDUCE_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_REDUCE_2K_SETUP_C
+|   |   |   |   |   +--->BN_MP_2EXPT_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_REDUCE_2K_C
+|   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_MONTGOMERY_CALC_NORMALIZATION_C
+|   |   |   |   |   +--->BN_MP_2EXPT_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_MULMOD_C
+|   |   |   |   |   +--->BN_MP_MUL_C
+|   |   |   |   |   |   +--->BN_MP_TOOM_MUL_C
+|   |   |   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_KARATSUBA_MUL_C
+|   |   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_S_MP_MUL_DIGS_C
+|   |   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   +--->BN_MP_MOD_C
+|   |   |   |   |   |   +--->BN_MP_DIV_C
+|   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_CMP_C
+|   |   |   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_MOD_C
+|   |   |   |   |   +--->BN_MP_DIV_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_C
+|   |   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_SQR_C
+|   |   |   |   |   +--->BN_MP_TOOM_SQR_C
+|   |   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_KARATSUBA_SQR_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_FAST_S_MP_SQR_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_S_MP_SQR_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_MUL_C
+|   |   |   |   |   +--->BN_MP_TOOM_MUL_C
+|   |   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_KARATSUBA_MUL_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_S_MP_MUL_DIGS_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_S_MP_EXPTMOD_C
+|   |   |   |   +--->BN_MP_COUNT_BITS_C
+|   |   |   |   +--->BN_MP_REDUCE_SETUP_C
+|   |   |   |   |   +--->BN_MP_2EXPT_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_DIV_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_C
+|   |   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_MOD_C
+|   |   |   |   |   +--->BN_MP_DIV_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_C
+|   |   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_SQR_C
+|   |   |   |   |   +--->BN_MP_TOOM_SQR_C
+|   |   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_KARATSUBA_SQR_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_FAST_S_MP_SQR_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_S_MP_SQR_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_REDUCE_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_MUL_C
+|   |   |   |   |   |   +--->BN_MP_TOOM_MUL_C
+|   |   |   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_KARATSUBA_MUL_C
+|   |   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_S_MP_MUL_DIGS_C
+|   |   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   +--->BN_S_MP_MUL_HIGH_DIGS_C
+|   |   |   |   |   |   +--->BN_FAST_S_MP_MUL_HIGH_DIGS_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   +--->BN_FAST_S_MP_MUL_HIGH_DIGS_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_S_MP_MUL_DIGS_C
+|   |   |   |   |   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_MUL_C
+|   |   |   |   |   +--->BN_MP_TOOM_MUL_C
+|   |   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_KARATSUBA_MUL_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_S_MP_MUL_DIGS_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_MP_SQRMOD_C
+|   |   |   +--->BN_MP_SQR_C
+|   |   |   |   +--->BN_MP_TOOM_SQR_C
+|   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   |   +--->BN_MP_KARATSUBA_SQR_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   |   +--->BN_FAST_S_MP_SQR_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_S_MP_SQR_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_MP_MOD_C
+|   |   |   |   +--->BN_MP_DIV_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   +--->BN_MP_COUNT_BITS_C
+|   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_CLEAR_C
++--->BN_MP_SUB_D_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_ADD_D_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_DIV_2_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_MUL_2_C
+|   +--->BN_MP_GROW_C
++--->BN_MP_ADD_D_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLAMP_C
+
+
+BN_MP_KARATSUBA_SQR_C
++--->BN_MP_INIT_SIZE_C
+|   +--->BN_MP_INIT_C
++--->BN_MP_CLAMP_C
++--->BN_MP_SQR_C
+|   +--->BN_MP_TOOM_SQR_C
+|   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   +--->BN_MP_INIT_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_MOD_2D_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_COPY_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_MUL_2_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_DIV_2_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_MUL_2D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   +--->BN_MP_MUL_D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_DIV_3_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   +--->BN_FAST_S_MP_SQR_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_S_MP_SQR_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_C
++--->BN_MP_SUB_C
+|   +--->BN_S_MP_ADD_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_CMP_MAG_C
+|   +--->BN_S_MP_SUB_C
+|   |   +--->BN_MP_GROW_C
++--->BN_S_MP_ADD_C
+|   +--->BN_MP_GROW_C
++--->BN_MP_LSHD_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_ZERO_C
++--->BN_MP_ADD_C
+|   +--->BN_MP_CMP_MAG_C
+|   +--->BN_S_MP_SUB_C
+|   |   +--->BN_MP_GROW_C
++--->BN_MP_CLEAR_C
+
+
+BN_MP_INIT_COPY_C
++--->BN_MP_COPY_C
+|   +--->BN_MP_GROW_C
+
+
+BN_MP_CLAMP_C
+
+
+BN_MP_TOOM_SQR_C
++--->BN_MP_INIT_MULTI_C
+|   +--->BN_MP_INIT_C
+|   +--->BN_MP_CLEAR_C
++--->BN_MP_MOD_2D_C
+|   +--->BN_MP_ZERO_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_COPY_C
+|   +--->BN_MP_GROW_C
++--->BN_MP_RSHD_C
+|   +--->BN_MP_ZERO_C
++--->BN_MP_SQR_C
+|   +--->BN_MP_KARATSUBA_SQR_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_INIT_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_FAST_S_MP_SQR_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_S_MP_SQR_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_INIT_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_C
++--->BN_MP_MUL_2_C
+|   +--->BN_MP_GROW_C
++--->BN_MP_ADD_C
+|   +--->BN_S_MP_ADD_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CMP_MAG_C
+|   +--->BN_S_MP_SUB_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
++--->BN_MP_SUB_C
+|   +--->BN_S_MP_ADD_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CMP_MAG_C
+|   +--->BN_S_MP_SUB_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
++--->BN_MP_DIV_2_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_MUL_2D_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_LSHD_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_MUL_D_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_DIV_3_C
+|   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_INIT_C
+|   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_EXCH_C
+|   +--->BN_MP_CLEAR_C
++--->BN_MP_LSHD_C
+|   +--->BN_MP_GROW_C
++--->BN_MP_CLEAR_MULTI_C
+|   +--->BN_MP_CLEAR_C
+
+
+BN_MP_MOD_C
++--->BN_MP_INIT_C
++--->BN_MP_DIV_C
+|   +--->BN_MP_CMP_MAG_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_ZERO_C
+|   +--->BN_MP_INIT_MULTI_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_SET_C
+|   +--->BN_MP_COUNT_BITS_C
+|   +--->BN_MP_MUL_2D_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CMP_C
+|   +--->BN_MP_SUB_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_ADD_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_DIV_2D_C
+|   |   +--->BN_MP_MOD_2D_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   +--->BN_MP_EXCH_C
+|   +--->BN_MP_CLEAR_MULTI_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_INIT_SIZE_C
+|   +--->BN_MP_INIT_COPY_C
+|   +--->BN_MP_LSHD_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_RSHD_C
+|   +--->BN_MP_RSHD_C
+|   +--->BN_MP_MUL_D_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CLEAR_C
++--->BN_MP_CLEAR_C
++--->BN_MP_ADD_C
+|   +--->BN_S_MP_ADD_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CMP_MAG_C
+|   +--->BN_S_MP_SUB_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
++--->BN_MP_EXCH_C
+
+
+BN_MP_INIT_C
+
+
+BN_MP_TOOM_MUL_C
++--->BN_MP_INIT_MULTI_C
+|   +--->BN_MP_INIT_C
+|   +--->BN_MP_CLEAR_C
++--->BN_MP_MOD_2D_C
+|   +--->BN_MP_ZERO_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_COPY_C
+|   +--->BN_MP_GROW_C
++--->BN_MP_RSHD_C
+|   +--->BN_MP_ZERO_C
++--->BN_MP_MUL_C
+|   +--->BN_MP_KARATSUBA_MUL_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_INIT_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_S_MP_MUL_DIGS_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_INIT_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_C
++--->BN_MP_MUL_2_C
+|   +--->BN_MP_GROW_C
++--->BN_MP_ADD_C
+|   +--->BN_S_MP_ADD_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CMP_MAG_C
+|   +--->BN_S_MP_SUB_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
++--->BN_MP_SUB_C
+|   +--->BN_S_MP_ADD_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CMP_MAG_C
+|   +--->BN_S_MP_SUB_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
++--->BN_MP_DIV_2_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_MUL_2D_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_LSHD_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_MUL_D_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_DIV_3_C
+|   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_INIT_C
+|   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_EXCH_C
+|   +--->BN_MP_CLEAR_C
++--->BN_MP_LSHD_C
+|   +--->BN_MP_GROW_C
++--->BN_MP_CLEAR_MULTI_C
+|   +--->BN_MP_CLEAR_C
+
+
+BN_MP_PRIME_IS_PRIME_C
++--->BN_MP_CMP_D_C
++--->BN_MP_PRIME_IS_DIVISIBLE_C
+|   +--->BN_MP_MOD_D_C
+|   |   +--->BN_MP_DIV_D_C
+|   |   |   +--->BN_MP_COPY_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_DIV_2D_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_INIT_C
+|   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_INIT_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_INIT_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_CLEAR_C
++--->BN_MP_INIT_C
++--->BN_MP_SET_C
+|   +--->BN_MP_ZERO_C
++--->BN_MP_PRIME_MILLER_RABIN_C
+|   +--->BN_MP_INIT_COPY_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_SUB_D_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ADD_D_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CNT_LSB_C
+|   +--->BN_MP_DIV_2D_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_MOD_2D_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   +--->BN_MP_EXPTMOD_C
+|   |   +--->BN_MP_INVMOD_C
+|   |   |   +--->BN_FAST_MP_INVMOD_C
+|   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ABS_C
+|   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_MP_INVMOD_SLOW_C
+|   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_ABS_C
+|   |   |   +--->BN_MP_COPY_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   +--->BN_MP_DR_IS_MODULUS_C
+|   |   +--->BN_MP_REDUCE_IS_2K_C
+|   |   |   +--->BN_MP_REDUCE_2K_C
+|   |   |   |   +--->BN_MP_COUNT_BITS_C
+|   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_COUNT_BITS_C
+|   |   +--->BN_MP_EXPTMOD_FAST_C
+|   |   |   +--->BN_MP_COUNT_BITS_C
+|   |   |   +--->BN_MP_MONTGOMERY_SETUP_C
+|   |   |   +--->BN_FAST_MP_MONTGOMERY_REDUCE_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_MONTGOMERY_REDUCE_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_DR_SETUP_C
+|   |   |   +--->BN_MP_DR_REDUCE_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_REDUCE_2K_SETUP_C
+|   |   |   |   +--->BN_MP_2EXPT_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_REDUCE_2K_C
+|   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_MONTGOMERY_CALC_NORMALIZATION_C
+|   |   |   |   +--->BN_MP_2EXPT_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_MULMOD_C
+|   |   |   |   +--->BN_MP_MUL_C
+|   |   |   |   |   +--->BN_MP_TOOM_MUL_C
+|   |   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_KARATSUBA_MUL_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_S_MP_MUL_DIGS_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_MOD_C
+|   |   |   |   |   +--->BN_MP_DIV_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_C
+|   |   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_MOD_C
+|   |   |   |   +--->BN_MP_DIV_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_COPY_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_SQR_C
+|   |   |   |   +--->BN_MP_TOOM_SQR_C
+|   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_KARATSUBA_SQR_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_FAST_S_MP_SQR_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_S_MP_SQR_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_MUL_C
+|   |   |   |   +--->BN_MP_TOOM_MUL_C
+|   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_KARATSUBA_MUL_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_S_MP_MUL_DIGS_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_S_MP_EXPTMOD_C
+|   |   |   +--->BN_MP_COUNT_BITS_C
+|   |   |   +--->BN_MP_REDUCE_SETUP_C
+|   |   |   |   +--->BN_MP_2EXPT_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_DIV_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_MOD_C
+|   |   |   |   +--->BN_MP_DIV_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_COPY_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_SQR_C
+|   |   |   |   +--->BN_MP_TOOM_SQR_C
+|   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_KARATSUBA_SQR_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_FAST_S_MP_SQR_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_S_MP_SQR_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_REDUCE_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_MUL_C
+|   |   |   |   |   +--->BN_MP_TOOM_MUL_C
+|   |   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_KARATSUBA_MUL_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_S_MP_MUL_DIGS_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_S_MP_MUL_HIGH_DIGS_C
+|   |   |   |   |   +--->BN_FAST_S_MP_MUL_HIGH_DIGS_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_FAST_S_MP_MUL_HIGH_DIGS_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_S_MP_MUL_DIGS_C
+|   |   |   |   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_MUL_C
+|   |   |   |   +--->BN_MP_TOOM_MUL_C
+|   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_KARATSUBA_MUL_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_S_MP_MUL_DIGS_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_EXCH_C
+|   +--->BN_MP_CMP_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   +--->BN_MP_SQRMOD_C
+|   |   +--->BN_MP_SQR_C
+|   |   |   +--->BN_MP_TOOM_SQR_C
+|   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_MP_KARATSUBA_SQR_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_FAST_S_MP_SQR_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_SQR_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_MOD_C
+|   |   |   +--->BN_MP_DIV_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   +--->BN_MP_COUNT_BITS_C
+|   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   +--->BN_MP_CLEAR_C
++--->BN_MP_CLEAR_C
+
+
+BN_MP_COPY_C
++--->BN_MP_GROW_C
+
+
+BN_S_MP_SUB_C
++--->BN_MP_GROW_C
++--->BN_MP_CLAMP_C
+
+
+BN_MP_READ_UNSIGNED_BIN_C
++--->BN_MP_GROW_C
++--->BN_MP_ZERO_C
++--->BN_MP_MUL_2D_C
+|   +--->BN_MP_COPY_C
+|   +--->BN_MP_LSHD_C
+|   |   +--->BN_MP_RSHD_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_CLAMP_C
+
+
+BN_MP_EXPTMOD_FAST_C
++--->BN_MP_COUNT_BITS_C
++--->BN_MP_INIT_C
++--->BN_MP_CLEAR_C
++--->BN_MP_MONTGOMERY_SETUP_C
++--->BN_FAST_MP_MONTGOMERY_REDUCE_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_ZERO_C
+|   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CMP_MAG_C
+|   +--->BN_S_MP_SUB_C
++--->BN_MP_MONTGOMERY_REDUCE_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_ZERO_C
+|   +--->BN_MP_CMP_MAG_C
+|   +--->BN_S_MP_SUB_C
++--->BN_MP_DR_SETUP_C
++--->BN_MP_DR_REDUCE_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CMP_MAG_C
+|   +--->BN_S_MP_SUB_C
++--->BN_MP_REDUCE_2K_SETUP_C
+|   +--->BN_MP_2EXPT_C
+|   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_S_MP_SUB_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
++--->BN_MP_REDUCE_2K_C
+|   +--->BN_MP_DIV_2D_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_MOD_2D_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   +--->BN_MP_MUL_D_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_S_MP_ADD_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CMP_MAG_C
+|   +--->BN_S_MP_SUB_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
++--->BN_MP_MONTGOMERY_CALC_NORMALIZATION_C
+|   +--->BN_MP_2EXPT_C
+|   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_SET_C
+|   |   +--->BN_MP_ZERO_C
+|   +--->BN_MP_MUL_2_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_CMP_MAG_C
+|   +--->BN_S_MP_SUB_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
++--->BN_MP_MULMOD_C
+|   +--->BN_MP_MUL_C
+|   |   +--->BN_MP_TOOM_MUL_C
+|   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_COPY_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_SUB_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   +--->BN_MP_KARATSUBA_MUL_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_SUB_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_S_MP_MUL_DIGS_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   +--->BN_MP_MOD_C
+|   |   +--->BN_MP_DIV_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_MP_COPY_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   +--->BN_MP_SET_C
+|   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_C
+|   |   |   +--->BN_MP_SUB_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_DIV_2D_C
+|   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_INIT_COPY_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
++--->BN_MP_SET_C
+|   +--->BN_MP_ZERO_C
++--->BN_MP_MOD_C
+|   +--->BN_MP_DIV_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_INIT_MULTI_C
+|   |   +--->BN_MP_MUL_2D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_2D_C
+|   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_INIT_COPY_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_MUL_D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_ADD_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_EXCH_C
++--->BN_MP_COPY_C
+|   +--->BN_MP_GROW_C
++--->BN_MP_SQR_C
+|   +--->BN_MP_TOOM_SQR_C
+|   |   +--->BN_MP_INIT_MULTI_C
+|   |   +--->BN_MP_MOD_2D_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_MUL_2_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_2_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MUL_2D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MUL_D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_3_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLEAR_MULTI_C
+|   +--->BN_MP_KARATSUBA_SQR_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   +--->BN_FAST_S_MP_SQR_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_S_MP_SQR_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
++--->BN_MP_MUL_C
+|   +--->BN_MP_TOOM_MUL_C
+|   |   +--->BN_MP_INIT_MULTI_C
+|   |   +--->BN_MP_MOD_2D_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_MUL_2_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_2_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MUL_2D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MUL_D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_3_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLEAR_MULTI_C
+|   +--->BN_MP_KARATSUBA_MUL_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_S_MP_MUL_DIGS_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
++--->BN_MP_EXCH_C
+
+
+BN_MP_TO_UNSIGNED_BIN_C
++--->BN_MP_INIT_COPY_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
++--->BN_MP_DIV_2D_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_ZERO_C
+|   +--->BN_MP_MOD_2D_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_RSHD_C
+|   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_EXCH_C
++--->BN_MP_CLEAR_C
+
+
+BN_MP_SET_INT_C
++--->BN_MP_ZERO_C
++--->BN_MP_MUL_2D_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_LSHD_C
+|   |   +--->BN_MP_RSHD_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_CLAMP_C
+
+
+BN_MP_MOD_D_C
++--->BN_MP_DIV_D_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_DIV_2D_C
+|   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_INIT_C
+|   |   +--->BN_MP_MOD_2D_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   +--->BN_MP_DIV_3_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_INIT_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_INIT_C
+|   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_EXCH_C
+|   +--->BN_MP_CLEAR_C
+
+
+BN_MP_SQR_C
++--->BN_MP_TOOM_SQR_C
+|   +--->BN_MP_INIT_MULTI_C
+|   |   +--->BN_MP_INIT_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_MOD_2D_C
+|   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_ZERO_C
+|   +--->BN_MP_MUL_2_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_ADD_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_SUB_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_DIV_2_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_MUL_2D_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_LSHD_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_MUL_D_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_DIV_3_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_INIT_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_LSHD_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLEAR_MULTI_C
+|   |   +--->BN_MP_CLEAR_C
++--->BN_MP_KARATSUBA_SQR_C
+|   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_INIT_C
+|   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_SUB_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   +--->BN_S_MP_ADD_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_LSHD_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_ZERO_C
+|   +--->BN_MP_ADD_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLEAR_C
++--->BN_FAST_S_MP_SQR_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLAMP_C
++--->BN_S_MP_SQR_C
+|   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_INIT_C
+|   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_EXCH_C
+|   +--->BN_MP_CLEAR_C
+
+
+BN_MP_MULMOD_C
++--->BN_MP_INIT_C
++--->BN_MP_MUL_C
+|   +--->BN_MP_TOOM_MUL_C
+|   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_MOD_2D_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_COPY_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_MUL_2_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_2_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MUL_2D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MUL_D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_3_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_KARATSUBA_MUL_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_S_MP_MUL_DIGS_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_C
++--->BN_MP_CLEAR_C
++--->BN_MP_MOD_C
+|   +--->BN_MP_DIV_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_INIT_MULTI_C
+|   |   +--->BN_MP_SET_C
+|   |   +--->BN_MP_COUNT_BITS_C
+|   |   +--->BN_MP_MUL_2D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_2D_C
+|   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_INIT_COPY_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_MUL_D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_ADD_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_EXCH_C
+
+
+BN_MP_DIV_2D_C
++--->BN_MP_COPY_C
+|   +--->BN_MP_GROW_C
++--->BN_MP_ZERO_C
++--->BN_MP_INIT_C
++--->BN_MP_MOD_2D_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_CLEAR_C
++--->BN_MP_RSHD_C
++--->BN_MP_CLAMP_C
++--->BN_MP_EXCH_C
+
+
+BN_S_MP_ADD_C
++--->BN_MP_GROW_C
++--->BN_MP_CLAMP_C
+
+
+BN_FAST_S_MP_SQR_C
++--->BN_MP_GROW_C
++--->BN_MP_CLAMP_C
+
+
+BN_S_MP_MUL_DIGS_C
++--->BN_FAST_S_MP_MUL_DIGS_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_INIT_SIZE_C
+|   +--->BN_MP_INIT_C
++--->BN_MP_CLAMP_C
++--->BN_MP_EXCH_C
++--->BN_MP_CLEAR_C
+
+
+BN_MP_XOR_C
++--->BN_MP_INIT_COPY_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
++--->BN_MP_CLAMP_C
++--->BN_MP_EXCH_C
++--->BN_MP_CLEAR_C
+
+
+BN_MP_RADIX_SMAP_C
+
+
+BN_MP_DR_IS_MODULUS_C
+
+
+BN_MP_MONTGOMERY_CALC_NORMALIZATION_C
++--->BN_MP_COUNT_BITS_C
++--->BN_MP_2EXPT_C
+|   +--->BN_MP_ZERO_C
+|   +--->BN_MP_GROW_C
++--->BN_MP_SET_C
+|   +--->BN_MP_ZERO_C
++--->BN_MP_MUL_2_C
+|   +--->BN_MP_GROW_C
++--->BN_MP_CMP_MAG_C
++--->BN_S_MP_SUB_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLAMP_C
+
+
+BN_MP_SUB_C
++--->BN_S_MP_ADD_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_CMP_MAG_C
++--->BN_S_MP_SUB_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLAMP_C
+
+
+BN_MP_INIT_MULTI_C
++--->BN_MP_INIT_C
++--->BN_MP_CLEAR_C
+
+
+BN_S_MP_MUL_HIGH_DIGS_C
++--->BN_FAST_S_MP_MUL_HIGH_DIGS_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_INIT_SIZE_C
+|   +--->BN_MP_INIT_C
++--->BN_MP_CLAMP_C
++--->BN_MP_EXCH_C
++--->BN_MP_CLEAR_C
+
+
+BN_MP_PRIME_NEXT_PRIME_C
++--->BN_MP_CMP_D_C
++--->BN_MP_SET_C
+|   +--->BN_MP_ZERO_C
++--->BN_MP_SUB_D_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_ADD_D_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_MOD_D_C
+|   +--->BN_MP_DIV_D_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_DIV_2D_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_INIT_C
+|   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_DIV_3_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_INIT_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_INIT_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_C
++--->BN_MP_INIT_C
++--->BN_MP_ADD_D_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_PRIME_MILLER_RABIN_C
+|   +--->BN_MP_INIT_COPY_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_CNT_LSB_C
+|   +--->BN_MP_DIV_2D_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_MOD_2D_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   +--->BN_MP_EXPTMOD_C
+|   |   +--->BN_MP_INVMOD_C
+|   |   |   +--->BN_FAST_MP_INVMOD_C
+|   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ABS_C
+|   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_MP_INVMOD_SLOW_C
+|   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_ABS_C
+|   |   |   +--->BN_MP_COPY_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   +--->BN_MP_DR_IS_MODULUS_C
+|   |   +--->BN_MP_REDUCE_IS_2K_C
+|   |   |   +--->BN_MP_REDUCE_2K_C
+|   |   |   |   +--->BN_MP_COUNT_BITS_C
+|   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_COUNT_BITS_C
+|   |   +--->BN_MP_EXPTMOD_FAST_C
+|   |   |   +--->BN_MP_COUNT_BITS_C
+|   |   |   +--->BN_MP_MONTGOMERY_SETUP_C
+|   |   |   +--->BN_FAST_MP_MONTGOMERY_REDUCE_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_MONTGOMERY_REDUCE_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_DR_SETUP_C
+|   |   |   +--->BN_MP_DR_REDUCE_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_REDUCE_2K_SETUP_C
+|   |   |   |   +--->BN_MP_2EXPT_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_REDUCE_2K_C
+|   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_MONTGOMERY_CALC_NORMALIZATION_C
+|   |   |   |   +--->BN_MP_2EXPT_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_MULMOD_C
+|   |   |   |   +--->BN_MP_MUL_C
+|   |   |   |   |   +--->BN_MP_TOOM_MUL_C
+|   |   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_KARATSUBA_MUL_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_S_MP_MUL_DIGS_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_MOD_C
+|   |   |   |   |   +--->BN_MP_DIV_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_C
+|   |   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_MOD_C
+|   |   |   |   +--->BN_MP_DIV_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_COPY_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_SQR_C
+|   |   |   |   +--->BN_MP_TOOM_SQR_C
+|   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_KARATSUBA_SQR_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_FAST_S_MP_SQR_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_S_MP_SQR_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_MUL_C
+|   |   |   |   +--->BN_MP_TOOM_MUL_C
+|   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_KARATSUBA_MUL_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_S_MP_MUL_DIGS_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_S_MP_EXPTMOD_C
+|   |   |   +--->BN_MP_COUNT_BITS_C
+|   |   |   +--->BN_MP_REDUCE_SETUP_C
+|   |   |   |   +--->BN_MP_2EXPT_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_DIV_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_MOD_C
+|   |   |   |   +--->BN_MP_DIV_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_COPY_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_SQR_C
+|   |   |   |   +--->BN_MP_TOOM_SQR_C
+|   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_KARATSUBA_SQR_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_FAST_S_MP_SQR_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_S_MP_SQR_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_REDUCE_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_MUL_C
+|   |   |   |   |   +--->BN_MP_TOOM_MUL_C
+|   |   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_KARATSUBA_MUL_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_S_MP_MUL_DIGS_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_S_MP_MUL_HIGH_DIGS_C
+|   |   |   |   |   +--->BN_FAST_S_MP_MUL_HIGH_DIGS_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_FAST_S_MP_MUL_HIGH_DIGS_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_S_MP_MUL_DIGS_C
+|   |   |   |   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_MUL_C
+|   |   |   |   +--->BN_MP_TOOM_MUL_C
+|   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_KARATSUBA_MUL_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_S_MP_MUL_DIGS_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_EXCH_C
+|   +--->BN_MP_CMP_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   +--->BN_MP_SQRMOD_C
+|   |   +--->BN_MP_SQR_C
+|   |   |   +--->BN_MP_TOOM_SQR_C
+|   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_MP_KARATSUBA_SQR_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_FAST_S_MP_SQR_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_SQR_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_MOD_C
+|   |   |   +--->BN_MP_DIV_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   +--->BN_MP_COUNT_BITS_C
+|   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   +--->BN_MP_CLEAR_C
++--->BN_MP_CLEAR_C
+
+
+BN_MP_SIGNED_BIN_SIZE_C
++--->BN_MP_UNSIGNED_BIN_SIZE_C
+|   +--->BN_MP_COUNT_BITS_C
+
+
+BN_MP_INVMOD_SLOW_C
++--->BN_MP_INIT_MULTI_C
+|   +--->BN_MP_INIT_C
+|   +--->BN_MP_CLEAR_C
++--->BN_MP_COPY_C
+|   +--->BN_MP_GROW_C
++--->BN_MP_SET_C
+|   +--->BN_MP_ZERO_C
++--->BN_MP_DIV_2_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_ADD_C
+|   +--->BN_S_MP_ADD_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CMP_MAG_C
+|   +--->BN_S_MP_SUB_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
++--->BN_MP_SUB_C
+|   +--->BN_S_MP_ADD_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CMP_MAG_C
+|   +--->BN_S_MP_SUB_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
++--->BN_MP_CMP_C
+|   +--->BN_MP_CMP_MAG_C
++--->BN_MP_CMP_D_C
++--->BN_MP_CMP_MAG_C
++--->BN_MP_EXCH_C
++--->BN_MP_CLEAR_MULTI_C
+|   +--->BN_MP_CLEAR_C
+
+
+BN_MP_LCM_C
++--->BN_MP_INIT_MULTI_C
+|   +--->BN_MP_INIT_C
+|   +--->BN_MP_CLEAR_C
++--->BN_MP_GCD_C
+|   +--->BN_MP_ABS_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_ZERO_C
+|   +--->BN_MP_INIT_COPY_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_CNT_LSB_C
+|   +--->BN_MP_DIV_2D_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_MOD_2D_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   +--->BN_MP_CMP_MAG_C
+|   +--->BN_MP_EXCH_C
+|   +--->BN_S_MP_SUB_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_MUL_2D_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CLEAR_C
++--->BN_MP_CMP_MAG_C
++--->BN_MP_DIV_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_ZERO_C
+|   +--->BN_MP_SET_C
+|   +--->BN_MP_COUNT_BITS_C
+|   +--->BN_MP_MUL_2D_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CMP_C
+|   +--->BN_MP_SUB_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_ADD_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_DIV_2D_C
+|   |   +--->BN_MP_INIT_C
+|   |   +--->BN_MP_MOD_2D_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   +--->BN_MP_EXCH_C
+|   +--->BN_MP_CLEAR_MULTI_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_INIT_C
+|   +--->BN_MP_INIT_C
+|   +--->BN_MP_INIT_COPY_C
+|   +--->BN_MP_LSHD_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_RSHD_C
+|   +--->BN_MP_RSHD_C
+|   +--->BN_MP_MUL_D_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CLEAR_C
++--->BN_MP_MUL_C
+|   +--->BN_MP_TOOM_MUL_C
+|   |   +--->BN_MP_MOD_2D_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_COPY_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_MUL_2_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_2_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MUL_2D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MUL_D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_3_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_INIT_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_KARATSUBA_MUL_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_INIT_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_S_MP_MUL_DIGS_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_INIT_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_C
++--->BN_MP_CLEAR_MULTI_C
+|   +--->BN_MP_CLEAR_C
+
+
+BN_REVERSE_C
+
+
+BN_MP_PRIME_IS_DIVISIBLE_C
++--->BN_MP_MOD_D_C
+|   +--->BN_MP_DIV_D_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_DIV_2D_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_INIT_C
+|   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_DIV_3_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_INIT_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_INIT_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_C
+
+
+BN_MP_SET_C
++--->BN_MP_ZERO_C
+
+
+BN_MP_GCD_C
++--->BN_MP_ABS_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
++--->BN_MP_ZERO_C
++--->BN_MP_INIT_COPY_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
++--->BN_MP_CNT_LSB_C
++--->BN_MP_DIV_2D_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_MOD_2D_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_RSHD_C
+|   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_EXCH_C
++--->BN_MP_CMP_MAG_C
++--->BN_MP_EXCH_C
++--->BN_S_MP_SUB_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_MUL_2D_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_LSHD_C
+|   |   +--->BN_MP_RSHD_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_CLEAR_C
+
+
+BN_MP_READ_RADIX_C
++--->BN_MP_ZERO_C
++--->BN_MP_MUL_D_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_ADD_D_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_SUB_D_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CLAMP_C
+
+
+BN_FAST_S_MP_MUL_HIGH_DIGS_C
++--->BN_MP_GROW_C
++--->BN_MP_CLAMP_C
+
+
+BN_FAST_MP_MONTGOMERY_REDUCE_C
++--->BN_MP_GROW_C
++--->BN_MP_RSHD_C
+|   +--->BN_MP_ZERO_C
++--->BN_MP_CLAMP_C
++--->BN_MP_CMP_MAG_C
++--->BN_S_MP_SUB_C
+
+
+BN_MP_DIV_D_C
++--->BN_MP_COPY_C
+|   +--->BN_MP_GROW_C
++--->BN_MP_DIV_2D_C
+|   +--->BN_MP_ZERO_C
+|   +--->BN_MP_INIT_C
+|   +--->BN_MP_MOD_2D_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_RSHD_C
+|   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_EXCH_C
++--->BN_MP_DIV_3_C
+|   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_INIT_C
+|   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_EXCH_C
+|   +--->BN_MP_CLEAR_C
++--->BN_MP_INIT_SIZE_C
+|   +--->BN_MP_INIT_C
++--->BN_MP_CLAMP_C
++--->BN_MP_EXCH_C
++--->BN_MP_CLEAR_C
+
+
+BN_MP_REDUCE_2K_SETUP_C
++--->BN_MP_INIT_C
++--->BN_MP_COUNT_BITS_C
++--->BN_MP_2EXPT_C
+|   +--->BN_MP_ZERO_C
+|   +--->BN_MP_GROW_C
++--->BN_MP_CLEAR_C
++--->BN_S_MP_SUB_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLAMP_C
+
+
+BN_MP_INIT_SET_C
++--->BN_MP_INIT_C
++--->BN_MP_SET_C
+|   +--->BN_MP_ZERO_C
+
+
+BN_MP_REDUCE_2K_C
++--->BN_MP_INIT_C
++--->BN_MP_COUNT_BITS_C
++--->BN_MP_DIV_2D_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_ZERO_C
+|   +--->BN_MP_MOD_2D_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_RSHD_C
+|   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_EXCH_C
++--->BN_MP_MUL_D_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLAMP_C
++--->BN_S_MP_ADD_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_CMP_MAG_C
++--->BN_S_MP_SUB_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_CLEAR_C
+
+
+BN_ERROR_C
+
+
+BN_MP_EXPT_D_C
++--->BN_MP_INIT_COPY_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
++--->BN_MP_SET_C
+|   +--->BN_MP_ZERO_C
++--->BN_MP_SQR_C
+|   +--->BN_MP_TOOM_SQR_C
+|   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_MOD_2D_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_COPY_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_MUL_2_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_2_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MUL_2D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MUL_D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_3_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_KARATSUBA_SQR_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_FAST_S_MP_SQR_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_S_MP_SQR_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_C
++--->BN_MP_CLEAR_C
++--->BN_MP_MUL_C
+|   +--->BN_MP_TOOM_MUL_C
+|   |   +--->BN_MP_INIT_MULTI_C
+|   |   +--->BN_MP_MOD_2D_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_COPY_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_MUL_2_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_2_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MUL_2D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MUL_D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_3_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLEAR_MULTI_C
+|   +--->BN_MP_KARATSUBA_MUL_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_S_MP_MUL_DIGS_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+
+
+BN_S_MP_EXPTMOD_C
++--->BN_MP_COUNT_BITS_C
++--->BN_MP_INIT_C
++--->BN_MP_CLEAR_C
++--->BN_MP_REDUCE_SETUP_C
+|   +--->BN_MP_2EXPT_C
+|   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_DIV_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_INIT_MULTI_C
+|   |   +--->BN_MP_SET_C
+|   |   +--->BN_MP_MUL_2D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_2D_C
+|   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_INIT_COPY_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_MUL_D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CLAMP_C
++--->BN_MP_MOD_C
+|   +--->BN_MP_DIV_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_INIT_MULTI_C
+|   |   +--->BN_MP_SET_C
+|   |   +--->BN_MP_MUL_2D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_2D_C
+|   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_INIT_COPY_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_MUL_D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_ADD_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_EXCH_C
++--->BN_MP_COPY_C
+|   +--->BN_MP_GROW_C
++--->BN_MP_SQR_C
+|   +--->BN_MP_TOOM_SQR_C
+|   |   +--->BN_MP_INIT_MULTI_C
+|   |   +--->BN_MP_MOD_2D_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_MUL_2_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_2_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MUL_2D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MUL_D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_3_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLEAR_MULTI_C
+|   +--->BN_MP_KARATSUBA_SQR_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   +--->BN_FAST_S_MP_SQR_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_S_MP_SQR_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
++--->BN_MP_REDUCE_C
+|   +--->BN_MP_INIT_COPY_C
+|   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_ZERO_C
+|   +--->BN_MP_MUL_C
+|   |   +--->BN_MP_TOOM_MUL_C
+|   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_SUB_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   +--->BN_MP_KARATSUBA_MUL_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_SUB_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_S_MP_MUL_DIGS_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   +--->BN_S_MP_MUL_HIGH_DIGS_C
+|   |   +--->BN_FAST_S_MP_MUL_HIGH_DIGS_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   +--->BN_FAST_S_MP_MUL_HIGH_DIGS_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_MOD_2D_C
+|   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_S_MP_MUL_DIGS_C
+|   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   +--->BN_MP_SUB_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CMP_D_C
+|   +--->BN_MP_SET_C
+|   |   +--->BN_MP_ZERO_C
+|   +--->BN_MP_LSHD_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_ADD_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CMP_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   +--->BN_S_MP_SUB_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
++--->BN_MP_MUL_C
+|   +--->BN_MP_TOOM_MUL_C
+|   |   +--->BN_MP_INIT_MULTI_C
+|   |   +--->BN_MP_MOD_2D_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_MUL_2_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_2_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MUL_2D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MUL_D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_3_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLEAR_MULTI_C
+|   +--->BN_MP_KARATSUBA_MUL_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_S_MP_MUL_DIGS_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
++--->BN_MP_SET_C
+|   +--->BN_MP_ZERO_C
++--->BN_MP_EXCH_C
+
+
+BN_MP_ABS_C
++--->BN_MP_COPY_C
+|   +--->BN_MP_GROW_C
+
+
+BN_MP_INIT_SET_INT_C
++--->BN_MP_INIT_C
++--->BN_MP_SET_INT_C
+|   +--->BN_MP_ZERO_C
+|   +--->BN_MP_MUL_2D_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CLAMP_C
+
+
+BN_MP_SUB_D_C
++--->BN_MP_GROW_C
++--->BN_MP_ADD_D_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_CLAMP_C
+
+
+BN_MP_TO_SIGNED_BIN_C
++--->BN_MP_TO_UNSIGNED_BIN_C
+|   +--->BN_MP_INIT_COPY_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_DIV_2D_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_MOD_2D_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   +--->BN_MP_CLEAR_C
+
+
+BN_MP_DIV_2_C
++--->BN_MP_GROW_C
++--->BN_MP_CLAMP_C
+
+
+BN_MP_REDUCE_IS_2K_C
++--->BN_MP_REDUCE_2K_C
+|   +--->BN_MP_INIT_C
+|   +--->BN_MP_COUNT_BITS_C
+|   +--->BN_MP_DIV_2D_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_MOD_2D_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   +--->BN_MP_MUL_D_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_S_MP_ADD_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CMP_MAG_C
+|   +--->BN_S_MP_SUB_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CLEAR_C
++--->BN_MP_COUNT_BITS_C
+
+
+BN_MP_INIT_SIZE_C
++--->BN_MP_INIT_C
+
+
+BN_MP_DIV_C
++--->BN_MP_CMP_MAG_C
++--->BN_MP_COPY_C
+|   +--->BN_MP_GROW_C
++--->BN_MP_ZERO_C
++--->BN_MP_INIT_MULTI_C
+|   +--->BN_MP_INIT_C
+|   +--->BN_MP_CLEAR_C
++--->BN_MP_SET_C
++--->BN_MP_COUNT_BITS_C
++--->BN_MP_MUL_2D_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_LSHD_C
+|   |   +--->BN_MP_RSHD_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_CMP_C
++--->BN_MP_SUB_C
+|   +--->BN_S_MP_ADD_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_S_MP_SUB_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
++--->BN_MP_ADD_C
+|   +--->BN_S_MP_ADD_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_S_MP_SUB_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
++--->BN_MP_DIV_2D_C
+|   +--->BN_MP_INIT_C
+|   +--->BN_MP_MOD_2D_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_RSHD_C
+|   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_EXCH_C
++--->BN_MP_EXCH_C
++--->BN_MP_CLEAR_MULTI_C
+|   +--->BN_MP_CLEAR_C
++--->BN_MP_INIT_SIZE_C
+|   +--->BN_MP_INIT_C
++--->BN_MP_INIT_C
++--->BN_MP_INIT_COPY_C
++--->BN_MP_LSHD_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_RSHD_C
++--->BN_MP_RSHD_C
++--->BN_MP_MUL_D_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_CLAMP_C
++--->BN_MP_CLEAR_C
+
+
+BN_MP_CLEAR_C
+
+
+BN_MP_MONTGOMERY_REDUCE_C
++--->BN_FAST_MP_MONTGOMERY_REDUCE_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_ZERO_C
+|   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CMP_MAG_C
+|   +--->BN_S_MP_SUB_C
++--->BN_MP_GROW_C
++--->BN_MP_CLAMP_C
++--->BN_MP_RSHD_C
+|   +--->BN_MP_ZERO_C
++--->BN_MP_CMP_MAG_C
++--->BN_S_MP_SUB_C
+
+
+BN_MP_MUL_2_C
++--->BN_MP_GROW_C
+
+
+BN_MP_UNSIGNED_BIN_SIZE_C
++--->BN_MP_COUNT_BITS_C
+
+
+BN_MP_ADDMOD_C
++--->BN_MP_INIT_C
++--->BN_MP_ADD_C
+|   +--->BN_S_MP_ADD_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CMP_MAG_C
+|   +--->BN_S_MP_SUB_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
++--->BN_MP_CLEAR_C
++--->BN_MP_MOD_C
+|   +--->BN_MP_DIV_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_INIT_MULTI_C
+|   |   +--->BN_MP_SET_C
+|   |   +--->BN_MP_COUNT_BITS_C
+|   |   +--->BN_MP_MUL_2D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_2D_C
+|   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_INIT_COPY_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_MUL_D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_EXCH_C
+
+
+BN_MP_ADD_C
++--->BN_S_MP_ADD_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_CMP_MAG_C
++--->BN_S_MP_SUB_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLAMP_C
+
+
+BN_MP_RAND_C
++--->BN_MP_ZERO_C
++--->BN_MP_ADD_D_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_SUB_D_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_LSHD_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_RSHD_C
+
+
+BN_MP_CNT_LSB_C
+
+
+BN_MP_2EXPT_C
++--->BN_MP_ZERO_C
++--->BN_MP_GROW_C
+
+
+BN_MP_RSHD_C
++--->BN_MP_ZERO_C
+
+
+BN_MP_SHRINK_C
+
+
+BN_MP_REDUCE_C
++--->BN_MP_REDUCE_SETUP_C
+|   +--->BN_MP_2EXPT_C
+|   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_DIV_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   +--->BN_MP_INIT_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_SET_C
+|   |   +--->BN_MP_COUNT_BITS_C
+|   |   +--->BN_MP_MUL_2D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_2D_C
+|   |   |   +--->BN_MP_INIT_C
+|   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_INIT_C
+|   |   +--->BN_MP_INIT_C
+|   |   +--->BN_MP_INIT_COPY_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_MUL_D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CLEAR_C
++--->BN_MP_INIT_COPY_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
++--->BN_MP_RSHD_C
+|   +--->BN_MP_ZERO_C
++--->BN_MP_MUL_C
+|   +--->BN_MP_TOOM_MUL_C
+|   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_MOD_2D_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_COPY_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_MUL_2_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_2_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MUL_2D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MUL_D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_3_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_KARATSUBA_MUL_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_S_MP_MUL_DIGS_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_C
++--->BN_S_MP_MUL_HIGH_DIGS_C
+|   +--->BN_FAST_S_MP_MUL_HIGH_DIGS_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_INIT_SIZE_C
+|   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_EXCH_C
+|   +--->BN_MP_CLEAR_C
++--->BN_FAST_S_MP_MUL_HIGH_DIGS_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_MOD_2D_C
+|   +--->BN_MP_ZERO_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLAMP_C
++--->BN_S_MP_MUL_DIGS_C
+|   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_INIT_SIZE_C
+|   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_EXCH_C
+|   +--->BN_MP_CLEAR_C
++--->BN_MP_SUB_C
+|   +--->BN_S_MP_ADD_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CMP_MAG_C
+|   +--->BN_S_MP_SUB_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
++--->BN_MP_CMP_D_C
++--->BN_MP_SET_C
+|   +--->BN_MP_ZERO_C
++--->BN_MP_LSHD_C
+|   +--->BN_MP_GROW_C
++--->BN_MP_ADD_C
+|   +--->BN_S_MP_ADD_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CMP_MAG_C
+|   +--->BN_S_MP_SUB_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
++--->BN_MP_CMP_C
+|   +--->BN_MP_CMP_MAG_C
++--->BN_S_MP_SUB_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_CLEAR_C
+
+
+BN_MP_MUL_2D_C
++--->BN_MP_COPY_C
+|   +--->BN_MP_GROW_C
++--->BN_MP_GROW_C
++--->BN_MP_LSHD_C
+|   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_ZERO_C
++--->BN_MP_CLAMP_C
+
+
+BN_MP_GET_INT_C
+
+
+BN_MP_JACOBI_C
++--->BN_MP_CMP_D_C
++--->BN_MP_INIT_COPY_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
++--->BN_MP_CNT_LSB_C
++--->BN_MP_DIV_2D_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_ZERO_C
+|   +--->BN_MP_MOD_2D_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_RSHD_C
+|   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_EXCH_C
++--->BN_MP_MOD_C
+|   +--->BN_MP_DIV_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_SET_C
+|   |   +--->BN_MP_COUNT_BITS_C
+|   |   +--->BN_MP_MUL_2D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_MUL_D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_ADD_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_EXCH_C
++--->BN_MP_CLEAR_C
+
+
+BN_MP_CLEAR_MULTI_C
++--->BN_MP_CLEAR_C
+
+
+BN_MP_MUL_C
++--->BN_MP_TOOM_MUL_C
+|   +--->BN_MP_INIT_MULTI_C
+|   |   +--->BN_MP_INIT_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_MOD_2D_C
+|   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_ZERO_C
+|   +--->BN_MP_MUL_2_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_ADD_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_SUB_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_DIV_2_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_MUL_2D_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_LSHD_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_MUL_D_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_DIV_3_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_INIT_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_LSHD_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLEAR_MULTI_C
+|   |   +--->BN_MP_CLEAR_C
++--->BN_MP_KARATSUBA_MUL_C
+|   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_INIT_C
+|   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_SUB_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_ADD_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CMP_MAG_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_LSHD_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_ZERO_C
+|   +--->BN_MP_CLEAR_C
++--->BN_FAST_S_MP_MUL_DIGS_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLAMP_C
++--->BN_S_MP_MUL_DIGS_C
+|   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_INIT_C
+|   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_EXCH_C
+|   +--->BN_MP_CLEAR_C
+
+
+BN_MP_EXTEUCLID_C
++--->BN_MP_INIT_MULTI_C
+|   +--->BN_MP_INIT_C
+|   +--->BN_MP_CLEAR_C
++--->BN_MP_SET_C
+|   +--->BN_MP_ZERO_C
++--->BN_MP_COPY_C
+|   +--->BN_MP_GROW_C
++--->BN_MP_DIV_C
+|   +--->BN_MP_CMP_MAG_C
+|   +--->BN_MP_ZERO_C
+|   +--->BN_MP_COUNT_BITS_C
+|   +--->BN_MP_MUL_2D_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CMP_C
+|   +--->BN_MP_SUB_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_ADD_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_DIV_2D_C
+|   |   +--->BN_MP_INIT_C
+|   |   +--->BN_MP_MOD_2D_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   +--->BN_MP_EXCH_C
+|   +--->BN_MP_CLEAR_MULTI_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_INIT_C
+|   +--->BN_MP_INIT_C
+|   +--->BN_MP_INIT_COPY_C
+|   +--->BN_MP_LSHD_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_RSHD_C
+|   +--->BN_MP_RSHD_C
+|   +--->BN_MP_MUL_D_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CLEAR_C
++--->BN_MP_MUL_C
+|   +--->BN_MP_TOOM_MUL_C
+|   |   +--->BN_MP_MOD_2D_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_MUL_2_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_2_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MUL_2D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MUL_D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_3_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_INIT_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_KARATSUBA_MUL_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_INIT_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_S_MP_MUL_DIGS_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_INIT_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_C
++--->BN_MP_SUB_C
+|   +--->BN_S_MP_ADD_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CMP_MAG_C
+|   +--->BN_S_MP_SUB_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
++--->BN_MP_EXCH_C
++--->BN_MP_CLEAR_MULTI_C
+|   +--->BN_MP_CLEAR_C
+
+
+BN_MP_DR_REDUCE_C
++--->BN_MP_GROW_C
++--->BN_MP_CLAMP_C
++--->BN_MP_CMP_MAG_C
++--->BN_S_MP_SUB_C
+
+
+BN_MP_FREAD_C
++--->BN_MP_ZERO_C
++--->BN_MP_MUL_D_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_ADD_D_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_SUB_D_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_CMP_D_C
+
+
+BN_MP_REDUCE_SETUP_C
++--->BN_MP_2EXPT_C
+|   +--->BN_MP_ZERO_C
+|   +--->BN_MP_GROW_C
++--->BN_MP_DIV_C
+|   +--->BN_MP_CMP_MAG_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_ZERO_C
+|   +--->BN_MP_INIT_MULTI_C
+|   |   +--->BN_MP_INIT_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_SET_C
+|   +--->BN_MP_COUNT_BITS_C
+|   +--->BN_MP_MUL_2D_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CMP_C
+|   +--->BN_MP_SUB_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_ADD_C
+|   |   +--->BN_S_MP_ADD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_S_MP_SUB_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_DIV_2D_C
+|   |   +--->BN_MP_INIT_C
+|   |   +--->BN_MP_MOD_2D_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   +--->BN_MP_EXCH_C
+|   +--->BN_MP_CLEAR_MULTI_C
+|   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_INIT_SIZE_C
+|   |   +--->BN_MP_INIT_C
+|   +--->BN_MP_INIT_C
+|   +--->BN_MP_INIT_COPY_C
+|   +--->BN_MP_LSHD_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_RSHD_C
+|   +--->BN_MP_RSHD_C
+|   +--->BN_MP_MUL_D_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CLEAR_C
+
+
+BN_MP_MONTGOMERY_SETUP_C
+
+
+BN_MP_KARATSUBA_MUL_C
++--->BN_MP_MUL_C
+|   +--->BN_MP_TOOM_MUL_C
+|   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   +--->BN_MP_INIT_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_MOD_2D_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_COPY_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_MUL_2_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_SUB_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_2_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MUL_2D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MUL_D_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_DIV_3_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_INIT_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_LSHD_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_S_MP_MUL_DIGS_C
+|   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_INIT_C
+|   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_CLEAR_C
++--->BN_MP_INIT_SIZE_C
+|   +--->BN_MP_INIT_C
++--->BN_MP_CLAMP_C
++--->BN_MP_SUB_C
+|   +--->BN_S_MP_ADD_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_CMP_MAG_C
+|   +--->BN_S_MP_SUB_C
+|   |   +--->BN_MP_GROW_C
++--->BN_MP_ADD_C
+|   +--->BN_S_MP_ADD_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_CMP_MAG_C
+|   +--->BN_S_MP_SUB_C
+|   |   +--->BN_MP_GROW_C
++--->BN_MP_LSHD_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_RSHD_C
+|   |   +--->BN_MP_ZERO_C
++--->BN_MP_CLEAR_C
+
+
+BN_MP_LSHD_C
++--->BN_MP_GROW_C
++--->BN_MP_RSHD_C
+|   +--->BN_MP_ZERO_C
+
+
+BN_MP_PRIME_MILLER_RABIN_C
++--->BN_MP_CMP_D_C
++--->BN_MP_INIT_COPY_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
++--->BN_MP_SUB_D_C
+|   +--->BN_MP_GROW_C
+|   +--->BN_MP_ADD_D_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CLAMP_C
++--->BN_MP_CNT_LSB_C
++--->BN_MP_DIV_2D_C
+|   +--->BN_MP_COPY_C
+|   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_ZERO_C
+|   +--->BN_MP_MOD_2D_C
+|   |   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_RSHD_C
+|   +--->BN_MP_CLAMP_C
+|   +--->BN_MP_EXCH_C
++--->BN_MP_EXPTMOD_C
+|   +--->BN_MP_INVMOD_C
+|   |   +--->BN_FAST_MP_INVMOD_C
+|   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_MP_COPY_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_ABS_C
+|   |   |   +--->BN_MP_SET_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_SUB_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_INVMOD_SLOW_C
+|   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_MP_COPY_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_SET_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_SUB_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_ABS_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   +--->BN_MP_CLEAR_MULTI_C
+|   +--->BN_MP_DR_IS_MODULUS_C
+|   +--->BN_MP_REDUCE_IS_2K_C
+|   |   +--->BN_MP_REDUCE_2K_C
+|   |   |   +--->BN_MP_COUNT_BITS_C
+|   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_COUNT_BITS_C
+|   +--->BN_MP_EXPTMOD_FAST_C
+|   |   +--->BN_MP_COUNT_BITS_C
+|   |   +--->BN_MP_MONTGOMERY_SETUP_C
+|   |   +--->BN_FAST_MP_MONTGOMERY_REDUCE_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   +--->BN_MP_MONTGOMERY_REDUCE_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   +--->BN_MP_DR_SETUP_C
+|   |   +--->BN_MP_DR_REDUCE_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   +--->BN_MP_REDUCE_2K_SETUP_C
+|   |   |   +--->BN_MP_2EXPT_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_REDUCE_2K_C
+|   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MONTGOMERY_CALC_NORMALIZATION_C
+|   |   |   +--->BN_MP_2EXPT_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_SET_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MULMOD_C
+|   |   |   +--->BN_MP_MUL_C
+|   |   |   |   +--->BN_MP_TOOM_MUL_C
+|   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_KARATSUBA_MUL_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_S_MP_MUL_DIGS_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_MOD_C
+|   |   |   |   +--->BN_MP_DIV_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   +--->BN_MP_SET_C
+|   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_SET_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_MOD_C
+|   |   |   +--->BN_MP_DIV_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_SQR_C
+|   |   |   +--->BN_MP_TOOM_SQR_C
+|   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_KARATSUBA_SQR_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_FAST_S_MP_SQR_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_SQR_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_MUL_C
+|   |   |   +--->BN_MP_TOOM_MUL_C
+|   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_KARATSUBA_MUL_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_MUL_DIGS_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_EXCH_C
+|   +--->BN_S_MP_EXPTMOD_C
+|   |   +--->BN_MP_COUNT_BITS_C
+|   |   +--->BN_MP_REDUCE_SETUP_C
+|   |   |   +--->BN_MP_2EXPT_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_DIV_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   +--->BN_MP_SET_C
+|   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MOD_C
+|   |   |   +--->BN_MP_DIV_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   +--->BN_MP_SET_C
+|   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_COPY_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   +--->BN_MP_SQR_C
+|   |   |   +--->BN_MP_TOOM_SQR_C
+|   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_KARATSUBA_SQR_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_FAST_S_MP_SQR_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_SQR_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_REDUCE_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_MUL_C
+|   |   |   |   +--->BN_MP_TOOM_MUL_C
+|   |   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_KARATSUBA_MUL_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_S_MP_MUL_DIGS_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_S_MP_MUL_HIGH_DIGS_C
+|   |   |   |   +--->BN_FAST_S_MP_MUL_HIGH_DIGS_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_FAST_S_MP_MUL_HIGH_DIGS_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_MUL_DIGS_C
+|   |   |   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_SUB_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_SET_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_MUL_C
+|   |   |   +--->BN_MP_TOOM_MUL_C
+|   |   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_KARATSUBA_MUL_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_SUB_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_ADD_C
+|   |   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_FAST_S_MP_MUL_DIGS_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_S_MP_MUL_DIGS_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   +--->BN_MP_SET_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   +--->BN_MP_EXCH_C
++--->BN_MP_CMP_C
+|   +--->BN_MP_CMP_MAG_C
++--->BN_MP_SQRMOD_C
+|   +--->BN_MP_SQR_C
+|   |   +--->BN_MP_TOOM_SQR_C
+|   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_MP_MOD_2D_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   |   +--->BN_MP_COPY_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_COPY_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_MUL_2_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_SUB_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_DIV_2_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_DIV_3_C
+|   |   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_MP_EXCH_C
+|   |   |   |   +--->BN_MP_CLEAR_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_MP_KARATSUBA_SQR_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_SUB_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   |   +--->BN_FAST_S_MP_SQR_C
+|   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_S_MP_SQR_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_CLEAR_C
+|   +--->BN_MP_MOD_C
+|   |   +--->BN_MP_DIV_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_MP_COPY_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   +--->BN_MP_ZERO_C
+|   |   |   +--->BN_MP_INIT_MULTI_C
+|   |   |   +--->BN_MP_SET_C
+|   |   |   +--->BN_MP_COUNT_BITS_C
+|   |   |   +--->BN_MP_MUL_2D_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_SUB_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_ADD_C
+|   |   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_EXCH_C
+|   |   |   +--->BN_MP_CLEAR_MULTI_C
+|   |   |   +--->BN_MP_INIT_SIZE_C
+|   |   |   +--->BN_MP_LSHD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_RSHD_C
+|   |   |   +--->BN_MP_MUL_D_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_ADD_C
+|   |   |   +--->BN_S_MP_ADD_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   |   +--->BN_MP_CMP_MAG_C
+|   |   |   +--->BN_S_MP_SUB_C
+|   |   |   |   +--->BN_MP_GROW_C
+|   |   |   |   +--->BN_MP_CLAMP_C
+|   |   +--->BN_MP_EXCH_C
++--->BN_MP_CLEAR_C
+
+
+BN_MP_DR_SETUP_C
+
+
+BN_MP_CMP_MAG_C
+
+
diff --git a/changes.txt b/changes.txt
index 63e9d617a02009b37eb155650088862ae219d1af..6a862096adb31f94773d0a07b205f2c345a87401 100644
--- a/changes.txt
+++ b/changes.txt
@@ -1,3 +1,27 @@
+October 29th, 2004
+v0.32  -- Added "makefile.shared" for shared object support
+       -- Added more to the build options/configs in the manual
+       -- Started the Depends framework, wrote dep.pl to scan deps and 
+          produce "callgraph.txt" ;-)
+       -- Wrote SC_RSA_1 which will enable close to the minimum required to perform
+          RSA on 32-bit [or 64-bit] platforms with LibTomCrypt
+       -- Merged in the small/slower mp_div replacement.  You can now toggle which
+          you want to use as your mp_div() at build time.  Saves roughly 8KB or so.
+       -- Renamed a few files and changed some comments to make depends system work better.
+          (No changes to function names)
+       -- Merged in new Combas that perform 2 reads per inner loop instead of the older 
+          3reads/2writes per inner loop of the old code.  Really though if you want speed
+          learn to use TomsFastMath ;-)
+
+August 9th, 2004
+v0.31  -- "profiled" builds now :-) new timings for Intel Northwoods
+       -- Added "pretty" build target
+       -- Update mp_init() to actually assign 0's instead of relying on calloc()
+       -- "Wolfgang Ehrhardt" <Wolfgang.Ehrhardt@munich.netsurf.de> found a bug in mp_mul() where if
+          you multiply a negative by zero you get negative zero as the result.  Oops.
+       -- J Harper from PeerSec let me toy with his AMD64 and I got 60-bit digits working properly
+          [this also means that I fixed a bug where if sizeof(int) < sizeof(mp_digit) it would bug]
+
 April 11th, 2004
 v0.30  -- Added "mp_toradix_n" which stores upto "n-1" least significant digits of an mp_int
        -- Johan Lindh sent a patch so MSVC wouldn't whine about redefining malloc [in weird dll modes]
diff --git a/demo/demo.c b/demo/demo.c
index 8014ea82d915ec0e0a75b0dd6d501eb2653d49eb..53eb3cfc6e4fa9e21c71addaeb9b99c187c76799 100644
--- a/demo/demo.c
+++ b/demo/demo.c
@@ -1,7 +1,5 @@
 #include <time.h>
 
-#define TESTING
-
 #ifdef IOWNANATHLON
 #include <unistd.h>
 #define SLEEP sleep(4)
@@ -11,49 +9,6 @@
 
 #include "tommath.h"
 
-#ifdef TIMER
-ulong64 _tt;
-
-#if defined(__i386__) || defined(_M_IX86) || defined(_M_AMD64)
-/* RDTSC from Scott Duplichan */
-static ulong64 TIMFUNC (void)
-   {
-   #if defined __GNUC__
-      #ifdef __i386__
-         ulong64 a;
-         __asm__ __volatile__ ("rdtsc ":"=A" (a));
-         return a;
-      #else /* gcc-IA64 version */
-         unsigned long result;
-         __asm__ __volatile__("mov %0=ar.itc" : "=r"(result) :: "memory");
-         while (__builtin_expect ((int) result == -1, 0))
-         __asm__ __volatile__("mov %0=ar.itc" : "=r"(result) :: "memory");
-         return result;
-      #endif
-
-   // Microsoft and Intel Windows compilers
-   #elif defined _M_IX86
-     __asm rdtsc
-   #elif defined _M_AMD64
-     return __rdtsc ();
-   #elif defined _M_IA64
-     #if defined __INTEL_COMPILER
-       #include <ia64intrin.h>
-     #endif
-      return __getReg (3116);
-   #else
-     #error need rdtsc function for this build
-   #endif
-   }
-#else
-#define TIMFUNC clock
-#endif
-
-ulong64 rdtsc(void) { return TIMFUNC() - _tt; }
-void reset(void) { _tt = TIMFUNC(); }
-
-#endif
-
 void ndraw(mp_int *a, char *name)
 {
    char buf[4096];
@@ -89,10 +44,6 @@ int myrng(unsigned char *dst, int len, void *dat)
 }
 
 
-#define DO2(x) x; x;
-#define DO4(x) DO2(x); DO2(x);
-#define DO8(x) DO4(x); DO4(x);
-#define DO(x)  DO8(x); DO8(x);
 
    char cmd[4096], buf[4096];
 int main(void)
@@ -103,10 +54,6 @@ int main(void)
    unsigned rr;
    int i, n, err, cnt, ix, old_kara_m, old_kara_s;
 
-#ifdef TIMER
-   ulong64 tt, CLK_PER_SEC;
-   FILE *log, *logb, *logc;
-#endif
 
    mp_init(&a);
    mp_init(&b);
@@ -117,11 +64,11 @@ int main(void)
 
    srand(time(NULL));
 
-#ifdef TESTING
+#if 0
   // test mp_get_int
   printf("Testing: mp_get_int\n");
   for(i=0;i<1000;++i) {
-    t = (unsigned long)rand()*rand()+1;
+    t = ((unsigned long)rand()*rand()+1)&0xFFFFFFFF;
     mp_set_int(&a,t);
     if (t!=mp_get_int(&a)) { 
       printf("mp_get_int() bad result!\n");
@@ -141,7 +88,7 @@ int main(void)
 
   // test mp_sqrt
   printf("Testing: mp_sqrt\n");
-  for (i=0;i<10000;++i) { 
+  for (i=0;i<1000;++i) { 
     printf("%6d\r", i); fflush(stdout);
     n = (rand()&15)+1;
     mp_rand(&a,n);
@@ -157,7 +104,7 @@ int main(void)
   }
 
   printf("\nTesting: mp_is_square\n");
-  for (i=0;i<100000;++i) {
+  for (i=0;i<1000;++i) {
     printf("%6d\r", i); fflush(stdout);
 
     /* test mp_is_square false negatives */
@@ -186,11 +133,9 @@ int main(void)
 
   }
   printf("\n\n");
-#endif
 
-#ifdef TESTING 
    /* test for size */
-   for (ix = 16; ix < 512; ix++) {
+   for (ix = 10; ix < 256; ix++) {
        printf("Testing (not safe-prime): %9d bits    \r", ix); fflush(stdout);
        err = mp_prime_random_ex(&a, 8, ix, (rand()&1)?LTM_PRIME_2MSB_OFF:LTM_PRIME_2MSB_ON, myrng, NULL);
        if (err != MP_OKAY) {
@@ -203,7 +148,7 @@ int main(void)
        }
    }
 
-   for (ix = 16; ix < 512; ix++) {
+   for (ix = 16; ix < 256; ix++) {
        printf("Testing (   safe-prime): %9d bits    \r", ix); fflush(stdout);
        err = mp_prime_random_ex(&a, 8, ix, ((rand()&1)?LTM_PRIME_2MSB_OFF:LTM_PRIME_2MSB_ON)|LTM_PRIME_SAFE, myrng, NULL);
        if (err != MP_OKAY) {
@@ -225,9 +170,7 @@ int main(void)
    }
 
    printf("\n\n");
-#endif
 
-#ifdef TESTING
    mp_read_radix(&a, "123456", 10);
    mp_toradix_n(&a, buf, 10, 3);
    printf("a == %s\n", buf);
@@ -235,7 +178,6 @@ int main(void)
    printf("a == %s\n", buf);
    mp_toradix_n(&a, buf, 10, 30);
    printf("a == %s\n", buf);
-#endif
 
 
 #if 0
@@ -248,22 +190,6 @@ int main(void)
    }
 #endif
 
-#if 0
-{
-   mp_word aa, bb;
-
-   for (;;) {
-       aa = abs(rand()) & MP_MASK;
-       bb = abs(rand()) & MP_MASK;
-      if (MULT(aa,bb) != (aa*bb)) {
-             printf("%llu * %llu == %llu or %llu?\n", aa, bb, (ulong64)MULT(aa,bb), (ulong64)(aa*bb));
-             return 0;
-          }
-   }
-}
-#endif
-
-#ifdef TESTING
    /* test mp_cnt_lsb */
    printf("testing mp_cnt_lsb...\n");
    mp_set(&a, 1);
@@ -274,12 +200,10 @@ int main(void)
        }
        mp_mul_2(&a, &a);
    }
-#endif
 
 /* test mp_reduce_2k */
-#ifdef TESTING
    printf("Testing mp_reduce_2k...\n");
-   for (cnt = 3; cnt <= 384; ++cnt) {
+   for (cnt = 3; cnt <= 128; ++cnt) {
        mp_digit tmp;
        mp_2expt(&a, cnt);
        mp_sub_d(&a, 2, &a);  /* a = 2**cnt - 2 */
@@ -289,7 +213,7 @@ int main(void)
        printf("(%d)", mp_reduce_is_2k(&a));
        mp_reduce_2k_setup(&a, &tmp);
        printf("(%d)", tmp);
-       for (ix = 0; ix < 10000; ix++) {
+       for (ix = 0; ix < 1000; ix++) {
            if (!(ix & 127)) {printf("."); fflush(stdout); }
            mp_rand(&b, (cnt/DIGIT_BIT  + 1) * 2);
            mp_copy(&c, &b);
@@ -301,14 +225,11 @@ int main(void)
            }
         }
     }
-#endif
-
 
 /* test mp_div_3  */
-#ifdef TESTING
    printf("Testing mp_div_3...\n");
    mp_set(&d, 3);
-   for (cnt = 0; cnt < 1000000; ) {
+   for (cnt = 0; cnt < 10000; ) {
       mp_digit r1, r2;
 
       if (!(++cnt & 127)) printf("%9d\r", cnt);
@@ -321,12 +242,10 @@ int main(void)
       }
    }
    printf("\n\nPassed div_3 testing\n");
-#endif
 
 /* test the DR reduction */
-#ifdef TESTING
    printf("testing mp_dr_reduce...\n");
-   for (cnt = 2; cnt < 128; cnt++) {
+   for (cnt = 2; cnt < 32; cnt++) {
        printf("%d digit modulus\n", cnt);
        mp_grow(&a, cnt);
        mp_zero(&a);
@@ -334,7 +253,7 @@ int main(void)
            a.dp[ix] = MP_MASK;
        }
        a.used = cnt;
-       mp_prime_next_prime(&a, 3, 0);
+       a.dp[0] = 3;
 
        mp_rand(&b, cnt - 1);
        mp_copy(&b, &c);
@@ -346,204 +265,15 @@ int main(void)
          mp_copy(&b, &c);
 
          mp_mod(&b, &a, &b);
-         mp_dr_reduce(&c, &a, (1<<DIGIT_BIT)-a.dp[0]);
+         mp_dr_reduce(&c, &a, (((mp_digit)1)<<DIGIT_BIT)-a.dp[0]);
 
          if (mp_cmp(&b, &c) != MP_EQ) {
             printf("Failed on trial %lu\n", rr); exit(-1);
 
          }
-      } while (++rr < 100000);
+      } while (++rr < 500);
       printf("Passed DR test for %d digits\n", cnt);
    }
-#endif
-
-#ifdef TIMER
-      /* temp. turn off TOOM */
-      TOOM_MUL_CUTOFF = TOOM_SQR_CUTOFF = 100000;
-
-      reset();
-      sleep(1);
-      CLK_PER_SEC = rdtsc();
-
-      printf("CLK_PER_SEC == %lu\n", CLK_PER_SEC);
-      
-
-      log = fopen("logs/add.log", "w");
-      for (cnt = 8; cnt <= 128; cnt += 8) {
-         SLEEP;
-         mp_rand(&a, cnt);
-         mp_rand(&b, cnt);
-         reset();
-         rr = 0;
-         do {
-            DO(mp_add(&a,&b,&c));
-            rr += 16;
-         } while (rdtsc() < (CLK_PER_SEC * 2));
-         tt = rdtsc();
-         printf("Adding\t\t%4d-bit => %9llu/sec, %9llu ticks\n", mp_count_bits(&a), (((ulong64)rr)*CLK_PER_SEC)/tt, tt);
-         fprintf(log, "%d %9llu\n", cnt*DIGIT_BIT, (((ulong64)rr)*CLK_PER_SEC)/tt); fflush(log);
-      }
-      fclose(log);
-
-      log = fopen("logs/sub.log", "w");
-      for (cnt = 8; cnt <= 128; cnt += 8) {
-         SLEEP;
-         mp_rand(&a, cnt);
-         mp_rand(&b, cnt);
-         reset();
-         rr = 0;
-         do {
-            DO(mp_sub(&a,&b,&c));
-            rr += 16;
-         } while (rdtsc() < (CLK_PER_SEC * 2));
-         tt = rdtsc();
-         printf("Subtracting\t\t%4d-bit => %9llu/sec, %9llu ticks\n", mp_count_bits(&a), (((ulong64)rr)*CLK_PER_SEC)/tt, tt);
-         fprintf(log, "%d %9llu\n", cnt*DIGIT_BIT, (((ulong64)rr)*CLK_PER_SEC)/tt);  fflush(log);
-      }
-      fclose(log);
-
-   /* do mult/square twice, first without karatsuba and second with */
-mult_test:   
-   old_kara_m = KARATSUBA_MUL_CUTOFF;
-   old_kara_s = KARATSUBA_SQR_CUTOFF;
-   for (ix = 0; ix < 2; ix++) {
-      printf("With%s Karatsuba\n", (ix==0)?"out":"");
-
-      KARATSUBA_MUL_CUTOFF = (ix==0)?9999:old_kara_m;
-      KARATSUBA_SQR_CUTOFF = (ix==0)?9999:old_kara_s;
-
-      log = fopen((ix==0)?"logs/mult.log":"logs/mult_kara.log", "w");
-      for (cnt = 32; cnt <= 288; cnt += 8) {
-         SLEEP;
-         mp_rand(&a, cnt);
-         mp_rand(&b, cnt);
-         reset();
-         rr = 0;
-         do {
-            DO(mp_mul(&a, &b, &c));
-            rr += 16;
-         } while (rdtsc() < (CLK_PER_SEC * 2));
-         tt = rdtsc();
-         printf("Multiplying\t%4d-bit => %9llu/sec, %9llu ticks\n", mp_count_bits(&a), (((ulong64)rr)*CLK_PER_SEC)/tt, tt);
-         fprintf(log, "%d %9llu\n", mp_count_bits(&a), (((ulong64)rr)*CLK_PER_SEC)/tt);  fflush(log);
-      }
-      fclose(log);
-
-      log = fopen((ix==0)?"logs/sqr.log":"logs/sqr_kara.log", "w");
-      for (cnt = 32; cnt <= 288; cnt += 8) {
-         SLEEP;
-         mp_rand(&a, cnt);
-         reset();
-         rr = 0;
-         do {
-            DO(mp_sqr(&a, &b));
-            rr += 16;
-         } while (rdtsc() < (CLK_PER_SEC * 2));
-         tt = rdtsc();
-         printf("Squaring\t%4d-bit => %9llu/sec, %9llu ticks\n", mp_count_bits(&a), (((ulong64)rr)*CLK_PER_SEC)/tt, tt);
-         fprintf(log, "%d %9llu\n", mp_count_bits(&a), (((ulong64)rr)*CLK_PER_SEC)/tt);  fflush(log);
-      }
-      fclose(log);
-
-   }
-expt_test:
-  {
-      char *primes[] = {
-         /* 2K moduli mersenne primes */
-         "6864797660130609714981900799081393217269435300143305409394463459185543183397656052122559640661454554977296311391480858037121987999716643812574028291115057151",
-         "531137992816767098689588206552468627329593117727031923199444138200403559860852242739162502265229285668889329486246501015346579337652707239409519978766587351943831270835393219031728127",
-         "10407932194664399081925240327364085538615262247266704805319112350403608059673360298012239441732324184842421613954281007791383566248323464908139906605677320762924129509389220345773183349661583550472959420547689811211693677147548478866962501384438260291732348885311160828538416585028255604666224831890918801847068222203140521026698435488732958028878050869736186900714720710555703168729087",
-         "1475979915214180235084898622737381736312066145333169775147771216478570297878078949377407337049389289382748507531496480477281264838760259191814463365330269540496961201113430156902396093989090226259326935025281409614983499388222831448598601834318536230923772641390209490231836446899608210795482963763094236630945410832793769905399982457186322944729636418890623372171723742105636440368218459649632948538696905872650486914434637457507280441823676813517852099348660847172579408422316678097670224011990280170474894487426924742108823536808485072502240519452587542875349976558572670229633962575212637477897785501552646522609988869914013540483809865681250419497686697771007",
-         "259117086013202627776246767922441530941818887553125427303974923161874019266586362086201209516800483406550695241733194177441689509238807017410377709597512042313066624082916353517952311186154862265604547691127595848775610568757931191017711408826252153849035830401185072116424747461823031471398340229288074545677907941037288235820705892351068433882986888616658650280927692080339605869308790500409503709875902119018371991620994002568935113136548829739112656797303241986517250116412703509705427773477972349821676443446668383119322540099648994051790241624056519054483690809616061625743042361721863339415852426431208737266591962061753535748892894599629195183082621860853400937932839420261866586142503251450773096274235376822938649407127700846077124211823080804139298087057504713825264571448379371125032081826126566649084251699453951887789613650248405739378594599444335231188280123660406262468609212150349937584782292237144339628858485938215738821232393687046160677362909315071",
-         "190797007524439073807468042969529173669356994749940177394741882673528979787005053706368049835514900244303495954950709725762186311224148828811920216904542206960744666169364221195289538436845390250168663932838805192055137154390912666527533007309292687539092257043362517857366624699975402375462954490293259233303137330643531556539739921926201438606439020075174723029056838272505051571967594608350063404495977660656269020823960825567012344189908927956646011998057988548630107637380993519826582389781888135705408653045219655801758081251164080554609057468028203308718724654081055323215860189611391296030471108443146745671967766308925858547271507311563765171008318248647110097614890313562856541784154881743146033909602737947385055355960331855614540900081456378659068370317267696980001187750995491090350108417050917991562167972281070161305972518044872048331306383715094854938415738549894606070722584737978176686422134354526989443028353644037187375385397838259511833166416134323695660367676897722287918773420968982326089026150031515424165462111337527431154890666327374921446276833564519776797633875503548665093914556482031482248883127023777039667707976559857333357013727342079099064400455741830654320379350833236245819348824064783585692924881021978332974949906122664421376034687815350484991",
-
-         /* DR moduli */
-         "14059105607947488696282932836518693308967803494693489478439861164411992439598399594747002144074658928593502845729752797260025831423419686528151609940203368612079",
-         "101745825697019260773923519755878567461315282017759829107608914364075275235254395622580447400994175578963163918967182013639660669771108475957692810857098847138903161308502419410142185759152435680068435915159402496058513611411688900243039",
-         "736335108039604595805923406147184530889923370574768772191969612422073040099331944991573923112581267542507986451953227192970402893063850485730703075899286013451337291468249027691733891486704001513279827771740183629161065194874727962517148100775228363421083691764065477590823919364012917984605619526140821797602431",
-         "38564998830736521417281865696453025806593491967131023221754800625044118265468851210705360385717536794615180260494208076605798671660719333199513807806252394423283413430106003596332513246682903994829528690198205120921557533726473585751382193953592127439965050261476810842071573684505878854588706623484573925925903505747545471088867712185004135201289273405614415899438276535626346098904241020877974002916168099951885406379295536200413493190419727789712076165162175783",
-         "542189391331696172661670440619180536749994166415993334151601745392193484590296600979602378676624808129613777993466242203025054573692562689251250471628358318743978285860720148446448885701001277560572526947619392551574490839286458454994488665744991822837769918095117129546414124448777033941223565831420390846864429504774477949153794689948747680362212954278693335653935890352619041936727463717926744868338358149568368643403037768649616778526013610493696186055899318268339432671541328195724261329606699831016666359440874843103020666106568222401047720269951530296879490444224546654729111504346660859907296364097126834834235287147",
-         "1487259134814709264092032648525971038895865645148901180585340454985524155135260217788758027400478312256339496385275012465661575576202252063145698732079880294664220579764848767704076761853197216563262660046602703973050798218246170835962005598561669706844469447435461092542265792444947706769615695252256130901271870341005768912974433684521436211263358097522726462083917939091760026658925757076733484173202927141441492573799914240222628795405623953109131594523623353044898339481494120112723445689647986475279242446083151413667587008191682564376412347964146113898565886683139407005941383669325997475076910488086663256335689181157957571445067490187939553165903773554290260531009121879044170766615232300936675369451260747671432073394867530820527479172464106442450727640226503746586340279816318821395210726268291535648506190714616083163403189943334431056876038286530365757187367147446004855912033137386225053275419626102417236133948503",
-         "1095121115716677802856811290392395128588168592409109494900178008967955253005183831872715423151551999734857184538199864469605657805519106717529655044054833197687459782636297255219742994736751541815269727940751860670268774903340296040006114013971309257028332849679096824800250742691718610670812374272414086863715763724622797509437062518082383056050144624962776302147890521249477060215148275163688301275847155316042279405557632639366066847442861422164832655874655824221577849928863023018366835675399949740429332468186340518172487073360822220449055340582568461568645259954873303616953776393853174845132081121976327462740354930744487429617202585015510744298530101547706821590188733515880733527449780963163909830077616357506845523215289297624086914545378511082534229620116563260168494523906566709418166011112754529766183554579321224940951177394088465596712620076240067370589036924024728375076210477267488679008016579588696191194060127319035195370137160936882402244399699172017835144537488486396906144217720028992863941288217185353914991583400421682751000603596655790990815525126154394344641336397793791497068253936771017031980867706707490224041075826337383538651825493679503771934836094655802776331664261631740148281763487765852746577808019633679",
-
-         /* generic unrestricted moduli */
-         "17933601194860113372237070562165128350027320072176844226673287945873370751245439587792371960615073855669274087805055507977323024886880985062002853331424203",
-         "2893527720709661239493896562339544088620375736490408468011883030469939904368086092336458298221245707898933583190713188177399401852627749210994595974791782790253946539043962213027074922559572312141181787434278708783207966459019479487",
-         "347743159439876626079252796797422223177535447388206607607181663903045907591201940478223621722118173270898487582987137708656414344685816179420855160986340457973820182883508387588163122354089264395604796675278966117567294812714812796820596564876450716066283126720010859041484786529056457896367683122960411136319",
-         "47266428956356393164697365098120418976400602706072312735924071745438532218237979333351774907308168340693326687317443721193266215155735814510792148768576498491199122744351399489453533553203833318691678263241941706256996197460424029012419012634671862283532342656309677173602509498417976091509154360039893165037637034737020327399910409885798185771003505320583967737293415979917317338985837385734747478364242020380416892056650841470869294527543597349250299539682430605173321029026555546832473048600327036845781970289288898317888427517364945316709081173840186150794397479045034008257793436817683392375274635794835245695887",
-         "436463808505957768574894870394349739623346440601945961161254440072143298152040105676491048248110146278752857839930515766167441407021501229924721335644557342265864606569000117714935185566842453630868849121480179691838399545644365571106757731317371758557990781880691336695584799313313687287468894148823761785582982549586183756806449017542622267874275103877481475534991201849912222670102069951687572917937634467778042874315463238062009202992087620963771759666448266532858079402669920025224220613419441069718482837399612644978839925207109870840278194042158748845445131729137117098529028886770063736487420613144045836803985635654192482395882603511950547826439092832800532152534003936926017612446606135655146445620623395788978726744728503058670046885876251527122350275750995227",
-         "11424167473351836398078306042624362277956429440521137061889702611766348760692206243140413411077394583180726863277012016602279290144126785129569474909173584789822341986742719230331946072730319555984484911716797058875905400999504305877245849119687509023232790273637466821052576859232452982061831009770786031785669030271542286603956118755585683996118896215213488875253101894663403069677745948305893849505434201763745232895780711972432011344857521691017896316861403206449421332243658855453435784006517202894181640562433575390821384210960117518650374602256601091379644034244332285065935413233557998331562749140202965844219336298970011513882564935538704289446968322281451907487362046511461221329799897350993370560697505809686438782036235372137015731304779072430260986460269894522159103008260495503005267165927542949439526272736586626709581721032189532726389643625590680105784844246152702670169304203783072275089194754889511973916207",
-         "1214855636816562637502584060163403830270705000634713483015101384881871978446801224798536155406895823305035467591632531067547890948695117172076954220727075688048751022421198712032848890056357845974246560748347918630050853933697792254955890439720297560693579400297062396904306270145886830719309296352765295712183040773146419022875165382778007040109957609739589875590885701126197906063620133954893216612678838507540777138437797705602453719559017633986486649523611975865005712371194067612263330335590526176087004421363598470302731349138773205901447704682181517904064735636518462452242791676541725292378925568296858010151852326316777511935037531017413910506921922450666933202278489024521263798482237150056835746454842662048692127173834433089016107854491097456725016327709663199738238442164843147132789153725513257167915555162094970853584447993125488607696008169807374736711297007473812256272245489405898470297178738029484459690836250560495461579533254473316340608217876781986188705928270735695752830825527963838355419762516246028680280988020401914551825487349990306976304093109384451438813251211051597392127491464898797406789175453067960072008590614886532333015881171367104445044718144312416815712216611576221546455968770801413440778423979",
-         NULL
-      };
-   log = fopen("logs/expt.log", "w");
-   logb = fopen("logs/expt_dr.log", "w");
-   logc = fopen("logs/expt_2k.log", "w");
-   for (n = 0; primes[n]; n++) {
-      SLEEP;
-      mp_read_radix(&a, primes[n], 10);
-      mp_zero(&b);
-      for (rr = 0; rr < mp_count_bits(&a); rr++) {
-         mp_mul_2(&b, &b);
-         b.dp[0] |= lbit();
-         b.used  += 1;
-      }
-      mp_sub_d(&a, 1, &c);
-      mp_mod(&b, &c, &b);
-      mp_set(&c, 3);
-      reset();
-      rr = 0;
-      do {
-         DO(mp_exptmod(&c, &b, &a, &d));
-         rr += 16;
-      } while (rdtsc() < (CLK_PER_SEC * 2));
-      tt = rdtsc();
-      mp_sub_d(&a, 1, &e);
-      mp_sub(&e, &b, &b);
-      mp_exptmod(&c, &b, &a, &e);  /* c^(p-1-b) mod a */
-      mp_mulmod(&e, &d, &a, &d);   /* c^b * c^(p-1-b) == c^p-1 == 1 */
-      if (mp_cmp_d(&d, 1)) {
-         printf("Different (%d)!!!\n", mp_count_bits(&a));
-         draw(&d);
-         exit(0);
-      }
-      printf("Exponentiating\t%4d-bit => %9llu/sec, %9llu ticks\n", mp_count_bits(&a), (((ulong64)rr)*CLK_PER_SEC)/tt, tt);
-      fprintf((n < 6) ? logc : (n < 13) ? logb : log, "%d %9llu\n", mp_count_bits(&a), (((ulong64)rr)*CLK_PER_SEC)/tt);
-   }
-   }
-   fclose(log);
-   fclose(logb);
-   fclose(logc);
-
-   log = fopen("logs/invmod.log", "w");
-   for (cnt = 4; cnt <= 128; cnt += 4) {
-      SLEEP;
-      mp_rand(&a, cnt);
-      mp_rand(&b, cnt);
-
-      do {
-         mp_add_d(&b, 1, &b);
-         mp_gcd(&a, &b, &c);
-      } while (mp_cmp_d(&c, 1) != MP_EQ);
-
-      reset();
-      rr = 0;
-      do {
-         DO(mp_invmod(&b, &a, &c));
-         rr += 16;
-      } while (rdtsc() < (CLK_PER_SEC * 2));
-      tt = rdtsc();
-      mp_mulmod(&b, &c, &a, &d);
-      if (mp_cmp_d(&d, 1) != MP_EQ) {
-         printf("Failed to invert\n");
-         return 0;
-      }
-      printf("Inverting mod\t%4d-bit => %9llu/sec, %9llu ticks\n", mp_count_bits(&a), (((ulong64)rr)*CLK_PER_SEC)/tt, tt);
-      fprintf(log, "%d %9llu\n", cnt*DIGIT_BIT, (((ulong64)rr)*CLK_PER_SEC)/tt);
-   }
-   fclose(log);
-
-   return 0;
 
 #endif
 
diff --git a/demo/timing.c b/demo/timing.c
new file mode 100644
index 0000000000000000000000000000000000000000..865c444aa1697208679a72ca44c24a6260508d5a
--- /dev/null
+++ b/demo/timing.c
@@ -0,0 +1,291 @@
+#include <tommath.h>
+#include <time.h>
+
+ulong64 _tt;
+
+#ifdef IOWNANATHLON
+#include <unistd.h>
+#define SLEEP sleep(4)
+#else
+#define SLEEP
+#endif
+
+
+void ndraw(mp_int *a, char *name)
+{
+   char buf[4096];
+   printf("%s: ", name);
+   mp_toradix(a, buf, 64);
+   printf("%s\n", buf);
+}
+
+static void draw(mp_int *a)
+{
+   ndraw(a, "");
+}
+
+
+unsigned long lfsr = 0xAAAAAAAAUL;
+
+int lbit(void)
+{
+   if (lfsr & 0x80000000UL) {
+      lfsr = ((lfsr << 1) ^ 0x8000001BUL) & 0xFFFFFFFFUL;
+      return 1;
+   } else {
+      lfsr <<= 1;
+      return 0;
+   }
+}
+
+#if defined(__i386__) || defined(_M_IX86) || defined(_M_AMD64)
+/* RDTSC from Scott Duplichan */
+static ulong64 TIMFUNC (void)
+   {
+   #if defined __GNUC__
+      #ifdef __i386__
+         ulong64 a;
+         __asm__ __volatile__ ("rdtsc ":"=A" (a));
+         return a;
+      #else /* gcc-IA64 version */
+         unsigned long result;
+         __asm__ __volatile__("mov %0=ar.itc" : "=r"(result) :: "memory");
+         while (__builtin_expect ((int) result == -1, 0))
+         __asm__ __volatile__("mov %0=ar.itc" : "=r"(result) :: "memory");
+         return result;
+      #endif
+
+   // Microsoft and Intel Windows compilers
+   #elif defined _M_IX86
+     __asm rdtsc
+   #elif defined _M_AMD64
+     return __rdtsc ();
+   #elif defined _M_IA64
+     #if defined __INTEL_COMPILER
+       #include <ia64intrin.h>
+     #endif
+      return __getReg (3116);
+   #else
+     #error need rdtsc function for this build
+   #endif
+   }
+#else
+#define TIMFUNC clock
+#endif
+
+#define DO(x) x; x;
+//#define DO4(x) DO2(x); DO2(x);
+//#define DO8(x) DO4(x); DO4(x);
+//#define DO(x)  DO8(x); DO8(x);
+
+int main(void)
+{
+   ulong64 tt, gg, CLK_PER_SEC;
+   FILE *log, *logb, *logc;
+   mp_int a, b, c, d, e, f;
+   int n, cnt, ix, old_kara_m, old_kara_s;
+   unsigned rr;
+
+   mp_init(&a);
+   mp_init(&b);
+   mp_init(&c);
+   mp_init(&d);
+   mp_init(&e);
+   mp_init(&f);
+
+   srand(time(NULL));
+ 
+
+      /* temp. turn off TOOM */
+      TOOM_MUL_CUTOFF = TOOM_SQR_CUTOFF = 100000;
+
+      CLK_PER_SEC = TIMFUNC();
+      sleep(1);
+      CLK_PER_SEC = TIMFUNC() - CLK_PER_SEC;
+
+      printf("CLK_PER_SEC == %llu\n", CLK_PER_SEC);
+      
+      log = fopen("logs/add.log", "w");
+      for (cnt = 8; cnt <= 128; cnt += 8) {
+         SLEEP;
+         mp_rand(&a, cnt);
+         mp_rand(&b, cnt);
+         rr = 0;
+         tt = -1;
+         do {
+            gg = TIMFUNC();
+            DO(mp_add(&a,&b,&c));
+            gg = (TIMFUNC() - gg)>>1;
+            if (tt > gg) tt = gg;
+         } while (++rr < 100000);
+         printf("Adding\t\t%4d-bit => %9llu/sec, %9llu cycles\n", mp_count_bits(&a), CLK_PER_SEC/tt, tt);
+         fprintf(log, "%d %9llu\n", cnt*DIGIT_BIT, tt); fflush(log);
+      }
+      fclose(log);
+
+      log = fopen("logs/sub.log", "w");
+      for (cnt = 8; cnt <= 128; cnt += 8) {
+         SLEEP;
+         mp_rand(&a, cnt);
+         mp_rand(&b, cnt);
+         rr = 0;
+         tt = -1;
+         do {
+            gg = TIMFUNC();
+            DO(mp_sub(&a,&b,&c));
+            gg = (TIMFUNC() - gg)>>1;
+            if (tt > gg) tt = gg;
+         } while (++rr < 100000);
+
+         printf("Subtracting\t\t%4d-bit => %9llu/sec, %9llu cycles\n", mp_count_bits(&a), CLK_PER_SEC/tt, tt);
+         fprintf(log, "%d %9llu\n", cnt*DIGIT_BIT, tt);  fflush(log);
+      }
+      fclose(log);
+
+   /* do mult/square twice, first without karatsuba and second with */
+   old_kara_m = KARATSUBA_MUL_CUTOFF;
+   old_kara_s = KARATSUBA_SQR_CUTOFF;
+   for (ix = 0; ix < 1; ix++) {
+      printf("With%s Karatsuba\n", (ix==0)?"out":"");
+
+      KARATSUBA_MUL_CUTOFF = (ix==0)?9999:old_kara_m;
+      KARATSUBA_SQR_CUTOFF = (ix==0)?9999:old_kara_s;
+
+      log = fopen((ix==0)?"logs/mult.log":"logs/mult_kara.log", "w");
+      for (cnt = 4; cnt <= 288; cnt += 2) {
+         SLEEP;
+         mp_rand(&a, cnt);
+         mp_rand(&b, cnt);
+         rr = 0;
+         tt = -1;
+         do {
+            gg = TIMFUNC();
+            DO(mp_mul(&a, &b, &c));
+            gg = (TIMFUNC() - gg)>>1;
+            if (tt > gg) tt = gg;
+         } while (++rr < 100);
+         printf("Multiplying\t%4d-bit => %9llu/sec, %9llu cycles\n", mp_count_bits(&a), CLK_PER_SEC/tt, tt);
+         fprintf(log, "%d %9llu\n", mp_count_bits(&a), tt);  fflush(log);
+      }
+      fclose(log);
+
+      log = fopen((ix==0)?"logs/sqr.log":"logs/sqr_kara.log", "w");
+      for (cnt = 4; cnt <= 288; cnt += 2) {
+         SLEEP;
+         mp_rand(&a, cnt);
+         rr = 0;
+         tt = -1;
+         do {
+            gg = TIMFUNC();
+            DO(mp_sqr(&a, &b));
+            gg = (TIMFUNC() - gg)>>1;
+            if (tt > gg) tt = gg;
+         } while (++rr < 100);
+         printf("Squaring\t%4d-bit => %9llu/sec, %9llu cycles\n", mp_count_bits(&a), CLK_PER_SEC/tt, tt);
+         fprintf(log, "%d %9llu\n", mp_count_bits(&a), tt);  fflush(log);
+      }
+      fclose(log);
+
+   }
+
+  {
+      char *primes[] = {
+         /* 2K moduli mersenne primes */
+         "6864797660130609714981900799081393217269435300143305409394463459185543183397656052122559640661454554977296311391480858037121987999716643812574028291115057151",
+         "531137992816767098689588206552468627329593117727031923199444138200403559860852242739162502265229285668889329486246501015346579337652707239409519978766587351943831270835393219031728127",
+         "10407932194664399081925240327364085538615262247266704805319112350403608059673360298012239441732324184842421613954281007791383566248323464908139906605677320762924129509389220345773183349661583550472959420547689811211693677147548478866962501384438260291732348885311160828538416585028255604666224831890918801847068222203140521026698435488732958028878050869736186900714720710555703168729087",
+         "1475979915214180235084898622737381736312066145333169775147771216478570297878078949377407337049389289382748507531496480477281264838760259191814463365330269540496961201113430156902396093989090226259326935025281409614983499388222831448598601834318536230923772641390209490231836446899608210795482963763094236630945410832793769905399982457186322944729636418890623372171723742105636440368218459649632948538696905872650486914434637457507280441823676813517852099348660847172579408422316678097670224011990280170474894487426924742108823536808485072502240519452587542875349976558572670229633962575212637477897785501552646522609988869914013540483809865681250419497686697771007",
+         "259117086013202627776246767922441530941818887553125427303974923161874019266586362086201209516800483406550695241733194177441689509238807017410377709597512042313066624082916353517952311186154862265604547691127595848775610568757931191017711408826252153849035830401185072116424747461823031471398340229288074545677907941037288235820705892351068433882986888616658650280927692080339605869308790500409503709875902119018371991620994002568935113136548829739112656797303241986517250116412703509705427773477972349821676443446668383119322540099648994051790241624056519054483690809616061625743042361721863339415852426431208737266591962061753535748892894599629195183082621860853400937932839420261866586142503251450773096274235376822938649407127700846077124211823080804139298087057504713825264571448379371125032081826126566649084251699453951887789613650248405739378594599444335231188280123660406262468609212150349937584782292237144339628858485938215738821232393687046160677362909315071",
+         "190797007524439073807468042969529173669356994749940177394741882673528979787005053706368049835514900244303495954950709725762186311224148828811920216904542206960744666169364221195289538436845390250168663932838805192055137154390912666527533007309292687539092257043362517857366624699975402375462954490293259233303137330643531556539739921926201438606439020075174723029056838272505051571967594608350063404495977660656269020823960825567012344189908927956646011998057988548630107637380993519826582389781888135705408653045219655801758081251164080554609057468028203308718724654081055323215860189611391296030471108443146745671967766308925858547271507311563765171008318248647110097614890313562856541784154881743146033909602737947385055355960331855614540900081456378659068370317267696980001187750995491090350108417050917991562167972281070161305972518044872048331306383715094854938415738549894606070722584737978176686422134354526989443028353644037187375385397838259511833166416134323695660367676897722287918773420968982326089026150031515424165462111337527431154890666327374921446276833564519776797633875503548665093914556482031482248883127023777039667707976559857333357013727342079099064400455741830654320379350833236245819348824064783585692924881021978332974949906122664421376034687815350484991",
+
+         /* DR moduli */
+         "14059105607947488696282932836518693308967803494693489478439861164411992439598399594747002144074658928593502845729752797260025831423419686528151609940203368612079",
+         "101745825697019260773923519755878567461315282017759829107608914364075275235254395622580447400994175578963163918967182013639660669771108475957692810857098847138903161308502419410142185759152435680068435915159402496058513611411688900243039",
+         "736335108039604595805923406147184530889923370574768772191969612422073040099331944991573923112581267542507986451953227192970402893063850485730703075899286013451337291468249027691733891486704001513279827771740183629161065194874727962517148100775228363421083691764065477590823919364012917984605619526140821797602431",
+         "38564998830736521417281865696453025806593491967131023221754800625044118265468851210705360385717536794615180260494208076605798671660719333199513807806252394423283413430106003596332513246682903994829528690198205120921557533726473585751382193953592127439965050261476810842071573684505878854588706623484573925925903505747545471088867712185004135201289273405614415899438276535626346098904241020877974002916168099951885406379295536200413493190419727789712076165162175783",
+         "542189391331696172661670440619180536749994166415993334151601745392193484590296600979602378676624808129613777993466242203025054573692562689251250471628358318743978285860720148446448885701001277560572526947619392551574490839286458454994488665744991822837769918095117129546414124448777033941223565831420390846864429504774477949153794689948747680362212954278693335653935890352619041936727463717926744868338358149568368643403037768649616778526013610493696186055899318268339432671541328195724261329606699831016666359440874843103020666106568222401047720269951530296879490444224546654729111504346660859907296364097126834834235287147",
+         "1487259134814709264092032648525971038895865645148901180585340454985524155135260217788758027400478312256339496385275012465661575576202252063145698732079880294664220579764848767704076761853197216563262660046602703973050798218246170835962005598561669706844469447435461092542265792444947706769615695252256130901271870341005768912974433684521436211263358097522726462083917939091760026658925757076733484173202927141441492573799914240222628795405623953109131594523623353044898339481494120112723445689647986475279242446083151413667587008191682564376412347964146113898565886683139407005941383669325997475076910488086663256335689181157957571445067490187939553165903773554290260531009121879044170766615232300936675369451260747671432073394867530820527479172464106442450727640226503746586340279816318821395210726268291535648506190714616083163403189943334431056876038286530365757187367147446004855912033137386225053275419626102417236133948503",
+         "1095121115716677802856811290392395128588168592409109494900178008967955253005183831872715423151551999734857184538199864469605657805519106717529655044054833197687459782636297255219742994736751541815269727940751860670268774903340296040006114013971309257028332849679096824800250742691718610670812374272414086863715763724622797509437062518082383056050144624962776302147890521249477060215148275163688301275847155316042279405557632639366066847442861422164832655874655824221577849928863023018366835675399949740429332468186340518172487073360822220449055340582568461568645259954873303616953776393853174845132081121976327462740354930744487429617202585015510744298530101547706821590188733515880733527449780963163909830077616357506845523215289297624086914545378511082534229620116563260168494523906566709418166011112754529766183554579321224940951177394088465596712620076240067370589036924024728375076210477267488679008016579588696191194060127319035195370137160936882402244399699172017835144537488486396906144217720028992863941288217185353914991583400421682751000603596655790990815525126154394344641336397793791497068253936771017031980867706707490224041075826337383538651825493679503771934836094655802776331664261631740148281763487765852746577808019633679",
+
+         /* generic unrestricted moduli */
+         "17933601194860113372237070562165128350027320072176844226673287945873370751245439587792371960615073855669274087805055507977323024886880985062002853331424203",
+         "2893527720709661239493896562339544088620375736490408468011883030469939904368086092336458298221245707898933583190713188177399401852627749210994595974791782790253946539043962213027074922559572312141181787434278708783207966459019479487",
+         "347743159439876626079252796797422223177535447388206607607181663903045907591201940478223621722118173270898487582987137708656414344685816179420855160986340457973820182883508387588163122354089264395604796675278966117567294812714812796820596564876450716066283126720010859041484786529056457896367683122960411136319",
+         "47266428956356393164697365098120418976400602706072312735924071745438532218237979333351774907308168340693326687317443721193266215155735814510792148768576498491199122744351399489453533553203833318691678263241941706256996197460424029012419012634671862283532342656309677173602509498417976091509154360039893165037637034737020327399910409885798185771003505320583967737293415979917317338985837385734747478364242020380416892056650841470869294527543597349250299539682430605173321029026555546832473048600327036845781970289288898317888427517364945316709081173840186150794397479045034008257793436817683392375274635794835245695887",
+         "436463808505957768574894870394349739623346440601945961161254440072143298152040105676491048248110146278752857839930515766167441407021501229924721335644557342265864606569000117714935185566842453630868849121480179691838399545644365571106757731317371758557990781880691336695584799313313687287468894148823761785582982549586183756806449017542622267874275103877481475534991201849912222670102069951687572917937634467778042874315463238062009202992087620963771759666448266532858079402669920025224220613419441069718482837399612644978839925207109870840278194042158748845445131729137117098529028886770063736487420613144045836803985635654192482395882603511950547826439092832800532152534003936926017612446606135655146445620623395788978726744728503058670046885876251527122350275750995227",
+         "11424167473351836398078306042624362277956429440521137061889702611766348760692206243140413411077394583180726863277012016602279290144126785129569474909173584789822341986742719230331946072730319555984484911716797058875905400999504305877245849119687509023232790273637466821052576859232452982061831009770786031785669030271542286603956118755585683996118896215213488875253101894663403069677745948305893849505434201763745232895780711972432011344857521691017896316861403206449421332243658855453435784006517202894181640562433575390821384210960117518650374602256601091379644034244332285065935413233557998331562749140202965844219336298970011513882564935538704289446968322281451907487362046511461221329799897350993370560697505809686438782036235372137015731304779072430260986460269894522159103008260495503005267165927542949439526272736586626709581721032189532726389643625590680105784844246152702670169304203783072275089194754889511973916207",
+         "1214855636816562637502584060163403830270705000634713483015101384881871978446801224798536155406895823305035467591632531067547890948695117172076954220727075688048751022421198712032848890056357845974246560748347918630050853933697792254955890439720297560693579400297062396904306270145886830719309296352765295712183040773146419022875165382778007040109957609739589875590885701126197906063620133954893216612678838507540777138437797705602453719559017633986486649523611975865005712371194067612263330335590526176087004421363598470302731349138773205901447704682181517904064735636518462452242791676541725292378925568296858010151852326316777511935037531017413910506921922450666933202278489024521263798482237150056835746454842662048692127173834433089016107854491097456725016327709663199738238442164843147132789153725513257167915555162094970853584447993125488607696008169807374736711297007473812256272245489405898470297178738029484459690836250560495461579533254473316340608217876781986188705928270735695752830825527963838355419762516246028680280988020401914551825487349990306976304093109384451438813251211051597392127491464898797406789175453067960072008590614886532333015881171367104445044718144312416815712216611576221546455968770801413440778423979",
+         NULL
+      };
+   log = fopen("logs/expt.log", "w");
+   logb = fopen("logs/expt_dr.log", "w");
+   logc = fopen("logs/expt_2k.log", "w");
+   for (n = 0; primes[n]; n++) {
+      SLEEP;
+      mp_read_radix(&a, primes[n], 10);
+      mp_zero(&b);
+      for (rr = 0; rr < (unsigned)mp_count_bits(&a); rr++) {
+         mp_mul_2(&b, &b);
+         b.dp[0] |= lbit();
+         b.used  += 1;
+      }
+      mp_sub_d(&a, 1, &c);
+      mp_mod(&b, &c, &b);
+      mp_set(&c, 3);
+         rr = 0;
+         tt = -1;
+         do {
+            gg = TIMFUNC();
+            DO(mp_exptmod(&c, &b, &a, &d));
+            gg = (TIMFUNC() - gg)>>1;
+            if (tt > gg) tt = gg;
+         } while (++rr < 10);
+      mp_sub_d(&a, 1, &e);
+      mp_sub(&e, &b, &b);
+      mp_exptmod(&c, &b, &a, &e);  /* c^(p-1-b) mod a */
+      mp_mulmod(&e, &d, &a, &d);   /* c^b * c^(p-1-b) == c^p-1 == 1 */
+      if (mp_cmp_d(&d, 1)) {
+         printf("Different (%d)!!!\n", mp_count_bits(&a));
+         draw(&d);
+         exit(0);
+      }
+      printf("Exponentiating\t%4d-bit => %9llu/sec, %9llu cycles\n", mp_count_bits(&a), CLK_PER_SEC/tt, tt);
+      fprintf((n < 6) ? logc : (n < 13) ? logb : log, "%d %9llu\n", mp_count_bits(&a), tt);
+   }
+   }
+   fclose(log);
+   fclose(logb);
+   fclose(logc);
+
+   log = fopen("logs/invmod.log", "w");
+   for (cnt = 4; cnt <= 128; cnt += 4) {
+      SLEEP;
+      mp_rand(&a, cnt);
+      mp_rand(&b, cnt);
+
+      do {
+         mp_add_d(&b, 1, &b);
+         mp_gcd(&a, &b, &c);
+      } while (mp_cmp_d(&c, 1) != MP_EQ);
+
+         rr = 0;
+         tt = -1;
+      do {
+         gg = TIMFUNC();
+         DO(mp_invmod(&b, &a, &c));
+         gg = (TIMFUNC() - gg)>>1;
+         if (tt > gg) tt = gg;
+      } while (++rr < 1000);
+      mp_mulmod(&b, &c, &a, &d);
+      if (mp_cmp_d(&d, 1) != MP_EQ) {
+         printf("Failed to invert\n");
+         return 0;
+      }
+      printf("Inverting mod\t%4d-bit => %9llu/sec, %9llu cycles\n", mp_count_bits(&a), CLK_PER_SEC/tt, tt);
+      fprintf(log, "%d %9llu\n", cnt*DIGIT_BIT, tt);
+   }
+   fclose(log);
+
+   return 0;
+}
+
diff --git a/dep.pl b/dep.pl
new file mode 100644
index 0000000000000000000000000000000000000000..22266e38a68b47692f21d2736ec7dd982aebae7c
--- /dev/null
+++ b/dep.pl
@@ -0,0 +1,121 @@
+#!/usr/bin/perl 
+#
+# Walk through source, add labels and make classes
+#
+#use strict;
+
+my %deplist;
+
+#open class file and write preamble 
+open(CLASS, ">tommath_class.h") or die "Couldn't open tommath_class.h for writing\n";
+print CLASS "#if !(defined(LTM1) && defined(LTM2) && defined(LTM3))\n#if defined(LTM2)\n#define LTM3\n#endif\n#if defined(LTM1)\n#define LTM2\n#endif\n#define LTM1\n\n#if defined(LTM_ALL)\n";
+
+foreach my $filename (glob "bn*.c") {
+   my $define = $filename;
+
+   # convert filename to upper case so we can use it as a define 
+   $define =~ tr/[a-z]/[A-Z]/;
+   $define =~ tr/\./_/;
+   print CLASS "#define $define\n";
+
+   # now copy text and apply #ifdef as required 
+   my $apply = 0;
+   open(SRC, "<$filename");
+   open(OUT, ">tmp");
+
+   # first line will be the #ifdef
+   my $line = <SRC>;
+   if ($line =~ /include/) {
+      print OUT $line;
+   } else {
+      print OUT "#include <tommath.h>\n#ifdef $define\n$line";
+      $apply = 1;
+   }
+   while (<SRC>) {
+      if (!($_ =~ /tommath\.h/)) {
+         print OUT $_;
+      }
+   }
+   if ($apply == 1) {
+      print OUT "#endif\n";
+   }
+   close SRC;
+   close OUT;
+
+   unlink($filename);
+   rename("tmp", $filename);
+}
+print CLASS "#endif\n\n";
+
+# now do classes 
+
+foreach my $filename (glob "bn*.c") {
+   open(SRC, "<$filename") or die "Can't open source file!\n"; 
+
+   # convert filename to upper case so we can use it as a define 
+   $filename =~ tr/[a-z]/[A-Z]/;
+   $filename =~ tr/\./_/;
+
+   print CLASS "#if defined($filename)\n";
+   my $list = $filename;
+
+   # scan for mp_* and make classes
+   while (<SRC>) {
+      my $line = $_;
+      while ($line =~ m/(fast_)*(s_)*mp\_[a-z_0-9]*/) {
+          $line = $';
+          # now $& is the match, we want to skip over LTM keywords like
+          # mp_int, mp_word, mp_digit
+          if (!($& eq "mp_digit") && !($& eq "mp_word") && !($& eq "mp_int")) {
+             my $a = $&;
+             $a =~ tr/[a-z]/[A-Z]/;
+             $a = "BN_" . $a . "_C";
+             if (!($list =~ /$a/)) {
+                print CLASS "   #define $a\n";
+             }
+             $list = $list . "," . $a;
+          }
+      }
+   }
+   @deplist{$filename} = $list;
+
+   print CLASS "#endif\n\n";
+   close SRC;
+}
+
+print CLASS "#ifdef LTM3\n#define LTM_LAST\n#endif\n#include <tommath_superclass.h>\n#include <tommath_class.h>\n#else\n#define LTM_LAST\n#endif\n";
+close CLASS;
+
+#now let's make a cool call graph... 
+
+open(OUT,">callgraph.txt");
+$indent = 0;
+foreach (keys %deplist) {
+   $list = "";
+   draw_func(@deplist{$_});
+   print OUT "\n\n";
+}
+close(OUT);
+
+sub draw_func()
+{
+   my @funcs = split(",", $_[0]);
+   if ($list =~ /@funcs[0]/) {
+      return;
+   } else {
+      $list = $list . @funcs[0];
+   }
+   if ($indent == 0) { }
+   elsif ($indent >= 1) { print OUT "|   " x ($indent - 1) . "+--->"; }
+   print OUT @funcs[0] . "\n";   
+   shift @funcs;
+      my $temp = $list;
+   foreach my $i (@funcs) {
+      ++$indent;
+      draw_func(@deplist{$i});
+      --$indent;
+   }
+      $list = $temp;
+}
+
+
diff --git a/etc/makefile b/etc/makefile
index 98ddb1cea9f587410f3a03161fe1c8cf4f8593ee..99154d864d9942bd01123def6307b7e420cf090b 100644
--- a/etc/makefile
+++ b/etc/makefile
@@ -46,4 +46,5 @@ mont: mont.o
 
         
 clean:
-	rm -f *.log *.o *.obj *.exe pprime tune mersenne drprime tune86 tune86l mont 2kprime pprime.dat
+	rm -f *.log *.o *.obj *.exe pprime tune mersenne drprime tune86 tune86l mont 2kprime pprime.dat \
+         *.da *.dyn *.dpi *~
diff --git a/etc/makefile.icc b/etc/makefile.icc
new file mode 100644
index 0000000000000000000000000000000000000000..0a507282220ad38ddf9ec25850ee41d36a5b710a
--- /dev/null
+++ b/etc/makefile.icc
@@ -0,0 +1,67 @@
+CC = icc
+
+CFLAGS += -I../
+
+# optimize for SPEED
+#
+# -mcpu= can be pentium, pentiumpro (covers PII through PIII) or pentium4
+# -ax?   specifies make code specifically for ? but compatible with IA-32
+# -x?    specifies compile solely for ? [not specifically IA-32 compatible]
+#
+# where ? is 
+#   K - PIII
+#   W - first P4 [Williamette]
+#   N - P4 Northwood
+#   P - P4 Prescott
+#   B - Blend of P4 and PM [mobile]
+#
+# Default to just generic max opts
+CFLAGS += -O3 -xN -ip
+
+# default lib name (requires install with root)
+# LIBNAME=-ltommath
+
+# libname when you can't install the lib with install
+LIBNAME=../libtommath.a
+
+#provable primes
+pprime: pprime.o
+	$(CC) pprime.o $(LIBNAME) -o pprime
+
+# portable [well requires clock()] tuning app
+tune: tune.o
+	$(CC) tune.o $(LIBNAME) -o tune
+	
+# same app but using RDTSC for higher precision [requires 80586+], coff based gcc installs [e.g. ming, cygwin, djgpp]
+tune86: tune.c
+	nasm -f coff timer.asm
+	$(CC) -DX86_TIMER $(CFLAGS) tune.c timer.o  $(LIBNAME) -o tune86
+	
+# for cygwin
+tune86c: tune.c
+	nasm -f gnuwin32 timer.asm
+	$(CC) -DX86_TIMER $(CFLAGS) tune.c timer.o  $(LIBNAME) -o tune86
+
+#make tune86 for linux or any ELF format
+tune86l: tune.c
+	nasm -f elf -DUSE_ELF timer.asm
+	$(CC) -DX86_TIMER $(CFLAGS) tune.c timer.o $(LIBNAME) -o tune86l
+        
+# spits out mersenne primes
+mersenne: mersenne.o
+	$(CC) mersenne.o $(LIBNAME) -o mersenne
+
+# fines DR safe primes for the given config
+drprime: drprime.o
+	$(CC) drprime.o $(LIBNAME) -o drprime
+	
+# fines 2k safe primes for the given config
+2kprime: 2kprime.o
+	$(CC) 2kprime.o $(LIBNAME) -o 2kprime
+
+mont: mont.o
+	$(CC) mont.o $(LIBNAME) -o mont
+
+        
+clean:
+	rm -f *.log *.o *.obj *.exe pprime tune mersenne drprime tune86 tune86l mont 2kprime pprime.dat *.il
diff --git a/etc/tune.c b/etc/tune.c
index 11ce2104432647e2b1d6cdc61e42648444455376..bc101be9bd6e78b594183f3261c0c40e25a5e661 100644
--- a/etc/tune.c
+++ b/etc/tune.c
@@ -8,7 +8,7 @@
 /* how many times todo each size mult.  Depends on your computer.  For slow computers
  * this can be low like 5 or 10.  For fast [re: Athlon] should be 25 - 50 or so 
  */
-#define TIMES 50
+#define TIMES (1UL<<14UL)
 
 
 #ifndef X86_TIMER
@@ -23,154 +23,85 @@ extern void t_start(void);
 extern ulong64 t_read(void);
 #endif
 
-ulong64
-time_mult (int max)
+ulong64 time_mult(int size, int s)
 {
-  int     x, y;
+  unsigned long     x;
   mp_int  a, b, c;
+  ulong64 t1;
 
   mp_init (&a);
   mp_init (&b);
   mp_init (&c);
 
+  mp_rand (&a, size);
+  mp_rand (&b, size);
+
+  if (s == 1) { 
+      KARATSUBA_MUL_CUTOFF = size;
+  } else {
+      KARATSUBA_MUL_CUTOFF = 100000;
+  }
+
   t_start();
-  for (x = 32; x <= max; x += 4) {
-    mp_rand (&a, x);
-    mp_rand (&b, x);
-    for (y = 0; y < TIMES; y++) {
-      mp_mul (&a, &b, &c);
-    }
+  for (x = 0; x < TIMES; x++) {
+      mp_mul(&a,&b,&c);
   }
+  t1 = t_read();
   mp_clear (&a);
   mp_clear (&b);
   mp_clear (&c);
-  return t_read();
+  return t1;
 }
 
-ulong64
-time_sqr (int max)
+ulong64 time_sqr(int size, int s)
 {
-  int     x, y;
+  unsigned long     x;
   mp_int  a, b;
+  ulong64 t1;
 
   mp_init (&a);
   mp_init (&b);
 
+  mp_rand (&a, size);
+
+  if (s == 1) { 
+      KARATSUBA_SQR_CUTOFF = size;
+  } else {
+      KARATSUBA_SQR_CUTOFF = 100000;
+  }
+
   t_start();
-  for (x = 32; x <= max; x += 4) {
-    mp_rand (&a, x);
-    for (y = 0; y < TIMES; y++) {
-      mp_sqr (&a, &b);
-    }
+  for (x = 0; x < TIMES; x++) {
+      mp_sqr(&a,&b);
   }
+  t1 = t_read();
   mp_clear (&a);
   mp_clear (&b);
-  return t_read();
+  return t1;
 }
 
 int
 main (void)
 {
-  int     best_kmult, best_tmult, best_ksquare, best_tsquare, counter;
-  ulong64 best, ti;
-  FILE   *log;
-
-  best_kmult = best_ksquare = best_tmult = best_tsquare = 0;
-  /* tune multiplication first */
-  
-  /* effectively turn TOOM off */
-  TOOM_SQR_CUTOFF = TOOM_MUL_CUTOFF = 100000;
-    
-  log = fopen ("mult.log", "w");
-  best = -1;
-  counter = 16;
-  for (KARATSUBA_MUL_CUTOFF = 8; KARATSUBA_MUL_CUTOFF <= 200; KARATSUBA_MUL_CUTOFF++) {
-    ti = time_mult (300);
-    printf ("%4d : %9llu            \r", KARATSUBA_MUL_CUTOFF, ti);
-    fprintf (log, "%d, %llu\n", KARATSUBA_MUL_CUTOFF, ti);
-    fflush (stdout);
-    if (ti < best) {
-      printf ("New best: %llu, %d         \r", ti, KARATSUBA_MUL_CUTOFF);
-      best = ti;
-      best_kmult = KARATSUBA_MUL_CUTOFF;
-      counter = 16;
-    } else if (--counter == 0) {
-       printf("No better found in 16 trials.\n");
-       break;
-    }
-  }
-  fclose (log);
-  printf("Karatsuba Multiplier Cutoff (KARATSUBA_MUL_CUTOFF) == %d\n", best_kmult);
-  
-  /* tune squaring */
-  log = fopen ("sqr.log", "w");
-  best = -1;
-  counter = 16;
-  for (KARATSUBA_SQR_CUTOFF = 8; KARATSUBA_SQR_CUTOFF <= 200; KARATSUBA_SQR_CUTOFF++) {
-    ti = time_sqr (300);
-    printf ("%4d : %9llu             \r", KARATSUBA_SQR_CUTOFF, ti);
-    fprintf (log, "%d, %llu\n", KARATSUBA_SQR_CUTOFF, ti);
-    fflush (stdout);
-    if (ti < best) {
-      printf ("New best: %llu, %d         \r", ti, KARATSUBA_SQR_CUTOFF);
-      best = ti;
-      best_ksquare = KARATSUBA_SQR_CUTOFF;
-      counter = 16;
-    } else if (--counter == 0) {
-       printf("No better found in 16 trials.\n");
-       break;
-    }
-  }
-  fclose (log);
-  printf("Karatsuba Squaring Cutoff (KARATSUBA_SQR_CUTOFF) == %d\n", best_ksquare);
-  
-  KARATSUBA_MUL_CUTOFF = best_kmult;
-  KARATSUBA_SQR_CUTOFF = best_ksquare;
-    
-  /* tune TOOM mult */
-  counter = 16;
-  log = fopen ("tmult.log", "w");
-  best = -1;
-  for (TOOM_MUL_CUTOFF = best_kmult*5; TOOM_MUL_CUTOFF <= 800; TOOM_MUL_CUTOFF++) {
-    ti = time_mult (1200);
-    printf ("%4d : %9llu          \r", TOOM_MUL_CUTOFF, ti);
-    fprintf (log, "%d, %llu\n", TOOM_MUL_CUTOFF, ti);
-    fflush (stdout);
-    if (ti < best) {
-      printf ("New best: %llu, %d         \r", ti, TOOM_MUL_CUTOFF);
-      best = ti;
-      best_tmult = TOOM_MUL_CUTOFF;
-      counter = 16;
-    } else if (--counter == 0) {
-       printf("No better found in 16 trials.\n");
-       break;
-    }
-  }
-  fclose (log);   
-  printf("Toom-Cook Multiplier Cutoff (TOOM_MUL_CUTOFF) == %d\n", best_tmult);
-  
-  /* tune TOOM sqr */
-  log = fopen ("tsqr.log", "w");
-  best = -1;
-  counter = 16;
-  for (TOOM_SQR_CUTOFF = best_ksquare*3; TOOM_SQR_CUTOFF <= 800; TOOM_SQR_CUTOFF++) {
-    ti = time_sqr (1200);
-    printf ("%4d : %9llu           \r", TOOM_SQR_CUTOFF, ti);
-    fprintf (log, "%d, %llu\n", TOOM_SQR_CUTOFF, ti);
-    fflush (stdout);
-    if (ti < best) {
-      printf ("New best: %llu, %d         \r", ti, TOOM_SQR_CUTOFF);
-      best = ti;
-      best_tsquare = TOOM_SQR_CUTOFF;
-      counter = 16;
-    } else if (--counter == 0) {
-       printf("No better found in 16 trials.\n");
-       break;
-    }
+  ulong64 t1, t2;
+  int x, y;
+
+  for (x = 8; ; x += 2) { 
+     t1 = time_mult(x, 0);
+     t2 = time_mult(x, 1);
+     printf("%d: %9llu %9llu, %9llu\n", x, t1, t2, t2 - t1);
+     if (t2 < t1) break;
   }
-  fclose (log);   
-  printf("Toom-Cook Squaring Cutoff (TOOM_SQR_CUTOFF) == %d\n", best_tsquare);
+  y = x;
 
+  for (x = 8; ; x += 2) { 
+     t1 = time_sqr(x, 0);
+     t2 = time_sqr(x, 1);
+     printf("%d: %9llu %9llu, %9llu\n", x, t1, t2, t2 - t1);
+     if (t2 < t1) break;
+  }
+  printf("KARATSUBA_MUL_CUTOFF = %d\n", y);
+  printf("KARATSUBA_SQR_CUTOFF = %d\n", x);
 
   return 0;
 }
diff --git a/logs/add.log b/logs/add.log
index e53b415d2c07beff4b7d31aad5970a7b7612cc4b..d44c4cde5ed53ed062e5a946b35d773f798f03ef 100644
--- a/logs/add.log
+++ b/logs/add.log
@@ -1,16 +1,16 @@
-224  20297071
-448  15151383
-672  13088682
-896  11111587
-1120   9240621
-1344   8221878
-1568   7227434
-1792   6718051
-2016   6042524
-2240   5685200
-2464   5240465
-2688   4818032
-2912   4412794
-3136   4155883
-3360   3927078
-3584   3722138
+224       222
+448       330
+672       436
+896       520
+1120       612
+1344       696
+1568       810
+1792       912
+2016      1006
+2240      1116
+2464      1152
+2688      1284
+2912      1348
+3136      1486
+3360      1580
+3584      1636
diff --git a/logs/addsub.png b/logs/addsub.png
index e733f8d7622be239c70d1708c6ef8c79ec12576d..a5679ac563e6cf347bccc70fb4694ef684375ffc 100644
Binary files a/logs/addsub.png and b/logs/addsub.png differ
diff --git a/logs/expt.log b/logs/expt.log
index 2597b48643d7dcfb22f9f81027e4dbe38b09ca36..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 100644
--- a/logs/expt.log
+++ b/logs/expt.log
@@ -1,7 +0,0 @@
-513       745
-769       282
-1025       130
-2049        20
-2561        11
-3073         6
-4097         2
diff --git a/logs/expt.png b/logs/expt.png
index 59bafa2a8b2fc8e1298799eccc4b65defb299530..9ee8bb769feaa56ee7cffbbd5c7d39ebdb4799d3 100644
Binary files a/logs/expt.png and b/logs/expt.png differ
diff --git a/logs/expt_2k.log b/logs/expt_2k.log
index f4c282c23194059e87f3e89fd96634292c256a41..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 100644
--- a/logs/expt_2k.log
+++ b/logs/expt_2k.log
@@ -1,6 +0,0 @@
-521       783
-607       585
-1279       138
-2203        39
-3217        15
-4253         6
diff --git a/logs/expt_dr.log b/logs/expt_dr.log
index c552e127d93865c142d32ebbb791121c581a62a7..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 100644
--- a/logs/expt_dr.log
+++ b/logs/expt_dr.log
@@ -1,7 +0,0 @@
-532      1296
-784       551
-1036       283
-1540       109
-2072        52
-3080        18
-4116         7
diff --git a/logs/graphs.dem b/logs/graphs.dem
index d5c9b8adfc96edb9ad329f010f58e5297e9a072e..dfaf6138a1b83dd9c3a5f0293b8ccb8afd7f7d20 100644
--- a/logs/graphs.dem
+++ b/logs/graphs.dem
@@ -1,17 +1,17 @@
-set terminal png
-set size 1.75
-set ylabel "Operations per Second"
-set xlabel "Operand size (bits)"
-
-set output "addsub.png"
-plot 'add.log' smooth bezier title "Addition", 'sub.log' smooth bezier title "Subtraction"
-
-set output "mult.png"
-plot 'sqr.log' smooth bezier title "Squaring (without Karatsuba)", 'sqr_kara.log' smooth bezier title "Squaring (Karatsuba)", 'mult.log' smooth bezier title "Multiplication (without Karatsuba)", 'mult_kara.log' smooth bezier title "Multiplication (Karatsuba)"
-
-set output "expt.png"
-plot 'expt.log' smooth bezier title "Exptmod (Montgomery)", 'expt_dr.log' smooth bezier title "Exptmod (Dimminished Radix)", 'expt_2k.log' smooth bezier title "Exptmod (2k Reduction)"
-
-set output "invmod.png"
-plot 'invmod.log' smooth bezier title "Modular Inverse"
-
+set terminal png
+set size 1.75
+set ylabel "Cycles per Operation"
+set xlabel "Operand size (bits)"
+
+set output "addsub.png"
+plot 'add.log' smooth bezier title "Addition", 'sub.log' smooth bezier title "Subtraction"
+
+set output "mult.png"
+plot 'sqr.log' smooth bezier title "Squaring (without Karatsuba)", 'sqr_kara.log' smooth bezier title "Squaring (Karatsuba)", 'mult.log' smooth bezier title "Multiplication (without Karatsuba)", 'mult_kara.log' smooth bezier title "Multiplication (Karatsuba)"
+
+set output "expt.png"
+plot 'expt.log' smooth bezier title "Exptmod (Montgomery)", 'expt_dr.log' smooth bezier title "Exptmod (Dimminished Radix)", 'expt_2k.log' smooth bezier title "Exptmod (2k Reduction)"
+
+set output "invmod.png"
+plot 'invmod.log' smooth bezier title "Modular Inverse"
+
diff --git a/logs/invmod.log b/logs/invmod.log
index c9294ef66683174f712d42f4779d0f4a58b659cb..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 100644
--- a/logs/invmod.log
+++ b/logs/invmod.log
@@ -1,32 +0,0 @@
-112     17364
-224      8643
-336      8867
-448      6228
-560      4737
-672      2259
-784      2899
-896      1497
-1008      1238
-1120      1010
-1232       870
-1344      1265
-1456      1102
-1568       981
-1680       539
-1792       484
-1904       722
-2016       392
-2128       604
-2240       551
-2352       511
-2464       469
-2576       263
-2688       247
-2800       227
-2912       354
-3024       336
-3136       312
-3248       296
-3360       166
-3472       155
-3584       248
diff --git a/logs/invmod.png b/logs/invmod.png
index baa287fe2085f953be78c2839236895241048624..0a8a4ad77170380b229a454283696e6cd4910fd0 100644
Binary files a/logs/invmod.png and b/logs/invmod.png differ
diff --git a/logs/k7/README b/logs/k7/README
deleted file mode 100644
index ea20c81379850287a65ba27de073306e13a7e720..0000000000000000000000000000000000000000
--- a/logs/k7/README
+++ /dev/null
@@ -1,13 +0,0 @@
-To use the pretty graphs you have to first build/run the ltmtest from the root directory of the package.  
-Todo this type 
-
-make timing ; ltmtest
-
-in the root.  It will run for a while [about ten minutes on most PCs] and produce a series of .log files in logs/.
-
-After doing that run "gnuplot graphs.dem" to make the PNGs.  If you managed todo that all so far just open index.html to view
-them all :-)
-
-Have fun
-
-Tom
\ No newline at end of file
diff --git a/logs/k7/add.log b/logs/k7/add.log
deleted file mode 100644
index 796ab485099db7082dc098553b5a022fe2d84d05..0000000000000000000000000000000000000000
--- a/logs/k7/add.log
+++ /dev/null
@@ -1,16 +0,0 @@
-224  11069160
-448   9156136
-672   8089755
-896   7399424
-1120   6389352
-1344   5818648
-1568   5257112
-1792   4982160
-2016   4527856
-2240   4325312
-2464   4051760
-2688   3767640
-2912   3612520
-3136   3415208
-3360   3258656
-3584   3113360
diff --git a/logs/k7/addsub.png b/logs/k7/addsub.png
deleted file mode 100644
index 56391d99eafc845be9268ec1dc9355f2efb5b63e..0000000000000000000000000000000000000000
Binary files a/logs/k7/addsub.png and /dev/null differ
diff --git a/logs/k7/expt.log b/logs/k7/expt.log
deleted file mode 100644
index 46bb50be6a50b7b54b5605e7bb3c0376b4703834..0000000000000000000000000000000000000000
--- a/logs/k7/expt.log
+++ /dev/null
@@ -1,7 +0,0 @@
-513       664
-769       256
-1025       117
-2049        17
-2561         9
-3073         5
-4097         2
diff --git a/logs/k7/expt.png b/logs/k7/expt.png
deleted file mode 100644
index fc82677a7f6c107c5d587ce659b9fb4e32ac3035..0000000000000000000000000000000000000000
Binary files a/logs/k7/expt.png and /dev/null differ
diff --git a/logs/k7/expt_dr.log b/logs/k7/expt_dr.log
deleted file mode 100644
index 7df658fcd7d5b85c4fab157ec9c63ec56fbe686b..0000000000000000000000000000000000000000
--- a/logs/k7/expt_dr.log
+++ /dev/null
@@ -1,7 +0,0 @@
-532      1088
-784       460
-1036       240
-1540        92
-2072        43
-3080        15
-4116         6
diff --git a/logs/k7/graphs.dem b/logs/k7/graphs.dem
deleted file mode 100644
index c580495c648e085b5c502c52b4a680d0bfdd02e8..0000000000000000000000000000000000000000
--- a/logs/k7/graphs.dem
+++ /dev/null
@@ -1,17 +0,0 @@
-set terminal png color
-set size 1.75
-set ylabel "Operations per Second"
-set xlabel "Operand size (bits)"
-
-set output "addsub.png"
-plot 'add.log' smooth bezier title "Addition", 'sub.log' smooth bezier title "Subtraction"
-
-set output "mult.png"
-plot 'sqr.log' smooth bezier title "Squaring (without Karatsuba)", 'sqr_kara.log' smooth bezier title "Squaring (Karatsuba)", 'mult.log' smooth bezier title "Multiplication (without Karatsuba)", 'mult_kara.log' smooth bezier title "Multiplication (Karatsuba)"
-
-set output "expt.png"
-plot 'expt.log' smooth bezier title "Exptmod (Montgomery)", 'expt_dr.log' smooth bezier title "Exptmod (Dimminished Radix)"
-
-set output "invmod.png"
-plot 'invmod.log' smooth bezier title "Modular Inverse"
-
diff --git a/logs/k7/index.html b/logs/k7/index.html
deleted file mode 100644
index 19fe403e96bece608d5639cc505a66b1c048fb12..0000000000000000000000000000000000000000
--- a/logs/k7/index.html
+++ /dev/null
@@ -1,24 +0,0 @@
-<html>
-<head>
-<title>LibTomMath Log Plots</title>
-</head>
-<body>
-
-<h1>Addition and Subtraction</h1>
-<center><img src=addsub.png></center>
-<hr>
-
-<h1>Multipliers</h1>
-<center><img src=mult.png></center>
-<hr>
-
-<h1>Exptmod</h1>
-<center><img src=expt.png></center>
-<hr>
-
-<h1>Modular Inverse</h1>
-<center><img src=invmod.png></center>
-<hr>
-
-</body>
-</html>
\ No newline at end of file
diff --git a/logs/k7/invmod.log b/logs/k7/invmod.log
deleted file mode 100644
index d1198fb94214a0eee7b0a1698ef69b288a656764..0000000000000000000000000000000000000000
--- a/logs/k7/invmod.log
+++ /dev/null
@@ -1,32 +0,0 @@
-112     16248
-224      8192
-336      5320
-448      3560
-560      2728
-672      2064
-784      1704
-896      2176
-1008      1184
-1120       976
-1232      1280
-1344      1176
-1456       624
-1568       912
-1680       504
-1792       452
-1904       658
-2016       608
-2128       336
-2240       312
-2352       288
-2464       264
-2576       408
-2688       376
-2800       354
-2912       198
-3024       307
-3136       173
-3248       162
-3360       256
-3472       145
-3584       226
diff --git a/logs/k7/invmod.png b/logs/k7/invmod.png
deleted file mode 100644
index a497a723f46328692f503d2732afb3c524e1af82..0000000000000000000000000000000000000000
Binary files a/logs/k7/invmod.png and /dev/null differ
diff --git a/logs/k7/mult.log b/logs/k7/mult.log
deleted file mode 100644
index 4b1bff38e0c96aabb8c74f4a313f6e91fbbb7ae2..0000000000000000000000000000000000000000
--- a/logs/k7/mult.log
+++ /dev/null
@@ -1,17 +0,0 @@
-896    322904
-1344    151592
-1792     90472
-2240     59984
-2688     42624
-3136     31872
-3584     24704
-4032     19704
-4480     16096
-4928     13376
-5376     11272
-5824      9616
-6272      8360
-6720      7304
-7168      1664
-7616      1472
-8064      1328
diff --git a/logs/k7/mult.png b/logs/k7/mult.png
deleted file mode 100644
index 3cd8a9349df44a189a69b8dc401e9140c21b28e5..0000000000000000000000000000000000000000
Binary files a/logs/k7/mult.png and /dev/null differ
diff --git a/logs/k7/mult_kara.log b/logs/k7/mult_kara.log
deleted file mode 100644
index 53c086470c50e310dc5f3d81b0bcd1919a124cd1..0000000000000000000000000000000000000000
--- a/logs/k7/mult_kara.log
+++ /dev/null
@@ -1,17 +0,0 @@
-896    322872
-1344    151688
-1792     90480
-2240     59984
-2688     42656
-3136     32144
-3584     25840
-4032     21328
-4480     17856
-4928     14928
-5376     12856
-5824     11256
-6272      9880
-6720      8984
-7168      7928
-7616      7200
-8064      6576
diff --git a/logs/k7/sqr.log b/logs/k7/sqr.log
deleted file mode 100644
index 2fb2e980318223dd43f460d2155dad9086c918e0..0000000000000000000000000000000000000000
--- a/logs/k7/sqr.log
+++ /dev/null
@@ -1,17 +0,0 @@
-896    415472
-1344    223736
-1792    141232
-2240     97624
-2688     71400
-3136     54800
-3584     16904
-4032     13528
-4480     10968
-4928      9128
-5376      7784
-5824      6672
-6272      5760
-6720      5056
-7168      4440
-7616      3952
-8064      3512
diff --git a/logs/k7/sqr_kara.log b/logs/k7/sqr_kara.log
deleted file mode 100644
index ba30f9ea3bd0e83ef3d976217051abbf8acf253b..0000000000000000000000000000000000000000
--- a/logs/k7/sqr_kara.log
+++ /dev/null
@@ -1,17 +0,0 @@
-896    420464
-1344    224800
-1792    142808
-2240     97704
-2688     71416
-3136     54504
-3584     38320
-4032     32360
-4480     27576
-4928     23840
-5376     20688
-5824     18264
-6272     16176
-6720     14440
-7168     11688
-7616     10752
-8064      9936
diff --git a/logs/k7/sub.log b/logs/k7/sub.log
deleted file mode 100644
index 91c7d65d8f9d3246ac5b633366ea0ddbd3ee4219..0000000000000000000000000000000000000000
--- a/logs/k7/sub.log
+++ /dev/null
@@ -1,16 +0,0 @@
-224   9728504
-448   8573648
-672   7488096
-896   6714064
-1120   5950472
-1344   5457400
-1568   5038896
-1792   4683632
-2016   4384656
-2240   4105976
-2464   3871608
-2688   3650680
-2912   3463552
-3136   3290016
-3360   3135272
-3584   2993848
diff --git a/logs/mult.log b/logs/mult.log
index d4f589981704304af1d228cdf680df7531dd5776..a2c9c1821fb33c24eb9a83c9d90e304b74908b05 100644
--- a/logs/mult.log
+++ b/logs/mult.log
@@ -1,33 +1,143 @@
-920    374785
-1142    242737
-1371    176704
-1596    134341
-1816    105537
-2044     85089
-2268     70051
-2490     58671
-2716     49851
-2937     42881
-3162     37288
-3387     32697
-3608     28915
-3836     25759
-4057     23088
-4284     20800
-4508     18827
-4730     17164
-4956     15689
-5180     14397
-5398     13260
-5628     12249
-5852     11346
-6071     10537
-6298      9812
-6522      9161
-6742      8572
-6971      8038
-7195      2915
-7419      2744
-7644      2587
-7866      2444
-8090      2311
+140      1272
+195      1428
+252      1996
+307      2586
+364      3464
+420      4420
+476      5260
+532      6430
+588      7692
+644      8704
+699     10226
+755     11670
+812     13190
+865     14834
+924     16738
+979     18362
+1036     20660
+1092     22776
+1148     24848
+1204     27168
+1260     29930
+1316     32258
+1370     35172
+1422     37534
+1482     40390
+1537     43990
+1589     46946
+1652     50438
+1703     52902
+1764     56646
+1820     59892
+1876     63248
+1932     66872
+1988     72596
+2042     74662
+2100     78512
+2156     82944
+2211     87444
+2268     92170
+2324     95534
+2380    100484
+2435    105024
+2491    109460
+2546    114154
+2603    118946
+2660    124110
+2716    129300
+2771    134274
+2828    139594
+2883    145234
+2939    150332
+2996    155750
+3048    161718
+3108    167492
+3162    173882
+3219    179766
+3276    185560
+3330    191826
+3388    197822
+3442    204176
+3500    210682
+3556    217236
+3612    223484
+3666    230714
+3724    237744
+3779    244080
+3835    250970
+3890    257914
+3947    265162
+4001    272128
+4060    279108
+4116    287606
+4171    294716
+4227    302806
+4284    310260
+4340    318564
+4395    326164
+4443    334034
+4508    342108
+4561    351810
+4618    358828
+4675    367332
+4732    376140
+4787    384172
+4841    393308
+4899    402036
+4955    411286
+5010    420290
+5067    429688
+5124    438810
+5180    448130
+5235    457264
+5290    467390
+5348    476586
+5404    486120
+5459    496512
+5516    506624
+5569    516346
+5628    526604
+5684    536544
+5740    546936
+5796    557284
+5852    568106
+5907    578824
+5963    589204
+6019    600176
+6076    610564
+6127    621972
+6188    633564
+6244    644730
+6300    655288
+6354    667402
+6412    678824
+6467    690594
+6522    702718
+6580    714148
+6636    725608
+6690    737834
+6747    750100
+6804    762202
+6860    774184
+6916    787298
+6971    798734
+7028    811162
+7083    824570
+7139    837738
+7196   2579488
+7245   2626714
+7308   2643582
+7364   2698746
+7416   2734106
+7476   2773372
+7530   2816738
+7588   2859204
+7643   2938596
+7698   2919716
+7754   2988542
+7812   3026520
+7867   3058304
+7924   3115790
+7977   3161450
+8035   3203138
+8092   3244056
diff --git a/logs/mult.png b/logs/mult.png
index d304db2ca210015c57a2d368cfe5efcf83d19295..4f7a4eed7b7f12e3713b6b3f7a2191305be31612 100644
Binary files a/logs/mult.png and b/logs/mult.png differ
diff --git a/logs/mult_kara.log b/logs/mult_kara.log
index 6edc43945af99e7b6eeba0680dc0ae3183d8158c..086feaff0039e90eb429ea835267a04eae7393b6 100644
--- a/logs/mult_kara.log
+++ b/logs/mult_kara.log
@@ -1,33 +1,33 @@
-924    374171
-1147    243163
-1371    177111
-1596    134465
-1819    105619
-2044     85145
-2266     70086
-2488     58717
-2715     49869
-2939     42894
-3164     37389
-3387     33510
-3610     29993
-3836     27205
-4060     24751
-4281     22576
-4508     20670
-4732     19019
-4954     17527
-5180     16217
-5404     15044
-5624     14003
-5849     13051
-6076     12067
-6300     11438
-6524     10772
-6748     10298
-6972      9715
-7195      9330
-7416      8836
-7644      8465
-7864      8042
-8091      7735
+924     16686
+1146     25334
+1371     35304
+1591     47122
+1820     61500
+2044     75254
+2266     91732
+2492    111656
+2716    129428
+2937    147508
+3164    167758
+3388    188248
+3612    210826
+3836    233814
+4059    256898
+4284    280210
+4508    310372
+4731    333902
+4955    376502
+5179    402854
+5404    432004
+5626    459010
+5849    491868
+6076    520550
+6300    547400
+6524    575968
+6747    608482
+6971    642850
+7196    673670
+7419    710680
+7644    743942
+7868    780394
+8092    817342
diff --git a/logs/p4/README b/logs/p4/README
deleted file mode 100644
index ea20c81379850287a65ba27de073306e13a7e720..0000000000000000000000000000000000000000
--- a/logs/p4/README
+++ /dev/null
@@ -1,13 +0,0 @@
-To use the pretty graphs you have to first build/run the ltmtest from the root directory of the package.  
-Todo this type 
-
-make timing ; ltmtest
-
-in the root.  It will run for a while [about ten minutes on most PCs] and produce a series of .log files in logs/.
-
-After doing that run "gnuplot graphs.dem" to make the PNGs.  If you managed todo that all so far just open index.html to view
-them all :-)
-
-Have fun
-
-Tom
\ No newline at end of file
diff --git a/logs/p4/add.log b/logs/p4/add.log
deleted file mode 100644
index 72b2506c65c6b9f95ba2d79f403110fae5095209..0000000000000000000000000000000000000000
--- a/logs/p4/add.log
+++ /dev/null
@@ -1,16 +0,0 @@
-224   8113248
-448   6585584
-672   5687678
-896   4761144
-1120   4111592
-1344   3995154
-1568   3532387
-1792   3225400
-2016   2963960
-2240   2720112
-2464   2533952
-2688   2307168
-2912   2287064
-3136   2150160
-3360   2035992
-3584   1936304
diff --git a/logs/p4/addsub.png b/logs/p4/addsub.png
deleted file mode 100644
index f4398cad4639e3747210de5308613b67f243ff28..0000000000000000000000000000000000000000
Binary files a/logs/p4/addsub.png and /dev/null differ
diff --git a/logs/p4/expt.log b/logs/p4/expt.log
deleted file mode 100644
index 3e6ffb850a8f5335706816ba7dd4d6870cab844b..0000000000000000000000000000000000000000
--- a/logs/p4/expt.log
+++ /dev/null
@@ -1,7 +0,0 @@
-513       195
-769        68
-1025        31
-2049         4
-2561         2
-3073         1
-4097         0
diff --git a/logs/p4/expt.png b/logs/p4/expt.png
deleted file mode 100644
index dac1ce2cbdbbbd56ff8213d3e4616e022b186988..0000000000000000000000000000000000000000
Binary files a/logs/p4/expt.png and /dev/null differ
diff --git a/logs/p4/expt_dr.log b/logs/p4/expt_dr.log
deleted file mode 100644
index 2f5f6a3a3890f2d913f28377ec5ef58953a188ce..0000000000000000000000000000000000000000
--- a/logs/p4/expt_dr.log
+++ /dev/null
@@ -1,7 +0,0 @@
-532       393
-784       158
-1036        79
-1540        27
-2072        12
-3080         4
-4116         1
diff --git a/logs/p4/graphs.dem b/logs/p4/graphs.dem
deleted file mode 100644
index c580495c648e085b5c502c52b4a680d0bfdd02e8..0000000000000000000000000000000000000000
--- a/logs/p4/graphs.dem
+++ /dev/null
@@ -1,17 +0,0 @@
-set terminal png color
-set size 1.75
-set ylabel "Operations per Second"
-set xlabel "Operand size (bits)"
-
-set output "addsub.png"
-plot 'add.log' smooth bezier title "Addition", 'sub.log' smooth bezier title "Subtraction"
-
-set output "mult.png"
-plot 'sqr.log' smooth bezier title "Squaring (without Karatsuba)", 'sqr_kara.log' smooth bezier title "Squaring (Karatsuba)", 'mult.log' smooth bezier title "Multiplication (without Karatsuba)", 'mult_kara.log' smooth bezier title "Multiplication (Karatsuba)"
-
-set output "expt.png"
-plot 'expt.log' smooth bezier title "Exptmod (Montgomery)", 'expt_dr.log' smooth bezier title "Exptmod (Dimminished Radix)"
-
-set output "invmod.png"
-plot 'invmod.log' smooth bezier title "Modular Inverse"
-
diff --git a/logs/p4/index.html b/logs/p4/index.html
deleted file mode 100644
index 19fe403e96bece608d5639cc505a66b1c048fb12..0000000000000000000000000000000000000000
--- a/logs/p4/index.html
+++ /dev/null
@@ -1,24 +0,0 @@
-<html>
-<head>
-<title>LibTomMath Log Plots</title>
-</head>
-<body>
-
-<h1>Addition and Subtraction</h1>
-<center><img src=addsub.png></center>
-<hr>
-
-<h1>Multipliers</h1>
-<center><img src=mult.png></center>
-<hr>
-
-<h1>Exptmod</h1>
-<center><img src=expt.png></center>
-<hr>
-
-<h1>Modular Inverse</h1>
-<center><img src=invmod.png></center>
-<hr>
-
-</body>
-</html>
\ No newline at end of file
diff --git a/logs/p4/invmod.log b/logs/p4/invmod.log
deleted file mode 100644
index 096087b0c4ea897ec0e80e00c69f677462a0d8ad..0000000000000000000000000000000000000000
--- a/logs/p4/invmod.log
+++ /dev/null
@@ -1,32 +0,0 @@
-112     13608
-224      6872
-336      4264
-448      2792
-560      2144
-672      1560
-784      1296
-896      1672
-1008       896
-1120       736
-1232      1024
-1344       888
-1456       472
-1568       680
-1680       373
-1792       328
-1904       484
-2016       436
-2128       232
-2240       211
-2352       200
-2464       177
-2576       293
-2688       262
-2800       251
-2912       137
-3024       216
-3136       117
-3248       113
-3360       181
-3472        98
-3584       158
diff --git a/logs/p4/invmod.png b/logs/p4/invmod.png
deleted file mode 100644
index 3b0580fe05b79a283f0880b807edfeadb8df35a3..0000000000000000000000000000000000000000
Binary files a/logs/p4/invmod.png and /dev/null differ
diff --git a/logs/p4/mult.log b/logs/p4/mult.log
deleted file mode 100644
index 6e43806cd9bf992fe1f75bf66700d864bbd44993..0000000000000000000000000000000000000000
--- a/logs/p4/mult.log
+++ /dev/null
@@ -1,17 +0,0 @@
-896     77600
-1344     35776
-1792     19688
-2240     13248
-2688      9424
-3136      7056
-3584      5464
-4032      4368
-4480      3568
-4928      2976
-5376      2520
-5824      2152
-6272      1872
-6720      1632
-7168       650
-7616       576
-8064       515
diff --git a/logs/p4/mult.png b/logs/p4/mult.png
deleted file mode 100644
index 8623558f5c4dc9bf4a529f8a7f7c652e9cb7d0a9..0000000000000000000000000000000000000000
Binary files a/logs/p4/mult.png and /dev/null differ
diff --git a/logs/p4/mult_kara.log b/logs/p4/mult_kara.log
deleted file mode 100644
index e1d50a65a8d11fefb6e570476fe083f69b842e99..0000000000000000000000000000000000000000
--- a/logs/p4/mult_kara.log
+++ /dev/null
@@ -1,17 +0,0 @@
-896     77752
-1344     35832
-1792     19688
-2240     14704
-2688     10832
-3136      8336
-3584      6600
-4032      5424
-4480      4648
-4928      3976
-5376      3448
-5824      3016
-6272      2664
-6720      2384
-7168      2120
-7616      1912
-8064      1752
diff --git a/logs/p4/sqr.log b/logs/p4/sqr.log
deleted file mode 100644
index b133fb3d558d4e9abab0e73eb288add9b047676a..0000000000000000000000000000000000000000
--- a/logs/p4/sqr.log
+++ /dev/null
@@ -1,17 +0,0 @@
-896    128088
-1344     63640
-1792     37968
-2240     25488
-2688     18176
-3136     13672
-3584      4920
-4032      3912
-4480      3160
-4928      2616
-5376      2216
-5824      1896
-6272      1624
-6720      1408
-7168      1240
-7616      1096
-8064       984
diff --git a/logs/p4/sqr_kara.log b/logs/p4/sqr_kara.log
deleted file mode 100644
index 13e4f3e5785c5260ed2082e2ac85d55cce34e58e..0000000000000000000000000000000000000000
--- a/logs/p4/sqr_kara.log
+++ /dev/null
@@ -1,17 +0,0 @@
-896    127456
-1344     63752
-1792     37920
-2240     25440
-2688     18200
-3136     13728
-3584     10968
-4032      9072
-4480      7608
-4928      6440
-5376      5528
-5824      4768
-6272      4328
-6720      3888
-7168      3504
-7616      3176
-8064      2896
diff --git a/logs/p4/sub.log b/logs/p4/sub.log
deleted file mode 100644
index 424de325a22b673d5719f2353d72de8c4234d5b4..0000000000000000000000000000000000000000
--- a/logs/p4/sub.log
+++ /dev/null
@@ -1,16 +0,0 @@
-224   7355896
-448   6162880
-672   5218984
-896   4622776
-1120   3999320
-1344   3629480
-1568   3290384
-1792   2954752
-2016   2737056
-2240   2563320
-2464   2451928
-2688   2310920
-2912   2139048
-3136   2034080
-3360   1890800
-3584   1808624
diff --git a/logs/sqr.log b/logs/sqr.log
index 81fa612fdf45a215a5578f5432ce03e9e3278275..3e175acd9630ca339d8dc643e8d99ed0e6f62783 100644
--- a/logs/sqr.log
+++ b/logs/sqr.log
@@ -1,33 +1,143 @@
-922    471095
-1147    337137
-1366    254327
-1596    199732
-1819    161225
-2044    132852
-2268    111493
-2490     94864
-2715     81745
-2940     71187
-3162     62575
-3387     55418
-3612     14540
-3836     12944
-4060     11627
-4281     10546
-4508      9502
-4730      8688
-4954      7937
-5180      7273
-5402      6701
-5627      6189
-5850      5733
-6076      5310
-6300      4933
-6522      4631
-6748      4313
-6971      4064
-7196      3801
-7420      3576
-7642      3388
-7868      3191
-8092      3020
+139       806
+195      1212
+252      1604
+307      2260
+364      2892
+420      3308
+476      4152
+532      4814
+588      5754
+644      6684
+700      7226
+756      8324
+808      9092
+866     10068
+924     11204
+976     12918
+1036     13656
+1092     15248
+1148     15956
+1204     17270
+1260     19894
+1316     20516
+1370     21864
+1428     25554
+1483     26138
+1540     27086
+1596     29246
+1652     32210
+1707     32704
+1764     35142
+1820     39050
+1876     39256
+1931     41574
+1985     45070
+2044     46352
+2099     48114
+2155     51332
+2212     53268
+2267     55890
+2324     59054
+2380     60206
+2434     63540
+2491     66084
+2547     68590
+2604     74332
+2660     74784
+2715     77974
+2772     79924
+2826     82914
+2884     87210
+2929     89076
+2996     92480
+3052     96814
+3108     99990
+3162    102550
+3219    105396
+3276    109284
+3332    113752
+3387    116628
+3444    120782
+3500    122938
+3556    127940
+3612    303656
+3667    312212
+3724    324376
+3779    329204
+3833    340910
+3892    353850
+3943    362348
+4003    367780
+4056    380448
+4114    393616
+4172    404104
+4227    415148
+4284    409770
+4339    436648
+4394    442970
+4451    463096
+4507    472056
+4564    485780
+4616    496286
+4675    507612
+4732    519524
+4788    536768
+4843    542754
+4899    553090
+4956    571986
+5012    586340
+5068    599606
+5124    613670
+5179    624256
+5235    636266
+5292    655518
+5348    668142
+5403    677266
+5460    696040
+5516    712772
+5570    723942
+5628    739052
+5684    755350
+5739    769962
+5790    775258
+5851    790128
+5908    814536
+5962    827278
+6018    844510
+6076    851606
+6130    865748
+6188    894752
+6244    900474
+6300    928174
+6356    928440
+6410    957758
+6468    981134
+6524    994088
+6580   1011124
+6636   1027178
+6692   1045466
+6747   1056910
+6804   1083784
+6860   1104706
+6915   1116450
+6972   1137894
+7028   1154670
+7084   1158064
+7138   1188734
+7196   1214218
+7249   1226822
+7307   1247528
+7363   1255338
+7420   1291104
+7475   1297940
+7532   1324994
+7587   1340274
+7644   1342596
+7698   1381418
+7756   1382904
+7812   1432588
+7867   1443632
+7922   1465092
+7979   1496804
+8036   1520142
+8092   1539566
diff --git a/logs/sqr_kara.log b/logs/sqr_kara.log
index 3b547cf347964164d8217469139e09485ce6ca45..cafe45824cb91152bc9e9944e304bf89e7817807 100644
--- a/logs/sqr_kara.log
+++ b/logs/sqr_kara.log
@@ -1,33 +1,33 @@
-922    470930
-1148    337217
-1372    254433
-1596    199827
-1820    161204
-2043    132871
-2267    111522
-2488     94932
-2714     81814
-2939     71231
-3164     62616
-3385     55467
-3611     44426
-3836     40695
-4060     37391
-4283     34371
-4508     31779
-4732     29499
-4956     27426
-5177     25598
-5403     23944
-5628     22416
-5851     21052
-6076     19781
-6299     18588
-6523     17539
-6746     16618
-6972     15705
-7196     13582
-7420     13004
-7643     12496
-7868     11963
-8092     11497
+922     11272
+1148     16004
+1370     21958
+1596     28684
+1817     37832
+2044     46386
+2262     56218
+2492     66388
+2716     77478
+2940     89380
+3163    103680
+3385    116274
+3612    135334
+3836    151332
+4057    164938
+4284    183178
+4508    198864
+4731    215222
+4954    231986
+5180    251660
+5404    269414
+5626    288454
+5850    307806
+6076    329458
+6299    347726
+6523    369864
+6748    387832
+6971    413010
+7194    453310
+7415    476936
+7643    497118
+7867    521394
+8091    540224
diff --git a/logs/sub.log b/logs/sub.log
index f1ade94b6af1af485feee808ed528a059287301c..cf2bcd6af333ebbef5b21e7dec5c4bda46e4afce 100644
--- a/logs/sub.log
+++ b/logs/sub.log
@@ -1,16 +1,16 @@
-224  16370431
-448  13327848
-672  11009401
-896   9125342
-1120   7930419
-1344   7114040
-1568   6506998
-1792   5899346
-2016   5435327
-2240   5038931
-2464   4696364
-2688   4425678
-2912   4134476
-3136   3913280
-3360   3692536
-3584   3505219
+224       216
+448       324
+672       428
+896       532
+1120       648
+1344       766
+1568       862
+1792       928
+2016      1070
+2240      1128
+2464      1250
+2688      1344
+2912      1436
+3136      1542
+3360      1628
+3584      1696
diff --git a/makefile b/makefile
index 07b7842b31ec87aa3670c0d056393a3c74136620..4fe2256906f816801bfa0a1593d91ff65d2a556d 100644
--- a/makefile
+++ b/makefile
@@ -12,7 +12,10 @@ CFLAGS += -O3 -funroll-loops
 #x86 optimizations [should be valid for any GCC install though]
 CFLAGS  += -fomit-frame-pointer
 
-VERSION=0.30
+#debug
+#CFLAGS += -g3
+
+VERSION=0.32
 
 default: libtommath.a
 
@@ -20,7 +23,7 @@ default: libtommath.a
 LIBNAME=libtommath.a
 HEADERS=tommath.h
 
-#LIBPATH-The directory for libtomcrypt to be installed to.
+#LIBPATH-The directory for libtommath to be installed to.
 #INCPATH-The directory to install the header files for libtommath.
 #DATAPATH-The directory to install the pdf docs.
 DESTDIR=
@@ -50,14 +53,38 @@ bn_mp_toom_mul.o bn_mp_toom_sqr.o bn_mp_div_3.o bn_s_mp_exptmod.o \
 bn_mp_reduce_2k.o bn_mp_reduce_is_2k.o bn_mp_reduce_2k_setup.o \
 bn_mp_radix_smap.o bn_mp_read_radix.o bn_mp_toradix.o bn_mp_radix_size.o \
 bn_mp_fread.o bn_mp_fwrite.o bn_mp_cnt_lsb.o bn_error.o \
-bn_mp_init_multi.o bn_mp_clear_multi.o bn_prime_sizes_tab.o bn_mp_exteuclid.o bn_mp_toradix_n.o \
+bn_mp_init_multi.o bn_mp_clear_multi.o bn_mp_exteuclid.o bn_mp_toradix_n.o \
 bn_mp_prime_random_ex.o bn_mp_get_int.o bn_mp_sqrt.o bn_mp_is_square.o bn_mp_init_set.o \
-bn_mp_init_set_int.o
+bn_mp_init_set_int.o bn_mp_invmod_slow.o bn_mp_prime_rabin_miller_trials.o
 
 libtommath.a:  $(OBJECTS)
 	$(AR) $(ARFLAGS) libtommath.a $(OBJECTS)
 	ranlib libtommath.a
 
+
+#make a profiled library (takes a while!!!)
+#
+# This will build the library with profile generation
+# then run the test demo and rebuild the library.
+# 
+# So far I've seen improvements in the MP math
+profiled:
+	make CFLAGS="$(CFLAGS) -fprofile-arcs -DTESTING" timing
+	./ltmtest
+	rm -f *.a *.o ltmtest
+	make CFLAGS="$(CFLAGS) -fbranch-probabilities"
+
+#make a single object profiled library 
+profiled_single:
+	perl gen.pl
+	$(CC) $(CFLAGS) -fprofile-arcs -DTESTING -c mpi.c -o mpi.o
+	$(CC) $(CFLAGS) -DTESTING -DTIMER demo/timing.c mpi.o -o ltmtest
+	./ltmtest
+	rm -f *.o ltmtest
+	$(CC) $(CFLAGS) -fbranch-probabilities -DTESTING -c mpi.c -o mpi.o
+	$(AR) $(ARFLAGS) libtommath.a mpi.o
+	ranlib libtommath.a	
+
 install: libtommath.a
 	install -d -g root -o root $(DESTDIR)$(LIBPATH)
 	install -d -g root -o root $(DESTDIR)$(INCPATH)
@@ -71,7 +98,7 @@ mtest: test
 	cd mtest ; $(CC) $(CFLAGS) mtest.c -o mtest -s
         
 timing: libtommath.a
-	$(CC) $(CFLAGS) -DTIMER demo/demo.c libtommath.a -o ltmtest -s
+	$(CC) $(CFLAGS) -DTIMER demo/timing.c libtommath.a -o ltmtest -s
 
 # makes the LTM book DVI file, requires tetex, perl and makeindex [part of tetex I think]
 docdvi: tommath.src
@@ -106,10 +133,14 @@ mandvi: bn.tex
 manual:	mandvi
 	pdflatex bn >/dev/null
 	rm -f bn.aux bn.dvi bn.log bn.idx bn.lof bn.out bn.toc
-	
+
+pretty: 
+	perl pretty.build
+
 clean:
 	rm -f *.bat *.pdf *.o *.a *.obj *.lib *.exe *.dll etclib/*.o demo/demo.o test ltmtest mpitest mtest/mtest mtest/mtest.exe \
-        *.idx *.toc *.log *.aux *.dvi *.lof *.ind *.ilg *.ps *.log *.s mpi.c 
+        *.idx *.toc *.log *.aux *.dvi *.lof *.ind *.ilg *.ps *.log *.s mpi.c *.da *.dyn *.dpi tommath.tex `find -type f | grep [~] | xargs` *.lo *.la
+	rm -rf .libs
 	cd etc ; make clean
 	cd pics ; make clean
 
diff --git a/makefile.bcc b/makefile.bcc
index 6874d2f3f87f8ed095a6905251c72717abdc2759..775e9ff2a3ad413be43ca65eb08c0d784823029c 100644
--- a/makefile.bcc
+++ b/makefile.bcc
@@ -29,8 +29,9 @@ bn_mp_toom_mul.obj bn_mp_toom_sqr.obj bn_mp_div_3.obj bn_s_mp_exptmod.obj \
 bn_mp_reduce_2k.obj bn_mp_reduce_is_2k.obj bn_mp_reduce_2k_setup.obj \
 bn_mp_radix_smap.obj bn_mp_read_radix.obj bn_mp_toradix.obj bn_mp_radix_size.obj \
 bn_mp_fread.obj bn_mp_fwrite.obj bn_mp_cnt_lsb.obj bn_error.obj \
-bn_mp_init_multi.obj bn_mp_clear_multi.obj bn_prime_sizes_tab.obj bn_mp_exteuclid.obj bn_mp_toradix_n.obj \
-bn_mp_prime_random_ex.obj bn_mp_get_int.obj bn_mp_sqrt.obj bn_mp_is_square.obj
+bn_mp_init_multi.obj bn_mp_clear_multi.obj bn_mp_exteuclid.obj bn_mp_toradix_n.obj \
+bn_mp_prime_random_ex.obj bn_mp_get_int.obj bn_mp_sqrt.obj bn_mp_is_square.obj \
+bn_mp_init_set.obj bn_mp_init_set_int.obj bn_mp_invmod_slow.obj bn_mp_prime_rabin_miller_trials.obj
 
 TARGET = libtommath.lib
 
diff --git a/makefile.cygwin_dll b/makefile.cygwin_dll
index e5ab814c4b99837dcd109519725880756f49ae77..c90e5d99f6174da55c2b6b663dc8e85fafc3cdd4 100644
--- a/makefile.cygwin_dll
+++ b/makefile.cygwin_dll
@@ -34,8 +34,9 @@ bn_mp_toom_mul.o bn_mp_toom_sqr.o bn_mp_div_3.o bn_s_mp_exptmod.o \
 bn_mp_reduce_2k.o bn_mp_reduce_is_2k.o bn_mp_reduce_2k_setup.o \
 bn_mp_radix_smap.o bn_mp_read_radix.o bn_mp_toradix.o bn_mp_radix_size.o \
 bn_mp_fread.o bn_mp_fwrite.o bn_mp_cnt_lsb.o bn_error.o \
-bn_mp_init_multi.o bn_mp_clear_multi.o bn_prime_sizes_tab.o bn_mp_exteuclid.o bn_mp_toradix_n.o \
-bn_mp_prime_random_ex.o bn_mp_get_int.o bn_mp_sqrt.o bn_mp_is_square.o
+bn_mp_init_multi.o bn_mp_clear_multi.o bn_mp_exteuclid.o bn_mp_toradix_n.o \
+bn_mp_prime_random_ex.o bn_mp_get_int.o bn_mp_sqrt.o bn_mp_is_square.o bn_mp_init_set.o \
+bn_mp_init_set_int.o bn_mp_invmod_slow.o bn_mp_prime_rabin_miller_trials.o
 
 # make a Windows DLL via Cygwin
 windll:  $(OBJECTS)
diff --git a/makefile.icc b/makefile.icc
new file mode 100644
index 0000000000000000000000000000000000000000..09117b7492a2d3901921dd5af2560897edfbf18e
--- /dev/null
+++ b/makefile.icc
@@ -0,0 +1,110 @@
+#Makefile for ICC
+#
+#Tom St Denis
+CC=icc
+
+CFLAGS  +=  -I./
+
+# optimize for SPEED
+#
+# -mcpu= can be pentium, pentiumpro (covers PII through PIII) or pentium4
+# -ax?   specifies make code specifically for ? but compatible with IA-32
+# -x?    specifies compile solely for ? [not specifically IA-32 compatible]
+#
+# where ? is 
+#   K - PIII
+#   W - first P4 [Williamette]
+#   N - P4 Northwood
+#   P - P4 Prescott
+#   B - Blend of P4 and PM [mobile]
+#
+# Default to just generic max opts
+CFLAGS += -O3 -xN
+
+default: libtommath.a
+
+#default files to install
+LIBNAME=libtommath.a
+HEADERS=tommath.h
+
+#LIBPATH-The directory for libtomcrypt to be installed to.
+#INCPATH-The directory to install the header files for libtommath.
+#DATAPATH-The directory to install the pdf docs.
+DESTDIR=
+LIBPATH=/usr/lib
+INCPATH=/usr/include
+DATAPATH=/usr/share/doc/libtommath/pdf
+
+OBJECTS=bncore.o bn_mp_init.o bn_mp_clear.o bn_mp_exch.o bn_mp_grow.o bn_mp_shrink.o \
+bn_mp_clamp.o bn_mp_zero.o  bn_mp_set.o bn_mp_set_int.o bn_mp_init_size.o bn_mp_copy.o \
+bn_mp_init_copy.o bn_mp_abs.o bn_mp_neg.o bn_mp_cmp_mag.o bn_mp_cmp.o bn_mp_cmp_d.o \
+bn_mp_rshd.o bn_mp_lshd.o bn_mp_mod_2d.o bn_mp_div_2d.o bn_mp_mul_2d.o bn_mp_div_2.o \
+bn_mp_mul_2.o bn_s_mp_add.o bn_s_mp_sub.o bn_fast_s_mp_mul_digs.o bn_s_mp_mul_digs.o \
+bn_fast_s_mp_mul_high_digs.o bn_s_mp_mul_high_digs.o bn_fast_s_mp_sqr.o bn_s_mp_sqr.o \
+bn_mp_add.o bn_mp_sub.o bn_mp_karatsuba_mul.o bn_mp_mul.o bn_mp_karatsuba_sqr.o \
+bn_mp_sqr.o bn_mp_div.o bn_mp_mod.o bn_mp_add_d.o bn_mp_sub_d.o bn_mp_mul_d.o \
+bn_mp_div_d.o bn_mp_mod_d.o bn_mp_expt_d.o bn_mp_addmod.o bn_mp_submod.o \
+bn_mp_mulmod.o bn_mp_sqrmod.o bn_mp_gcd.o bn_mp_lcm.o bn_fast_mp_invmod.o bn_mp_invmod.o \
+bn_mp_reduce.o bn_mp_montgomery_setup.o bn_fast_mp_montgomery_reduce.o bn_mp_montgomery_reduce.o \
+bn_mp_exptmod_fast.o bn_mp_exptmod.o bn_mp_2expt.o bn_mp_n_root.o bn_mp_jacobi.o bn_reverse.o \
+bn_mp_count_bits.o bn_mp_read_unsigned_bin.o bn_mp_read_signed_bin.o bn_mp_to_unsigned_bin.o \
+bn_mp_to_signed_bin.o bn_mp_unsigned_bin_size.o bn_mp_signed_bin_size.o  \
+bn_mp_xor.o bn_mp_and.o bn_mp_or.o bn_mp_rand.o bn_mp_montgomery_calc_normalization.o \
+bn_mp_prime_is_divisible.o bn_prime_tab.o bn_mp_prime_fermat.o bn_mp_prime_miller_rabin.o \
+bn_mp_prime_is_prime.o bn_mp_prime_next_prime.o bn_mp_dr_reduce.o \
+bn_mp_dr_is_modulus.o bn_mp_dr_setup.o bn_mp_reduce_setup.o \
+bn_mp_toom_mul.o bn_mp_toom_sqr.o bn_mp_div_3.o bn_s_mp_exptmod.o \
+bn_mp_reduce_2k.o bn_mp_reduce_is_2k.o bn_mp_reduce_2k_setup.o \
+bn_mp_radix_smap.o bn_mp_read_radix.o bn_mp_toradix.o bn_mp_radix_size.o \
+bn_mp_fread.o bn_mp_fwrite.o bn_mp_cnt_lsb.o bn_error.o \
+bn_mp_init_multi.o bn_mp_clear_multi.o bn_mp_exteuclid.o bn_mp_toradix_n.o \
+bn_mp_prime_random_ex.o bn_mp_get_int.o bn_mp_sqrt.o bn_mp_is_square.o bn_mp_init_set.o \
+bn_mp_init_set_int.o bn_mp_invmod_slow.o bn_mp_prime_rabin_miller_trials.o
+
+libtommath.a:  $(OBJECTS)
+	$(AR) $(ARFLAGS) libtommath.a $(OBJECTS)
+	ranlib libtommath.a
+
+#make a profiled library (takes a while!!!)
+#
+# This will build the library with profile generation
+# then run the test demo and rebuild the library.
+# 
+# So far I've seen improvements in the MP math
+profiled:
+	make -f makefile.icc CFLAGS="$(CFLAGS) -prof_gen -DTESTING" timing
+	./ltmtest
+	rm -f *.a *.o ltmtest
+	make -f makefile.icc CFLAGS="$(CFLAGS) -prof_use"
+
+#make a single object profiled library 
+profiled_single:
+	perl gen.pl
+	$(CC) $(CFLAGS) -prof_gen -DTESTING -c mpi.c -o mpi.o
+	$(CC) $(CFLAGS) -DTESTING -DTIMER demo/demo.c mpi.o -o ltmtest
+	./ltmtest
+	rm -f *.o ltmtest
+	$(CC) $(CFLAGS) -prof_use -ip -DTESTING -c mpi.c -o mpi.o
+	$(AR) $(ARFLAGS) libtommath.a mpi.o
+	ranlib libtommath.a	
+
+install: libtommath.a
+	install -d -g root -o root $(DESTDIR)$(LIBPATH)
+	install -d -g root -o root $(DESTDIR)$(INCPATH)
+	install -g root -o root $(LIBNAME) $(DESTDIR)$(LIBPATH)
+	install -g root -o root $(HEADERS) $(DESTDIR)$(INCPATH)
+
+test: libtommath.a demo/demo.o
+	$(CC) demo/demo.o libtommath.a -o test
+	
+mtest: test	
+	cd mtest ; $(CC) $(CFLAGS) mtest.c -o mtest
+        
+timing: libtommath.a
+	$(CC) $(CFLAGS) -DTIMER demo/timing.c libtommath.a -o ltmtest
+
+clean:
+	rm -f *.bat *.pdf *.o *.a *.obj *.lib *.exe *.dll etclib/*.o demo/demo.o test ltmtest mpitest mtest/mtest mtest/mtest.exe \
+        *.idx *.toc *.log *.aux *.dvi *.lof *.ind *.ilg *.ps *.log *.s mpi.c *.il etc/*.il *.dyn
+	cd etc ; make clean
+	cd pics ; make clean
diff --git a/makefile.msvc b/makefile.msvc
index beeb77e4985d072f129c012c6ae51fc04f455d5d..cf59943812cfcbfc312fce19ce7fb120b41ebaa7 100644
--- a/makefile.msvc
+++ b/makefile.msvc
@@ -28,8 +28,9 @@ bn_mp_toom_mul.obj bn_mp_toom_sqr.obj bn_mp_div_3.obj bn_s_mp_exptmod.obj \
 bn_mp_reduce_2k.obj bn_mp_reduce_is_2k.obj bn_mp_reduce_2k_setup.obj \
 bn_mp_radix_smap.obj bn_mp_read_radix.obj bn_mp_toradix.obj bn_mp_radix_size.obj \
 bn_mp_fread.obj bn_mp_fwrite.obj bn_mp_cnt_lsb.obj bn_error.obj \
-bn_mp_init_multi.obj bn_mp_clear_multi.obj bn_prime_sizes_tab.obj bn_mp_exteuclid.obj bn_mp_toradix_n.obj \
-bn_mp_prime_random_ex.obj bn_mp_get_int.obj bn_mp_sqrt.obj bn_mp_is_square.obj
+bn_mp_init_multi.obj bn_mp_clear_multi.obj bn_mp_exteuclid.obj bn_mp_toradix_n.obj \
+bn_mp_prime_random_ex.obj bn_mp_get_int.obj bn_mp_sqrt.obj bn_mp_is_square.obj \
+bn_mp_init_set.obj bn_mp_init_set_int.obj bn_mp_invmod_slow.obj bn_mp_prime_rabin_miller_trials.obj
 
 library: $(OBJECTS)
 	lib /out:tommath.lib $(OBJECTS)
diff --git a/makefile.shared b/makefile.shared
new file mode 100644
index 0000000000000000000000000000000000000000..96bbf3297cd0f953537955e279cc71af3ea1c15e
--- /dev/null
+++ b/makefile.shared
@@ -0,0 +1,74 @@
+#Makefile for GCC
+#
+#Tom St Denis
+VERSION=0:32
+
+CC = libtool --mode=compile gcc
+
+CFLAGS  +=  -I./ -Wall -W -Wshadow -Wsign-compare
+
+#for speed 
+CFLAGS += -O3 -funroll-loops
+
+#for size 
+#CFLAGS += -Os
+
+#x86 optimizations [should be valid for any GCC install though]
+CFLAGS  += -fomit-frame-pointer
+
+default: libtommath.la
+
+#default files to install
+LIBNAME=libtommath.la
+HEADERS=tommath.h
+
+#LIBPATH-The directory for libtommath to be installed to.
+#INCPATH-The directory to install the header files for libtommath.
+#DATAPATH-The directory to install the pdf docs.
+DESTDIR=
+LIBPATH=/usr/lib
+INCPATH=/usr/include
+DATAPATH=/usr/share/doc/libtommath/pdf
+
+OBJECTS=bncore.o bn_mp_init.o bn_mp_clear.o bn_mp_exch.o bn_mp_grow.o bn_mp_shrink.o \
+bn_mp_clamp.o bn_mp_zero.o  bn_mp_set.o bn_mp_set_int.o bn_mp_init_size.o bn_mp_copy.o \
+bn_mp_init_copy.o bn_mp_abs.o bn_mp_neg.o bn_mp_cmp_mag.o bn_mp_cmp.o bn_mp_cmp_d.o \
+bn_mp_rshd.o bn_mp_lshd.o bn_mp_mod_2d.o bn_mp_div_2d.o bn_mp_mul_2d.o bn_mp_div_2.o \
+bn_mp_mul_2.o bn_s_mp_add.o bn_s_mp_sub.o bn_fast_s_mp_mul_digs.o bn_s_mp_mul_digs.o \
+bn_fast_s_mp_mul_high_digs.o bn_s_mp_mul_high_digs.o bn_fast_s_mp_sqr.o bn_s_mp_sqr.o \
+bn_mp_add.o bn_mp_sub.o bn_mp_karatsuba_mul.o bn_mp_mul.o bn_mp_karatsuba_sqr.o \
+bn_mp_sqr.o bn_mp_div.o bn_mp_mod.o bn_mp_add_d.o bn_mp_sub_d.o bn_mp_mul_d.o \
+bn_mp_div_d.o bn_mp_mod_d.o bn_mp_expt_d.o bn_mp_addmod.o bn_mp_submod.o \
+bn_mp_mulmod.o bn_mp_sqrmod.o bn_mp_gcd.o bn_mp_lcm.o bn_fast_mp_invmod.o bn_mp_invmod.o \
+bn_mp_reduce.o bn_mp_montgomery_setup.o bn_fast_mp_montgomery_reduce.o bn_mp_montgomery_reduce.o \
+bn_mp_exptmod_fast.o bn_mp_exptmod.o bn_mp_2expt.o bn_mp_n_root.o bn_mp_jacobi.o bn_reverse.o \
+bn_mp_count_bits.o bn_mp_read_unsigned_bin.o bn_mp_read_signed_bin.o bn_mp_to_unsigned_bin.o \
+bn_mp_to_signed_bin.o bn_mp_unsigned_bin_size.o bn_mp_signed_bin_size.o  \
+bn_mp_xor.o bn_mp_and.o bn_mp_or.o bn_mp_rand.o bn_mp_montgomery_calc_normalization.o \
+bn_mp_prime_is_divisible.o bn_prime_tab.o bn_mp_prime_fermat.o bn_mp_prime_miller_rabin.o \
+bn_mp_prime_is_prime.o bn_mp_prime_next_prime.o bn_mp_dr_reduce.o \
+bn_mp_dr_is_modulus.o bn_mp_dr_setup.o bn_mp_reduce_setup.o \
+bn_mp_toom_mul.o bn_mp_toom_sqr.o bn_mp_div_3.o bn_s_mp_exptmod.o \
+bn_mp_reduce_2k.o bn_mp_reduce_is_2k.o bn_mp_reduce_2k_setup.o \
+bn_mp_radix_smap.o bn_mp_read_radix.o bn_mp_toradix.o bn_mp_radix_size.o \
+bn_mp_fread.o bn_mp_fwrite.o bn_mp_cnt_lsb.o bn_error.o \
+bn_mp_init_multi.o bn_mp_clear_multi.o bn_mp_exteuclid.o bn_mp_toradix_n.o \
+bn_mp_prime_random_ex.o bn_mp_get_int.o bn_mp_sqrt.o bn_mp_is_square.o bn_mp_init_set.o \
+bn_mp_init_set_int.o bn_mp_invmod_slow.o bn_mp_prime_rabin_miller_trials.o
+
+libtommath.la:  $(OBJECTS)
+	libtool --mode=link gcc *.lo -o libtommath.la -rpath $(LIBPATH) -version-info $(VERSION)
+	libtool --mode=link gcc *.o -o libtommath.a 
+	libtool --mode=install install -c libtommath.la $(LIBPATH)/libtommath.la
+	install -d -g root -o root $(DESTDIR)$(INCPATH)
+	install -g root -o root $(HEADERS) $(DESTDIR)$(INCPATH)
+
+test: libtommath.a demo/demo.o
+	gcc $(CFLAGS) -c demo/demo.c -o demo/demo.o
+	libtool --mode=link gcc -o test demo/demo.o libtommath.la
+	
+mtest: test	
+	cd mtest ; gcc $(CFLAGS) mtest.c -o mtest -s
+        
+timing: libtommath.la
+	gcc $(CFLAGS) -DTIMER demo/timing.c libtommath.a -o ltmtest -s
diff --git a/mtest/mtest.c b/mtest/mtest.c
index 2c248256cb6b5e20065a1f46de8c0fe03166192d..ef0e0933b2c741fc2d09f1a54c72b6a55c76d467 100644
--- a/mtest/mtest.c
+++ b/mtest/mtest.c
@@ -58,7 +58,7 @@ void rand_num2(mp_int *a)
    int n, size;
    unsigned char buf[2048];
 
-   size = 1 + ((fgetc(rng)<<8) + fgetc(rng)) % 97;
+   size = 10 + ((fgetc(rng)<<8) + fgetc(rng)) % 97;
    buf[0] = (fgetc(rng)&1)?1:0;
    fread(buf+1, 1, size, rng);
    while (buf[1] == 0) buf[1] = fgetc(rng);
@@ -109,11 +109,12 @@ int main(void)
 
    t1 = clock();
    for (;;) {
+#if 0
       if (clock() - t1 > CLOCKS_PER_SEC) {
          sleep(2);
          t1 = clock();
       }
-
+#endif
        n = fgetc(rng) % 15;
 
    if (n == 0) {
diff --git a/poster.pdf b/poster.pdf
index 3731bd2bdaa53bb2052ef8258e5d97368b1b413c..60999daef1a6991036d5a8836971cca7180f9aac 100644
Binary files a/poster.pdf and b/poster.pdf differ
diff --git a/pre_gen/mpi.c b/pre_gen/mpi.c
index 1f9997f1ae43f9b3283549f676c07501e05294d9..78a73f09073e6dc12ee0f51a40cdf8d54ee7494f 100644
--- a/pre_gen/mpi.c
+++ b/pre_gen/mpi.c
@@ -1,4 +1,6 @@
 /* Start: bn_error.c */
+#include <tommath.h>
+#ifdef BN_ERROR_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -13,7 +15,6 @@
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 static const struct {
      int code;
@@ -40,10 +41,13 @@ char *mp_error_to_string(int code)
    return "Invalid error code";
 }
 
+#endif
 
 /* End: bn_error.c */
 
 /* Start: bn_fast_mp_invmod.c */
+#include <tommath.h>
+#ifdef BN_FAST_MP_INVMOD_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -58,12 +62,11 @@ char *mp_error_to_string(int code)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* computes the modular inverse via binary extended euclidean algorithm, 
  * that is c = 1/a mod b 
  *
- * Based on mp_invmod except this is optimized for the case where b is 
+ * Based on slow invmod except this is optimized for the case where b is 
  * odd as per HAC Note 14.64 on pp. 610
  */
 int
@@ -187,10 +190,13 @@ top:
 __ERR:mp_clear_multi (&x, &y, &u, &v, &B, &D, NULL);
   return res;
 }
+#endif
 
 /* End: bn_fast_mp_invmod.c */
 
 /* Start: bn_fast_mp_montgomery_reduce.c */
+#include <tommath.h>
+#ifdef BN_FAST_MP_MONTGOMERY_REDUCE_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -205,11 +211,10 @@ __ERR:mp_clear_multi (&x, &y, &u, &v, &B, &D, NULL);
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* computes xR**-1 == x (mod N) via Montgomery Reduction
  *
- * This is an optimized implementation of mp_montgomery_reduce
+ * This is an optimized implementation of montgomery_reduce
  * which uses the comba method to quickly calculate the columns of the
  * reduction.
  *
@@ -358,10 +363,13 @@ fast_mp_montgomery_reduce (mp_int * x, mp_int * n, mp_digit rho)
   }
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_fast_mp_montgomery_reduce.c */
 
 /* Start: bn_fast_s_mp_mul_digs.c */
+#include <tommath.h>
+#ifdef BN_FAST_S_MP_MUL_DIGS_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -376,7 +384,6 @@ fast_mp_montgomery_reduce (mp_int * x, mp_int * n, mp_digit rho)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* Fast (comba) multiplier
  *
@@ -397,8 +404,9 @@ fast_mp_montgomery_reduce (mp_int * x, mp_int * n, mp_digit rho)
 int
 fast_s_mp_mul_digs (mp_int * a, mp_int * b, mp_int * c, int digs)
 {
-  int     olduse, res, pa, ix;
-  mp_word W[MP_WARRAY];
+  int     olduse, res, pa, ix, iz;
+  mp_digit W[MP_WARRAY];
+  register mp_word  _W;
 
   /* grow the destination as required */
   if (c->alloc < digs) {
@@ -407,82 +415,52 @@ fast_s_mp_mul_digs (mp_int * a, mp_int * b, mp_int * c, int digs)
     }
   }
 
-  /* clear temp buf (the columns) */
-  memset (W, 0, sizeof (mp_word) * digs);
-
-  /* calculate the columns */
-  pa = a->used;
-  for (ix = 0; ix < pa; ix++) {
-    /* this multiplier has been modified to allow you to 
-     * control how many digits of output are produced.  
-     * So at most we want to make upto "digs" digits of output.
-     *
-     * this adds products to distinct columns (at ix+iy) of W
-     * note that each step through the loop is not dependent on
-     * the previous which means the compiler can easily unroll
-     * the loop without scheduling problems
-     */
-    {
-      register mp_digit tmpx, *tmpy;
-      register mp_word *_W;
-      register int iy, pb;
+  /* number of output digits to produce */
+  pa = MIN(digs, a->used + b->used);
 
-      /* alias for the the word on the left e.g. A[ix] * A[iy] */
-      tmpx = a->dp[ix];
+  /* clear the carry */
+  _W = 0;
+  for (ix = 0; ix <= pa; ix++) { 
+      int      tx, ty;
+      int      iy;
+      mp_digit *tmpx, *tmpy;
 
-      /* alias for the right side */
-      tmpy = b->dp;
+      /* get offsets into the two bignums */
+      ty = MIN(b->used-1, ix);
+      tx = ix - ty;
 
-      /* alias for the columns, each step through the loop adds a new
-         term to each column
-       */
-      _W = W + ix;
+      /* setup temp aliases */
+      tmpx = a->dp + tx;
+      tmpy = b->dp + ty;
 
-      /* the number of digits is limited by their placement.  E.g.
-         we avoid multiplying digits that will end up above the # of
-         digits of precision requested
+      /* this is the number of times the loop will iterrate, essentially its 
+         while (tx++ < a->used && ty-- >= 0) { ... }
        */
-      pb = MIN (b->used, digs - ix);
+      iy = MIN(a->used-tx, ty+1);
 
-      for (iy = 0; iy < pb; iy++) {
-        *_W++ += ((mp_word)tmpx) * ((mp_word)*tmpy++);
+      /* execute loop */
+      for (iz = 0; iz < iy; ++iz) {
+         _W += ((mp_word)*tmpx++)*((mp_word)*tmpy--);
       }
-    }
 
+      /* store term */
+      W[ix] = ((mp_digit)_W) & MP_MASK;
+
+      /* make next carry */
+      _W = _W >> ((mp_word)DIGIT_BIT);
   }
 
   /* setup dest */
-  olduse = c->used;
+  olduse  = c->used;
   c->used = digs;
 
   {
     register mp_digit *tmpc;
-
-    /* At this point W[] contains the sums of each column.  To get the
-     * correct result we must take the extra bits from each column and
-     * carry them down
-     *
-     * Note that while this adds extra code to the multiplier it 
-     * saves time since the carry propagation is removed from the 
-     * above nested loop.This has the effect of reducing the work 
-     * from N*(N+N*c)==N**2 + c*N**2 to N**2 + N*c where c is the 
-     * cost of the shifting.  On very small numbers this is slower 
-     * but on most cryptographic size numbers it is faster.
-     *
-     * In this particular implementation we feed the carries from
-     * behind which means when the loop terminates we still have one
-     * last digit to copy
-     */
     tmpc = c->dp;
-    for (ix = 1; ix < digs; ix++) {
-      /* forward the carry from the previous temp */
-      W[ix] += (W[ix - 1] >> ((mp_word) DIGIT_BIT));
-
+    for (ix = 0; ix < digs; ix++) {
       /* now extract the previous digit [below the carry] */
-      *tmpc++ = (mp_digit) (W[ix - 1] & ((mp_word) MP_MASK));
+      *tmpc++ = W[ix];
     }
-    /* fetch the last digit */
-    *tmpc++ = (mp_digit) (W[digs - 1] & ((mp_word) MP_MASK));
 
     /* clear unused digits [that existed in the old copy of c] */
     for (; ix < olduse; ix++) {
@@ -492,10 +470,13 @@ fast_s_mp_mul_digs (mp_int * a, mp_int * b, mp_int * c, int digs)
   mp_clamp (c);
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_fast_s_mp_mul_digs.c */
 
 /* Start: bn_fast_s_mp_mul_high_digs.c */
+#include <tommath.h>
+#ifdef BN_FAST_S_MP_MUL_HIGH_DIGS_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -510,10 +491,9 @@ fast_s_mp_mul_digs (mp_int * a, mp_int * b, mp_int * c, int digs)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
- #include <tommath.h>
 
-/* this is a modified version of fast_s_mp_mul_digs that only produces
- * output digits *above* digs.  See the comments for fast_s_mp_mul_digs
+/* this is a modified version of fast_s_mul_digs that only produces
+ * output digits *above* digs.  See the comments for fast_s_mul_digs
  * to see how it works.
  *
  * This is used in the Barrett reduction since for one of the multiplications
@@ -524,80 +504,78 @@ fast_s_mp_mul_digs (mp_int * a, mp_int * b, mp_int * c, int digs)
 int
 fast_s_mp_mul_high_digs (mp_int * a, mp_int * b, mp_int * c, int digs)
 {
-  int     oldused, newused, res, pa, pb, ix;
-  mp_word W[MP_WARRAY];
+  int     olduse, res, pa, ix, iz;
+  mp_digit W[MP_WARRAY];
+  mp_word  _W;
 
-  /* calculate size of product and allocate more space if required */
-  newused = a->used + b->used + 1;
-  if (c->alloc < newused) {
-    if ((res = mp_grow (c, newused)) != MP_OKAY) {
+  /* grow the destination as required */
+  pa = a->used + b->used;
+  if (c->alloc < pa) {
+    if ((res = mp_grow (c, pa)) != MP_OKAY) {
       return res;
     }
   }
 
-  /* like the other comba method we compute the columns first */
-  pa = a->used;
-  pb = b->used;
-  memset (W + digs, 0, (pa + pb + 1 - digs) * sizeof (mp_word));
-  for (ix = 0; ix < pa; ix++) {
-    {
-      register mp_digit tmpx, *tmpy;
-      register int iy;
-      register mp_word *_W;
+  /* number of output digits to produce */
+  pa = a->used + b->used;
+  _W = 0;
+  for (ix = digs; ix <= pa; ix++) { 
+      int      tx, ty, iy;
+      mp_digit *tmpx, *tmpy;
 
-      /* work todo, that is we only calculate digits that are at "digs" or above  */
-      iy = digs - ix;
+      /* get offsets into the two bignums */
+      ty = MIN(b->used-1, ix);
+      tx = ix - ty;
 
-      /* copy of word on the left of A[ix] * B[iy] */
-      tmpx = a->dp[ix];
+      /* setup temp aliases */
+      tmpx = a->dp + tx;
+      tmpy = b->dp + ty;
 
-      /* alias for right side */
-      tmpy = b->dp + iy;
-     
-      /* alias for the columns of output.  Offset to be equal to or above the 
-       * smallest digit place requested 
+      /* this is the number of times the loop will iterrate, essentially its 
+         while (tx++ < a->used && ty-- >= 0) { ... }
        */
-      _W = W + digs;     
-      
-      /* skip cases below zero where ix > digs */
-      if (iy < 0) {
-         iy    = abs(iy);
-         tmpy += iy;
-         _W   += iy;
-         iy    = 0;
-      }
+      iy = MIN(a->used-tx, ty+1);
 
-      /* compute column products for digits above the minimum */
-      for (; iy < pb; iy++) {
-         *_W++ += ((mp_word) tmpx) * ((mp_word)*tmpy++);
+      /* execute loop */
+      for (iz = 0; iz < iy; iz++) {
+         _W += ((mp_word)*tmpx++)*((mp_word)*tmpy--);
       }
-    }
+
+      /* store term */
+      W[ix] = ((mp_digit)_W) & MP_MASK;
+
+      /* make next carry */
+      _W = _W >> ((mp_word)DIGIT_BIT);
   }
 
   /* setup dest */
-  oldused = c->used;
-  c->used = newused;
+  olduse  = c->used;
+  c->used = pa;
 
-  /* now convert the array W downto what we need
-   *
-   * See comments in bn_fast_s_mp_mul_digs.c
-   */
-  for (ix = digs + 1; ix < newused; ix++) {
-    W[ix] += (W[ix - 1] >> ((mp_word) DIGIT_BIT));
-    c->dp[ix - 1] = (mp_digit) (W[ix - 1] & ((mp_word) MP_MASK));
-  }
-  c->dp[newused - 1] = (mp_digit) (W[newused - 1] & ((mp_word) MP_MASK));
+  {
+    register mp_digit *tmpc;
+
+    tmpc = c->dp + digs;
+    for (ix = digs; ix <= pa; ix++) {
+      /* now extract the previous digit [below the carry] */
+      *tmpc++ = W[ix];
+    }
 
-  for (; ix < oldused; ix++) {
-    c->dp[ix] = 0;
+    /* clear unused digits [that existed in the old copy of c] */
+    for (; ix < olduse; ix++) {
+      *tmpc++ = 0;
+    }
   }
   mp_clamp (c);
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_fast_s_mp_mul_high_digs.c */
 
 /* Start: bn_fast_s_mp_sqr.c */
+#include <tommath.h>
+#ifdef BN_FAST_S_MP_SQR_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -612,7 +590,6 @@ fast_s_mp_mul_high_digs (mp_int * a, mp_int * b, mp_int * c, int digs)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* fast squaring
  *
@@ -631,116 +608,107 @@ fast_s_mp_mul_high_digs (mp_int * a, mp_int * b, mp_int * c, int digs)
  * Based on Algorithm 14.16 on pp.597 of HAC.
  *
  */
+/* the jist of squaring...
+
+you do like mult except the offset of the tmpx [one that starts closer to zero]
+can't equal the offset of tmpy.  So basically you set up iy like before then you min it with
+(ty-tx) so that it never happens.  You double all those you add in the inner loop
+
+After that loop you do the squares and add them in.
+
+Remove W2 and don't memset W
+
+*/
+
 int fast_s_mp_sqr (mp_int * a, mp_int * b)
 {
-  int     olduse, newused, res, ix, pa;
-  mp_word W2[MP_WARRAY], W[MP_WARRAY];
+  int       olduse, res, pa, ix, iz;
+  mp_digit   W[MP_WARRAY], *tmpx;
+  mp_word   W1;
 
-  /* calculate size of product and allocate as required */
-  pa = a->used;
-  newused = pa + pa + 1;
-  if (b->alloc < newused) {
-    if ((res = mp_grow (b, newused)) != MP_OKAY) {
+  /* grow the destination as required */
+  pa = a->used + a->used;
+  if (b->alloc < pa) {
+    if ((res = mp_grow (b, pa)) != MP_OKAY) {
       return res;
     }
   }
 
-  /* zero temp buffer (columns)
-   * Note that there are two buffers.  Since squaring requires
-   * a outer and inner product and the inner product requires
-   * computing a product and doubling it (a relatively expensive
-   * op to perform n**2 times if you don't have to) the inner and
-   * outer products are computed in different buffers.  This way
-   * the inner product can be doubled using n doublings instead of
-   * n**2
-   */
-  memset (W,  0, newused * sizeof (mp_word));
-  memset (W2, 0, newused * sizeof (mp_word));
+  /* number of output digits to produce */
+  W1 = 0;
+  for (ix = 0; ix <= pa; ix++) { 
+      int      tx, ty, iy;
+      mp_word  _W;
+      mp_digit *tmpy;
 
-  /* This computes the inner product.  To simplify the inner N**2 loop
-   * the multiplication by two is done afterwards in the N loop.
-   */
-  for (ix = 0; ix < pa; ix++) {
-    /* compute the outer product
-     *
-     * Note that every outer product is computed
-     * for a particular column only once which means that
-     * there is no need todo a double precision addition
-     * into the W2[] array.
-     */
-    W2[ix + ix] = ((mp_word)a->dp[ix]) * ((mp_word)a->dp[ix]);
+      /* clear counter */
+      _W = 0;
 
-    {
-      register mp_digit tmpx, *tmpy;
-      register mp_word *_W;
-      register int iy;
+      /* get offsets into the two bignums */
+      ty = MIN(a->used-1, ix);
+      tx = ix - ty;
+
+      /* setup temp aliases */
+      tmpx = a->dp + tx;
+      tmpy = a->dp + ty;
 
-      /* copy of left side */
-      tmpx = a->dp[ix];
+      /* this is the number of times the loop will iterrate, essentially its 
+         while (tx++ < a->used && ty-- >= 0) { ... }
+       */
+      iy = MIN(a->used-tx, ty+1);
+
+      /* now for squaring tx can never equal ty 
+       * we halve the distance since they approach at a rate of 2x
+       * and we have to round because odd cases need to be executed
+       */
+      iy = MIN(iy, (ty-tx+1)>>1);
 
-      /* alias for right side */
-      tmpy = a->dp + (ix + 1);
+      /* execute loop */
+      for (iz = 0; iz < iy; iz++) {
+         _W += ((mp_word)*tmpx++)*((mp_word)*tmpy--);
+      }
 
-      /* the column to store the result in */
-      _W = W + (ix + ix + 1);
+      /* double the inner product and add carry */
+      _W = _W + _W + W1;
 
-      /* inner products */
-      for (iy = ix + 1; iy < pa; iy++) {
-          *_W++ += ((mp_word)tmpx) * ((mp_word)*tmpy++);
+      /* even columns have the square term in them */
+      if ((ix&1) == 0) {
+         _W += ((mp_word)a->dp[ix>>1])*((mp_word)a->dp[ix>>1]);
       }
-    }
+
+      /* store it */
+      W[ix] = _W;
+
+      /* make next carry */
+      W1 = _W >> ((mp_word)DIGIT_BIT);
   }
 
   /* setup dest */
   olduse  = b->used;
-  b->used = newused;
+  b->used = a->used+a->used;
 
-  /* now compute digits
-   *
-   * We have to double the inner product sums, add in the
-   * outer product sums, propagate carries and convert
-   * to single precision.
-   */
   {
-    register mp_digit *tmpb;
-
-    /* double first value, since the inner products are
-     * half of what they should be
-     */
-    W[0] += W[0] + W2[0];
-
+    mp_digit *tmpb;
     tmpb = b->dp;
-    for (ix = 1; ix < newused; ix++) {
-      /* double/add next digit */
-      W[ix] += W[ix] + W2[ix];
-
-      /* propagate carry forwards [from the previous digit] */
-      W[ix] = W[ix] + (W[ix - 1] >> ((mp_word) DIGIT_BIT));
-
-      /* store the current digit now that the carry isn't
-       * needed
-       */
-      *tmpb++ = (mp_digit) (W[ix - 1] & ((mp_word) MP_MASK));
+    for (ix = 0; ix < pa; ix++) {
+      *tmpb++ = W[ix] & MP_MASK;
     }
-    /* set the last value.  Note even if the carry is zero
-     * this is required since the next step will not zero
-     * it if b originally had a value at b->dp[2*a.used]
-     */
-    *tmpb++ = (mp_digit) (W[(newused) - 1] & ((mp_word) MP_MASK));
 
-    /* clear high digits of b if there were any originally */
+    /* clear unused digits [that existed in the old copy of c] */
     for (; ix < olduse; ix++) {
       *tmpb++ = 0;
     }
   }
-
   mp_clamp (b);
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_fast_s_mp_sqr.c */
 
 /* Start: bn_mp_2expt.c */
+#include <tommath.h>
+#ifdef BN_MP_2EXPT_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -755,7 +723,6 @@ int fast_s_mp_sqr (mp_int * a, mp_int * b)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* computes a = 2**b 
  *
@@ -779,14 +746,17 @@ mp_2expt (mp_int * a, int b)
   a->used = b / DIGIT_BIT + 1;
 
   /* put the single bit in its place */
-  a->dp[b / DIGIT_BIT] = 1 << (b % DIGIT_BIT);
+  a->dp[b / DIGIT_BIT] = ((mp_digit)1) << (b % DIGIT_BIT);
 
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_mp_2expt.c */
 
 /* Start: bn_mp_abs.c */
+#include <tommath.h>
+#ifdef BN_MP_ABS_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -801,7 +771,6 @@ mp_2expt (mp_int * a, int b)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* b = |a| 
  *
@@ -824,10 +793,13 @@ mp_abs (mp_int * a, mp_int * b)
 
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_mp_abs.c */
 
 /* Start: bn_mp_add.c */
+#include <tommath.h>
+#ifdef BN_MP_ADD_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -842,7 +814,6 @@ mp_abs (mp_int * a, mp_int * b)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* high level addition (handles signs) */
 int mp_add (mp_int * a, mp_int * b, mp_int * c)
@@ -875,10 +846,13 @@ int mp_add (mp_int * a, mp_int * b, mp_int * c)
   return res;
 }
 
+#endif
 
 /* End: bn_mp_add.c */
 
 /* Start: bn_mp_add_d.c */
+#include <tommath.h>
+#ifdef BN_MP_ADD_D_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -893,7 +867,6 @@ int mp_add (mp_int * a, mp_int * b, mp_int * c)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* single digit addition */
 int
@@ -982,10 +955,13 @@ mp_add_d (mp_int * a, mp_digit b, mp_int * c)
   return MP_OKAY;
 }
 
+#endif
 
 /* End: bn_mp_add_d.c */
 
 /* Start: bn_mp_addmod.c */
+#include <tommath.h>
+#ifdef BN_MP_ADDMOD_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -1000,7 +976,6 @@ mp_add_d (mp_int * a, mp_digit b, mp_int * c)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* d = a + b (mod c) */
 int
@@ -1021,10 +996,13 @@ mp_addmod (mp_int * a, mp_int * b, mp_int * c, mp_int * d)
   mp_clear (&t);
   return res;
 }
+#endif
 
 /* End: bn_mp_addmod.c */
 
 /* Start: bn_mp_and.c */
+#include <tommath.h>
+#ifdef BN_MP_AND_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -1039,7 +1017,6 @@ mp_addmod (mp_int * a, mp_int * b, mp_int * c, mp_int * d)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* AND two ints together */
 int
@@ -1076,10 +1053,13 @@ mp_and (mp_int * a, mp_int * b, mp_int * c)
   mp_clear (&t);
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_mp_and.c */
 
 /* Start: bn_mp_clamp.c */
+#include <tommath.h>
+#ifdef BN_MP_CLAMP_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -1094,7 +1074,6 @@ mp_and (mp_int * a, mp_int * b, mp_int * c)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* trim unused digits 
  *
@@ -1118,10 +1097,13 @@ mp_clamp (mp_int * a)
     a->sign = MP_ZPOS;
   }
 }
+#endif
 
 /* End: bn_mp_clamp.c */
 
 /* Start: bn_mp_clear.c */
+#include <tommath.h>
+#ifdef BN_MP_CLEAR_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -1136,16 +1118,19 @@ mp_clamp (mp_int * a)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* clear one (frees)  */
 void
 mp_clear (mp_int * a)
 {
+  int i;
+
   /* only do anything if a hasn't been freed previously */
   if (a->dp != NULL) {
     /* first zero the digits */
-    memset (a->dp, 0, sizeof (mp_digit) * a->used);
+    for (i = 0; i < a->used; i++) {
+        a->dp[i] = 0;
+    }
 
     /* free ram */
     XFREE(a->dp);
@@ -1156,10 +1141,13 @@ mp_clear (mp_int * a)
     a->sign  = MP_ZPOS;
   }
 }
+#endif
 
 /* End: bn_mp_clear.c */
 
 /* Start: bn_mp_clear_multi.c */
+#include <tommath.h>
+#ifdef BN_MP_CLEAR_MULTI_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -1174,7 +1162,6 @@ mp_clear (mp_int * a)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 #include <stdarg.h>
 
 void mp_clear_multi(mp_int *mp, ...) 
@@ -1188,10 +1175,13 @@ void mp_clear_multi(mp_int *mp, ...)
     }
     va_end(args);
 }
+#endif
 
 /* End: bn_mp_clear_multi.c */
 
 /* Start: bn_mp_cmp.c */
+#include <tommath.h>
+#ifdef BN_MP_CMP_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -1206,7 +1196,6 @@ void mp_clear_multi(mp_int *mp, ...)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* compare two ints (signed)*/
 int
@@ -1229,10 +1218,13 @@ mp_cmp (mp_int * a, mp_int * b)
      return mp_cmp_mag(a, b);
   }
 }
+#endif
 
 /* End: bn_mp_cmp.c */
 
 /* Start: bn_mp_cmp_d.c */
+#include <tommath.h>
+#ifdef BN_MP_CMP_D_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -1247,7 +1239,6 @@ mp_cmp (mp_int * a, mp_int * b)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* compare a digit */
 int mp_cmp_d(mp_int * a, mp_digit b)
@@ -1271,10 +1262,13 @@ int mp_cmp_d(mp_int * a, mp_digit b)
     return MP_EQ;
   }
 }
+#endif
 
 /* End: bn_mp_cmp_d.c */
 
 /* Start: bn_mp_cmp_mag.c */
+#include <tommath.h>
+#ifdef BN_MP_CMP_MAG_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -1289,7 +1283,6 @@ int mp_cmp_d(mp_int * a, mp_digit b)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* compare maginitude of two ints (unsigned) */
 int mp_cmp_mag (mp_int * a, mp_int * b)
@@ -1324,10 +1317,13 @@ int mp_cmp_mag (mp_int * a, mp_int * b)
   }
   return MP_EQ;
 }
+#endif
 
 /* End: bn_mp_cmp_mag.c */
 
 /* Start: bn_mp_cnt_lsb.c */
+#include <tommath.h>
+#ifdef BN_MP_CNT_LSB_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -1342,7 +1338,6 @@ int mp_cmp_mag (mp_int * a, mp_int * b)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 static const int lnz[16] = { 
    4, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0
@@ -1375,10 +1370,13 @@ int mp_cnt_lsb(mp_int *a)
    return x;
 }
 
+#endif
 
 /* End: bn_mp_cnt_lsb.c */
 
 /* Start: bn_mp_copy.c */
+#include <tommath.h>
+#ifdef BN_MP_COPY_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -1393,7 +1391,6 @@ int mp_cnt_lsb(mp_int *a)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* copy, b = a */
 int
@@ -1441,10 +1438,13 @@ mp_copy (mp_int * a, mp_int * b)
   b->sign = a->sign;
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_mp_copy.c */
 
 /* Start: bn_mp_count_bits.c */
+#include <tommath.h>
+#ifdef BN_MP_COUNT_BITS_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -1459,7 +1459,6 @@ mp_copy (mp_int * a, mp_int * b)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* returns the number of bits in an int */
 int
@@ -1484,10 +1483,13 @@ mp_count_bits (mp_int * a)
   }
   return r;
 }
+#endif
 
 /* End: bn_mp_count_bits.c */
 
 /* Start: bn_mp_div.c */
+#include <tommath.h>
+#ifdef BN_MP_DIV_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -1502,7 +1504,78 @@ mp_count_bits (mp_int * a)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
+
+#ifdef BN_MP_DIV_SMALL
+
+/* slower bit-bang division... also smaller */
+int mp_div(mp_int * a, mp_int * b, mp_int * c, mp_int * d)
+{
+   mp_int ta, tb, tq, q;
+   int    res, n, n2;
+
+  /* is divisor zero ? */
+  if (mp_iszero (b) == 1) {
+    return MP_VAL;
+  }
+
+  /* if a < b then q=0, r = a */
+  if (mp_cmp_mag (a, b) == MP_LT) {
+    if (d != NULL) {
+      res = mp_copy (a, d);
+    } else {
+      res = MP_OKAY;
+    }
+    if (c != NULL) {
+      mp_zero (c);
+    }
+    return res;
+  }
+	
+  /* init our temps */
+  if ((res = mp_init_multi(&ta, &tb, &tq, &q, NULL) != MP_OKAY)) {
+     return res;
+  }
+
+
+  mp_set(&tq, 1);
+  n = mp_count_bits(a) - mp_count_bits(b);
+  if (((res = mp_copy(a, &ta)) != MP_OKAY) ||
+      ((res = mp_copy(b, &tb)) != MP_OKAY) || 
+      ((res = mp_mul_2d(&tb, n, &tb)) != MP_OKAY) ||
+      ((res = mp_mul_2d(&tq, n, &tq)) != MP_OKAY)) {
+      goto __ERR;
+  }
+
+  while (n-- >= 0) {
+     if (mp_cmp(&tb, &ta) != MP_GT) {
+        if (((res = mp_sub(&ta, &tb, &ta)) != MP_OKAY) ||
+            ((res = mp_add(&q, &tq, &q)) != MP_OKAY)) {
+           goto __ERR;
+        }
+     }
+     if (((res = mp_div_2d(&tb, 1, &tb, NULL)) != MP_OKAY) ||
+         ((res = mp_div_2d(&tq, 1, &tq, NULL)) != MP_OKAY)) {
+           goto __ERR;
+     }
+  }
+
+  /* now q == quotient and ta == remainder */
+  n  = a->sign;
+  n2 = (a->sign == b->sign ? MP_ZPOS : MP_NEG);
+  if (c != NULL) {
+     mp_exch(c, &q);
+     c->sign  = n2;
+  }
+  if (d != NULL) {
+     mp_exch(d, &ta);
+     d->sign = n;
+  }
+__ERR:
+   mp_clear_multi(&ta, &tb, &tq, &q, NULL);
+   return res;
+}
+
+#else
 
 /* integer signed division. 
  * c*b + d == a [e.g. a/b, c=quotient, d=remainder]
@@ -1677,7 +1750,7 @@ int mp_div (mp_int * a, mp_int * b, mp_int * c, mp_int * d)
    */
   
   /* get sign before writing to c */
-  x.sign = a->sign;
+  x.sign = x.used == 0 ? MP_ZPOS : a->sign;
 
   if (c != NULL) {
     mp_clamp (&q);
@@ -1700,9 +1773,15 @@ __Q:mp_clear (&q);
   return res;
 }
 
+#endif
+
+#endif
+
 /* End: bn_mp_div.c */
 
 /* Start: bn_mp_div_2.c */
+#include <tommath.h>
+#ifdef BN_MP_DIV_2_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -1717,7 +1796,6 @@ __Q:mp_clear (&q);
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* b = a/2 */
 int mp_div_2(mp_int * a, mp_int * b)
@@ -1765,10 +1843,13 @@ int mp_div_2(mp_int * a, mp_int * b)
   mp_clamp (b);
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_mp_div_2.c */
 
 /* Start: bn_mp_div_2d.c */
+#include <tommath.h>
+#ifdef BN_MP_DIV_2D_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -1783,7 +1864,6 @@ int mp_div_2(mp_int * a, mp_int * b)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* shift right by a certain bit count (store quotient in c, optional remainder in d) */
 int mp_div_2d (mp_int * a, int b, mp_int * c, mp_int * d)
@@ -1860,10 +1940,13 @@ int mp_div_2d (mp_int * a, int b, mp_int * c, mp_int * d)
   mp_clear (&t);
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_mp_div_2d.c */
 
 /* Start: bn_mp_div_3.c */
+#include <tommath.h>
+#ifdef BN_MP_DIV_3_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -1878,7 +1961,6 @@ int mp_div_2d (mp_int * a, int b, mp_int * c, mp_int * d)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* divide by three (based on routine from MPI and the GMP manual) */
 int
@@ -1937,10 +2019,13 @@ mp_div_3 (mp_int * a, mp_int *c, mp_digit * d)
   return res;
 }
 
+#endif
 
 /* End: bn_mp_div_3.c */
 
 /* Start: bn_mp_div_d.c */
+#include <tommath.h>
+#ifdef BN_MP_DIV_D_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -1955,7 +2040,6 @@ mp_div_3 (mp_int * a, mp_int *c, mp_digit * d)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 static int s_is_power_of_two(mp_digit b, int *p)
 {
@@ -1997,7 +2081,7 @@ int mp_div_d (mp_int * a, mp_digit b, mp_int * c, mp_digit * d)
   /* power of two ? */
   if (s_is_power_of_two(b, &ix) == 1) {
      if (d != NULL) {
-        *d = a->dp[0] & ((1<<ix) - 1);
+        *d = a->dp[0] & ((((mp_digit)1)<<ix) - 1);
      }
      if (c != NULL) {
         return mp_div_2d(a, ix, c, NULL);
@@ -2005,10 +2089,12 @@ int mp_div_d (mp_int * a, mp_digit b, mp_int * c, mp_digit * d)
      return MP_OKAY;
   }
 
+#ifdef BN_MP_DIV_3_C
   /* three? */
   if (b == 3) {
      return mp_div_3(a, c, d);
   }
+#endif
 
   /* no easy answer [c'est la vie].  Just division */
   if ((res = mp_init_size(&q, a->used)) != MP_OKAY) {
@@ -2043,10 +2129,13 @@ int mp_div_d (mp_int * a, mp_digit b, mp_int * c, mp_digit * d)
   return res;
 }
 
+#endif
 
 /* End: bn_mp_div_d.c */
 
 /* Start: bn_mp_dr_is_modulus.c */
+#include <tommath.h>
+#ifdef BN_MP_DR_IS_MODULUS_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -2061,7 +2150,6 @@ int mp_div_d (mp_int * a, mp_digit b, mp_int * c, mp_digit * d)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* determines if a number is a valid DR modulus */
 int mp_dr_is_modulus(mp_int *a)
@@ -2084,10 +2172,13 @@ int mp_dr_is_modulus(mp_int *a)
    return 1;
 }
 
+#endif
 
 /* End: bn_mp_dr_is_modulus.c */
 
 /* Start: bn_mp_dr_reduce.c */
+#include <tommath.h>
+#ifdef BN_MP_DR_REDUCE_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -2102,7 +2193,6 @@ int mp_dr_is_modulus(mp_int *a)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* reduce "x" in place modulo "n" using the Diminished Radix algorithm.
  *
@@ -2176,10 +2266,13 @@ top:
   }
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_mp_dr_reduce.c */
 
 /* Start: bn_mp_dr_setup.c */
+#include <tommath.h>
+#ifdef BN_MP_DR_SETUP_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -2194,7 +2287,6 @@ top:
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* determines the setup value */
 void mp_dr_setup(mp_int *a, mp_digit *d)
@@ -2206,10 +2298,13 @@ void mp_dr_setup(mp_int *a, mp_digit *d)
         ((mp_word)a->dp[0]));
 }
 
+#endif
 
 /* End: bn_mp_dr_setup.c */
 
 /* Start: bn_mp_exch.c */
+#include <tommath.h>
+#ifdef BN_MP_EXCH_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -2224,7 +2319,6 @@ void mp_dr_setup(mp_int *a, mp_digit *d)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* swap the elements of two integers, for cases where you can't simply swap the 
  * mp_int pointers around
@@ -2238,10 +2332,13 @@ mp_exch (mp_int * a, mp_int * b)
   *a = *b;
   *b = t;
 }
+#endif
 
 /* End: bn_mp_exch.c */
 
 /* Start: bn_mp_expt_d.c */
+#include <tommath.h>
+#ifdef BN_MP_EXPT_D_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -2256,7 +2353,6 @@ mp_exch (mp_int * a, mp_int * b)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* calculate c = a**b  using a square-multiply algorithm */
 int mp_expt_d (mp_int * a, mp_digit b, mp_int * c)
@@ -2293,10 +2389,13 @@ int mp_expt_d (mp_int * a, mp_digit b, mp_int * c)
   mp_clear (&g);
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_mp_expt_d.c */
 
 /* Start: bn_mp_exptmod.c */
+#include <tommath.h>
+#ifdef BN_MP_EXPTMOD_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -2311,7 +2410,6 @@ int mp_expt_d (mp_int * a, mp_digit b, mp_int * c)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 
 /* this is a shell function that calls either the normal or Montgomery
@@ -2330,6 +2428,7 @@ int mp_exptmod (mp_int * G, mp_int * X, mp_int * P, mp_int * Y)
 
   /* if exponent X is negative we have to recurse */
   if (X->sign == MP_NEG) {
+#ifdef BN_MP_INVMOD_C
      mp_int tmpG, tmpX;
      int err;
 
@@ -2356,29 +2455,51 @@ int mp_exptmod (mp_int * G, mp_int * X, mp_int * P, mp_int * Y)
      err = mp_exptmod(&tmpG, &tmpX, P, Y);
      mp_clear_multi(&tmpG, &tmpX, NULL);
      return err;
+#else 
+     /* no invmod */
+     return MP_VAL
+#endif
   }
 
+#ifdef BN_MP_DR_IS_MODULUS_C
   /* is it a DR modulus? */
   dr = mp_dr_is_modulus(P);
+#else
+  dr = 0;
+#endif
 
+#ifdef BN_MP_REDUCE_IS_2K_C
   /* if not, is it a uDR modulus? */
   if (dr == 0) {
      dr = mp_reduce_is_2k(P) << 1;
   }
+#endif
     
   /* if the modulus is odd or dr != 0 use the fast method */
+#ifdef BN_MP_EXPTMOD_FAST_C
   if (mp_isodd (P) == 1 || dr !=  0) {
     return mp_exptmod_fast (G, X, P, Y, dr);
   } else {
+#endif
+#ifdef BN_S_MP_EXPTMOD_C
     /* otherwise use the generic Barrett reduction technique */
     return s_mp_exptmod (G, X, P, Y);
+#else
+    /* no exptmod for evens */
+    return MP_VAL;
+#endif
+#ifdef BN_MP_EXPTMOD_FAST_C
   }
+#endif
 }
 
+#endif
 
 /* End: bn_mp_exptmod.c */
 
 /* Start: bn_mp_exptmod_fast.c */
+#include <tommath.h>
+#ifdef BN_MP_EXPTMOD_FAST_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -2393,7 +2514,6 @@ int mp_exptmod (mp_int * G, mp_int * X, mp_int * P, mp_int * Y)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* computes Y == G**X mod P, HAC pp.616, Algorithm 14.85
  *
@@ -2465,29 +2585,52 @@ mp_exptmod_fast (mp_int * G, mp_int * X, mp_int * P, mp_int * Y, int redmode)
 
   /* determine and setup reduction code */
   if (redmode == 0) {
+#ifdef BN_MP_MONTGOMERY_SETUP_C     
      /* now setup montgomery  */
      if ((err = mp_montgomery_setup (P, &mp)) != MP_OKAY) {
         goto __M;
      }
+#else
+     err = MP_VAL;
+     goto __M;
+#endif
 
      /* automatically pick the comba one if available (saves quite a few calls/ifs) */
+#ifdef BN_FAST_MP_MONTGOMERY_REDUCE_C
      if (((P->used * 2 + 1) < MP_WARRAY) &&
           P->used < (1 << ((CHAR_BIT * sizeof (mp_word)) - (2 * DIGIT_BIT)))) {
         redux = fast_mp_montgomery_reduce;
-     } else {
+     } else 
+#endif
+     {
+#ifdef BN_MP_MONTGOMERY_REDUCE_C
         /* use slower baseline Montgomery method */
         redux = mp_montgomery_reduce;
+#else
+        err = MP_VAL;
+        goto __M;
+#endif
      }
   } else if (redmode == 1) {
+#if defined(BN_MP_DR_SETUP_C) && defined(BN_MP_DR_REDUCE_C)
      /* setup DR reduction for moduli of the form B**k - b */
      mp_dr_setup(P, &mp);
      redux = mp_dr_reduce;
+#else
+     err = MP_VAL;
+     goto __M;
+#endif
   } else {
+#if defined(BN_MP_REDUCE_2K_SETUP_C) && defined(BN_MP_REDUCE_2K_C)
      /* setup DR reduction for moduli of the form 2**k - b */
      if ((err = mp_reduce_2k_setup(P, &mp)) != MP_OKAY) {
         goto __M;
      }
      redux = mp_reduce_2k;
+#else
+     err = MP_VAL;
+     goto __M;
+#endif
   }
 
   /* setup result */
@@ -2497,16 +2640,21 @@ mp_exptmod_fast (mp_int * G, mp_int * X, mp_int * P, mp_int * Y, int redmode)
 
   /* create M table
    *
-   * The M table contains powers of the input base, e.g. M[x] = G^x mod P
+
    *
    * The first half of the table is not computed though accept for M[0] and M[1]
    */
 
   if (redmode == 0) {
+#ifdef BN_MP_MONTGOMERY_CALC_NORMALIZATION_C
      /* now we need R mod m */
      if ((err = mp_montgomery_calc_normalization (&res, P)) != MP_OKAY) {
        goto __RES;
      }
+#else 
+     err = MP_VAL;
+     goto __RES;
+#endif
 
      /* now set M[1] to G * R mod m */
      if ((err = mp_mulmod (G, &res, P, &M[1])) != MP_OKAY) {
@@ -2650,7 +2798,7 @@ mp_exptmod_fast (mp_int * G, mp_int * X, mp_int * P, mp_int * Y, int redmode)
       * to reduce one more time to cancel out the factor
       * of R.
       */
-     if ((err = mp_montgomery_reduce (&res, P, mp)) != MP_OKAY) {
+     if ((err = redux(&res, P, mp)) != MP_OKAY) {
        goto __RES;
      }
   }
@@ -2666,10 +2814,14 @@ __M:
   }
   return err;
 }
+#endif
+
 
 /* End: bn_mp_exptmod_fast.c */
 
 /* Start: bn_mp_exteuclid.c */
+#include <tommath.h>
+#ifdef BN_MP_EXTEUCLID_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -2684,7 +2836,6 @@ __M:
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* Extended euclidean algorithm of (a, b) produces 
    a*u1 + b*u2 = u3
@@ -2739,10 +2890,13 @@ int mp_exteuclid(mp_int *a, mp_int *b, mp_int *U1, mp_int *U2, mp_int *U3)
 _ERR: mp_clear_multi(&u1, &u2, &u3, &v1, &v2, &v3, &t1, &t2, &t3, &q, &tmp, NULL);
    return err;
 }
+#endif
 
 /* End: bn_mp_exteuclid.c */
 
 /* Start: bn_mp_fread.c */
+#include <tommath.h>
+#ifdef BN_MP_FREAD_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -2757,7 +2911,6 @@ _ERR: mp_clear_multi(&u1, &u2, &u3, &v1, &v2, &v3, &t1, &t2, &t3, &q, &tmp, NULL
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* read a bigint from a file stream in ASCII */
 int mp_fread(mp_int *a, int radix, FILE *stream)
@@ -2804,10 +2957,13 @@ int mp_fread(mp_int *a, int radix, FILE *stream)
    return MP_OKAY;
 }
 
+#endif
 
 /* End: bn_mp_fread.c */
 
 /* Start: bn_mp_fwrite.c */
+#include <tommath.h>
+#ifdef BN_MP_FWRITE_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -2822,7 +2978,6 @@ int mp_fread(mp_int *a, int radix, FILE *stream)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 int mp_fwrite(mp_int *a, int radix, FILE *stream)
 {
@@ -2854,10 +3009,13 @@ int mp_fwrite(mp_int *a, int radix, FILE *stream)
    return MP_OKAY;
 }
 
+#endif
 
 /* End: bn_mp_fwrite.c */
 
 /* Start: bn_mp_gcd.c */
+#include <tommath.h>
+#ifdef BN_MP_GCD_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -2872,7 +3030,6 @@ int mp_fwrite(mp_int *a, int radix, FILE *stream)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* Greatest Common Divisor using the binary method */
 int mp_gcd (mp_int * a, mp_int * b, mp_int * c)
@@ -2965,10 +3122,13 @@ __V:mp_clear (&u);
 __U:mp_clear (&v);
   return res;
 }
+#endif
 
 /* End: bn_mp_gcd.c */
 
 /* Start: bn_mp_get_int.c */
+#include <tommath.h>
+#ifdef BN_MP_GET_INT_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -2983,7 +3143,6 @@ __U:mp_clear (&v);
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* get the lower 32-bits of an mp_int */
 unsigned long mp_get_int(mp_int * a) 
@@ -3008,10 +3167,13 @@ unsigned long mp_get_int(mp_int * a)
   /* force result to 32-bits always so it is consistent on non 32-bit platforms */
   return res & 0xFFFFFFFFUL;
 }
+#endif
 
 /* End: bn_mp_get_int.c */
 
 /* Start: bn_mp_grow.c */
+#include <tommath.h>
+#ifdef BN_MP_GROW_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -3026,7 +3188,6 @@ unsigned long mp_get_int(mp_int * a)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* grow as required */
 int mp_grow (mp_int * a, int size)
@@ -3063,10 +3224,13 @@ int mp_grow (mp_int * a, int size)
   }
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_mp_grow.c */
 
 /* Start: bn_mp_init.c */
+#include <tommath.h>
+#ifdef BN_MP_INIT_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -3081,17 +3245,23 @@ int mp_grow (mp_int * a, int size)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
-/* init a new bigint */
+/* init a new mp_int */
 int mp_init (mp_int * a)
 {
+  int i;
+
   /* allocate memory required and clear it */
-  a->dp = OPT_CAST(mp_digit) XCALLOC (sizeof (mp_digit), MP_PREC);
+  a->dp = OPT_CAST(mp_digit) XMALLOC (sizeof (mp_digit) * MP_PREC);
   if (a->dp == NULL) {
     return MP_MEM;
   }
 
+  /* set the digits to zero */
+  for (i = 0; i < MP_PREC; i++) {
+      a->dp[i] = 0;
+  }
+
   /* set the used to zero, allocated digits to the default precision
    * and sign to positive */
   a->used  = 0;
@@ -3100,10 +3270,13 @@ int mp_init (mp_int * a)
 
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_mp_init.c */
 
 /* Start: bn_mp_init_copy.c */
+#include <tommath.h>
+#ifdef BN_MP_INIT_COPY_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -3118,7 +3291,6 @@ int mp_init (mp_int * a)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* creates "a" then copies b into it */
 int mp_init_copy (mp_int * a, mp_int * b)
@@ -3130,10 +3302,13 @@ int mp_init_copy (mp_int * a, mp_int * b)
   }
   return mp_copy (b, a);
 }
+#endif
 
 /* End: bn_mp_init_copy.c */
 
 /* Start: bn_mp_init_multi.c */
+#include <tommath.h>
+#ifdef BN_MP_INIT_MULTI_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -3148,7 +3323,6 @@ int mp_init_copy (mp_int * a, mp_int * b)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 #include <stdarg.h>
 
 int mp_init_multi(mp_int *mp, ...) 
@@ -3187,10 +3361,13 @@ int mp_init_multi(mp_int *mp, ...)
     return res;                /* Assumed ok, if error flagged above. */
 }
 
+#endif
 
 /* End: bn_mp_init_multi.c */
 
 /* Start: bn_mp_init_set.c */
+#include <tommath.h>
+#ifdef BN_MP_INIT_SET_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -3205,7 +3382,6 @@ int mp_init_multi(mp_int *mp, ...)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* initialize and set a digit */
 int mp_init_set (mp_int * a, mp_digit b)
@@ -3217,10 +3393,13 @@ int mp_init_set (mp_int * a, mp_digit b)
   mp_set(a, b);
   return err;
 }
+#endif
 
 /* End: bn_mp_init_set.c */
 
 /* Start: bn_mp_init_set_int.c */
+#include <tommath.h>
+#ifdef BN_MP_INIT_SET_INT_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -3235,7 +3414,6 @@ int mp_init_set (mp_int * a, mp_digit b)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* initialize and set a digit */
 int mp_init_set_int (mp_int * a, unsigned long b)
@@ -3246,10 +3424,13 @@ int mp_init_set_int (mp_int * a, unsigned long b)
   }
   return mp_set_int(a, b);
 }
+#endif
 
 /* End: bn_mp_init_set_int.c */
 
 /* Start: bn_mp_init_size.c */
+#include <tommath.h>
+#ifdef BN_MP_INIT_SIZE_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -3264,29 +3445,40 @@ int mp_init_set_int (mp_int * a, unsigned long b)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* init an mp_init for a given size */
 int mp_init_size (mp_int * a, int size)
 {
+  int x;
+
   /* pad size so there are always extra digits */
   size += (MP_PREC * 2) - (size % MP_PREC);	
   
   /* alloc mem */
-  a->dp = OPT_CAST(mp_digit) XCALLOC (sizeof (mp_digit), size);
+  a->dp = OPT_CAST(mp_digit) XMALLOC (sizeof (mp_digit) * size);
   if (a->dp == NULL) {
     return MP_MEM;
   }
+
+  /* set the members */
   a->used  = 0;
   a->alloc = size;
   a->sign  = MP_ZPOS;
 
+  /* zero the digits */
+  for (x = 0; x < size; x++) {
+      a->dp[x] = 0;
+  }
+
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_mp_init_size.c */
 
 /* Start: bn_mp_invmod.c */
+#include <tommath.h>
+#ifdef BN_MP_INVMOD_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -3301,24 +3493,61 @@ int mp_init_size (mp_int * a, int size)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* hac 14.61, pp608 */
 int mp_invmod (mp_int * a, mp_int * b, mp_int * c)
 {
-  mp_int  x, y, u, v, A, B, C, D;
-  int     res;
-
   /* b cannot be negative */
   if (b->sign == MP_NEG || mp_iszero(b) == 1) {
     return MP_VAL;
   }
 
+#ifdef BN_FAST_MP_INVMOD_C
   /* if the modulus is odd we can use a faster routine instead */
   if (mp_isodd (b) == 1) {
     return fast_mp_invmod (a, b, c);
   }
-  
+#endif
+
+#ifdef BN_MP_INVMOD_SLOW_C
+  return mp_invmod_slow(a, b, c);
+#endif
+
+  return MP_VAL;
+}
+#endif
+
+/* End: bn_mp_invmod.c */
+
+/* Start: bn_mp_invmod_slow.c */
+#include <tommath.h>
+#ifdef BN_MP_INVMOD_SLOW_C
+/* LibTomMath, multiple-precision integer library -- Tom St Denis
+ *
+ * LibTomMath is a library that provides multiple-precision
+ * integer arithmetic as well as number theoretic functionality.
+ *
+ * The library was designed directly after the MPI library by
+ * Michael Fromberger but has been written from scratch with
+ * additional optimizations in place.
+ *
+ * The library is free for all purposes without any express
+ * guarantee it works.
+ *
+ * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
+ */
+
+/* hac 14.61, pp608 */
+int mp_invmod_slow (mp_int * a, mp_int * b, mp_int * c)
+{
+  mp_int  x, y, u, v, A, B, C, D;
+  int     res;
+
+  /* b cannot be negative */
+  if (b->sign == MP_NEG || mp_iszero(b) == 1) {
+    return MP_VAL;
+  }
+
   /* init temps */
   if ((res = mp_init_multi(&x, &y, &u, &v, 
                            &A, &B, &C, &D, NULL)) != MP_OKAY) {
@@ -3461,10 +3690,13 @@ top:
 __ERR:mp_clear_multi (&x, &y, &u, &v, &A, &B, &C, &D, NULL);
   return res;
 }
+#endif
 
-/* End: bn_mp_invmod.c */
+/* End: bn_mp_invmod_slow.c */
 
 /* Start: bn_mp_is_square.c */
+#include <tommath.h>
+#ifdef BN_MP_IS_SQUARE_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -3479,7 +3711,6 @@ __ERR:mp_clear_multi (&x, &y, &u, &v, &A, &B, &C, &D, NULL);
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* Check if remainders are possible squares - fast exclude non-squares */
 static const char rem_128[128] = {
@@ -3536,7 +3767,7 @@ int mp_is_square(mp_int *arg,int *ret)
      return MP_OKAY;
   }
 
-  /* product of primes less than 2^31 */
+
   if ((res = mp_init_set_int(&t,11L*13L*17L*19L*23L*29L*31L)) != MP_OKAY) {
      return res;
   }
@@ -3568,10 +3799,13 @@ int mp_is_square(mp_int *arg,int *ret)
 ERR:mp_clear(&t);
   return res;
 }
+#endif
 
 /* End: bn_mp_is_square.c */
 
 /* Start: bn_mp_jacobi.c */
+#include <tommath.h>
+#ifdef BN_MP_JACOBI_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -3586,7 +3820,6 @@ ERR:mp_clear(&t);
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* computes the jacobi c = (a | n) (or Legendre if n is prime)
  * HAC pp. 73 Algorithm 2.149
@@ -3671,10 +3904,13 @@ __P1:mp_clear (&p1);
 __A1:mp_clear (&a1);
   return res;
 }
+#endif
 
 /* End: bn_mp_jacobi.c */
 
 /* Start: bn_mp_karatsuba_mul.c */
+#include <tommath.h>
+#ifdef BN_MP_KARATSUBA_MUL_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -3689,7 +3925,6 @@ __A1:mp_clear (&a1);
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* c = |a| * |b| using Karatsuba Multiplication using 
  * three half size multiplications
@@ -3753,9 +3988,6 @@ int mp_karatsuba_mul (mp_int * a, mp_int * b, mp_int * c)
     goto X0Y0;
 
   /* now shift the digits */
-  x0.sign = x1.sign = a->sign;
-  y0.sign = y1.sign = b->sign;
-
   x0.used = y0.used = B;
   x1.used = a->used - B;
   y1.used = b->used - B;
@@ -3839,10 +4071,13 @@ X0:mp_clear (&x0);
 ERR:
   return err;
 }
+#endif
 
 /* End: bn_mp_karatsuba_mul.c */
 
 /* Start: bn_mp_karatsuba_sqr.c */
+#include <tommath.h>
+#ifdef BN_MP_KARATSUBA_SQR_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -3857,12 +4092,11 @@ ERR:
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* Karatsuba squaring, computes b = a*a using three 
  * half size squarings
  *
- * See comments of mp_karatsuba_mul for details.  It 
+ * See comments of karatsuba_mul for details.  It 
  * is essentially the same algorithm but merely 
  * tuned to perform recursive squarings.
  */
@@ -3958,10 +4192,13 @@ X0:mp_clear (&x0);
 ERR:
   return err;
 }
+#endif
 
 /* End: bn_mp_karatsuba_sqr.c */
 
 /* Start: bn_mp_lcm.c */
+#include <tommath.h>
+#ifdef BN_MP_LCM_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -3976,7 +4213,6 @@ ERR:
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* computes least common multiple as |a*b|/(a, b) */
 int mp_lcm (mp_int * a, mp_int * b, mp_int * c)
@@ -4016,10 +4252,13 @@ __T:
   mp_clear_multi (&t1, &t2, NULL);
   return res;
 }
+#endif
 
 /* End: bn_mp_lcm.c */
 
 /* Start: bn_mp_lshd.c */
+#include <tommath.h>
+#ifdef BN_MP_LSHD_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -4034,7 +4273,6 @@ __T:
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* shift left a certain amount of digits */
 int mp_lshd (mp_int * a, int b)
@@ -4081,10 +4319,13 @@ int mp_lshd (mp_int * a, int b)
   }
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_mp_lshd.c */
 
 /* Start: bn_mp_mod.c */
+#include <tommath.h>
+#ifdef BN_MP_MOD_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -4099,7 +4340,6 @@ int mp_lshd (mp_int * a, int b)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* c = a mod b, 0 <= c < b */
 int
@@ -4127,10 +4367,13 @@ mp_mod (mp_int * a, mp_int * b, mp_int * c)
   mp_clear (&t);
   return res;
 }
+#endif
 
 /* End: bn_mp_mod.c */
 
 /* Start: bn_mp_mod_2d.c */
+#include <tommath.h>
+#ifdef BN_MP_MOD_2D_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -4145,7 +4388,6 @@ mp_mod (mp_int * a, mp_int * b, mp_int * c)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* calc a value mod 2**b */
 int
@@ -4180,10 +4422,13 @@ mp_mod_2d (mp_int * a, int b, mp_int * c)
   mp_clamp (c);
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_mp_mod_2d.c */
 
 /* Start: bn_mp_mod_d.c */
+#include <tommath.h>
+#ifdef BN_MP_MOD_D_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -4198,17 +4443,19 @@ mp_mod_2d (mp_int * a, int b, mp_int * c)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 int
 mp_mod_d (mp_int * a, mp_digit b, mp_digit * c)
 {
   return mp_div_d(a, b, NULL, c);
 }
+#endif
 
 /* End: bn_mp_mod_d.c */
 
 /* Start: bn_mp_montgomery_calc_normalization.c */
+#include <tommath.h>
+#ifdef BN_MP_MONTGOMERY_CALC_NORMALIZATION_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -4223,31 +4470,31 @@ mp_mod_d (mp_int * a, mp_digit b, mp_digit * c)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
-/* calculates a = B^n mod b for Montgomery reduction
- * Where B is the base [e.g. 2^DIGIT_BIT].
- * B^n mod b is computed by first computing
- * A = B^(n-1) which doesn't require a reduction but a simple OR.
- * then C = A * B = B^n is computed by performing upto DIGIT_BIT
+/*
  * shifts with subtractions when the result is greater than b.
  *
  * The method is slightly modified to shift B unconditionally upto just under
  * the leading bit of b.  This saves alot of multiple precision shifting.
  */
-int
-mp_montgomery_calc_normalization (mp_int * a, mp_int * b)
+int mp_montgomery_calc_normalization (mp_int * a, mp_int * b)
 {
   int     x, bits, res;
 
   /* how many bits of last digit does b use */
   bits = mp_count_bits (b) % DIGIT_BIT;
 
-  /* compute A = B^(n-1) * 2^(bits-1) */
-  if ((res = mp_2expt (a, (b->used - 1) * DIGIT_BIT + bits - 1)) != MP_OKAY) {
-    return res;
+
+  if (b->used > 1) {
+     if ((res = mp_2expt (a, (b->used - 1) * DIGIT_BIT + bits - 1)) != MP_OKAY) {
+        return res;
+     }
+  } else {
+     mp_set(a, 1);
+     bits = 1;
   }
 
+
   /* now compute C = A * B mod b */
   for (x = bits - 1; x < (int)DIGIT_BIT; x++) {
     if ((res = mp_mul_2 (a, a)) != MP_OKAY) {
@@ -4262,10 +4509,13 @@ mp_montgomery_calc_normalization (mp_int * a, mp_int * b)
 
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_mp_montgomery_calc_normalization.c */
 
 /* Start: bn_mp_montgomery_reduce.c */
+#include <tommath.h>
+#ifdef BN_MP_MONTGOMERY_REDUCE_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -4280,7 +4530,6 @@ mp_montgomery_calc_normalization (mp_int * a, mp_int * b)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* computes xR**-1 == x (mod N) via Montgomery Reduction */
 int
@@ -4291,7 +4540,7 @@ mp_montgomery_reduce (mp_int * x, mp_int * n, mp_digit rho)
 
   /* can the fast reduction [comba] method be used?
    *
-   * Note that unlike in mp_mul you're safely allowed *less*
+   * Note that unlike in mul you're safely allowed *less*
    * than the available columns [255 per default] since carries
    * are fixed up in the inner loop.
    */
@@ -4314,7 +4563,7 @@ mp_montgomery_reduce (mp_int * x, mp_int * n, mp_digit rho)
     /* mu = ai * rho mod b
      *
      * The value of rho must be precalculated via
-     * bn_mp_montgomery_setup() such that
+     * montgomery_setup() such that
      * it equals -1/n0 mod b this allows the
      * following inner loop to reduce the
      * input one digit at a time
@@ -4378,10 +4627,13 @@ mp_montgomery_reduce (mp_int * x, mp_int * n, mp_digit rho)
 
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_mp_montgomery_reduce.c */
 
 /* Start: bn_mp_montgomery_setup.c */
+#include <tommath.h>
+#ifdef BN_MP_MONTGOMERY_SETUP_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -4396,7 +4648,6 @@ mp_montgomery_reduce (mp_int * x, mp_int * n, mp_digit rho)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* setups the montgomery reduction stuff */
 int
@@ -4431,14 +4682,17 @@ mp_montgomery_setup (mp_int * n, mp_digit * rho)
 #endif
 
   /* rho = -1/m mod b */
-  *rho = (((mp_digit) 1 << ((mp_digit) DIGIT_BIT)) - x) & MP_MASK;
+  *rho = (((mp_word)1 << ((mp_word) DIGIT_BIT)) - x) & MP_MASK;
 
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_mp_montgomery_setup.c */
 
 /* Start: bn_mp_mul.c */
+#include <tommath.h>
+#ifdef BN_MP_MUL_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -4453,7 +4707,6 @@ mp_montgomery_setup (mp_int * n, mp_digit * rho)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* high level multiplication (handles sign) */
 int mp_mul (mp_int * a, mp_int * b, mp_int * c)
@@ -4462,12 +4715,18 @@ int mp_mul (mp_int * a, mp_int * b, mp_int * c)
   neg = (a->sign == b->sign) ? MP_ZPOS : MP_NEG;
 
   /* use Toom-Cook? */
+#ifdef BN_MP_TOOM_MUL_C
   if (MIN (a->used, b->used) >= TOOM_MUL_CUTOFF) {
     res = mp_toom_mul(a, b, c);
+  } else 
+#endif
+#ifdef BN_MP_KARATSUBA_MUL_C
   /* use Karatsuba? */
-  } else if (MIN (a->used, b->used) >= KARATSUBA_MUL_CUTOFF) {
+  if (MIN (a->used, b->used) >= KARATSUBA_MUL_CUTOFF) {
     res = mp_karatsuba_mul (a, b, c);
-  } else {
+  } else 
+#endif
+  {
     /* can we use the fast multiplier?
      *
      * The fast multiplier can be used if the output will 
@@ -4476,21 +4735,30 @@ int mp_mul (mp_int * a, mp_int * b, mp_int * c)
      */
     int     digs = a->used + b->used + 1;
 
+#ifdef BN_FAST_S_MP_MUL_DIGS_C
     if ((digs < MP_WARRAY) &&
         MIN(a->used, b->used) <= 
         (1 << ((CHAR_BIT * sizeof (mp_word)) - (2 * DIGIT_BIT)))) {
       res = fast_s_mp_mul_digs (a, b, c, digs);
-    } else {
-      res = s_mp_mul (a, b, c);
-    }
+    } else 
+#endif
+#ifdef BN_S_MP_MUL_DIGS_C
+      res = s_mp_mul (a, b, c); /* uses s_mp_mul_digs */
+#else
+      res = MP_VAL;
+#endif
+
   }
-  c->sign = neg;
+  c->sign = (c->used > 0) ? neg : MP_ZPOS;
   return res;
 }
+#endif
 
 /* End: bn_mp_mul.c */
 
 /* Start: bn_mp_mul_2.c */
+#include <tommath.h>
+#ifdef BN_MP_MUL_2_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -4505,7 +4773,6 @@ int mp_mul (mp_int * a, mp_int * b, mp_int * c)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* b = a*2 */
 int mp_mul_2(mp_int * a, mp_int * b)
@@ -4567,10 +4834,13 @@ int mp_mul_2(mp_int * a, mp_int * b)
   b->sign = a->sign;
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_mp_mul_2.c */
 
 /* Start: bn_mp_mul_2d.c */
+#include <tommath.h>
+#ifdef BN_MP_MUL_2D_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -4585,7 +4855,6 @@ int mp_mul_2(mp_int * a, mp_int * b)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* shift left by a certain bit count */
 int mp_mul_2d (mp_int * a, int b, mp_int * c)
@@ -4650,10 +4919,13 @@ int mp_mul_2d (mp_int * a, int b, mp_int * c)
   mp_clamp (c);
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_mp_mul_2d.c */
 
 /* Start: bn_mp_mul_d.c */
+#include <tommath.h>
+#ifdef BN_MP_MUL_D_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -4668,7 +4940,6 @@ int mp_mul_2d (mp_int * a, int b, mp_int * c)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* multiply by a digit */
 int
@@ -4726,10 +4997,13 @@ mp_mul_d (mp_int * a, mp_digit b, mp_int * c)
 
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_mp_mul_d.c */
 
 /* Start: bn_mp_mulmod.c */
+#include <tommath.h>
+#ifdef BN_MP_MULMOD_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -4744,7 +5018,6 @@ mp_mul_d (mp_int * a, mp_digit b, mp_int * c)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* d = a * b (mod c) */
 int
@@ -4765,10 +5038,13 @@ mp_mulmod (mp_int * a, mp_int * b, mp_int * c, mp_int * d)
   mp_clear (&t);
   return res;
 }
+#endif
 
 /* End: bn_mp_mulmod.c */
 
 /* Start: bn_mp_n_root.c */
+#include <tommath.h>
+#ifdef BN_MP_N_ROOT_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -4783,7 +5059,6 @@ mp_mulmod (mp_int * a, mp_int * b, mp_int * c, mp_int * d)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* find the n'th root of an integer 
  *
@@ -4895,10 +5170,13 @@ __T2:mp_clear (&t2);
 __T1:mp_clear (&t1);
   return res;
 }
+#endif
 
 /* End: bn_mp_n_root.c */
 
 /* Start: bn_mp_neg.c */
+#include <tommath.h>
+#ifdef BN_MP_NEG_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -4913,7 +5191,6 @@ __T1:mp_clear (&t1);
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* b = -a */
 int mp_neg (mp_int * a, mp_int * b)
@@ -4927,10 +5204,13 @@ int mp_neg (mp_int * a, mp_int * b)
   }
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_mp_neg.c */
 
 /* Start: bn_mp_or.c */
+#include <tommath.h>
+#ifdef BN_MP_OR_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -4945,7 +5225,6 @@ int mp_neg (mp_int * a, mp_int * b)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* OR two ints together */
 int mp_or (mp_int * a, mp_int * b, mp_int * c)
@@ -4975,10 +5254,13 @@ int mp_or (mp_int * a, mp_int * b, mp_int * c)
   mp_clear (&t);
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_mp_or.c */
 
 /* Start: bn_mp_prime_fermat.c */
+#include <tommath.h>
+#ifdef BN_MP_PRIME_FERMAT_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -4993,7 +5275,6 @@ int mp_or (mp_int * a, mp_int * b, mp_int * c)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* performs one Fermat test.
  * 
@@ -5035,10 +5316,13 @@ int mp_prime_fermat (mp_int * a, mp_int * b, int *result)
 __T:mp_clear (&t);
   return err;
 }
+#endif
 
 /* End: bn_mp_prime_fermat.c */
 
 /* Start: bn_mp_prime_is_divisible.c */
+#include <tommath.h>
+#ifdef BN_MP_PRIME_IS_DIVISIBLE_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -5053,7 +5337,6 @@ __T:mp_clear (&t);
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* determines if an integers is divisible by one 
  * of the first PRIME_SIZE primes or not
@@ -5083,10 +5366,13 @@ int mp_prime_is_divisible (mp_int * a, int *result)
 
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_mp_prime_is_divisible.c */
 
 /* Start: bn_mp_prime_is_prime.c */
+#include <tommath.h>
+#ifdef BN_MP_PRIME_IS_PRIME_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -5101,12 +5387,11 @@ int mp_prime_is_divisible (mp_int * a, int *result)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* performs a variable number of rounds of Miller-Rabin
  *
  * Probability of error after t rounds is no more than
- * (1/4)^t when 1 <= t <= PRIME_SIZE
+
  *
  * Sets result to 1 if probably prime, 0 otherwise
  */
@@ -5164,10 +5449,13 @@ int mp_prime_is_prime (mp_int * a, int t, int *result)
 __B:mp_clear (&b);
   return err;
 }
+#endif
 
 /* End: bn_mp_prime_is_prime.c */
 
 /* Start: bn_mp_prime_miller_rabin.c */
+#include <tommath.h>
+#ifdef BN_MP_PRIME_MILLER_RABIN_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -5182,7 +5470,6 @@ __B:mp_clear (&b);
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* Miller-Rabin test of "a" to the base of "b" as described in 
  * HAC pp. 139 Algorithm 4.24
@@ -5265,10 +5552,13 @@ __R:mp_clear (&r);
 __N1:mp_clear (&n1);
   return err;
 }
+#endif
 
 /* End: bn_mp_prime_miller_rabin.c */
 
 /* Start: bn_mp_prime_next_prime.c */
+#include <tommath.h>
+#ifdef BN_MP_PRIME_NEXT_PRIME_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -5283,7 +5573,6 @@ __N1:mp_clear (&n1);
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* finds the next prime after the number "a" using "t" trials
  * of Miller-Rabin.
@@ -5433,10 +5722,13 @@ __ERR:
    return err;
 }
 
+#endif
 
 /* End: bn_mp_prime_next_prime.c */
 
-/* Start: bn_mp_prime_random_ex.c */
+/* Start: bn_mp_prime_rabin_miller_trials.c */
+#include <tommath.h>
+#ifdef BN_MP_PRIME_RABIN_MILLER_TRIALS_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -5451,7 +5743,58 @@ __ERR:
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
+
+
+static const struct {
+   int k, t;
+} sizes[] = {
+{   128,    28 },
+{   256,    16 },
+{   384,    10 },
+{   512,     7 },
+{   640,     6 },
+{   768,     5 },
+{   896,     4 },
+{  1024,     4 }
+};
+
+/* returns # of RM trials required for a given bit size */
+int mp_prime_rabin_miller_trials(int size)
+{
+   int x;
+
+   for (x = 0; x < (int)(sizeof(sizes)/(sizeof(sizes[0]))); x++) {
+       if (sizes[x].k == size) {
+          return sizes[x].t;
+       } else if (sizes[x].k > size) {
+          return (x == 0) ? sizes[0].t : sizes[x - 1].t;
+       }
+   }
+   return sizes[x-1].t + 1;
+}
+
+
+#endif
+
+/* End: bn_mp_prime_rabin_miller_trials.c */
+
+/* Start: bn_mp_prime_random_ex.c */
 #include <tommath.h>
+#ifdef BN_MP_PRIME_RANDOM_EX_C
+/* LibTomMath, multiple-precision integer library -- Tom St Denis
+ *
+ * LibTomMath is a library that provides multiple-precision
+ * integer arithmetic as well as number theoretic functionality.
+ *
+ * The library was designed directly after the MPI library by
+ * Michael Fromberger but has been written from scratch with
+ * additional optimizations in place.
+ *
+ * The library is free for all purposes without any express
+ * guarantee it works.
+ *
+ * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
+ */
 
 /* makes a truly random prime of a given size (bits),
  *
@@ -5531,6 +5874,9 @@ int mp_prime_random_ex(mp_int *a, int t, int size, int flags, ltm_prime_callback
 
       /* is it prime? */
       if ((err = mp_prime_is_prime(a, t, &res)) != MP_OKAY)           { goto error; }
+      if (res == MP_NO) {  
+         continue;
+      }
 
       if (flags & LTM_PRIME_SAFE) {
          /* see if (a-1)/2 is prime */
@@ -5555,10 +5901,13 @@ error:
 }
 
 
+#endif
 
 /* End: bn_mp_prime_random_ex.c */
 
 /* Start: bn_mp_radix_size.c */
+#include <tommath.h>
+#ifdef BN_MP_RADIX_SIZE_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -5573,7 +5922,6 @@ error:
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* returns size of ASCII reprensentation */
 int mp_radix_size (mp_int * a, int radix, int *size)
@@ -5624,10 +5972,13 @@ int mp_radix_size (mp_int * a, int radix, int *size)
   return MP_OKAY;
 }
 
+#endif
 
 /* End: bn_mp_radix_size.c */
 
 /* Start: bn_mp_radix_smap.c */
+#include <tommath.h>
+#ifdef BN_MP_RADIX_SMAP_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -5642,14 +5993,16 @@ int mp_radix_size (mp_int * a, int radix, int *size)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* chars used in radix conversions */
 const char *mp_s_rmap = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz+/";
+#endif
 
 /* End: bn_mp_radix_smap.c */
 
 /* Start: bn_mp_rand.c */
+#include <tommath.h>
+#ifdef BN_MP_RAND_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -5664,7 +6017,6 @@ const char *mp_s_rmap = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrs
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* makes a pseudo-random int of a given size */
 int
@@ -5699,10 +6051,13 @@ mp_rand (mp_int * a, int digits)
 
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_mp_rand.c */
 
 /* Start: bn_mp_read_radix.c */
+#include <tommath.h>
+#ifdef BN_MP_READ_RADIX_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -5717,7 +6072,6 @@ mp_rand (mp_int * a, int digits)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* read a string [ASCII] in a given radix */
 int mp_read_radix (mp_int * a, char *str, int radix)
@@ -5779,10 +6133,13 @@ int mp_read_radix (mp_int * a, char *str, int radix)
   }
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_mp_read_radix.c */
 
 /* Start: bn_mp_read_signed_bin.c */
+#include <tommath.h>
+#ifdef BN_MP_READ_SIGNED_BIN_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -5797,7 +6154,6 @@ int mp_read_radix (mp_int * a, char *str, int radix)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* read signed bin, big endian, first byte is 0==positive or 1==negative */
 int
@@ -5819,10 +6175,13 @@ mp_read_signed_bin (mp_int * a, unsigned char *b, int c)
 
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_mp_read_signed_bin.c */
 
 /* Start: bn_mp_read_unsigned_bin.c */
+#include <tommath.h>
+#ifdef BN_MP_READ_UNSIGNED_BIN_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -5837,7 +6196,6 @@ mp_read_signed_bin (mp_int * a, unsigned char *b, int c)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* reads a unsigned char array, assumes the msb is stored first [big endian] */
 int
@@ -5873,10 +6231,13 @@ mp_read_unsigned_bin (mp_int * a, unsigned char *b, int c)
   mp_clamp (a);
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_mp_read_unsigned_bin.c */
 
 /* Start: bn_mp_reduce.c */
+#include <tommath.h>
+#ifdef BN_MP_REDUCE_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -5891,7 +6252,6 @@ mp_read_unsigned_bin (mp_int * a, unsigned char *b, int c)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* reduces x mod m, assumes 0 < x < m**2, mu is 
  * precomputed via mp_reduce_setup.
@@ -5917,9 +6277,20 @@ mp_reduce (mp_int * x, mp_int * m, mp_int * mu)
       goto CLEANUP;
     }
   } else {
+#ifdef BN_S_MP_MUL_HIGH_DIGS_C
     if ((res = s_mp_mul_high_digs (&q, mu, &q, um - 1)) != MP_OKAY) {
       goto CLEANUP;
     }
+#elif defined(BN_FAST_S_MP_MUL_HIGH_DIGS_C)
+    if ((res = fast_s_mp_mul_high_digs (&q, mu, &q, um - 1)) != MP_OKAY) {
+      goto CLEANUP;
+    }
+#else 
+    { 
+      res = MP_VAL;
+      goto CLEANUP;
+    }
+#endif
   }
 
   /* q3 = q2 / b**(k+1) */
@@ -5961,10 +6332,13 @@ CLEANUP:
 
   return res;
 }
+#endif
 
 /* End: bn_mp_reduce.c */
 
 /* Start: bn_mp_reduce_2k.c */
+#include <tommath.h>
+#ifdef BN_MP_REDUCE_2K_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -5979,7 +6353,6 @@ CLEANUP:
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* reduces a modulo n where n is of the form 2**p - d */
 int
@@ -6021,10 +6394,13 @@ ERR:
    return res;
 }
 
+#endif
 
 /* End: bn_mp_reduce_2k.c */
 
 /* Start: bn_mp_reduce_2k_setup.c */
+#include <tommath.h>
+#ifdef BN_MP_REDUCE_2K_SETUP_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -6039,7 +6415,6 @@ ERR:
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* determines the setup value */
 int 
@@ -6067,10 +6442,13 @@ mp_reduce_2k_setup(mp_int *a, mp_digit *d)
    mp_clear(&tmp);
    return MP_OKAY;
 }
+#endif
 
 /* End: bn_mp_reduce_2k_setup.c */
 
 /* Start: bn_mp_reduce_is_2k.c */
+#include <tommath.h>
+#ifdef BN_MP_REDUCE_IS_2K_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -6085,12 +6463,12 @@ mp_reduce_2k_setup(mp_int *a, mp_digit *d)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* determines if mp_reduce_2k can be used */
 int mp_reduce_is_2k(mp_int *a)
 {
-   int ix, iy, iz, iw;
+   int ix, iy, iw;
+   mp_digit iz;
    
    if (a->used == 0) {
       return 0;
@@ -6107,7 +6485,7 @@ int mp_reduce_is_2k(mp_int *a)
              return 0;
           }
           iz <<= 1;
-          if (iz > (int)MP_MASK) {
+          if (iz > (mp_digit)MP_MASK) {
              ++iw;
              iz = 1;
           }
@@ -6116,10 +6494,13 @@ int mp_reduce_is_2k(mp_int *a)
    return 1;
 }
 
+#endif
 
 /* End: bn_mp_reduce_is_2k.c */
 
 /* Start: bn_mp_reduce_setup.c */
+#include <tommath.h>
+#ifdef BN_MP_REDUCE_SETUP_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -6134,13 +6515,11 @@ int mp_reduce_is_2k(mp_int *a)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* pre-calculate the value required for Barrett reduction
  * For a given modulus "b" it calulates the value required in "a"
  */
-int
-mp_reduce_setup (mp_int * a, mp_int * b)
+int mp_reduce_setup (mp_int * a, mp_int * b)
 {
   int     res;
   
@@ -6149,10 +6528,13 @@ mp_reduce_setup (mp_int * a, mp_int * b)
   }
   return mp_div (a, b, a, NULL);
 }
+#endif
 
 /* End: bn_mp_reduce_setup.c */
 
 /* Start: bn_mp_rshd.c */
+#include <tommath.h>
+#ifdef BN_MP_RSHD_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -6167,7 +6549,6 @@ mp_reduce_setup (mp_int * a, mp_int * b)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* shift right a certain amount of digits */
 void mp_rshd (mp_int * a, int b)
@@ -6219,10 +6600,13 @@ void mp_rshd (mp_int * a, int b)
   /* remove excess digits */
   a->used -= b;
 }
+#endif
 
 /* End: bn_mp_rshd.c */
 
 /* Start: bn_mp_set.c */
+#include <tommath.h>
+#ifdef BN_MP_SET_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -6237,7 +6621,6 @@ void mp_rshd (mp_int * a, int b)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* set to a digit */
 void mp_set (mp_int * a, mp_digit b)
@@ -6246,10 +6629,13 @@ void mp_set (mp_int * a, mp_digit b)
   a->dp[0] = b & MP_MASK;
   a->used  = (a->dp[0] != 0) ? 1 : 0;
 }
+#endif
 
 /* End: bn_mp_set.c */
 
 /* Start: bn_mp_set_int.c */
+#include <tommath.h>
+#ifdef BN_MP_SET_INT_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -6264,7 +6650,6 @@ void mp_set (mp_int * a, mp_digit b)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* set a 32-bit const */
 int mp_set_int (mp_int * a, unsigned long b)
@@ -6292,10 +6677,13 @@ int mp_set_int (mp_int * a, unsigned long b)
   mp_clamp (a);
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_mp_set_int.c */
 
 /* Start: bn_mp_shrink.c */
+#include <tommath.h>
+#ifdef BN_MP_SHRINK_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -6310,7 +6698,6 @@ int mp_set_int (mp_int * a, unsigned long b)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* shrink a bignum */
 int mp_shrink (mp_int * a)
@@ -6325,10 +6712,13 @@ int mp_shrink (mp_int * a)
   }
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_mp_shrink.c */
 
 /* Start: bn_mp_signed_bin_size.c */
+#include <tommath.h>
+#ifdef BN_MP_SIGNED_BIN_SIZE_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -6343,17 +6733,19 @@ int mp_shrink (mp_int * a)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* get the size for an signed equivalent */
 int mp_signed_bin_size (mp_int * a)
 {
   return 1 + mp_unsigned_bin_size (a);
 }
+#endif
 
 /* End: bn_mp_signed_bin_size.c */
 
 /* Start: bn_mp_sqr.c */
+#include <tommath.h>
+#ifdef BN_MP_SQR_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -6368,7 +6760,6 @@ int mp_signed_bin_size (mp_int * a)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* computes b = a*a */
 int
@@ -6376,29 +6767,43 @@ mp_sqr (mp_int * a, mp_int * b)
 {
   int     res;
 
+#ifdef BN_MP_TOOM_SQR_C
   /* use Toom-Cook? */
   if (a->used >= TOOM_SQR_CUTOFF) {
     res = mp_toom_sqr(a, b);
   /* Karatsuba? */
-  } else if (a->used >= KARATSUBA_SQR_CUTOFF) {
+  } else 
+#endif
+#ifdef BN_MP_KARATSUBA_SQR_C
+if (a->used >= KARATSUBA_SQR_CUTOFF) {
     res = mp_karatsuba_sqr (a, b);
-  } else {
+  } else 
+#endif
+  {
+#ifdef BN_FAST_S_MP_SQR_C
     /* can we use the fast comba multiplier? */
     if ((a->used * 2 + 1) < MP_WARRAY && 
          a->used < 
          (1 << (sizeof(mp_word) * CHAR_BIT - 2*DIGIT_BIT - 1))) {
       res = fast_s_mp_sqr (a, b);
-    } else {
+    } else
+#endif
+#ifdef BN_S_MP_SQR_C
       res = s_mp_sqr (a, b);
-    }
+#else
+      res = MP_VAL;
+#endif
   }
   b->sign = MP_ZPOS;
   return res;
 }
+#endif
 
 /* End: bn_mp_sqr.c */
 
 /* Start: bn_mp_sqrmod.c */
+#include <tommath.h>
+#ifdef BN_MP_SQRMOD_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -6413,7 +6818,6 @@ mp_sqr (mp_int * a, mp_int * b)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* c = a * a (mod b) */
 int
@@ -6434,10 +6838,13 @@ mp_sqrmod (mp_int * a, mp_int * b, mp_int * c)
   mp_clear (&t);
   return res;
 }
+#endif
 
 /* End: bn_mp_sqrmod.c */
 
 /* Start: bn_mp_sqrt.c */
+#include <tommath.h>
+#ifdef BN_MP_SQRT_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -6452,7 +6859,6 @@ mp_sqrmod (mp_int * a, mp_int * b, mp_int * c)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* this function is less generic than mp_n_root, simpler and faster */
 int mp_sqrt(mp_int *arg, mp_int *ret) 
@@ -6513,10 +6919,13 @@ E2: mp_clear(&t1);
   return res;
 }
 
+#endif
 
 /* End: bn_mp_sqrt.c */
 
 /* Start: bn_mp_sub.c */
+#include <tommath.h>
+#ifdef BN_MP_SUB_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -6531,7 +6940,6 @@ E2: mp_clear(&t1);
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* high level subtraction (handles signs) */
 int
@@ -6570,10 +6978,13 @@ mp_sub (mp_int * a, mp_int * b, mp_int * c)
   return res;
 }
 
+#endif
 
 /* End: bn_mp_sub.c */
 
 /* Start: bn_mp_sub_d.c */
+#include <tommath.h>
+#ifdef BN_MP_SUB_D_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -6588,7 +6999,6 @@ mp_sub (mp_int * a, mp_int * b, mp_int * c)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* single digit subtraction */
 int
@@ -6657,10 +7067,13 @@ mp_sub_d (mp_int * a, mp_digit b, mp_int * c)
   return MP_OKAY;
 }
 
+#endif
 
 /* End: bn_mp_sub_d.c */
 
 /* Start: bn_mp_submod.c */
+#include <tommath.h>
+#ifdef BN_MP_SUBMOD_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -6675,7 +7088,6 @@ mp_sub_d (mp_int * a, mp_digit b, mp_int * c)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* d = a - b (mod c) */
 int
@@ -6697,10 +7109,13 @@ mp_submod (mp_int * a, mp_int * b, mp_int * c, mp_int * d)
   mp_clear (&t);
   return res;
 }
+#endif
 
 /* End: bn_mp_submod.c */
 
 /* Start: bn_mp_to_signed_bin.c */
+#include <tommath.h>
+#ifdef BN_MP_TO_SIGNED_BIN_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -6715,7 +7130,6 @@ mp_submod (mp_int * a, mp_int * b, mp_int * c, mp_int * d)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* store in signed [big endian] format */
 int
@@ -6729,10 +7143,13 @@ mp_to_signed_bin (mp_int * a, unsigned char *b)
   b[0] = (unsigned char) ((a->sign == MP_ZPOS) ? 0 : 1);
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_mp_to_signed_bin.c */
 
 /* Start: bn_mp_to_unsigned_bin.c */
+#include <tommath.h>
+#ifdef BN_MP_TO_UNSIGNED_BIN_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -6747,7 +7164,6 @@ mp_to_signed_bin (mp_int * a, unsigned char *b)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* store in unsigned [big endian] format */
 int
@@ -6776,10 +7192,13 @@ mp_to_unsigned_bin (mp_int * a, unsigned char *b)
   mp_clear (&t);
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_mp_to_unsigned_bin.c */
 
 /* Start: bn_mp_toom_mul.c */
+#include <tommath.h>
+#ifdef BN_MP_TOOM_MUL_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -6794,9 +7213,13 @@ mp_to_unsigned_bin (mp_int * a, unsigned char *b)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
-/* multiplication using the Toom-Cook 3-way algorithm */
+/* multiplication using the Toom-Cook 3-way algorithm 
+ *
+ * Much more complicated than Karatsuba but has a lower asymptotic running time of 
+ * O(N**1.464).  This algorithm is only particularly useful on VERY large
+ * inputs (we're talking 1000s of digits here...).
+*/
 int mp_toom_mul(mp_int *a, mp_int *b, mp_int *c)
 {
     mp_int w0, w1, w2, w3, w4, tmp1, tmp2, a0, a1, a2, b0, b1, b2;
@@ -7052,10 +7475,13 @@ ERR:
      return res;
 }     
      
+#endif
 
 /* End: bn_mp_toom_mul.c */
 
 /* Start: bn_mp_toom_sqr.c */
+#include <tommath.h>
+#ifdef BN_MP_TOOM_SQR_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -7070,7 +7496,6 @@ ERR:
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* squaring using Toom-Cook 3-way algorithm */
 int
@@ -7276,10 +7701,13 @@ ERR:
      return res;
 }
 
+#endif
 
 /* End: bn_mp_toom_sqr.c */
 
 /* Start: bn_mp_toradix.c */
+#include <tommath.h>
+#ifdef BN_MP_TORADIX_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -7294,7 +7722,6 @@ ERR:
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* stores a bignum as a ASCII string in a given radix (2..64) */
 int mp_toradix (mp_int * a, char *str, int radix)
@@ -7349,10 +7776,13 @@ int mp_toradix (mp_int * a, char *str, int radix)
   return MP_OKAY;
 }
 
+#endif
 
 /* End: bn_mp_toradix.c */
 
 /* Start: bn_mp_toradix_n.c */
+#include <tommath.h>
+#ifdef BN_MP_TORADIX_N_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -7367,7 +7797,6 @@ int mp_toradix (mp_int * a, char *str, int radix)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* stores a bignum as a ASCII string in a given radix (2..64) 
  *
@@ -7436,10 +7865,13 @@ int mp_toradix_n(mp_int * a, char *str, int radix, int maxlen)
   return MP_OKAY;
 }
 
+#endif
 
 /* End: bn_mp_toradix_n.c */
 
 /* Start: bn_mp_unsigned_bin_size.c */
+#include <tommath.h>
+#ifdef BN_MP_UNSIGNED_BIN_SIZE_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -7454,7 +7886,6 @@ int mp_toradix_n(mp_int * a, char *str, int radix, int maxlen)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* get the size for an unsigned equivalent */
 int
@@ -7463,10 +7894,13 @@ mp_unsigned_bin_size (mp_int * a)
   int     size = mp_count_bits (a);
   return (size / 8 + ((size & 7) != 0 ? 1 : 0));
 }
+#endif
 
 /* End: bn_mp_unsigned_bin_size.c */
 
 /* Start: bn_mp_xor.c */
+#include <tommath.h>
+#ifdef BN_MP_XOR_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -7481,7 +7915,6 @@ mp_unsigned_bin_size (mp_int * a)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* XOR two ints together */
 int
@@ -7505,17 +7938,20 @@ mp_xor (mp_int * a, mp_int * b, mp_int * c)
   }
 
   for (ix = 0; ix < px; ix++) {
-    t.dp[ix] ^= x->dp[ix];
+
   }
   mp_clamp (&t);
   mp_exch (c, &t);
   mp_clear (&t);
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_mp_xor.c */
 
 /* Start: bn_mp_zero.c */
+#include <tommath.h>
+#ifdef BN_MP_ZERO_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -7530,7 +7966,6 @@ mp_xor (mp_int * a, mp_int * b, mp_int * c)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* set to zero */
 void
@@ -7540,65 +7975,13 @@ mp_zero (mp_int * a)
   a->used = 0;
   memset (a->dp, 0, sizeof (mp_digit) * a->alloc);
 }
+#endif
 
 /* End: bn_mp_zero.c */
 
-/* Start: bn_prime_sizes_tab.c */
-/* LibTomMath, multiple-precision integer library -- Tom St Denis
- *
- * LibTomMath is a library that provides multiple-precision
- * integer arithmetic as well as number theoretic functionality.
- *
- * The library was designed directly after the MPI library by
- * Michael Fromberger but has been written from scratch with
- * additional optimizations in place.
- *
- * The library is free for all purposes without any express
- * guarantee it works.
- *
- * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
- */
-#include <tommath.h>
-
-/* this table gives the # of rabin miller trials for a prob of failure lower than 2^-96 */
-static const struct {
-   int k, t;
-} sizes[] = {
-{   128,    28 },
-{   256,    16 },
-{   384,    10 },
-{   512,     7 },
-{   640,     6 },
-{   768,     5 },
-{   896,     4 },
-{  1024,     4 },
-{  1152,     3 },
-{  1280,     3 },
-{  1408,     3 },
-{  1536,     3 },
-{  1664,     3 },
-{  1792,     2 } };
-
-/* returns # of RM trials required for a given bit size */
-int mp_prime_rabin_miller_trials(int size)
-{
-   int x;
-
-   for (x = 0; x < (int)(sizeof(sizes)/(sizeof(sizes[0]))); x++) {
-       if (sizes[x].k == size) {
-          return sizes[x].t;
-       } else if (sizes[x].k > size) {
-          return (x == 0) ? sizes[0].t : sizes[x - 1].t;
-       }
-   }
-   return 1;
-}
-
-
-
-/* End: bn_prime_sizes_tab.c */
-
 /* Start: bn_prime_tab.c */
+#include <tommath.h>
+#ifdef BN_PRIME_TAB_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -7613,7 +7996,6 @@ int mp_prime_rabin_miller_trials(int size)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 const mp_digit __prime_tab[] = {
   0x0002, 0x0003, 0x0005, 0x0007, 0x000B, 0x000D, 0x0011, 0x0013,
   0x0017, 0x001D, 0x001F, 0x0025, 0x0029, 0x002B, 0x002F, 0x0035,
@@ -7654,10 +8036,13 @@ const mp_digit __prime_tab[] = {
   0x062B, 0x062F, 0x063D, 0x0641, 0x0647, 0x0649, 0x064D, 0x0653
 #endif
 };
+#endif
 
 /* End: bn_prime_tab.c */
 
 /* Start: bn_reverse.c */
+#include <tommath.h>
+#ifdef BN_REVERSE_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -7672,7 +8057,6 @@ const mp_digit __prime_tab[] = {
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* reverse an array, used for radix code */
 void
@@ -7691,10 +8075,13 @@ bn_reverse (unsigned char *s, int len)
     --iy;
   }
 }
+#endif
 
 /* End: bn_reverse.c */
 
 /* Start: bn_s_mp_add.c */
+#include <tommath.h>
+#ifdef BN_S_MP_ADD_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -7709,7 +8096,6 @@ bn_reverse (unsigned char *s, int len)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* low level addition, based on HAC pp.594, Algorithm 14.7 */
 int
@@ -7798,10 +8184,13 @@ s_mp_add (mp_int * a, mp_int * b, mp_int * c)
   mp_clamp (c);
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_s_mp_add.c */
 
 /* Start: bn_s_mp_exptmod.c */
+#include <tommath.h>
+#ifdef BN_S_MP_EXPTMOD_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -7816,7 +8205,6 @@ s_mp_add (mp_int * a, mp_int * b, mp_int * c)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 #ifdef MP_LOW_MEM
    #define TAB_SIZE 32
@@ -8036,10 +8424,13 @@ __M:
   }
   return err;
 }
+#endif
 
 /* End: bn_s_mp_exptmod.c */
 
 /* Start: bn_s_mp_mul_digs.c */
+#include <tommath.h>
+#ifdef BN_S_MP_MUL_DIGS_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -8054,7 +8445,6 @@ __M:
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* multiplies |a| * |b| and only computes upto digs digits of result
  * HAC pp. 595, Algorithm 14.12  Modified so you can control how 
@@ -8125,10 +8515,13 @@ s_mp_mul_digs (mp_int * a, mp_int * b, mp_int * c, int digs)
   mp_clear (&t);
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_s_mp_mul_digs.c */
 
 /* Start: bn_s_mp_mul_high_digs.c */
+#include <tommath.h>
+#ifdef BN_S_MP_MUL_HIGH_DIGS_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -8143,7 +8536,6 @@ s_mp_mul_digs (mp_int * a, mp_int * b, mp_int * c, int digs)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* multiplies |a| * |b| and does not compute the lower digs digits
  * [meant to get the higher part of the product]
@@ -8158,10 +8550,12 @@ s_mp_mul_high_digs (mp_int * a, mp_int * b, mp_int * c, int digs)
   mp_digit tmpx, *tmpt, *tmpy;
 
   /* can we use the fast multiplier? */
+#ifdef BN_FAST_S_MP_MUL_HIGH_DIGS_C
   if (((a->used + b->used + 1) < MP_WARRAY)
       && MIN (a->used, b->used) < (1 << ((CHAR_BIT * sizeof (mp_word)) - (2 * DIGIT_BIT)))) {
     return fast_s_mp_mul_high_digs (a, b, c, digs);
   }
+#endif
 
   if ((res = mp_init_size (&t, a->used + b->used + 1)) != MP_OKAY) {
     return res;
@@ -8202,10 +8596,13 @@ s_mp_mul_high_digs (mp_int * a, mp_int * b, mp_int * c, int digs)
   mp_clear (&t);
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_s_mp_mul_high_digs.c */
 
 /* Start: bn_s_mp_sqr.c */
+#include <tommath.h>
+#ifdef BN_S_MP_SQR_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -8220,7 +8617,6 @@ s_mp_mul_high_digs (mp_int * a, mp_int * b, mp_int * c, int digs)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* low level squaring, b = a*a, HAC pp.596-597, Algorithm 14.16 */
 int
@@ -8285,10 +8681,13 @@ s_mp_sqr (mp_int * a, mp_int * b)
   mp_clear (&t);
   return MP_OKAY;
 }
+#endif
 
 /* End: bn_s_mp_sqr.c */
 
 /* Start: bn_s_mp_sub.c */
+#include <tommath.h>
+#ifdef BN_S_MP_SUB_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -8303,7 +8702,6 @@ s_mp_sqr (mp_int * a, mp_int * b)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* low level subtraction (assumes |a| > |b|), HAC pp.595 Algorithm 14.9 */
 int
@@ -8372,10 +8770,13 @@ s_mp_sub (mp_int * a, mp_int * b, mp_int * c)
   return MP_OKAY;
 }
 
+#endif
 
 /* End: bn_s_mp_sub.c */
 
 /* Start: bncore.c */
+#include <tommath.h>
+#ifdef BNCORE_C
 /* LibTomMath, multiple-precision integer library -- Tom St Denis
  *
  * LibTomMath is a library that provides multiple-precision
@@ -8390,23 +8791,21 @@ s_mp_sub (mp_int * a, mp_int * b, mp_int * c)
  *
  * Tom St Denis, tomstdenis@iahu.ca, http://math.libtomcrypt.org
  */
-#include <tommath.h>
 
 /* Known optimal configurations
 
  CPU                    /Compiler     /MUL CUTOFF/SQR CUTOFF
 -------------------------------------------------------------
- Intel P4               /GCC v3.2     /        70/       108
- AMD Athlon XP          /GCC v3.2     /       109/       127
-
+ Intel P4 Northwood     /GCC v3.4.1   /        88/       128/LTM 0.32 ;-)
+ 
 */
 
-/* configured for a AMD XP Thoroughbred core with etc/tune.c */
-int     KARATSUBA_MUL_CUTOFF = 109,      /* Min. number of digits before Karatsuba multiplication is used. */
-        KARATSUBA_SQR_CUTOFF = 127,      /* Min. number of digits before Karatsuba squaring is used. */
+int     KARATSUBA_MUL_CUTOFF = 88,      /* Min. number of digits before Karatsuba multiplication is used. */
+        KARATSUBA_SQR_CUTOFF = 128,     /* Min. number of digits before Karatsuba squaring is used. */
         
         TOOM_MUL_CUTOFF      = 350,      /* no optimal values of these are known yet so set em high */
         TOOM_SQR_CUTOFF      = 400; 
+#endif
 
 /* End: bncore.c */
 
diff --git a/pretty.build b/pretty.build
new file mode 100644
index 0000000000000000000000000000000000000000..a708b8af221d4b1ee5745204e4e3d699be96ce76
--- /dev/null
+++ b/pretty.build
@@ -0,0 +1,66 @@
+#!/bin/perl -w
+#
+# Cute little builder for perl 
+# Total waste of development time...
+#
+# This will build all the object files and then the archive .a file
+# requires GCC, GNU make and a sense of humour.
+#
+# Tom St Denis
+use strict;
+
+my $count = 0;
+my $starttime = time;
+my $rate  = 0;
+print "Scanning for source files...\n";
+foreach my $filename (glob "*.c") {
+       ++$count;
+}
+print "Source files to build: $count\nBuilding...\n";
+my $i = 0;
+my $lines = 0;
+my $filesbuilt = 0;
+foreach my $filename (glob "*.c") {
+       printf("Building %3.2f%%, ", (++$i/$count)*100.0);
+       if ($i % 4 == 0) { print "/, "; }
+       if ($i % 4 == 1) { print "-, "; }
+       if ($i % 4 == 2) { print "\\, "; }
+       if ($i % 4 == 3) { print "|, "; }
+       if ($rate > 0) {
+           my $tleft = ($count - $i) / $rate;
+           my $tsec  = $tleft%60;
+           my $tmin  = ($tleft/60)%60;
+           my $thour = ($tleft/3600)%60;
+           printf("%2d:%02d:%02d left, ", $thour, $tmin, $tsec);
+       }
+       my $cnt = ($i/$count)*30.0;
+       my $x   = 0;
+       print "[";
+       for (; $x < $cnt; $x++) { print "#"; }
+       for (; $x < 30; $x++)   { print " "; }
+       print "]\r";
+       my $tmp = $filename;
+       $tmp =~ s/\.c/".o"/ge;
+       if (open(SRC, "<$tmp")) {
+          close SRC;
+       } else {
+          !system("make $tmp > /dev/null 2>/dev/null") or die "\nERROR: Failed to make $tmp!!!\n";
+          open( SRC, "<$filename" ) or die "Couldn't open $filename for reading: $!";
+          ++$lines while (<SRC>);
+          close SRC or die "Error closing $filename after reading: $!";
+          ++$filesbuilt;
+       }      
+
+       # update timer 
+       if (time != $starttime) {
+          my $delay = time - $starttime;
+          $rate = $i/$delay;
+       }
+}
+
+# finish building the library 
+printf("\nFinished building source (%d seconds, %3.2f files per second).\n", time - $starttime, $rate);
+print "Compiled approximately $filesbuilt files and $lines lines of code.\n";
+print "Doing final make (building archive...)\n";
+!system("make > /dev/null 2>/dev/null") or die "\nERROR: Failed to perform last make command!!!\n";
+print "done.\n";
\ No newline at end of file
diff --git a/tommath.h b/tommath.h
index 00299947f7518d21408b1c83cd1f8f55fd723628..896d389a65c36ea18b18acbb71ceaf9ab8e6836e 100644
--- a/tommath.h
+++ b/tommath.h
@@ -21,6 +21,8 @@
 #include <ctype.h>
 #include <limits.h>
 
+#include <tommath_class.h>
+
 #undef MIN
 #define MIN(x,y) ((x)<(y)?(x):(y))
 #undef MAX
@@ -39,6 +41,14 @@ extern "C" {
 
 #endif
 
+
+/* detect 64-bit mode if possible */
+#if defined(__x86_64__) 
+   #if !(defined(MP_64BIT) && defined(MP_16BIT) && defined(MP_8BIT))
+      #define MP_64BIT
+   #endif
+#endif
+
 /* some default configurations.
  *
  * A "mp_digit" must be able to hold DIGIT_BIT + 1 bits
@@ -60,7 +70,7 @@ extern "C" {
    typedef signed long long   long64;
 #endif
 
-   typedef ulong64            mp_digit;
+   typedef unsigned long      mp_digit;
    typedef unsigned long      mp_word __attribute__ ((mode(TI)));
 
    #define DIGIT_BIT          60
@@ -153,7 +163,7 @@ extern int KARATSUBA_MUL_CUTOFF,
 
 /* default precision */
 #ifndef MP_PREC
-   #ifdef MP_LOW_MEM
+   #ifndef MP_LOW_MEM
       #define MP_PREC                 64     /* default digits of precision */
    #else
       #define MP_PREC                 8      /* default digits of precision */
@@ -541,6 +551,7 @@ int mp_toom_mul(mp_int *a, mp_int *b, mp_int *c);
 int mp_karatsuba_sqr(mp_int *a, mp_int *b);
 int mp_toom_sqr(mp_int *a, mp_int *b);
 int fast_mp_invmod(mp_int *a, mp_int *b, mp_int *c);
+int mp_invmod_slow (mp_int * a, mp_int * b, mp_int * c);
 int fast_mp_montgomery_reduce(mp_int *a, mp_int *m, mp_digit mp);
 int mp_exptmod_fast(mp_int *G, mp_int *X, mp_int *P, mp_int *Y, int mode);
 int s_mp_exptmod (mp_int * G, mp_int * X, mp_int * P, mp_int * Y);
diff --git a/tommath.pdf b/tommath.pdf
index fc3301a306fc0e63f9012d97b879bbed326ebf54..18cac6f2db318bf5fb3dae69c62855dfbce54bbe 100644
Binary files a/tommath.pdf and b/tommath.pdf differ
diff --git a/tommath.src b/tommath.src
index 0389831c83dd1cac7ff2ac2d325ef5a72eb265b3..6ee842d7dd77cfb75be92faf0eefbb93fecf6da2 100644
--- a/tommath.src
+++ b/tommath.src
@@ -258,7 +258,7 @@ floating point is meant to be implemented in hardware the precision of the manti
 a mantissa of much larger precision than hardware alone can efficiently support.  This approach could be useful where 
 scientific applications must minimize the total output error over long calculations.
 
-Another use for large integers is within arithmetic on polynomials of large characteristic (i.e. $GF(p)[x]$ for large $p$).
+Yet another use for large integers is within arithmetic on polynomials of large characteristic (i.e. $GF(p)[x]$ for large $p$).
 In fact the library discussed within this text has already been used to form a polynomial basis library\footnote{See \url{http://poly.libtomcrypt.org} for more details.}.
 
 \subsection{Benefits of Multiple Precision Arithmetic}
@@ -316,7 +316,7 @@ the reader how the algorithms fit together as well as where to start on various
 
 \section{Discussion and Notation}
 \subsection{Notation}
-A multiple precision integer of $n$-digits shall be denoted as $x = (x_{n-1} ... x_1 x_0)_{ \beta }$ and represent
+A multiple precision integer of $n$-digits shall be denoted as $x = (x_{n-1}, \ldots, x_1, x_0)_{ \beta }$ and represent
 the integer $x \equiv \sum_{i=0}^{n-1} x_i\beta^i$.  The elements of the array $x$ are said to be the radix $\beta$ digits 
 of the integer.  For example, $x = (1,2,3)_{10}$ would represent the integer 
 $1\cdot 10^2 + 2\cdot10^1 + 3\cdot10^0 = 123$.  
@@ -339,12 +339,11 @@ algorithms will be used to establish the relevant theory which will subsequently
 precision algorithm to solve the same problem.  
 
 \subsection{Precision Notation}
-For the purposes of this text a single precision variable must be able to represent integers in the range 
-$0 \le x < q \beta$ while a double precision variable must be able to represent integers in the range 
-$0 \le x < q \beta^2$.  The variable $\beta$ represents the radix of a single digit of a multiple precision integer and 
-must be of the form $q^p$ for $q, p \in \Z^+$.  The extra radix-$q$ factor allows additions and subtractions to proceed 
-without truncation of the carry.  Since all modern computers are binary, it is assumed that $q$ is two, for all intents 
-and purposes.
+The variable $\beta$ represents the radix of a single digit of a multiple precision integer and 
+must be of the form $q^p$ for $q, p \in \Z^+$.  A single precision variable must be able to represent integers in 
+the range $0 \le x < q \beta$ while a double precision variable must be able to represent integers in the range 
+$0 \le x < q \beta^2$.  The extra radix-$q$ factor allows additions and subtractions to proceed without truncation of the 
+carry.  Since all modern computers are binary, it is assumed that $q$ is two.
 
 \index{mp\_digit} \index{mp\_word}
 Within the source code that will be presented for each algorithm, the data type \textbf{mp\_digit} will represent 
@@ -376,7 +375,7 @@ the $/$ division symbol is used the intention is to perform an integer division
 $5/2 = 2$ which will often be written as $\lfloor 5/2 \rfloor = 2$ for clarity.  When an expression is written as a 
 fraction a real value division is implied, for example ${5 \over 2} = 2.5$.  
 
-The norm of a multiple precision integer, for example, $\vert \vert x \vert \vert$ will be used to represent the number of digits in the representation
+The norm of a multiple precision integer, for example $\vert \vert x \vert \vert$, will be used to represent the number of digits in the representation
 of the integer.  For example, $\vert \vert 123 \vert \vert = 3$ and $\vert \vert 79452 \vert \vert = 5$.  
 
 \subsection{Work Effort}
@@ -569,7 +568,7 @@ By building outwards from a base foundation instead of using a parallel design m
 highly modular.  Being highly modular is a desirable property of any project as it often means the resulting product
 has a small footprint and updates are easy to perform.  
 
-Usually when I start a project I will begin with the header file.  I define the data types I think I will need and 
+Usually when I start a project I will begin with the header files.  I define the data types I think I will need and 
 prototype the initial functions that are not dependent on other functions (within the library).  After I 
 implement these base functions I prototype more dependent functions and implement them.   The process repeats until
 I implement all of the functions I require.  For example, in the case of LibTomMath I implemented functions such as 
@@ -619,14 +618,26 @@ any such data type but it does provide for making composite data types known as
 used within LibTomMath.
 
 \index{mp\_int}
-\begin{verbatim}
-typedef struct  {
-    int used, alloc, sign;
-    mp_digit *dp;
-} mp_int;
-\end{verbatim}
+\begin{figure}[here]
+\begin{center}
+\begin{small}
+%\begin{verbatim}
+\begin{tabular}{|l|}
+\hline
+typedef struct \{ \\
+\hspace{3mm}int used, alloc, sign;\\
+\hspace{3mm}mp\_digit *dp;\\
+\} \textbf{mp\_int}; \\
+\hline
+\end{tabular}
+%\end{verbatim}
+\end{small}
+\caption{The mp\_int Structure}
+\label{fig:mpint}
+\end{center}
+\end{figure}
 
-The mp\_int structure can be broken down as follows.
+The mp\_int structure (fig. \ref{fig:mpint}) can be broken down as follows.
 
 \begin{enumerate}
 \item The \textbf{used} parameter denotes how many digits of the array \textbf{dp} contain the digits used to represent
@@ -701,9 +712,10 @@ fault by dereferencing memory not owned by the application.
 In the case of LibTomMath the only errors that are checked for are related to inappropriate inputs (division by zero for 
 instance) and memory allocation errors.  It will not check that the mp\_int passed to any function is valid nor 
 will it check pointers for validity.  Any function that can cause a runtime error will return an error code as an 
-\textbf{int} data type with one of the following values.
+\textbf{int} data type with one of the following values (fig \ref{fig:errcodes}).
 
 \index{MP\_OKAY} \index{MP\_VAL} \index{MP\_MEM}
+\begin{figure}[here]
 \begin{center}
 \begin{tabular}{|l|l|}
 \hline \textbf{Value} & \textbf{Meaning} \\
@@ -713,6 +725,9 @@ will it check pointers for validity.  Any function that can cause a runtime erro
 \hline
 \end{tabular}
 \end{center}
+\caption{LibTomMath Error Codes}
+\label{fig:errcodes}
+\end{figure}
 
 When an error is detected within a function it should free any memory it allocated, often during the initialization of
 temporary mp\_ints, and return as soon as possible.  The goal is to leave the system in the same state it was when the 
@@ -748,6 +763,7 @@ to zero.  The \textbf{used} count set to zero and \textbf{sign} set to \textbf{M
 An mp\_int is said to be initialized if it is set to a valid, preferably default, state such that all of the members of the
 structure are set to valid values.  The mp\_init algorithm will perform such an action.
 
+\index{mp\_init}
 \begin{figure}[here]
 \begin{center}
 \begin{tabular}{l}
@@ -770,17 +786,23 @@ structure are set to valid values.  The mp\_init algorithm will perform such an
 \end{figure}
 
 \textbf{Algorithm mp\_init.}
-The \textbf{MP\_PREC} name represents a constant\footnote{Defined in the ``tommath.h'' header file within LibTomMath.} 
-used to dictate the minimum precision of allocated mp\_int integers.  Ideally, it is at least equal to $32$ since for most
-purposes that will be more than enough.
+The purpose of this function is to initialize an mp\_int structure so that the rest of the library can properly
+manipulte it.  It is assumed that the input may not have had any of its members previously initialized which is certainly
+a valid assumption if the input resides on the stack.  
 
-Memory for the default number of digits is allocated first.  If the allocation fails the algorithm returns immediately
-with the \textbf{MP\_MEM} error code.  If the allocation succeeds the remaining members of the mp\_int structure
-must be initialized to reflect the default initial state.
+Before any of the members such as \textbf{sign}, \textbf{used} or \textbf{alloc} are initialized the memory for
+the digits is allocated.  If this fails the function returns before setting any of the other members.  The \textbf{MP\_PREC} 
+name represents a constant\footnote{Defined in the ``tommath.h'' header file within LibTomMath.} 
+used to dictate the minimum precision of newly initialized mp\_int integers.  Ideally, it is at least equal to the smallest
+precision number you'll be working with.
 
-The allocated digits are all set to zero (step three) to ensure they are in a known state.  The \textbf{sign}, \textbf{used}
-and \textbf{alloc} are subsequently initialized to represent the zero integer.  By step seven the algorithm returns a success 
-code and the mp\_int $a$ has been successfully initialized to a valid state representing the integer zero.  
+Allocating a block of digits at first instead of a single digit has the benefit of lowering the number of usually slow
+heap operations later functions will have to perform in the future.  If \textbf{MP\_PREC} is set correctly the slack 
+memory and the number of heap operations will be trivial.
+
+Once the allocation has been made the digits have to be set to zero as well as the \textbf{used}, \textbf{sign} and
+\textbf{alloc} members initialized.  This ensures that the mp\_int will always represent the default state of zero regardless
+of the original condition of the input.
 
 \textbf{Remark.}
 This function introduces the idiosyncrasy that all iterative loops, commonly initiated with the ``for'' keyword, iterate incrementally
@@ -796,19 +818,21 @@ One immediate observation of this initializtion function is that it does not ret
 is assumed that the caller has already allocated memory for the mp\_int structure, typically on the application stack.  The 
 call to mp\_init() is used only to initialize the members of the structure to a known default state.  
 
-Before any of the other members of the structure are initialized memory from the application heap is allocated with
-the calloc() function (line @22,calloc@).  The size of the allocated memory is large enough to hold \textbf{MP\_PREC} 
-mp\_digit variables.  The calloc() function is used instead\footnote{calloc() will allocate memory in the same
-manner as malloc() except that it also sets the contents to zero upon successfully allocating the memory.} of malloc() 
-since digits have to be set to zero for the function to finish correctly.  The \textbf{OPT\_CAST} token is a macro 
-definition which will turn into a cast from void * to mp\_digit * for C++ compilers.  It is not required for C compilers.
+Here we see (line @23,XMALLOC@) the memory allocation is performed first.  This allows us to exit cleanly and quickly
+if there is an error.  If the allocation fails the routine will return \textbf{MP\_MEM} to the caller to indicate there
+was a memory error.  The function XMALLOC is what actually allocates the memory.  Technically XMALLOC is not a function
+but a macro defined in ``tommath.h``.  By default, XMALLOC will evaluate to malloc() which is the C library's built--in
+memory allocation routine.
 
-After the memory has been successfully allocated the remainder of the members are initialized 
-(lines @29,used@ through @31,sign@) to their respective default states.  At this point the algorithm has succeeded and
-a success code is returned to the calling function.
+In order to assure the mp\_int is in a known state the digits must be set to zero.  On most platforms this could have been
+accomplished by using calloc() instead of malloc().  However,  to correctly initialize a integer type to a given value in a 
+portable fashion you have to actually assign the value.  The for loop (line @28,for@) performs this required
+operation.
 
-If this function returns \textbf{MP\_OKAY} it is safe to assume the mp\_int structure has been properly initialized and
-is safe to use with other functions within the library.  
+After the memory has been successfully initialized the remainder of the members are initialized 
+(lines @29,used@ through @31,sign@) to their respective default states.  At this point the algorithm has succeeded and
+a success code is returned to the calling function.  If this function returns \textbf{MP\_OKAY} it is safe to assume the 
+mp\_int structure has been properly initialized and is safe to use with other functions within the library.  
 
 \subsection{Clearing an mp\_int}
 When an mp\_int is no longer required by the application, the memory that has been allocated for its digits must be 
@@ -819,7 +843,7 @@ returned to the application's memory pool with the mp\_clear algorithm.
 \begin{tabular}{l}
 \hline Algorithm \textbf{mp\_clear}. \\
 \textbf{Input}.   An mp\_int $a$ \\
-\textbf{Output}.  The memory for $a$ is freed for reuse.  \\
+\textbf{Output}.  The memory for $a$ shall be deallocated.  \\
 \hline \\
 1.  If $a$ has been previously freed then return(\textit{MP\_OKAY}). \\
 2.  for $n$ from 0 to $a.used - 1$ do \\
@@ -836,32 +860,31 @@ returned to the application's memory pool with the mp\_clear algorithm.
 \end{figure}
 
 \textbf{Algorithm mp\_clear.}
-This algorithm releases the memory allocated for an mp\_int back into the memory pool for reuse.  It is designed
-such that a given mp\_int structure can be cleared multiple times between initializations without attempting to 
-free the memory twice\footnote{In ISO C for example, calling free() twice on the same memory block causes undefinied
-behaviour.}.  
-
-The first step determines if the mp\_int structure has been marked as free already.  If it has, the algorithm returns
-success immediately as no further actions are required.  Otherwise, the algorithm will proceed to put the structure 
-in a known empty and otherwise invalid state.  First the digits of the mp\_int are set to zero.  The memory that has been allocated for the 
-digits is then freed.  The \textbf{used} and \textbf{alloc} counts are both set to zero and the \textbf{sign} set to 
-\textbf{MP\_ZPOS}.  This known fixed state for cleared mp\_int structures will make debuging easier for the end 
-developer.  That is, if they spot (via their debugger) an mp\_int they are using that is in this state it will be 
-obvious that they erroneously and prematurely cleared the mp\_int structure.
-
-Note that once an mp\_int has been cleared the mp\_int structure is no longer in a valid state for any other algorithm
+This algorithm accomplishes two goals.  First, it clears the digits and the other mp\_int members.  This ensures that 
+if a developer accidentally re-uses a cleared structure it is less likely to cause problems.  The second goal
+is to free the allocated memory.
+
+The logic behind the algorithm is extended by marking cleared mp\_int structures so that subsequent calls to this
+algorithm will not try to free the memory multiple times.  Cleared mp\_ints are detectable by having a pre-defined invalid 
+digit pointer \textbf{dp} setting.
+
+Once an mp\_int has been cleared the mp\_int structure is no longer in a valid state for any other algorithm
 with the exception of algorithms mp\_init, mp\_init\_copy, mp\_init\_size and mp\_clear.
 
 EXAM,bn_mp_clear.c
 
-The ``if'' statement (line @21,a->dp != NULL@) prevents the heap from being corrupted if a user double-frees an 
-mp\_int.  This is because once the memory is freed the pointer is set to \textbf{NULL} (line @30,NULL@).  
+The algorithm only operates on the mp\_int if it hasn't been previously cleared.  The if statement (line @23,a->dp != NULL@)
+checks to see if the \textbf{dp} member is not \textbf{NULL}.  If the mp\_int is a valid mp\_int then \textbf{dp} cannot be
+\textbf{NULL} in which case the if statement will evaluate to true.
+
+The digits of the mp\_int are cleared by the for loop (line @25,for@) which assigns a zero to every digit.  Similar to mp\_init()
+the digits are assigned zero instead of using block memory operations (such as memset()) since this is more portable.  
 
-Without the check, code that accidentally calls mp\_clear twice for a given mp\_int structure would try to free the memory 
-allocated for the digits twice.  This may cause some C libraries to signal a fault.  By setting the pointer to 
-\textbf{NULL} it helps debug code that may inadvertently free the mp\_int before it is truly not needed, because attempts 
-to reference digits should fail immediately. The allocated digits are set to zero before being freed (line @24,memset@).  
-This is ideal for cryptographic situations where the integer that the mp\_int represents might need to be kept a secret.
+The digits are deallocated off the heap via the XFREE macro.  Similar to XMALLOC the XFREE macro actually evaluates to
+a standard C library function.  In this case the free() function.  Since free() only deallocates the memory the pointer
+still has to be reset to \textbf{NULL} manually (line @33,NULL@).  
+
+Now that the digits have been cleared and deallocated the other members are set to their final values (lines @34,= 0@ and @35,ZPOS@).
 
 \section{Maintenance Algorithms}
 
@@ -889,7 +912,7 @@ must be re-sized appropriately to accomodate the result.  The mp\_grow algorithm
 1.  if $a.alloc \ge b$ then return(\textit{MP\_OKAY}) \\
 2.  $u \leftarrow b\mbox{ (mod }MP\_PREC\mbox{)}$ \\
 3.  $v \leftarrow b + 2 \cdot MP\_PREC - u$ \\
-4.  Re-Allocate the array of digits $a$ to size $v$ \\
+4.  Re-allocate the array of digits $a$ to size $v$ \\
 5.  If the allocation failed then return(\textit{MP\_MEM}). \\
 6.  for n from a.alloc to $v - 1$ do  \\
 \hspace{+3mm}6.1  $a_n \leftarrow 0$ \\
@@ -914,15 +937,19 @@ assumed to contain undefined values they are initially set to zero.
 
 EXAM,bn_mp_grow.c
 
-The first step is to see if we actually need to perform a re-allocation at all (line @24,a->alloc < size@).  If a reallocation
-must occur the digit count is padded upwards to help prevent many trivial reallocations (line @28,size@).  Next the reallocation is performed
-and the return of realloc() is stored in a temporary pointer named $tmp$ (line @36,realloc@).  The return is stored in a temporary
-instead of $a.dp$ to prevent the code from losing the original pointer in case the reallocation fails.  Had the return been stored 
-in $a.dp$ instead there would be no way to reclaim the heap originally used.
+A quick optimization is to first determine if a memory re-allocation is required at all.  The if statement (line @23,if@) checks
+if the \textbf{alloc} member of the mp\_int is smaller than the requested digit count.  If the count is not larger than \textbf{alloc}
+the function skips the re-allocation part thus saving time.
+
+When a re-allocation is performed it is turned into an optimal request to save time in the future.  The requested digit count is
+padded upwards to 2nd multiple of \textbf{MP\_PREC} larger than \textbf{alloc} (line @25, size@).  The XREALLOC function is used
+to re-allocate the memory.  As per the other functions XREALLOC is actually a macro which evaluates to realloc by default.  The realloc
+function leaves the base of the allocation intact which means the first \textbf{alloc} digits of the mp\_int are the same as before
+the re-allocation.  All	that is left is to clear the newly allocated digits and return.
 
-If the reallocation fails the function will return \textbf{MP\_MEM} (line @39,return@), otherwise, the value of $tmp$ is assigned
-to the pointer $a.dp$ and the function continues.  A simple for loop from line @48,a->alloc@ to line @50,}@ will zero all digits 
-that were above the old \textbf{alloc} limit to make sure the integer is in a known state.
+Note that the re-allocation result is actually stored in a temporary pointer $tmp$.  This is to allow this function to return
+an error with a valid pointer.  Earlier releases of the library stored the result of XREALLOC into the mp\_int $a$.  That would
+result in a memory leak if XREALLOC ever failed.  
 
 \subsection{Initializing Variable Precision mp\_ints}
 Occasionally the number of digits required will be known in advance of an initialization, based on, for example, the size 
@@ -970,7 +997,7 @@ The number of digits $b$ requested is padded (line @22,MP_PREC@) by first augmen
 mp\_int is placed in a default state representing the integer zero.  Otherwise, the error code \textbf{MP\_MEM} will be 
 returned (line @27,return@).  
 
-The digits are allocated and set to zero at the same time with the calloc() function (line @25,calloc@).  The 
+The digits are allocated and set to zero at the same time with the calloc() function (line @25,XCALLOC@).  The 
 \textbf{used} count is set to zero, the \textbf{alloc} count set to the padded digit count and the \textbf{sign} flag set 
 to \textbf{MP\_ZPOS} to achieve a default valid mp\_int state (lines @29,used@, @30,alloc@ and @31,sign@).  If the function 
 returns succesfully then it is correct to assume that the mp\_int structure is in a valid state for the remainder of the 
diff --git a/tommath.tex b/tommath.tex
index 629edbad3091c9066f77d47766866c6a7f7fab24..d0ac9474c6a0f328dbdf1b83d3f7e385c90359f0 100644
--- a/tommath.tex
+++ b/tommath.tex
@@ -258,7 +258,7 @@ floating point is meant to be implemented in hardware the precision of the manti
 a mantissa of much larger precision than hardware alone can efficiently support.  This approach could be useful where 
 scientific applications must minimize the total output error over long calculations.
 
-Another use for large integers is within arithmetic on polynomials of large characteristic (i.e. $GF(p)[x]$ for large $p$).
+Yet another use for large integers is within arithmetic on polynomials of large characteristic (i.e. $GF(p)[x]$ for large $p$).
 In fact the library discussed within this text has already been used to form a polynomial basis library\footnote{See \url{http://poly.libtomcrypt.org} for more details.}.
 
 \subsection{Benefits of Multiple Precision Arithmetic}
@@ -316,7 +316,7 @@ the reader how the algorithms fit together as well as where to start on various
 
 \section{Discussion and Notation}
 \subsection{Notation}
-A multiple precision integer of $n$-digits shall be denoted as $x = (x_{n-1} ... x_1 x_0)_{ \beta }$ and represent
+A multiple precision integer of $n$-digits shall be denoted as $x = (x_{n-1}, \ldots, x_1, x_0)_{ \beta }$ and represent
 the integer $x \equiv \sum_{i=0}^{n-1} x_i\beta^i$.  The elements of the array $x$ are said to be the radix $\beta$ digits 
 of the integer.  For example, $x = (1,2,3)_{10}$ would represent the integer 
 $1\cdot 10^2 + 2\cdot10^1 + 3\cdot10^0 = 123$.  
@@ -339,12 +339,11 @@ algorithms will be used to establish the relevant theory which will subsequently
 precision algorithm to solve the same problem.  
 
 \subsection{Precision Notation}
-For the purposes of this text a single precision variable must be able to represent integers in the range 
-$0 \le x < q \beta$ while a double precision variable must be able to represent integers in the range 
-$0 \le x < q \beta^2$.  The variable $\beta$ represents the radix of a single digit of a multiple precision integer and 
-must be of the form $q^p$ for $q, p \in \Z^+$.  The extra radix-$q$ factor allows additions and subtractions to proceed 
-without truncation of the carry.  Since all modern computers are binary, it is assumed that $q$ is two, for all intents 
-and purposes.
+The variable $\beta$ represents the radix of a single digit of a multiple precision integer and 
+must be of the form $q^p$ for $q, p \in \Z^+$.  A single precision variable must be able to represent integers in 
+the range $0 \le x < q \beta$ while a double precision variable must be able to represent integers in the range 
+$0 \le x < q \beta^2$.  The extra radix-$q$ factor allows additions and subtractions to proceed without truncation of the 
+carry.  Since all modern computers are binary, it is assumed that $q$ is two.
 
 \index{mp\_digit} \index{mp\_word}
 Within the source code that will be presented for each algorithm, the data type \textbf{mp\_digit} will represent 
@@ -376,7 +375,7 @@ the $/$ division symbol is used the intention is to perform an integer division
 $5/2 = 2$ which will often be written as $\lfloor 5/2 \rfloor = 2$ for clarity.  When an expression is written as a 
 fraction a real value division is implied, for example ${5 \over 2} = 2.5$.  
 
-The norm of a multiple precision integer, for example, $\vert \vert x \vert \vert$ will be used to represent the number of digits in the representation
+The norm of a multiple precision integer, for example $\vert \vert x \vert \vert$, will be used to represent the number of digits in the representation
 of the integer.  For example, $\vert \vert 123 \vert \vert = 3$ and $\vert \vert 79452 \vert \vert = 5$.  
 
 \subsection{Work Effort}
@@ -569,7 +568,7 @@ By building outwards from a base foundation instead of using a parallel design m
 highly modular.  Being highly modular is a desirable property of any project as it often means the resulting product
 has a small footprint and updates are easy to perform.  
 
-Usually when I start a project I will begin with the header file.  I define the data types I think I will need and 
+Usually when I start a project I will begin with the header files.  I define the data types I think I will need and 
 prototype the initial functions that are not dependent on other functions (within the library).  After I 
 implement these base functions I prototype more dependent functions and implement them.   The process repeats until
 I implement all of the functions I require.  For example, in the case of LibTomMath I implemented functions such as 
@@ -625,14 +624,26 @@ any such data type but it does provide for making composite data types known as
 used within LibTomMath.
 
 \index{mp\_int}
-\begin{verbatim}
-typedef struct  {
-    int used, alloc, sign;
-    mp_digit *dp;
-} mp_int;
-\end{verbatim}
+\begin{figure}[here]
+\begin{center}
+\begin{small}
+%\begin{verbatim}
+\begin{tabular}{|l|}
+\hline
+typedef struct \{ \\
+\hspace{3mm}int used, alloc, sign;\\
+\hspace{3mm}mp\_digit *dp;\\
+\} \textbf{mp\_int}; \\
+\hline
+\end{tabular}
+%\end{verbatim}
+\end{small}
+\caption{The mp\_int Structure}
+\label{fig:mpint}
+\end{center}
+\end{figure}
 
-The mp\_int structure can be broken down as follows.
+The mp\_int structure (fig. \ref{fig:mpint}) can be broken down as follows.
 
 \begin{enumerate}
 \item The \textbf{used} parameter denotes how many digits of the array \textbf{dp} contain the digits used to represent
@@ -707,9 +718,10 @@ fault by dereferencing memory not owned by the application.
 In the case of LibTomMath the only errors that are checked for are related to inappropriate inputs (division by zero for 
 instance) and memory allocation errors.  It will not check that the mp\_int passed to any function is valid nor 
 will it check pointers for validity.  Any function that can cause a runtime error will return an error code as an 
-\textbf{int} data type with one of the following values.
+\textbf{int} data type with one of the following values (fig \ref{fig:errcodes}).
 
 \index{MP\_OKAY} \index{MP\_VAL} \index{MP\_MEM}
+\begin{figure}[here]
 \begin{center}
 \begin{tabular}{|l|l|}
 \hline \textbf{Value} & \textbf{Meaning} \\
@@ -719,6 +731,9 @@ will it check pointers for validity.  Any function that can cause a runtime erro
 \hline
 \end{tabular}
 \end{center}
+\caption{LibTomMath Error Codes}
+\label{fig:errcodes}
+\end{figure}
 
 When an error is detected within a function it should free any memory it allocated, often during the initialization of
 temporary mp\_ints, and return as soon as possible.  The goal is to leave the system in the same state it was when the 
@@ -754,6 +769,7 @@ to zero.  The \textbf{used} count set to zero and \textbf{sign} set to \textbf{M
 An mp\_int is said to be initialized if it is set to a valid, preferably default, state such that all of the members of the
 structure are set to valid values.  The mp\_init algorithm will perform such an action.
 
+\index{mp\_init}
 \begin{figure}[here]
 \begin{center}
 \begin{tabular}{l}
@@ -776,17 +792,23 @@ structure are set to valid values.  The mp\_init algorithm will perform such an
 \end{figure}
 
 \textbf{Algorithm mp\_init.}
-The \textbf{MP\_PREC} name represents a constant\footnote{Defined in the ``tommath.h'' header file within LibTomMath.} 
-used to dictate the minimum precision of allocated mp\_int integers.  Ideally, it is at least equal to $32$ since for most
-purposes that will be more than enough.
+The purpose of this function is to initialize an mp\_int structure so that the rest of the library can properly
+manipulte it.  It is assumed that the input may not have had any of its members previously initialized which is certainly
+a valid assumption if the input resides on the stack.  
+
+Before any of the members such as \textbf{sign}, \textbf{used} or \textbf{alloc} are initialized the memory for
+the digits is allocated.  If this fails the function returns before setting any of the other members.  The \textbf{MP\_PREC} 
+name represents a constant\footnote{Defined in the ``tommath.h'' header file within LibTomMath.} 
+used to dictate the minimum precision of newly initialized mp\_int integers.  Ideally, it is at least equal to the smallest
+precision number you'll be working with.
 
-Memory for the default number of digits is allocated first.  If the allocation fails the algorithm returns immediately
-with the \textbf{MP\_MEM} error code.  If the allocation succeeds the remaining members of the mp\_int structure
-must be initialized to reflect the default initial state.
+Allocating a block of digits at first instead of a single digit has the benefit of lowering the number of usually slow
+heap operations later functions will have to perform in the future.  If \textbf{MP\_PREC} is set correctly the slack 
+memory and the number of heap operations will be trivial.
 
-The allocated digits are all set to zero (step three) to ensure they are in a known state.  The \textbf{sign}, \textbf{used}
-and \textbf{alloc} are subsequently initialized to represent the zero integer.  By step seven the algorithm returns a success 
-code and the mp\_int $a$ has been successfully initialized to a valid state representing the integer zero.  
+Once the allocation has been made the digits have to be set to zero as well as the \textbf{used}, \textbf{sign} and
+\textbf{alloc} members initialized.  This ensures that the mp\_int will always represent the default state of zero regardless
+of the original condition of the input.
 
 \textbf{Remark.}
 This function introduces the idiosyncrasy that all iterative loops, commonly initiated with the ``for'' keyword, iterate incrementally
@@ -801,23 +823,31 @@ decrementally.
 \vspace{-3mm}
 \begin{alltt}
 016   
-017   /* init a new bigint */
+017   /* init a new mp_int */
 018   int mp_init (mp_int * a)
 019   \{
-020     /* allocate memory required and clear it */
-021     a->dp = OPT_CAST(mp_digit) XCALLOC (sizeof (mp_digit), MP_PREC);
-022     if (a->dp == NULL) \{
-023       return MP_MEM;
-024     \}
-025   
-026     /* set the used to zero, allocated digits to the default precision
-027      * and sign to positive */
-028     a->used  = 0;
-029     a->alloc = MP_PREC;
-030     a->sign  = MP_ZPOS;
-031   
-032     return MP_OKAY;
-033   \}
+020     int i;
+021   
+022     /* allocate memory required and clear it */
+023     a->dp = OPT_CAST(mp_digit) XMALLOC (sizeof (mp_digit) * MP_PREC);
+024     if (a->dp == NULL) \{
+025       return MP_MEM;
+026     \}
+027   
+028     /* set the digits to zero */
+029     for (i = 0; i < MP_PREC; i++) \{
+030         a->dp[i] = 0;
+031     \}
+032   
+033     /* set the used to zero, allocated digits to the default precision
+034      * and sign to positive */
+035     a->used  = 0;
+036     a->alloc = MP_PREC;
+037     a->sign  = MP_ZPOS;
+038   
+039     return MP_OKAY;
+040   \}
+041   #endif
 \end{alltt}
 \end{small}
 
@@ -825,19 +855,21 @@ One immediate observation of this initializtion function is that it does not ret
 is assumed that the caller has already allocated memory for the mp\_int structure, typically on the application stack.  The 
 call to mp\_init() is used only to initialize the members of the structure to a known default state.  
 
-Before any of the other members of the structure are initialized memory from the application heap is allocated with
-the calloc() function (line @22,calloc@).  The size of the allocated memory is large enough to hold \textbf{MP\_PREC} 
-mp\_digit variables.  The calloc() function is used instead\footnote{calloc() will allocate memory in the same
-manner as malloc() except that it also sets the contents to zero upon successfully allocating the memory.} of malloc() 
-since digits have to be set to zero for the function to finish correctly.  The \textbf{OPT\_CAST} token is a macro 
-definition which will turn into a cast from void * to mp\_digit * for C++ compilers.  It is not required for C compilers.
+Here we see (line 23) the memory allocation is performed first.  This allows us to exit cleanly and quickly
+if there is an error.  If the allocation fails the routine will return \textbf{MP\_MEM} to the caller to indicate there
+was a memory error.  The function XMALLOC is what actually allocates the memory.  Technically XMALLOC is not a function
+but a macro defined in ``tommath.h``.  By default, XMALLOC will evaluate to malloc() which is the C library's built--in
+memory allocation routine.
 
-After the memory has been successfully allocated the remainder of the members are initialized 
-(lines 28 through 30) to their respective default states.  At this point the algorithm has succeeded and
-a success code is returned to the calling function.
+In order to assure the mp\_int is in a known state the digits must be set to zero.  On most platforms this could have been
+accomplished by using calloc() instead of malloc().  However,  to correctly initialize a integer type to a given value in a 
+portable fashion you have to actually assign the value.  The for loop (line 29) performs this required
+operation.
 
-If this function returns \textbf{MP\_OKAY} it is safe to assume the mp\_int structure has been properly initialized and
-is safe to use with other functions within the library.  
+After the memory has been successfully initialized the remainder of the members are initialized 
+(lines 33 through 34) to their respective default states.  At this point the algorithm has succeeded and
+a success code is returned to the calling function.  If this function returns \textbf{MP\_OKAY} it is safe to assume the 
+mp\_int structure has been properly initialized and is safe to use with other functions within the library.  
 
 \subsection{Clearing an mp\_int}
 When an mp\_int is no longer required by the application, the memory that has been allocated for its digits must be 
@@ -848,7 +880,7 @@ returned to the application's memory pool with the mp\_clear algorithm.
 \begin{tabular}{l}
 \hline Algorithm \textbf{mp\_clear}. \\
 \textbf{Input}.   An mp\_int $a$ \\
-\textbf{Output}.  The memory for $a$ is freed for reuse.  \\
+\textbf{Output}.  The memory for $a$ shall be deallocated.  \\
 \hline \\
 1.  If $a$ has been previously freed then return(\textit{MP\_OKAY}). \\
 2.  for $n$ from 0 to $a.used - 1$ do \\
@@ -865,20 +897,15 @@ returned to the application's memory pool with the mp\_clear algorithm.
 \end{figure}
 
 \textbf{Algorithm mp\_clear.}
-This algorithm releases the memory allocated for an mp\_int back into the memory pool for reuse.  It is designed
-such that a given mp\_int structure can be cleared multiple times between initializations without attempting to 
-free the memory twice\footnote{In ISO C for example, calling free() twice on the same memory block causes undefinied
-behaviour.}.  
-
-The first step determines if the mp\_int structure has been marked as free already.  If it has, the algorithm returns
-success immediately as no further actions are required.  Otherwise, the algorithm will proceed to put the structure 
-in a known empty and otherwise invalid state.  First the digits of the mp\_int are set to zero.  The memory that has been allocated for the 
-digits is then freed.  The \textbf{used} and \textbf{alloc} counts are both set to zero and the \textbf{sign} set to 
-\textbf{MP\_ZPOS}.  This known fixed state for cleared mp\_int structures will make debuging easier for the end 
-developer.  That is, if they spot (via their debugger) an mp\_int they are using that is in this state it will be 
-obvious that they erroneously and prematurely cleared the mp\_int structure.
-
-Note that once an mp\_int has been cleared the mp\_int structure is no longer in a valid state for any other algorithm
+This algorithm accomplishes two goals.  First, it clears the digits and the other mp\_int members.  This ensures that 
+if a developer accidentally re-uses a cleared structure it is less likely to cause problems.  The second goal
+is to free the allocated memory.
+
+The logic behind the algorithm is extended by marking cleared mp\_int structures so that subsequent calls to this
+algorithm will not try to free the memory multiple times.  Cleared mp\_ints are detectable by having a pre-defined invalid 
+digit pointer \textbf{dp} setting.
+
+Once an mp\_int has been cleared the mp\_int structure is no longer in a valid state for any other algorithm
 with the exception of algorithms mp\_init, mp\_init\_copy, mp\_init\_size and mp\_clear.
 
 \vspace{+3mm}\begin{small}
@@ -890,31 +917,40 @@ with the exception of algorithms mp\_init, mp\_init\_copy, mp\_init\_size and mp
 018   void
 019   mp_clear (mp_int * a)
 020   \{
-021     /* only do anything if a hasn't been freed previously */
-022     if (a->dp != NULL) \{
-023       /* first zero the digits */
-024       memset (a->dp, 0, sizeof (mp_digit) * a->used);
-025   
-026       /* free ram */
-027       XFREE(a->dp);
-028   
-029       /* reset members to make debugging easier */
-030       a->dp    = NULL;
-031       a->alloc = a->used = 0;
-032       a->sign  = MP_ZPOS;
-033     \}
-034   \}
+021     int i;
+022   
+023     /* only do anything if a hasn't been freed previously */
+024     if (a->dp != NULL) \{
+025       /* first zero the digits */
+026       for (i = 0; i < a->used; i++) \{
+027           a->dp[i] = 0;
+028       \}
+029   
+030       /* free ram */
+031       XFREE(a->dp);
+032   
+033       /* reset members to make debugging easier */
+034       a->dp    = NULL;
+035       a->alloc = a->used = 0;
+036       a->sign  = MP_ZPOS;
+037     \}
+038   \}
+039   #endif
 \end{alltt}
 \end{small}
 
-The ``if'' statement (line 22) prevents the heap from being corrupted if a user double-frees an 
-mp\_int.  This is because once the memory is freed the pointer is set to \textbf{NULL} (line 30).  
+The algorithm only operates on the mp\_int if it hasn't been previously cleared.  The if statement (line 24)
+checks to see if the \textbf{dp} member is not \textbf{NULL}.  If the mp\_int is a valid mp\_int then \textbf{dp} cannot be
+\textbf{NULL} in which case the if statement will evaluate to true.
 
-Without the check, code that accidentally calls mp\_clear twice for a given mp\_int structure would try to free the memory 
-allocated for the digits twice.  This may cause some C libraries to signal a fault.  By setting the pointer to 
-\textbf{NULL} it helps debug code that may inadvertently free the mp\_int before it is truly not needed, because attempts 
-to reference digits should fail immediately. The allocated digits are set to zero before being freed (line 24).  
-This is ideal for cryptographic situations where the integer that the mp\_int represents might need to be kept a secret.
+The digits of the mp\_int are cleared by the for loop (line 26) which assigns a zero to every digit.  Similar to mp\_init()
+the digits are assigned zero instead of using block memory operations (such as memset()) since this is more portable.  
+
+The digits are deallocated off the heap via the XFREE macro.  Similar to XMALLOC the XFREE macro actually evaluates to
+a standard C library function.  In this case the free() function.  Since free() only deallocates the memory the pointer
+still has to be reset to \textbf{NULL} manually (line 34).  
+
+Now that the digits have been cleared and deallocated the other members are set to their final values (lines 35 and 36).
 
 \section{Maintenance Algorithms}
 
@@ -942,7 +978,7 @@ must be re-sized appropriately to accomodate the result.  The mp\_grow algorithm
 1.  if $a.alloc \ge b$ then return(\textit{MP\_OKAY}) \\
 2.  $u \leftarrow b\mbox{ (mod }MP\_PREC\mbox{)}$ \\
 3.  $v \leftarrow b + 2 \cdot MP\_PREC - u$ \\
-4.  Re-Allocate the array of digits $a$ to size $v$ \\
+4.  Re-allocate the array of digits $a$ to size $v$ \\
 5.  If the allocation failed then return(\textit{MP\_MEM}). \\
 6.  for n from a.alloc to $v - 1$ do  \\
 \hspace{+3mm}6.1  $a_n \leftarrow 0$ \\
@@ -1005,18 +1041,23 @@ assumed to contain undefined values they are initially set to zero.
 049     \}
 050     return MP_OKAY;
 051   \}
+052   #endif
 \end{alltt}
 \end{small}
 
-The first step is to see if we actually need to perform a re-allocation at all (line 24).  If a reallocation
-must occur the digit count is padded upwards to help prevent many trivial reallocations (line 26).  Next the reallocation is performed
-and the return of realloc() is stored in a temporary pointer named $tmp$ (line 36).  The return is stored in a temporary
-instead of $a.dp$ to prevent the code from losing the original pointer in case the reallocation fails.  Had the return been stored 
-in $a.dp$ instead there would be no way to reclaim the heap originally used.
+A quick optimization is to first determine if a memory re-allocation is required at all.  The if statement (line 23) checks
+if the \textbf{alloc} member of the mp\_int is smaller than the requested digit count.  If the count is not larger than \textbf{alloc}
+the function skips the re-allocation part thus saving time.
+
+When a re-allocation is performed it is turned into an optimal request to save time in the future.  The requested digit count is
+padded upwards to 2nd multiple of \textbf{MP\_PREC} larger than \textbf{alloc} (line 26).  The XREALLOC function is used
+to re-allocate the memory.  As per the other functions XREALLOC is actually a macro which evaluates to realloc by default.  The realloc
+function leaves the base of the allocation intact which means the first \textbf{alloc} digits of the mp\_int are the same as before
+the re-allocation.  All	that is left is to clear the newly allocated digits and return.
 
-If the reallocation fails the function will return \textbf{MP\_MEM} (line 37), otherwise, the value of $tmp$ is assigned
-to the pointer $a.dp$ and the function continues.  A simple for loop from line 46 to line 51 will zero all digits 
-that were above the old \textbf{alloc} limit to make sure the integer is in a known state.
+Note that the re-allocation result is actually stored in a temporary pointer $tmp$.  This is to allow this function to return
+an error with a valid pointer.  Earlier releases of the library stored the result of XREALLOC into the mp\_int $a$.  That would
+result in a memory leak if XREALLOC ever failed.  
 
 \subsection{Initializing Variable Precision mp\_ints}
 Occasionally the number of digits required will be known in advance of an initialization, based on, for example, the size 
@@ -1065,31 +1106,41 @@ correct no further memory re-allocations are required to work with the mp\_int.
 017   /* init an mp_init for a given size */
 018   int mp_init_size (mp_int * a, int size)
 019   \{
-020     /* pad size so there are always extra digits */
-021     size += (MP_PREC * 2) - (size % MP_PREC);    
-022     
-023     /* alloc mem */
-024     a->dp = OPT_CAST(mp_digit) XCALLOC (sizeof (mp_digit), size);
-025     if (a->dp == NULL) \{
-026       return MP_MEM;
-027     \}
-028     a->used  = 0;
-029     a->alloc = size;
-030     a->sign  = MP_ZPOS;
-031   
-032     return MP_OKAY;
-033   \}
+020     int x;
+021   
+022     /* pad size so there are always extra digits */
+023     size += (MP_PREC * 2) - (size % MP_PREC);    
+024     
+025     /* alloc mem */
+026     a->dp = OPT_CAST(mp_digit) XMALLOC (sizeof (mp_digit) * size);
+027     if (a->dp == NULL) \{
+028       return MP_MEM;
+029     \}
+030   
+031     /* set the members */
+032     a->used  = 0;
+033     a->alloc = size;
+034     a->sign  = MP_ZPOS;
+035   
+036     /* zero the digits */
+037     for (x = 0; x < size; x++) \{
+038         a->dp[x] = 0;
+039     \}
+040   
+041     return MP_OKAY;
+042   \}
+043   #endif
 \end{alltt}
 \end{small}
 
-The number of digits $b$ requested is padded (line 21) by first augmenting it to the next multiple of 
+The number of digits $b$ requested is padded (line 23) by first augmenting it to the next multiple of 
 \textbf{MP\_PREC} and then adding \textbf{MP\_PREC} to the result.  If the memory can be successfully allocated the 
 mp\_int is placed in a default state representing the integer zero.  Otherwise, the error code \textbf{MP\_MEM} will be 
-returned (line 26).  
+returned (line 28).  
 
-The digits are allocated and set to zero at the same time with the calloc() function (line @25,calloc@).  The 
+The digits are allocated and set to zero at the same time with the calloc() function (line @25,XCALLOC@).  The 
 \textbf{used} count is set to zero, the \textbf{alloc} count set to the padded digit count and the \textbf{sign} flag set 
-to \textbf{MP\_ZPOS} to achieve a default valid mp\_int state (lines 28, 29 and 30).  If the function 
+to \textbf{MP\_ZPOS} to achieve a default valid mp\_int state (lines 32, 33 and 34).  If the function 
 returns succesfully then it is correct to assume that the mp\_int structure is in a valid state for the remainder of the 
 functions to work with.
 
@@ -1165,6 +1216,7 @@ initialization which allows for quick recovery from runtime errors.
 051       return res;                /* Assumed ok, if error flagged above. */
 052   \}
 053   
+054   #endif
 \end{alltt}
 \end{small}
 
@@ -1249,6 +1301,7 @@ when all of the digits are zero to ensure that the mp\_int is valid at all times
 036       a->sign = MP_ZPOS;
 037     \}
 038   \}
+039   #endif
 \end{alltt}
 \end{small}
 
@@ -1385,6 +1438,7 @@ implement the pseudo-code.
 060     b->sign = a->sign;
 061     return MP_OKAY;
 062   \}
+063   #endif
 \end{alltt}
 \end{small}
 
@@ -1498,6 +1552,7 @@ such this algorithm will perform two operations in one step.
 024     \}
 025     return mp_copy (b, a);
 026   \}
+027   #endif
 \end{alltt}
 \end{small}
 
@@ -1542,6 +1597,7 @@ This algorithm simply resets a mp\_int to the default state.
 022     a->used = 0;
 023     memset (a->dp, 0, sizeof (mp_digit) * a->alloc);
 024   \}
+025   #endif
 \end{alltt}
 \end{small}
 
@@ -1602,6 +1658,7 @@ logic to handle it.
 035   
 036     return MP_OKAY;
 037   \}
+038   #endif
 \end{alltt}
 \end{small}
 
@@ -1653,6 +1710,7 @@ zero as negative.
 026     \}
 027     return MP_OKAY;
 028   \}
+029   #endif
 \end{alltt}
 \end{small}
 
@@ -1695,6 +1753,7 @@ single digit is set (\textit{modulo $\beta$}) and the \textbf{used} count is adj
 021     a->dp[0] = b & MP_MASK;
 022     a->used  = (a->dp[0] != 0) ? 1 : 0;
 023   \}
+024   #endif
 \end{alltt}
 \end{small}
 
@@ -1771,6 +1830,7 @@ Excess zero digits are trimmed in steps 2.1 and 3 by using higher level algorith
 040     mp_clamp (a);
 041     return MP_OKAY;
 042   \}
+043   #endif
 \end{alltt}
 \end{small}
 
@@ -1872,6 +1932,7 @@ the zero'th digit.  If after all of the digits have been compared, no difference
 047     \}
 048     return MP_EQ;
 049   \}
+050   #endif
 \end{alltt}
 \end{small}
 
@@ -1935,6 +1996,7 @@ $\vert a \vert < \vert b \vert$.  Step number four will compare the two when the
 035        return mp_cmp_mag(a, b);
 036     \}
 037   \}
+038   #endif
 \end{alltt}
 \end{small}
 
@@ -2152,6 +2214,7 @@ The final carry is stored in $c_{max}$ and digits above $max$ upto $oldused$ are
 101     mp_clamp (c);
 102     return MP_OKAY;
 103   \}
+104   #endif
 \end{alltt}
 \end{small}
 
@@ -2320,6 +2383,7 @@ If $b$ has a smaller magnitude than $a$ then step 9 will force the carry and cop
 081     return MP_OKAY;
 082   \}
 083   
+084   #endif
 \end{alltt}
 \end{small}
 
@@ -2451,6 +2515,7 @@ within algorithm s\_mp\_add will force $-0$ to become $0$.
 045     return res;
 046   \}
 047   
+048   #endif
 \end{alltt}
 \end{small}
 
@@ -2562,6 +2627,7 @@ algorithm from producing $-a - -a = -0$ as a result.
 051     return res;
 052   \}
 053   
+054   #endif
 \end{alltt}
 \end{small}
 
@@ -2695,6 +2761,7 @@ Step 8 clears any leading digits of $b$ in case it originally had a larger magni
 074     b->sign = a->sign;
 075     return MP_OKAY;
 076   \}
+077   #endif
 \end{alltt}
 \end{small}
 
@@ -2794,6 +2861,7 @@ least significant bit not the most significant bit.
 060     mp_clamp (b);
 061     return MP_OKAY;
 062   \}
+063   #endif
 \end{alltt}
 \end{small}
 
@@ -2913,6 +2981,7 @@ step 8 sets the lower $b$ digits to zero.
 059     \}
 060     return MP_OKAY;
 061   \}
+062   #endif
 \end{alltt}
 \end{small}
 
@@ -3022,6 +3091,7 @@ Once the window copy is complete the upper digits must be zeroed and the \textbf
 064     /* remove excess digits */
 065     a->used -= b;
 066   \}
+067   #endif
 \end{alltt}
 \end{small}
 
@@ -3154,6 +3224,7 @@ complete.  It is possible to optimize this algorithm down to a $O(n)$ algorithm
 077     mp_clamp (c);
 078     return MP_OKAY;
 079   \}
+080   #endif
 \end{alltt}
 \end{small}
 
@@ -3281,6 +3352,7 @@ by using algorithm mp\_mod\_2d.
 089     mp_clear (&t);
 090     return MP_OKAY;
 091   \}
+092   #endif
 \end{alltt}
 \end{small}
 
@@ -3370,6 +3442,7 @@ is copied to $b$, leading digits are removed and the remaining leading digit is
 047     mp_clamp (c);
 048     return MP_OKAY;
 049   \}
+050   #endif
 \end{alltt}
 \end{small}
 
@@ -3603,6 +3676,7 @@ exceed the precision requested.
 083     mp_clear (&t);
 084     return MP_OKAY;
 085   \}
+086   #endif
 \end{alltt}
 \end{small}
 
@@ -3806,109 +3880,82 @@ and addition operations in the nested loop in parallel.
 033   int
 034   fast_s_mp_mul_digs (mp_int * a, mp_int * b, mp_int * c, int digs)
 035   \{
-036     int     olduse, res, pa, ix;
-037     mp_word W[MP_WARRAY];
-038   
-039     /* grow the destination as required */
-040     if (c->alloc < digs) \{
-041       if ((res = mp_grow (c, digs)) != MP_OKAY) \{
-042         return res;
-043       \}
-044     \}
-045   
-046     /* clear temp buf (the columns) */
-047     memset (W, 0, sizeof (mp_word) * digs);
-048   
-049     /* calculate the columns */
-050     pa = a->used;
-051     for (ix = 0; ix < pa; ix++) \{
-052       /* this multiplier has been modified to allow you to 
-053        * control how many digits of output are produced.  
-054        * So at most we want to make upto "digs" digits of output.
-055        *
-056        * this adds products to distinct columns (at ix+iy) of W
-057        * note that each step through the loop is not dependent on
-058        * the previous which means the compiler can easily unroll
-059        * the loop without scheduling problems
-060        */
-061       \{
-062         register mp_digit tmpx, *tmpy;
-063         register mp_word *_W;
-064         register int iy, pb;
-065   
-066         /* alias for the the word on the left e.g. A[ix] * A[iy] */
-067         tmpx = a->dp[ix];
-068   
-069         /* alias for the right side */
-070         tmpy = b->dp;
-071   
-072         /* alias for the columns, each step through the loop adds a new
-073            term to each column
-074          */
-075         _W = W + ix;
-076   
-077         /* the number of digits is limited by their placement.  E.g.
-078            we avoid multiplying digits that will end up above the # of
-079            digits of precision requested
-080          */
-081         pb = MIN (b->used, digs - ix);
-082   
-083         for (iy = 0; iy < pb; iy++) \{
-084           *_W++ += ((mp_word)tmpx) * ((mp_word)*tmpy++);
-085         \}
-086       \}
-087   
-088     \}
-089   
-090     /* setup dest */
-091     olduse = c->used;
-092     c->used = digs;
+036     int     olduse, res, pa, ix, iz;
+037     mp_digit W[MP_WARRAY];
+038     register mp_word  _W;
+039   
+040     /* grow the destination as required */
+041     if (c->alloc < digs) \{
+042       if ((res = mp_grow (c, digs)) != MP_OKAY) \{
+043         return res;
+044       \}
+045     \}
+046   
+047     /* number of output digits to produce */
+048     pa = MIN(digs, a->used + b->used);
+049   
+050     /* clear the carry */
+051     _W = 0;
+052     for (ix = 0; ix <= pa; ix++) \{ 
+053         int      tx, ty;
+054         int      iy;
+055         mp_digit *tmpx, *tmpy;
+056   
+057         /* get offsets into the two bignums */
+058         ty = MIN(b->used-1, ix);
+059         tx = ix - ty;
+060   
+061         /* setup temp aliases */
+062         tmpx = a->dp + tx;
+063         tmpy = b->dp + ty;
+064   
+065         /* this is the number of times the loop will iterrate, essentially its
+       
+066            while (tx++ < a->used && ty-- >= 0) \{ ... \}
+067          */
+068         iy = MIN(a->used-tx, ty+1);
+069   
+070         /* execute loop */
+071         for (iz = 0; iz < iy; ++iz) \{
+072            _W += ((mp_word)*tmpx++)*((mp_word)*tmpy--);
+073         \}
+074   
+075         /* store term */
+076         W[ix] = ((mp_digit)_W) & MP_MASK;
+077   
+078         /* make next carry */
+079         _W = _W >> ((mp_word)DIGIT_BIT);
+080     \}
+081   
+082     /* setup dest */
+083     olduse  = c->used;
+084     c->used = digs;
+085   
+086     \{
+087       register mp_digit *tmpc;
+088       tmpc = c->dp;
+089       for (ix = 0; ix < digs; ix++) \{
+090         /* now extract the previous digit [below the carry] */
+091         *tmpc++ = W[ix];
+092       \}
 093   
-094     \{
-095       register mp_digit *tmpc;
-096   
-097       /* At this point W[] contains the sums of each column.  To get the
-098        * correct result we must take the extra bits from each column and
-099        * carry them down
-100        *
-101        * Note that while this adds extra code to the multiplier it 
-102        * saves time since the carry propagation is removed from the 
-103        * above nested loop.This has the effect of reducing the work 
-104        * from N*(N+N*c)==N**2 + c*N**2 to N**2 + N*c where c is the 
-105        * cost of the shifting.  On very small numbers this is slower 
-106        * but on most cryptographic size numbers it is faster.
-107        *
-108        * In this particular implementation we feed the carries from
-109        * behind which means when the loop terminates we still have one
-110        * last digit to copy
-111        */
-112       tmpc = c->dp;
-113       for (ix = 1; ix < digs; ix++) \{
-114         /* forward the carry from the previous temp */
-115         W[ix] += (W[ix - 1] >> ((mp_word) DIGIT_BIT));
-116   
-117         /* now extract the previous digit [below the carry] */
-118         *tmpc++ = (mp_digit) (W[ix - 1] & ((mp_word) MP_MASK));
-119       \}
-120       /* fetch the last digit */
-121       *tmpc++ = (mp_digit) (W[digs - 1] & ((mp_word) MP_MASK));
-122   
-123       /* clear unused digits [that existed in the old copy of c] */
-124       for (; ix < olduse; ix++) \{
-125         *tmpc++ = 0;
-126       \}
-127     \}
-128     mp_clamp (c);
-129     return MP_OKAY;
-130   \}
+094       /* clear unused digits [that existed in the old copy of c] */
+095       for (; ix < olduse; ix++) \{
+096         *tmpc++ = 0;
+097       \}
+098     \}
+099     mp_clamp (c);
+100     return MP_OKAY;
+101   \}
+102   #endif
 \end{alltt}
 \end{small}
 
-The memset on line 47 clears the initial $\hat W$ array to zero in a single step. Like the slower baseline multiplication
-implementation a series of aliases (\textit{lines 67, 70 and 75}) are used to simplify the inner $O(n^2)$ loop.  
+The memset on line @47,memset@ clears the initial $\hat W$ array to zero in a single step. Like the slower baseline multiplication
+implementation a series of aliases (\textit{lines 62, 63 and 76}) are used to simplify the inner $O(n^2)$ loop.  
 In this case a new alias $\_\hat W$ has been added which refers to the double precision columns offset by $ix$ in each pass.  
 
-The inner loop on lines 83, 84 and 85 is where the algorithm will spend the majority of the time, which is why it has been 
+The inner loop on lines 89, 79 and 80 is where the algorithm will spend the majority of the time, which is why it has been 
 stripped to the bones of any extra baggage\footnote{Hence the pointer aliases.}.  On x86 processors the multiplication and additions amount to at the 
 very least five instructions (\textit{two loads, two additions, one multiply}) while on the ARMv4 processors they amount to only three 
 (\textit{one load, one store, one multiply-add}).   For both of the x86 and ARMv4 processors the GCC compiler performs a good job at unrolling the loop 
@@ -4168,92 +4215,90 @@ The remaining steps 13 through 18 compute the Karatsuba polynomial through a var
 076       goto X0Y0;
 077   
 078     /* now shift the digits */
-079     x0.sign = x1.sign = a->sign;
-080     y0.sign = y1.sign = b->sign;
-081   
-082     x0.used = y0.used = B;
-083     x1.used = a->used - B;
-084     y1.used = b->used - B;
-085   
-086     \{
-087       register int x;
-088       register mp_digit *tmpa, *tmpb, *tmpx, *tmpy;
-089   
-090       /* we copy the digits directly instead of using higher level functions
-091        * since we also need to shift the digits
-092        */
-093       tmpa = a->dp;
-094       tmpb = b->dp;
-095   
-096       tmpx = x0.dp;
-097       tmpy = y0.dp;
-098       for (x = 0; x < B; x++) \{
-099         *tmpx++ = *tmpa++;
-100         *tmpy++ = *tmpb++;
-101       \}
-102   
-103       tmpx = x1.dp;
-104       for (x = B; x < a->used; x++) \{
-105         *tmpx++ = *tmpa++;
-106       \}
-107   
-108       tmpy = y1.dp;
-109       for (x = B; x < b->used; x++) \{
-110         *tmpy++ = *tmpb++;
-111       \}
-112     \}
-113   
-114     /* only need to clamp the lower words since by definition the 
-115      * upper words x1/y1 must have a known number of digits
-116      */
-117     mp_clamp (&x0);
-118     mp_clamp (&y0);
-119   
-120     /* now calc the products x0y0 and x1y1 */
-121     /* after this x0 is no longer required, free temp [x0==t2]! */
-122     if (mp_mul (&x0, &y0, &x0y0) != MP_OKAY)  
-123       goto X1Y1;          /* x0y0 = x0*y0 */
-124     if (mp_mul (&x1, &y1, &x1y1) != MP_OKAY)
-125       goto X1Y1;          /* x1y1 = x1*y1 */
-126   
-127     /* now calc x1-x0 and y1-y0 */
-128     if (mp_sub (&x1, &x0, &t1) != MP_OKAY)
-129       goto X1Y1;          /* t1 = x1 - x0 */
-130     if (mp_sub (&y1, &y0, &x0) != MP_OKAY)
-131       goto X1Y1;          /* t2 = y1 - y0 */
-132     if (mp_mul (&t1, &x0, &t1) != MP_OKAY)
-133       goto X1Y1;          /* t1 = (x1 - x0) * (y1 - y0) */
-134   
-135     /* add x0y0 */
-136     if (mp_add (&x0y0, &x1y1, &x0) != MP_OKAY)
-137       goto X1Y1;          /* t2 = x0y0 + x1y1 */
-138     if (mp_sub (&x0, &t1, &t1) != MP_OKAY)
-139       goto X1Y1;          /* t1 = x0y0 + x1y1 - (x1-x0)*(y1-y0) */
-140   
-141     /* shift by B */
-142     if (mp_lshd (&t1, B) != MP_OKAY)
-143       goto X1Y1;          /* t1 = (x0y0 + x1y1 - (x1-x0)*(y1-y0))<<B */
-144     if (mp_lshd (&x1y1, B * 2) != MP_OKAY)
-145       goto X1Y1;          /* x1y1 = x1y1 << 2*B */
-146   
-147     if (mp_add (&x0y0, &t1, &t1) != MP_OKAY)
-148       goto X1Y1;          /* t1 = x0y0 + t1 */
-149     if (mp_add (&t1, &x1y1, c) != MP_OKAY)
-150       goto X1Y1;          /* t1 = x0y0 + t1 + x1y1 */
+079     x0.used = y0.used = B;
+080     x1.used = a->used - B;
+081     y1.used = b->used - B;
+082   
+083     \{
+084       register int x;
+085       register mp_digit *tmpa, *tmpb, *tmpx, *tmpy;
+086   
+087       /* we copy the digits directly instead of using higher level functions
+088        * since we also need to shift the digits
+089        */
+090       tmpa = a->dp;
+091       tmpb = b->dp;
+092   
+093       tmpx = x0.dp;
+094       tmpy = y0.dp;
+095       for (x = 0; x < B; x++) \{
+096         *tmpx++ = *tmpa++;
+097         *tmpy++ = *tmpb++;
+098       \}
+099   
+100       tmpx = x1.dp;
+101       for (x = B; x < a->used; x++) \{
+102         *tmpx++ = *tmpa++;
+103       \}
+104   
+105       tmpy = y1.dp;
+106       for (x = B; x < b->used; x++) \{
+107         *tmpy++ = *tmpb++;
+108       \}
+109     \}
+110   
+111     /* only need to clamp the lower words since by definition the 
+112      * upper words x1/y1 must have a known number of digits
+113      */
+114     mp_clamp (&x0);
+115     mp_clamp (&y0);
+116   
+117     /* now calc the products x0y0 and x1y1 */
+118     /* after this x0 is no longer required, free temp [x0==t2]! */
+119     if (mp_mul (&x0, &y0, &x0y0) != MP_OKAY)  
+120       goto X1Y1;          /* x0y0 = x0*y0 */
+121     if (mp_mul (&x1, &y1, &x1y1) != MP_OKAY)
+122       goto X1Y1;          /* x1y1 = x1*y1 */
+123   
+124     /* now calc x1-x0 and y1-y0 */
+125     if (mp_sub (&x1, &x0, &t1) != MP_OKAY)
+126       goto X1Y1;          /* t1 = x1 - x0 */
+127     if (mp_sub (&y1, &y0, &x0) != MP_OKAY)
+128       goto X1Y1;          /* t2 = y1 - y0 */
+129     if (mp_mul (&t1, &x0, &t1) != MP_OKAY)
+130       goto X1Y1;          /* t1 = (x1 - x0) * (y1 - y0) */
+131   
+132     /* add x0y0 */
+133     if (mp_add (&x0y0, &x1y1, &x0) != MP_OKAY)
+134       goto X1Y1;          /* t2 = x0y0 + x1y1 */
+135     if (mp_sub (&x0, &t1, &t1) != MP_OKAY)
+136       goto X1Y1;          /* t1 = x0y0 + x1y1 - (x1-x0)*(y1-y0) */
+137   
+138     /* shift by B */
+139     if (mp_lshd (&t1, B) != MP_OKAY)
+140       goto X1Y1;          /* t1 = (x0y0 + x1y1 - (x1-x0)*(y1-y0))<<B */
+141     if (mp_lshd (&x1y1, B * 2) != MP_OKAY)
+142       goto X1Y1;          /* x1y1 = x1y1 << 2*B */
+143   
+144     if (mp_add (&x0y0, &t1, &t1) != MP_OKAY)
+145       goto X1Y1;          /* t1 = x0y0 + t1 */
+146     if (mp_add (&t1, &x1y1, c) != MP_OKAY)
+147       goto X1Y1;          /* t1 = x0y0 + t1 + x1y1 */
+148   
+149     /* Algorithm succeeded set the return code to MP_OKAY */
+150     err = MP_OKAY;
 151   
-152     /* Algorithm succeeded set the return code to MP_OKAY */
-153     err = MP_OKAY;
-154   
-155   X1Y1:mp_clear (&x1y1);
-156   X0Y0:mp_clear (&x0y0);
-157   T1:mp_clear (&t1);
-158   Y1:mp_clear (&y1);
-159   Y0:mp_clear (&y0);
-160   X1:mp_clear (&x1);
-161   X0:mp_clear (&x0);
-162   ERR:
-163     return err;
-164   \}
+152   X1Y1:mp_clear (&x1y1);
+153   X0Y0:mp_clear (&x0y0);
+154   T1:mp_clear (&t1);
+155   Y1:mp_clear (&y1);
+156   Y0:mp_clear (&y0);
+157   X1:mp_clear (&x1);
+158   X0:mp_clear (&x0);
+159   ERR:
+160     return err;
+161   \}
+162   #endif
 \end{alltt}
 \end{small}
 
@@ -4269,13 +4314,13 @@ number of digits for the next section of code.
 
 The first algebraic portion of the algorithm is to split the two inputs into their halves.  However, instead of using mp\_mod\_2d and mp\_rshd
 to extract the halves, the respective code has been placed inline within the body of the function.  To initialize the halves, the \textbf{used} and 
-\textbf{sign} members are copied first.  The first for loop on line 98 copies the lower halves.  Since they are both the same magnitude it 
-is simpler to calculate both lower halves in a single loop.  The for loop on lines 104 and 109 calculate the upper halves $x1$ and 
+\textbf{sign} members are copied first.  The first for loop on line 101 copies the lower halves.  Since they are both the same magnitude it 
+is simpler to calculate both lower halves in a single loop.  The for loop on lines 106 and 106 calculate the upper halves $x1$ and 
 $y1$ respectively.
 
 By inlining the calculation of the halves, the Karatsuba multiplier has a slightly lower overhead and can be used for smaller magnitude inputs.
 
-When line 153 is reached, the algorithm has completed succesfully.  The ``error status'' variable $err$ is set to \textbf{MP\_OKAY} so that
+When line 150 is reached, the algorithm has completed succesfully.  The ``error status'' variable $err$ is set to \textbf{MP\_OKAY} so that
 the same code that handles errors can be used to clear the temporary variables and return.  
 
 \subsection{Toom-Cook $3$-Way Multiplication}
@@ -4394,262 +4439,269 @@ result $a \cdot b$ is produced.
 \vspace{-3mm}
 \begin{alltt}
 016   
-017   /* multiplication using the Toom-Cook 3-way algorithm */
-018   int mp_toom_mul(mp_int *a, mp_int *b, mp_int *c)
-019   \{
-020       mp_int w0, w1, w2, w3, w4, tmp1, tmp2, a0, a1, a2, b0, b1, b2;
-021       int res, B;
-022           
-023       /* init temps */
-024       if ((res = mp_init_multi(&w0, &w1, &w2, &w3, &w4, 
-025                                &a0, &a1, &a2, &b0, &b1, 
-026                                &b2, &tmp1, &tmp2, NULL)) != MP_OKAY) \{
-027          return res;
-028       \}
-029       
-030       /* B */
-031       B = MIN(a->used, b->used) / 3;
-032       
-033       /* a = a2 * B**2 + a1 * B + a0 */
-034       if ((res = mp_mod_2d(a, DIGIT_BIT * B, &a0)) != MP_OKAY) \{
-035          goto ERR;
-036       \}
-037   
-038       if ((res = mp_copy(a, &a1)) != MP_OKAY) \{
-039          goto ERR;
-040       \}
-041       mp_rshd(&a1, B);
-042       mp_mod_2d(&a1, DIGIT_BIT * B, &a1);
-043   
-044       if ((res = mp_copy(a, &a2)) != MP_OKAY) \{
-045          goto ERR;
-046       \}
-047       mp_rshd(&a2, B*2);
-048       
-049       /* b = b2 * B**2 + b1 * B + b0 */
-050       if ((res = mp_mod_2d(b, DIGIT_BIT * B, &b0)) != MP_OKAY) \{
-051          goto ERR;
-052       \}
-053   
-054       if ((res = mp_copy(b, &b1)) != MP_OKAY) \{
-055          goto ERR;
-056       \}
-057       mp_rshd(&b1, B);
-058       mp_mod_2d(&b1, DIGIT_BIT * B, &b1);
-059   
-060       if ((res = mp_copy(b, &b2)) != MP_OKAY) \{
-061          goto ERR;
-062       \}
-063       mp_rshd(&b2, B*2);
-064       
-065       /* w0 = a0*b0 */
-066       if ((res = mp_mul(&a0, &b0, &w0)) != MP_OKAY) \{
-067          goto ERR;
-068       \}
+017   /* multiplication using the Toom-Cook 3-way algorithm 
+018    *
+019    * Much more complicated than Karatsuba but has a lower asymptotic running t
+      ime of 
+020    * O(N**1.464).  This algorithm is only particularly useful on VERY large
+021    * inputs (we're talking 1000s of digits here...).
+022   */
+023   int mp_toom_mul(mp_int *a, mp_int *b, mp_int *c)
+024   \{
+025       mp_int w0, w1, w2, w3, w4, tmp1, tmp2, a0, a1, a2, b0, b1, b2;
+026       int res, B;
+027           
+028       /* init temps */
+029       if ((res = mp_init_multi(&w0, &w1, &w2, &w3, &w4, 
+030                                &a0, &a1, &a2, &b0, &b1, 
+031                                &b2, &tmp1, &tmp2, NULL)) != MP_OKAY) \{
+032          return res;
+033       \}
+034       
+035       /* B */
+036       B = MIN(a->used, b->used) / 3;
+037       
+038       /* a = a2 * B**2 + a1 * B + a0 */
+039       if ((res = mp_mod_2d(a, DIGIT_BIT * B, &a0)) != MP_OKAY) \{
+040          goto ERR;
+041       \}
+042   
+043       if ((res = mp_copy(a, &a1)) != MP_OKAY) \{
+044          goto ERR;
+045       \}
+046       mp_rshd(&a1, B);
+047       mp_mod_2d(&a1, DIGIT_BIT * B, &a1);
+048   
+049       if ((res = mp_copy(a, &a2)) != MP_OKAY) \{
+050          goto ERR;
+051       \}
+052       mp_rshd(&a2, B*2);
+053       
+054       /* b = b2 * B**2 + b1 * B + b0 */
+055       if ((res = mp_mod_2d(b, DIGIT_BIT * B, &b0)) != MP_OKAY) \{
+056          goto ERR;
+057       \}
+058   
+059       if ((res = mp_copy(b, &b1)) != MP_OKAY) \{
+060          goto ERR;
+061       \}
+062       mp_rshd(&b1, B);
+063       mp_mod_2d(&b1, DIGIT_BIT * B, &b1);
+064   
+065       if ((res = mp_copy(b, &b2)) != MP_OKAY) \{
+066          goto ERR;
+067       \}
+068       mp_rshd(&b2, B*2);
 069       
-070       /* w4 = a2 * b2 */
-071       if ((res = mp_mul(&a2, &b2, &w4)) != MP_OKAY) \{
+070       /* w0 = a0*b0 */
+071       if ((res = mp_mul(&a0, &b0, &w0)) != MP_OKAY) \{
 072          goto ERR;
 073       \}
 074       
-075       /* w1 = (a2 + 2(a1 + 2a0))(b2 + 2(b1 + 2b0)) */
-076       if ((res = mp_mul_2(&a0, &tmp1)) != MP_OKAY) \{
+075       /* w4 = a2 * b2 */
+076       if ((res = mp_mul(&a2, &b2, &w4)) != MP_OKAY) \{
 077          goto ERR;
 078       \}
-079       if ((res = mp_add(&tmp1, &a1, &tmp1)) != MP_OKAY) \{
-080          goto ERR;
-081       \}
-082       if ((res = mp_mul_2(&tmp1, &tmp1)) != MP_OKAY) \{
-083          goto ERR;
-084       \}
-085       if ((res = mp_add(&tmp1, &a2, &tmp1)) != MP_OKAY) \{
-086          goto ERR;
-087       \}
-088       
-089       if ((res = mp_mul_2(&b0, &tmp2)) != MP_OKAY) \{
-090          goto ERR;
-091       \}
-092       if ((res = mp_add(&tmp2, &b1, &tmp2)) != MP_OKAY) \{
-093          goto ERR;
-094       \}
-095       if ((res = mp_mul_2(&tmp2, &tmp2)) != MP_OKAY) \{
-096          goto ERR;
-097       \}
-098       if ((res = mp_add(&tmp2, &b2, &tmp2)) != MP_OKAY) \{
-099          goto ERR;
-100       \}
-101       
-102       if ((res = mp_mul(&tmp1, &tmp2, &w1)) != MP_OKAY) \{
-103          goto ERR;
-104       \}
-105       
-106       /* w3 = (a0 + 2(a1 + 2a2))(b0 + 2(b1 + 2b2)) */
-107       if ((res = mp_mul_2(&a2, &tmp1)) != MP_OKAY) \{
+079       
+080       /* w1 = (a2 + 2(a1 + 2a0))(b2 + 2(b1 + 2b0)) */
+081       if ((res = mp_mul_2(&a0, &tmp1)) != MP_OKAY) \{
+082          goto ERR;
+083       \}
+084       if ((res = mp_add(&tmp1, &a1, &tmp1)) != MP_OKAY) \{
+085          goto ERR;
+086       \}
+087       if ((res = mp_mul_2(&tmp1, &tmp1)) != MP_OKAY) \{
+088          goto ERR;
+089       \}
+090       if ((res = mp_add(&tmp1, &a2, &tmp1)) != MP_OKAY) \{
+091          goto ERR;
+092       \}
+093       
+094       if ((res = mp_mul_2(&b0, &tmp2)) != MP_OKAY) \{
+095          goto ERR;
+096       \}
+097       if ((res = mp_add(&tmp2, &b1, &tmp2)) != MP_OKAY) \{
+098          goto ERR;
+099       \}
+100       if ((res = mp_mul_2(&tmp2, &tmp2)) != MP_OKAY) \{
+101          goto ERR;
+102       \}
+103       if ((res = mp_add(&tmp2, &b2, &tmp2)) != MP_OKAY) \{
+104          goto ERR;
+105       \}
+106       
+107       if ((res = mp_mul(&tmp1, &tmp2, &w1)) != MP_OKAY) \{
 108          goto ERR;
 109       \}
-110       if ((res = mp_add(&tmp1, &a1, &tmp1)) != MP_OKAY) \{
-111          goto ERR;
-112       \}
-113       if ((res = mp_mul_2(&tmp1, &tmp1)) != MP_OKAY) \{
-114          goto ERR;
-115       \}
-116       if ((res = mp_add(&tmp1, &a0, &tmp1)) != MP_OKAY) \{
-117          goto ERR;
-118       \}
-119       
-120       if ((res = mp_mul_2(&b2, &tmp2)) != MP_OKAY) \{
-121          goto ERR;
-122       \}
-123       if ((res = mp_add(&tmp2, &b1, &tmp2)) != MP_OKAY) \{
-124          goto ERR;
-125       \}
-126       if ((res = mp_mul_2(&tmp2, &tmp2)) != MP_OKAY) \{
-127          goto ERR;
-128       \}
-129       if ((res = mp_add(&tmp2, &b0, &tmp2)) != MP_OKAY) \{
-130          goto ERR;
-131       \}
-132       
-133       if ((res = mp_mul(&tmp1, &tmp2, &w3)) != MP_OKAY) \{
-134          goto ERR;
-135       \}
-136       
-137   
-138       /* w2 = (a2 + a1 + a0)(b2 + b1 + b0) */
-139       if ((res = mp_add(&a2, &a1, &tmp1)) != MP_OKAY) \{
-140          goto ERR;
-141       \}
-142       if ((res = mp_add(&tmp1, &a0, &tmp1)) != MP_OKAY) \{
-143          goto ERR;
-144       \}
-145       if ((res = mp_add(&b2, &b1, &tmp2)) != MP_OKAY) \{
-146          goto ERR;
-147       \}
-148       if ((res = mp_add(&tmp2, &b0, &tmp2)) != MP_OKAY) \{
-149          goto ERR;
-150       \}
-151       if ((res = mp_mul(&tmp1, &tmp2, &w2)) != MP_OKAY) \{
-152          goto ERR;
-153       \}
-154       
-155       /* now solve the matrix 
-156       
-157          0  0  0  0  1
-158          1  2  4  8  16
-159          1  1  1  1  1
-160          16 8  4  2  1
-161          1  0  0  0  0
-162          
-163          using 12 subtractions, 4 shifts, 
-164                 2 small divisions and 1 small multiplication 
-165        */
-166        
-167        /* r1 - r4 */
-168        if ((res = mp_sub(&w1, &w4, &w1)) != MP_OKAY) \{
-169           goto ERR;
-170        \}
-171        /* r3 - r0 */
-172        if ((res = mp_sub(&w3, &w0, &w3)) != MP_OKAY) \{
-173           goto ERR;
-174        \}
-175        /* r1/2 */
-176        if ((res = mp_div_2(&w1, &w1)) != MP_OKAY) \{
-177           goto ERR;
-178        \}
-179        /* r3/2 */
-180        if ((res = mp_div_2(&w3, &w3)) != MP_OKAY) \{
-181           goto ERR;
-182        \}
-183        /* r2 - r0 - r4 */
-184        if ((res = mp_sub(&w2, &w0, &w2)) != MP_OKAY) \{
-185           goto ERR;
-186        \}
-187        if ((res = mp_sub(&w2, &w4, &w2)) != MP_OKAY) \{
-188           goto ERR;
-189        \}
-190        /* r1 - r2 */
-191        if ((res = mp_sub(&w1, &w2, &w1)) != MP_OKAY) \{
-192           goto ERR;
-193        \}
-194        /* r3 - r2 */
-195        if ((res = mp_sub(&w3, &w2, &w3)) != MP_OKAY) \{
-196           goto ERR;
-197        \}
-198        /* r1 - 8r0 */
-199        if ((res = mp_mul_2d(&w0, 3, &tmp1)) != MP_OKAY) \{
-200           goto ERR;
-201        \}
-202        if ((res = mp_sub(&w1, &tmp1, &w1)) != MP_OKAY) \{
-203           goto ERR;
-204        \}
-205        /* r3 - 8r4 */
-206        if ((res = mp_mul_2d(&w4, 3, &tmp1)) != MP_OKAY) \{
-207           goto ERR;
-208        \}
-209        if ((res = mp_sub(&w3, &tmp1, &w3)) != MP_OKAY) \{
-210           goto ERR;
-211        \}
-212        /* 3r2 - r1 - r3 */
-213        if ((res = mp_mul_d(&w2, 3, &w2)) != MP_OKAY) \{
-214           goto ERR;
-215        \}
-216        if ((res = mp_sub(&w2, &w1, &w2)) != MP_OKAY) \{
-217           goto ERR;
-218        \}
-219        if ((res = mp_sub(&w2, &w3, &w2)) != MP_OKAY) \{
-220           goto ERR;
-221        \}
-222        /* r1 - r2 */
-223        if ((res = mp_sub(&w1, &w2, &w1)) != MP_OKAY) \{
-224           goto ERR;
-225        \}
-226        /* r3 - r2 */
-227        if ((res = mp_sub(&w3, &w2, &w3)) != MP_OKAY) \{
-228           goto ERR;
-229        \}
-230        /* r1/3 */
-231        if ((res = mp_div_3(&w1, &w1, NULL)) != MP_OKAY) \{
-232           goto ERR;
-233        \}
-234        /* r3/3 */
-235        if ((res = mp_div_3(&w3, &w3, NULL)) != MP_OKAY) \{
-236           goto ERR;
-237        \}
-238        
-239        /* at this point shift W[n] by B*n */
-240        if ((res = mp_lshd(&w1, 1*B)) != MP_OKAY) \{
+110       
+111       /* w3 = (a0 + 2(a1 + 2a2))(b0 + 2(b1 + 2b2)) */
+112       if ((res = mp_mul_2(&a2, &tmp1)) != MP_OKAY) \{
+113          goto ERR;
+114       \}
+115       if ((res = mp_add(&tmp1, &a1, &tmp1)) != MP_OKAY) \{
+116          goto ERR;
+117       \}
+118       if ((res = mp_mul_2(&tmp1, &tmp1)) != MP_OKAY) \{
+119          goto ERR;
+120       \}
+121       if ((res = mp_add(&tmp1, &a0, &tmp1)) != MP_OKAY) \{
+122          goto ERR;
+123       \}
+124       
+125       if ((res = mp_mul_2(&b2, &tmp2)) != MP_OKAY) \{
+126          goto ERR;
+127       \}
+128       if ((res = mp_add(&tmp2, &b1, &tmp2)) != MP_OKAY) \{
+129          goto ERR;
+130       \}
+131       if ((res = mp_mul_2(&tmp2, &tmp2)) != MP_OKAY) \{
+132          goto ERR;
+133       \}
+134       if ((res = mp_add(&tmp2, &b0, &tmp2)) != MP_OKAY) \{
+135          goto ERR;
+136       \}
+137       
+138       if ((res = mp_mul(&tmp1, &tmp2, &w3)) != MP_OKAY) \{
+139          goto ERR;
+140       \}
+141       
+142   
+143       /* w2 = (a2 + a1 + a0)(b2 + b1 + b0) */
+144       if ((res = mp_add(&a2, &a1, &tmp1)) != MP_OKAY) \{
+145          goto ERR;
+146       \}
+147       if ((res = mp_add(&tmp1, &a0, &tmp1)) != MP_OKAY) \{
+148          goto ERR;
+149       \}
+150       if ((res = mp_add(&b2, &b1, &tmp2)) != MP_OKAY) \{
+151          goto ERR;
+152       \}
+153       if ((res = mp_add(&tmp2, &b0, &tmp2)) != MP_OKAY) \{
+154          goto ERR;
+155       \}
+156       if ((res = mp_mul(&tmp1, &tmp2, &w2)) != MP_OKAY) \{
+157          goto ERR;
+158       \}
+159       
+160       /* now solve the matrix 
+161       
+162          0  0  0  0  1
+163          1  2  4  8  16
+164          1  1  1  1  1
+165          16 8  4  2  1
+166          1  0  0  0  0
+167          
+168          using 12 subtractions, 4 shifts, 
+169                 2 small divisions and 1 small multiplication 
+170        */
+171        
+172        /* r1 - r4 */
+173        if ((res = mp_sub(&w1, &w4, &w1)) != MP_OKAY) \{
+174           goto ERR;
+175        \}
+176        /* r3 - r0 */
+177        if ((res = mp_sub(&w3, &w0, &w3)) != MP_OKAY) \{
+178           goto ERR;
+179        \}
+180        /* r1/2 */
+181        if ((res = mp_div_2(&w1, &w1)) != MP_OKAY) \{
+182           goto ERR;
+183        \}
+184        /* r3/2 */
+185        if ((res = mp_div_2(&w3, &w3)) != MP_OKAY) \{
+186           goto ERR;
+187        \}
+188        /* r2 - r0 - r4 */
+189        if ((res = mp_sub(&w2, &w0, &w2)) != MP_OKAY) \{
+190           goto ERR;
+191        \}
+192        if ((res = mp_sub(&w2, &w4, &w2)) != MP_OKAY) \{
+193           goto ERR;
+194        \}
+195        /* r1 - r2 */
+196        if ((res = mp_sub(&w1, &w2, &w1)) != MP_OKAY) \{
+197           goto ERR;
+198        \}
+199        /* r3 - r2 */
+200        if ((res = mp_sub(&w3, &w2, &w3)) != MP_OKAY) \{
+201           goto ERR;
+202        \}
+203        /* r1 - 8r0 */
+204        if ((res = mp_mul_2d(&w0, 3, &tmp1)) != MP_OKAY) \{
+205           goto ERR;
+206        \}
+207        if ((res = mp_sub(&w1, &tmp1, &w1)) != MP_OKAY) \{
+208           goto ERR;
+209        \}
+210        /* r3 - 8r4 */
+211        if ((res = mp_mul_2d(&w4, 3, &tmp1)) != MP_OKAY) \{
+212           goto ERR;
+213        \}
+214        if ((res = mp_sub(&w3, &tmp1, &w3)) != MP_OKAY) \{
+215           goto ERR;
+216        \}
+217        /* 3r2 - r1 - r3 */
+218        if ((res = mp_mul_d(&w2, 3, &w2)) != MP_OKAY) \{
+219           goto ERR;
+220        \}
+221        if ((res = mp_sub(&w2, &w1, &w2)) != MP_OKAY) \{
+222           goto ERR;
+223        \}
+224        if ((res = mp_sub(&w2, &w3, &w2)) != MP_OKAY) \{
+225           goto ERR;
+226        \}
+227        /* r1 - r2 */
+228        if ((res = mp_sub(&w1, &w2, &w1)) != MP_OKAY) \{
+229           goto ERR;
+230        \}
+231        /* r3 - r2 */
+232        if ((res = mp_sub(&w3, &w2, &w3)) != MP_OKAY) \{
+233           goto ERR;
+234        \}
+235        /* r1/3 */
+236        if ((res = mp_div_3(&w1, &w1, NULL)) != MP_OKAY) \{
+237           goto ERR;
+238        \}
+239        /* r3/3 */
+240        if ((res = mp_div_3(&w3, &w3, NULL)) != MP_OKAY) \{
 241           goto ERR;
 242        \}
-243        if ((res = mp_lshd(&w2, 2*B)) != MP_OKAY) \{
-244           goto ERR;
-245        \}
-246        if ((res = mp_lshd(&w3, 3*B)) != MP_OKAY) \{
-247           goto ERR;
-248        \}
-249        if ((res = mp_lshd(&w4, 4*B)) != MP_OKAY) \{
-250           goto ERR;
-251        \}     
-252        
-253        if ((res = mp_add(&w0, &w1, c)) != MP_OKAY) \{
-254           goto ERR;
-255        \}
-256        if ((res = mp_add(&w2, &w3, &tmp1)) != MP_OKAY) \{
-257           goto ERR;
-258        \}
-259        if ((res = mp_add(&w4, &tmp1, &tmp1)) != MP_OKAY) \{
-260           goto ERR;
-261        \}
-262        if ((res = mp_add(&tmp1, c, c)) != MP_OKAY) \{
-263           goto ERR;
-264        \}     
-265        
-266   ERR:
-267        mp_clear_multi(&w0, &w1, &w2, &w3, &w4, 
-268                       &a0, &a1, &a2, &b0, &b1, 
-269                       &b2, &tmp1, &tmp2, NULL);
-270        return res;
-271   \}     
-272        
+243        
+244        /* at this point shift W[n] by B*n */
+245        if ((res = mp_lshd(&w1, 1*B)) != MP_OKAY) \{
+246           goto ERR;
+247        \}
+248        if ((res = mp_lshd(&w2, 2*B)) != MP_OKAY) \{
+249           goto ERR;
+250        \}
+251        if ((res = mp_lshd(&w3, 3*B)) != MP_OKAY) \{
+252           goto ERR;
+253        \}
+254        if ((res = mp_lshd(&w4, 4*B)) != MP_OKAY) \{
+255           goto ERR;
+256        \}     
+257        
+258        if ((res = mp_add(&w0, &w1, c)) != MP_OKAY) \{
+259           goto ERR;
+260        \}
+261        if ((res = mp_add(&w2, &w3, &tmp1)) != MP_OKAY) \{
+262           goto ERR;
+263        \}
+264        if ((res = mp_add(&w4, &tmp1, &tmp1)) != MP_OKAY) \{
+265           goto ERR;
+266        \}
+267        if ((res = mp_add(&tmp1, c, c)) != MP_OKAY) \{
+268           goto ERR;
+269        \}     
+270        
+271   ERR:
+272        mp_clear_multi(&w0, &w1, &w2, &w3, &w4, 
+273                       &a0, &a1, &a2, &b0, &b1, 
+274                       &b2, &tmp1, &tmp2, NULL);
+275        return res;
+276   \}     
+277        
+278   #endif
 \end{alltt}
 \end{small}
 
@@ -4707,36 +4759,49 @@ s\_mp\_mul\_digs will clear it.
 021     neg = (a->sign == b->sign) ? MP_ZPOS : MP_NEG;
 022   
 023     /* use Toom-Cook? */
-024     if (MIN (a->used, b->used) >= TOOM_MUL_CUTOFF) \{
-025       res = mp_toom_mul(a, b, c);
-026     /* use Karatsuba? */
-027     \} else if (MIN (a->used, b->used) >= KARATSUBA_MUL_CUTOFF) \{
-028       res = mp_karatsuba_mul (a, b, c);
-029     \} else \{
-030       /* can we use the fast multiplier?
-031        *
-032        * The fast multiplier can be used if the output will 
-033        * have less than MP_WARRAY digits and the number of 
-034        * digits won't affect carry propagation
-035        */
-036       int     digs = a->used + b->used + 1;
-037   
-038       if ((digs < MP_WARRAY) &&
-039           MIN(a->used, b->used) <= 
-040           (1 << ((CHAR_BIT * sizeof (mp_word)) - (2 * DIGIT_BIT)))) \{
-041         res = fast_s_mp_mul_digs (a, b, c, digs);
-042       \} else \{
-043         res = s_mp_mul (a, b, c);
-044       \}
-045     \}
-046     c->sign = neg;
-047     return res;
-048   \}
+024   #ifdef BN_MP_TOOM_MUL_C
+025     if (MIN (a->used, b->used) >= TOOM_MUL_CUTOFF) \{
+026       res = mp_toom_mul(a, b, c);
+027     \} else 
+028   #endif
+029   #ifdef BN_MP_KARATSUBA_MUL_C
+030     /* use Karatsuba? */
+031     if (MIN (a->used, b->used) >= KARATSUBA_MUL_CUTOFF) \{
+032       res = mp_karatsuba_mul (a, b, c);
+033     \} else 
+034   #endif
+035     \{
+036       /* can we use the fast multiplier?
+037        *
+038        * The fast multiplier can be used if the output will 
+039        * have less than MP_WARRAY digits and the number of 
+040        * digits won't affect carry propagation
+041        */
+042       int     digs = a->used + b->used + 1;
+043   
+044   #ifdef BN_FAST_S_MP_MUL_DIGS_C
+045       if ((digs < MP_WARRAY) &&
+046           MIN(a->used, b->used) <= 
+047           (1 << ((CHAR_BIT * sizeof (mp_word)) - (2 * DIGIT_BIT)))) \{
+048         res = fast_s_mp_mul_digs (a, b, c, digs);
+049       \} else 
+050   #endif
+051   #ifdef BN_S_MP_MUL_DIGS_C
+052         res = s_mp_mul (a, b, c); /* uses s_mp_mul_digs */
+053   #else
+054         res = MP_VAL;
+055   #endif
+056   
+057     \}
+058     c->sign = (c->used > 0) ? neg : MP_ZPOS;
+059     return res;
+060   \}
+061   #endif
 \end{alltt}
 \end{small}
 
 The implementation is rather simplistic and is not particularly noteworthy.  Line 23 computes the sign of the result using the ``?'' 
-operator from the C programming language.  Line 40 computes $\delta$ using the fact that $1 << k$ is equal to $2^k$.  
+operator from the C programming language.  Line 47 computes $\delta$ using the fact that $1 << k$ is equal to $2^k$.  
 
 \section{Squaring}
 \label{sec:basesquare}
@@ -4901,6 +4966,7 @@ results calculated so far.  This involves expensive carry propagation which will
 077     mp_clear (&t);
 078     return MP_OKAY;
 079   \}
+080   #endif
 \end{alltt}
 \end{small}
 
@@ -5003,112 +5069,105 @@ squares in place.
 031    * Based on Algorithm 14.16 on pp.597 of HAC.
 032    *
 033    */
-034   int fast_s_mp_sqr (mp_int * a, mp_int * b)
-035   \{
-036     int     olduse, newused, res, ix, pa;
-037     mp_word W2[MP_WARRAY], W[MP_WARRAY];
-038   
-039     /* calculate size of product and allocate as required */
-040     pa = a->used;
-041     newused = pa + pa + 1;
-042     if (b->alloc < newused) \{
-043       if ((res = mp_grow (b, newused)) != MP_OKAY) \{
-044         return res;
-045       \}
-046     \}
-047   
-048     /* zero temp buffer (columns)
-049      * Note that there are two buffers.  Since squaring requires
-050      * a outer and inner product and the inner product requires
-051      * computing a product and doubling it (a relatively expensive
-052      * op to perform n**2 times if you don't have to) the inner and
-053      * outer products are computed in different buffers.  This way
-054      * the inner product can be doubled using n doublings instead of
-055      * n**2
-056      */
-057     memset (W,  0, newused * sizeof (mp_word));
-058     memset (W2, 0, newused * sizeof (mp_word));
+034   /* the jist of squaring...
+035   
+036   you do like mult except the offset of the tmpx [one that starts closer to ze
+      ro]
+037   can't equal the offset of tmpy.  So basically you set up iy like before then
+       you min it with
+038   (ty-tx) so that it never happens.  You double all those you add in the inner
+       loop
+039   
+040   After that loop you do the squares and add them in.
+041   
+042   Remove W2 and don't memset W
+043   
+044   */
+045   
+046   int fast_s_mp_sqr (mp_int * a, mp_int * b)
+047   \{
+048     int       olduse, res, pa, ix, iz;
+049     mp_digit   W[MP_WARRAY], *tmpx;
+050     mp_word   W1;
+051   
+052     /* grow the destination as required */
+053     pa = a->used + a->used;
+054     if (b->alloc < pa) \{
+055       if ((res = mp_grow (b, pa)) != MP_OKAY) \{
+056         return res;
+057       \}
+058     \}
 059   
-060     /* This computes the inner product.  To simplify the inner N**2 loop
-061      * the multiplication by two is done afterwards in the N loop.
-062      */
-063     for (ix = 0; ix < pa; ix++) \{
-064       /* compute the outer product
-065        *
-066        * Note that every outer product is computed
-067        * for a particular column only once which means that
-068        * there is no need todo a double precision addition
-069        * into the W2[] array.
-070        */
-071       W2[ix + ix] = ((mp_word)a->dp[ix]) * ((mp_word)a->dp[ix]);
-072   
-073       \{
-074         register mp_digit tmpx, *tmpy;
-075         register mp_word *_W;
-076         register int iy;
+060     /* number of output digits to produce */
+061     W1 = 0;
+062     for (ix = 0; ix <= pa; ix++) \{ 
+063         int      tx, ty, iy;
+064         mp_word  _W;
+065         mp_digit *tmpy;
+066   
+067         /* clear counter */
+068         _W = 0;
+069   
+070         /* get offsets into the two bignums */
+071         ty = MIN(a->used-1, ix);
+072         tx = ix - ty;
+073   
+074         /* setup temp aliases */
+075         tmpx = a->dp + tx;
+076         tmpy = a->dp + ty;
 077   
-078         /* copy of left side */
-079         tmpx = a->dp[ix];
-080   
-081         /* alias for right side */
-082         tmpy = a->dp + (ix + 1);
-083   
-084         /* the column to store the result in */
-085         _W = W + (ix + ix + 1);
-086   
-087         /* inner products */
-088         for (iy = ix + 1; iy < pa; iy++) \{
-089             *_W++ += ((mp_word)tmpx) * ((mp_word)*tmpy++);
-090         \}
-091       \}
-092     \}
+078         /* this is the number of times the loop will iterrate, essentially its
+       
+079            while (tx++ < a->used && ty-- >= 0) \{ ... \}
+080          */
+081         iy = MIN(a->used-tx, ty+1);
+082   
+083         /* now for squaring tx can never equal ty 
+084          * we halve the distance since they approach at a rate of 2x
+085          * and we have to round because odd cases need to be executed
+086          */
+087         iy = MIN(iy, (ty-tx+1)>>1);
+088   
+089         /* execute loop */
+090         for (iz = 0; iz < iy; iz++) \{
+091            _W += ((mp_word)*tmpx++)*((mp_word)*tmpy--);
+092         \}
 093   
-094     /* setup dest */
-095     olduse  = b->used;
-096     b->used = newused;
-097   
-098     /* now compute digits
-099      *
-100      * We have to double the inner product sums, add in the
-101      * outer product sums, propagate carries and convert
-102      * to single precision.
-103      */
-104     \{
-105       register mp_digit *tmpb;
-106   
-107       /* double first value, since the inner products are
-108        * half of what they should be
-109        */
-110       W[0] += W[0] + W2[0];
-111   
-112       tmpb = b->dp;
-113       for (ix = 1; ix < newused; ix++) \{
-114         /* double/add next digit */
-115         W[ix] += W[ix] + W2[ix];
-116   
-117         /* propagate carry forwards [from the previous digit] */
-118         W[ix] = W[ix] + (W[ix - 1] >> ((mp_word) DIGIT_BIT));
+094         /* double the inner product and add carry */
+095         _W = _W + _W + W1;
+096   
+097         /* even columns have the square term in them */
+098         if ((ix&1) == 0) \{
+099            _W += ((mp_word)a->dp[ix>>1])*((mp_word)a->dp[ix>>1]);
+100         \}
+101   
+102         /* store it */
+103         W[ix] = _W;
+104   
+105         /* make next carry */
+106         W1 = _W >> ((mp_word)DIGIT_BIT);
+107     \}
+108   
+109     /* setup dest */
+110     olduse  = b->used;
+111     b->used = a->used+a->used;
+112   
+113     \{
+114       mp_digit *tmpb;
+115       tmpb = b->dp;
+116       for (ix = 0; ix < pa; ix++) \{
+117         *tmpb++ = W[ix] & MP_MASK;
+118       \}
 119   
-120         /* store the current digit now that the carry isn't
-121          * needed
-122          */
-123         *tmpb++ = (mp_digit) (W[ix - 1] & ((mp_word) MP_MASK));
-124       \}
-125       /* set the last value.  Note even if the carry is zero
-126        * this is required since the next step will not zero
-127        * it if b originally had a value at b->dp[2*a.used]
-128        */
-129       *tmpb++ = (mp_digit) (W[(newused) - 1] & ((mp_word) MP_MASK));
-130   
-131       /* clear high digits of b if there were any originally */
-132       for (; ix < olduse; ix++) \{
-133         *tmpb++ = 0;
-134       \}
-135     \}
-136   
-137     mp_clamp (b);
-138     return MP_OKAY;
-139   \}
+120       /* clear unused digits [that existed in the old copy of c] */
+121       for (; ix < olduse; ix++) \{
+122         *tmpb++ = 0;
+123       \}
+124     \}
+125     mp_clamp (b);
+126     return MP_OKAY;
+127   \}
+128   #endif
 \end{alltt}
 \end{small}
 
@@ -5221,7 +5280,7 @@ ratio of 1:7.  } than simpler operations such as addition.
 017   /* Karatsuba squaring, computes b = a*a using three 
 018    * half size squarings
 019    *
-020    * See comments of mp_karatsuba_mul for details.  It 
+020    * See comments of karatsuba_mul for details.  It 
 021    * is essentially the same algorithm but merely 
 022    * tuned to perform recursive squarings.
 023    */
@@ -5317,6 +5376,7 @@ ratio of 1:7.  } than simpler operations such as addition.
 113   ERR:
 114     return err;
 115   \}
+116   #endif
 \end{alltt}
 \end{small}
 
@@ -5383,25 +5443,37 @@ neither of the polynomial basis algorithms should be used then either the Comba
 020   \{
 021     int     res;
 022   
-023     /* use Toom-Cook? */
-024     if (a->used >= TOOM_SQR_CUTOFF) \{
-025       res = mp_toom_sqr(a, b);
-026     /* Karatsuba? */
-027     \} else if (a->used >= KARATSUBA_SQR_CUTOFF) \{
-028       res = mp_karatsuba_sqr (a, b);
-029     \} else \{
-030       /* can we use the fast comba multiplier? */
-031       if ((a->used * 2 + 1) < MP_WARRAY && 
-032            a->used < 
-033            (1 << (sizeof(mp_word) * CHAR_BIT - 2*DIGIT_BIT - 1))) \{
-034         res = fast_s_mp_sqr (a, b);
-035       \} else \{
-036         res = s_mp_sqr (a, b);
-037       \}
-038     \}
-039     b->sign = MP_ZPOS;
-040     return res;
-041   \}
+023   #ifdef BN_MP_TOOM_SQR_C
+024     /* use Toom-Cook? */
+025     if (a->used >= TOOM_SQR_CUTOFF) \{
+026       res = mp_toom_sqr(a, b);
+027     /* Karatsuba? */
+028     \} else 
+029   #endif
+030   #ifdef BN_MP_KARATSUBA_SQR_C
+031   if (a->used >= KARATSUBA_SQR_CUTOFF) \{
+032       res = mp_karatsuba_sqr (a, b);
+033     \} else 
+034   #endif
+035     \{
+036   #ifdef BN_FAST_S_MP_SQR_C
+037       /* can we use the fast comba multiplier? */
+038       if ((a->used * 2 + 1) < MP_WARRAY && 
+039            a->used < 
+040            (1 << (sizeof(mp_word) * CHAR_BIT - 2*DIGIT_BIT - 1))) \{
+041         res = fast_s_mp_sqr (a, b);
+042       \} else
+043   #endif
+044   #ifdef BN_S_MP_SQR_C
+045         res = s_mp_sqr (a, b);
+046   #else
+047         res = MP_VAL;
+048   #endif
+049     \}
+050     b->sign = MP_ZPOS;
+051     return res;
+052   \}
+053   #endif
 \end{alltt}
 \end{small}
 
@@ -5675,50 +5747,62 @@ performed at most twice, and on average once. However, if $a \ge b^2$ than it wi
 038         goto CLEANUP;
 039       \}
 040     \} else \{
-041       if ((res = s_mp_mul_high_digs (&q, mu, &q, um - 1)) != MP_OKAY) \{
-042         goto CLEANUP;
-043       \}
-044     \}
-045   
-046     /* q3 = q2 / b**(k+1) */
-047     mp_rshd (&q, um + 1);         
-048   
-049     /* x = x mod b**(k+1), quick (no division) */
-050     if ((res = mp_mod_2d (x, DIGIT_BIT * (um + 1), x)) != MP_OKAY) \{
-051       goto CLEANUP;
-052     \}
-053   
-054     /* q = q * m mod b**(k+1), quick (no division) */
-055     if ((res = s_mp_mul_digs (&q, m, &q, um + 1)) != MP_OKAY) \{
-056       goto CLEANUP;
-057     \}
-058   
-059     /* x = x - q */
-060     if ((res = mp_sub (x, &q, x)) != MP_OKAY) \{
-061       goto CLEANUP;
-062     \}
-063   
-064     /* If x < 0, add b**(k+1) to it */
-065     if (mp_cmp_d (x, 0) == MP_LT) \{
-066       mp_set (&q, 1);
-067       if ((res = mp_lshd (&q, um + 1)) != MP_OKAY)
-068         goto CLEANUP;
-069       if ((res = mp_add (x, &q, x)) != MP_OKAY)
-070         goto CLEANUP;
-071     \}
-072   
-073     /* Back off if it's too big */
-074     while (mp_cmp (x, m) != MP_LT) \{
-075       if ((res = s_mp_sub (x, m, x)) != MP_OKAY) \{
-076         goto CLEANUP;
-077       \}
-078     \}
-079     
-080   CLEANUP:
-081     mp_clear (&q);
-082   
-083     return res;
-084   \}
+041   #ifdef BN_S_MP_MUL_HIGH_DIGS_C
+042       if ((res = s_mp_mul_high_digs (&q, mu, &q, um - 1)) != MP_OKAY) \{
+043         goto CLEANUP;
+044       \}
+045   #elif defined(BN_FAST_S_MP_MUL_HIGH_DIGS_C)
+046       if ((res = fast_s_mp_mul_high_digs (&q, mu, &q, um - 1)) != MP_OKAY) \{
+047         goto CLEANUP;
+048       \}
+049   #else 
+050       \{ 
+051         res = MP_VAL;
+052         goto CLEANUP;
+053       \}
+054   #endif
+055     \}
+056   
+057     /* q3 = q2 / b**(k+1) */
+058     mp_rshd (&q, um + 1);         
+059   
+060     /* x = x mod b**(k+1), quick (no division) */
+061     if ((res = mp_mod_2d (x, DIGIT_BIT * (um + 1), x)) != MP_OKAY) \{
+062       goto CLEANUP;
+063     \}
+064   
+065     /* q = q * m mod b**(k+1), quick (no division) */
+066     if ((res = s_mp_mul_digs (&q, m, &q, um + 1)) != MP_OKAY) \{
+067       goto CLEANUP;
+068     \}
+069   
+070     /* x = x - q */
+071     if ((res = mp_sub (x, &q, x)) != MP_OKAY) \{
+072       goto CLEANUP;
+073     \}
+074   
+075     /* If x < 0, add b**(k+1) to it */
+076     if (mp_cmp_d (x, 0) == MP_LT) \{
+077       mp_set (&q, 1);
+078       if ((res = mp_lshd (&q, um + 1)) != MP_OKAY)
+079         goto CLEANUP;
+080       if ((res = mp_add (x, &q, x)) != MP_OKAY)
+081         goto CLEANUP;
+082     \}
+083   
+084     /* Back off if it's too big */
+085     while (mp_cmp (x, m) != MP_LT) \{
+086       if ((res = s_mp_sub (x, m, x)) != MP_OKAY) \{
+087         goto CLEANUP;
+088       \}
+089     \}
+090     
+091   CLEANUP:
+092     mp_clear (&q);
+093   
+094     return res;
+095   \}
+096   #endif
 \end{alltt}
 \end{small}
 
@@ -5761,16 +5845,16 @@ is equivalent and much faster.  The final value is computed by taking the intege
 017   /* pre-calculate the value required for Barrett reduction
 018    * For a given modulus "b" it calulates the value required in "a"
 019    */
-020   int
-021   mp_reduce_setup (mp_int * a, mp_int * b)
-022   \{
-023     int     res;
-024     
-025     if ((res = mp_2expt (a, b->used * 2 * DIGIT_BIT)) != MP_OKAY) \{
-026       return res;
-027     \}
-028     return mp_div (a, b, a, NULL);
-029   \}
+020   int mp_reduce_setup (mp_int * a, mp_int * b)
+021   \{
+022     int     res;
+023     
+024     if ((res = mp_2expt (a, b->used * 2 * DIGIT_BIT)) != MP_OKAY) \{
+025       return res;
+026     \}
+027     return mp_div (a, b, a, NULL);
+028   \}
+029   #endif
 \end{alltt}
 \end{small}
 
@@ -6038,7 +6122,7 @@ multiplications.
 023   
 024     /* can the fast reduction [comba] method be used?
 025      *
-026      * Note that unlike in mp_mul you're safely allowed *less*
+026      * Note that unlike in mul you're safely allowed *less*
 027      * than the available columns [255 per default] since carries
 028      * are fixed up in the inner loop.
 029      */
@@ -6061,7 +6145,7 @@ multiplications.
 046       /* mu = ai * rho mod b
 047        *
 048        * The value of rho must be precalculated via
-049        * bn_mp_montgomery_setup() such that
+049        * montgomery_setup() such that
 050        * it equals -1/n0 mod b this allows the
 051        * following inner loop to reduce the
 052        * input one digit at a time
@@ -6125,6 +6209,7 @@ multiplications.
 110   
 111     return MP_OKAY;
 112   \}
+113   #endif
 \end{alltt}
 \end{small}
 
@@ -6220,7 +6305,7 @@ stored in the destination $x$.
 016   
 017   /* computes xR**-1 == x (mod N) via Montgomery Reduction
 018    *
-019    * This is an optimized implementation of mp_montgomery_reduce
+019    * This is an optimized implementation of montgomery_reduce
 020    * which uses the comba method to quickly calculate the columns of the
 021    * reduction.
 022    *
@@ -6369,6 +6454,7 @@ stored in the destination $x$.
 165     \}
 166     return MP_OKAY;
 167   \}
+168   #endif
 \end{alltt}
 \end{small}
 
@@ -6450,10 +6536,11 @@ to calculate $1/n_0$ when $\beta$ is a power of two.
 047   #endif
 048   
 049     /* rho = -1/m mod b */
-050     *rho = (((mp_digit) 1 << ((mp_digit) DIGIT_BIT)) - x) & MP_MASK;
+050     *rho = (((mp_word)1 << ((mp_word) DIGIT_BIT)) - x) & MP_MASK;
 051   
 052     return MP_OKAY;
 053   \}
+054   #endif
 \end{alltt}
 \end{small}
 
@@ -6719,6 +6806,7 @@ at step 3.
 086     \}
 087     return MP_OKAY;
 088   \}
+089   #endif
 \end{alltt}
 \end{small}
 
@@ -6773,6 +6861,7 @@ completeness.
 024           ((mp_word)a->dp[0]));
 025   \}
 026   
+027   #endif
 \end{alltt}
 \end{small}
 
@@ -6830,6 +6919,7 @@ step 3 then $n$ must be of Diminished Radix form.
 035      return 1;
 036   \}
 037   
+038   #endif
 \end{alltt}
 \end{small}
 
@@ -6914,6 +7004,7 @@ shift which makes the algorithm fairly inexpensive to use.
 054      return res;
 055   \}
 056   
+057   #endif
 \end{alltt}
 \end{small}
 
@@ -6983,6 +7074,7 @@ is sufficient to solve for $k$.  Alternatively if $n$ has more than one digit th
 040      mp_clear(&tmp);
 041      return MP_OKAY;
 042   \}
+043   #endif
 \end{alltt}
 \end{small}
 
@@ -7031,32 +7123,34 @@ This algorithm quickly determines if a modulus is of the form required for algor
 017   /* determines if mp_reduce_2k can be used */
 018   int mp_reduce_is_2k(mp_int *a)
 019   \{
-020      int ix, iy, iz, iw;
-021      
-022      if (a->used == 0) \{
-023         return 0;
-024      \} else if (a->used == 1) \{
-025         return 1;
-026      \} else if (a->used > 1) \{
-027         iy = mp_count_bits(a);
-028         iz = 1;
-029         iw = 1;
-030       
-031         /* Test every bit from the second digit up, must be 1 */
-032         for (ix = DIGIT_BIT; ix < iy; ix++) \{
-033             if ((a->dp[iw] & iz) == 0) \{
-034                return 0;
-035             \}
-036             iz <<= 1;
-037             if (iz > (int)MP_MASK) \{
-038                ++iw;
-039                iz = 1;
-040             \}
-041         \}
-042      \}
-043      return 1;
-044   \}
-045   
+020      int ix, iy, iw;
+021      mp_digit iz;
+022      
+023      if (a->used == 0) \{
+024         return 0;
+025      \} else if (a->used == 1) \{
+026         return 1;
+027      \} else if (a->used > 1) \{
+028         iy = mp_count_bits(a);
+029         iz = 1;
+030         iw = 1;
+031       
+032         /* Test every bit from the second digit up, must be 1 */
+033         for (ix = DIGIT_BIT; ix < iy; ix++) \{
+034             if ((a->dp[iw] & iz) == 0) \{
+035                return 0;
+036             \}
+037             iz <<= 1;
+038             if (iz > (mp_digit)MP_MASK) \{
+039                ++iw;
+040                iz = 1;
+041             \}
+042         \}
+043      \}
+044      return 1;
+045   \}
+046   
+047   #endif
 \end{alltt}
 \end{small}
 
@@ -7265,6 +7359,7 @@ iteration of the loop moves the bits of the exponent $b$ upwards to the most sig
 049     mp_clear (&g);
 050     return MP_OKAY;
 051   \}
+052   #endif
 \end{alltt}
 \end{small}
 
@@ -7470,51 +7565,72 @@ algorithm since their arguments are essentially the same (\textit{two mp\_ints a
 031   
 032     /* if exponent X is negative we have to recurse */
 033     if (X->sign == MP_NEG) \{
-034        mp_int tmpG, tmpX;
-035        int err;
-036   
-037        /* first compute 1/G mod P */
-038        if ((err = mp_init(&tmpG)) != MP_OKAY) \{
-039           return err;
-040        \}
-041        if ((err = mp_invmod(G, P, &tmpG)) != MP_OKAY) \{
-042           mp_clear(&tmpG);
-043           return err;
-044        \}
-045   
-046        /* now get |X| */
-047        if ((err = mp_init(&tmpX)) != MP_OKAY) \{
-048           mp_clear(&tmpG);
-049           return err;
-050        \}
-051        if ((err = mp_abs(X, &tmpX)) != MP_OKAY) \{
-052           mp_clear_multi(&tmpG, &tmpX, NULL);
-053           return err;
-054        \}
-055   
-056        /* and now compute (1/G)**|X| instead of G**X [X < 0] */
-057        err = mp_exptmod(&tmpG, &tmpX, P, Y);
-058        mp_clear_multi(&tmpG, &tmpX, NULL);
-059        return err;
-060     \}
-061   
-062     /* is it a DR modulus? */
-063     dr = mp_dr_is_modulus(P);
-064   
-065     /* if not, is it a uDR modulus? */
-066     if (dr == 0) \{
-067        dr = mp_reduce_is_2k(P) << 1;
-068     \}
-069       
-070     /* if the modulus is odd or dr != 0 use the fast method */
-071     if (mp_isodd (P) == 1 || dr !=  0) \{
-072       return mp_exptmod_fast (G, X, P, Y, dr);
-073     \} else \{
-074       /* otherwise use the generic Barrett reduction technique */
-075       return s_mp_exptmod (G, X, P, Y);
-076     \}
-077   \}
-078   
+034   #ifdef BN_MP_INVMOD_C
+035        mp_int tmpG, tmpX;
+036        int err;
+037   
+038        /* first compute 1/G mod P */
+039        if ((err = mp_init(&tmpG)) != MP_OKAY) \{
+040           return err;
+041        \}
+042        if ((err = mp_invmod(G, P, &tmpG)) != MP_OKAY) \{
+043           mp_clear(&tmpG);
+044           return err;
+045        \}
+046   
+047        /* now get |X| */
+048        if ((err = mp_init(&tmpX)) != MP_OKAY) \{
+049           mp_clear(&tmpG);
+050           return err;
+051        \}
+052        if ((err = mp_abs(X, &tmpX)) != MP_OKAY) \{
+053           mp_clear_multi(&tmpG, &tmpX, NULL);
+054           return err;
+055        \}
+056   
+057        /* and now compute (1/G)**|X| instead of G**X [X < 0] */
+058        err = mp_exptmod(&tmpG, &tmpX, P, Y);
+059        mp_clear_multi(&tmpG, &tmpX, NULL);
+060        return err;
+061   #else 
+062        /* no invmod */
+063        return MP_VAL
+064   #endif
+065     \}
+066   
+067   #ifdef BN_MP_DR_IS_MODULUS_C
+068     /* is it a DR modulus? */
+069     dr = mp_dr_is_modulus(P);
+070   #else
+071     dr = 0;
+072   #endif
+073   
+074   #ifdef BN_MP_REDUCE_IS_2K_C
+075     /* if not, is it a uDR modulus? */
+076     if (dr == 0) \{
+077        dr = mp_reduce_is_2k(P) << 1;
+078     \}
+079   #endif
+080       
+081     /* if the modulus is odd or dr != 0 use the fast method */
+082   #ifdef BN_MP_EXPTMOD_FAST_C
+083     if (mp_isodd (P) == 1 || dr !=  0) \{
+084       return mp_exptmod_fast (G, X, P, Y, dr);
+085     \} else \{
+086   #endif
+087   #ifdef BN_S_MP_EXPTMOD_C
+088       /* otherwise use the generic Barrett reduction technique */
+089       return s_mp_exptmod (G, X, P, Y);
+090   #else
+091       /* no exptmod for evens */
+092       return MP_VAL;
+093   #endif
+094   #ifdef BN_MP_EXPTMOD_FAST_C
+095     \}
+096   #endif
+097   \}
+098   
+099   #endif
 \end{alltt}
 \end{small}
 
@@ -7523,8 +7639,8 @@ negative the algorithm tries to perform a modular exponentiation with the modula
 the modular inverse of $G$ and $tmpX$ is assigned the absolute value of $X$.  The algorithm will recuse with these new values with a positive
 exponent.
 
-If the exponent is positive the algorithm resumes the exponentiation.  Line 63 determines if the modulus is of the restricted Diminished Radix 
-form.  If it is not line 67 attempts to determine if it is of a unrestricted Diminished Radix form.  The integer $dr$ will take on one
+If the exponent is positive the algorithm resumes the exponentiation.  Line 69 determines if the modulus is of the restricted Diminished Radix 
+form.  If it is not line 77 attempts to determine if it is of a unrestricted Diminished Radix form.  The integer $dr$ will take on one
 of three values.
 
 \begin{enumerate}
@@ -7533,7 +7649,7 @@ of three values.
 \item $dr = 2$ means that the modulus is of unrestricted Diminished Radix form.
 \end{enumerate}
 
-Line 70 determines if the fast modular exponentiation algorithm can be used.  It is allowed if $dr \ne 0$ or if the modulus is odd.  Otherwise,
+Line 67 determines if the fast modular exponentiation algorithm can be used.  It is allowed if $dr \ne 0$ or if the modulus is odd.  Otherwise,
 the slower s\_mp\_exptmod algorithm is used which uses Barrett reduction.  
 
 \subsection{Barrett Modular Exponentiation}
@@ -7913,6 +8029,7 @@ a Left-to-Right algorithm is used to process the remaining few bits.
 232     \}
 233     return err;
 234   \}
+235   #endif
 \end{alltt}
 \end{small}
 
@@ -7981,10 +8098,11 @@ equivalent to $m \cdot 2^k$.  By this logic when $m = 1$ a quick power of two ca
 036     a->used = b / DIGIT_BIT + 1;
 037   
 038     /* put the single bit in its place */
-039     a->dp[b / DIGIT_BIT] = 1 << (b % DIGIT_BIT);
+039     a->dp[b / DIGIT_BIT] = ((mp_digit)1) << (b % DIGIT_BIT);
 040   
 041     return MP_OKAY;
 042   \}
+043   #endif
 \end{alltt}
 \end{small}
 
@@ -8234,201 +8352,277 @@ respectively be replaced with a zero.
 \vspace{-3mm}
 \begin{alltt}
 016   
-017   /* integer signed division. 
-018    * c*b + d == a [e.g. a/b, c=quotient, d=remainder]
-019    * HAC pp.598 Algorithm 14.20
-020    *
-021    * Note that the description in HAC is horribly 
-022    * incomplete.  For example, it doesn't consider 
-023    * the case where digits are removed from 'x' in 
-024    * the inner loop.  It also doesn't consider the 
-025    * case that y has fewer than three digits, etc..
-026    *
-027    * The overall algorithm is as described as 
-028    * 14.20 from HAC but fixed to treat these cases.
-029   */
-030   int mp_div (mp_int * a, mp_int * b, mp_int * c, mp_int * d)
-031   \{
-032     mp_int  q, x, y, t1, t2;
-033     int     res, n, t, i, norm, neg;
-034   
-035     /* is divisor zero ? */
-036     if (mp_iszero (b) == 1) \{
-037       return MP_VAL;
-038     \}
-039   
-040     /* if a < b then q=0, r = a */
-041     if (mp_cmp_mag (a, b) == MP_LT) \{
-042       if (d != NULL) \{
-043         res = mp_copy (a, d);
-044       \} else \{
-045         res = MP_OKAY;
-046       \}
-047       if (c != NULL) \{
-048         mp_zero (c);
-049       \}
-050       return res;
-051     \}
-052   
-053     if ((res = mp_init_size (&q, a->used + 2)) != MP_OKAY) \{
-054       return res;
-055     \}
-056     q.used = a->used + 2;
+017   #ifdef BN_MP_DIV_SMALL
+018   
+019   /* slower bit-bang division... also smaller */
+020   int mp_div(mp_int * a, mp_int * b, mp_int * c, mp_int * d)
+021   \{
+022      mp_int ta, tb, tq, q;
+023      int    res, n, n2;
+024   
+025     /* is divisor zero ? */
+026     if (mp_iszero (b) == 1) \{
+027       return MP_VAL;
+028     \}
+029   
+030     /* if a < b then q=0, r = a */
+031     if (mp_cmp_mag (a, b) == MP_LT) \{
+032       if (d != NULL) \{
+033         res = mp_copy (a, d);
+034       \} else \{
+035         res = MP_OKAY;
+036       \}
+037       if (c != NULL) \{
+038         mp_zero (c);
+039       \}
+040       return res;
+041     \}
+042       
+043     /* init our temps */
+044     if ((res = mp_init_multi(&ta, &tb, &tq, &q, NULL) != MP_OKAY)) \{
+045        return res;
+046     \}
+047   
+048   
+049     mp_set(&tq, 1);
+050     n = mp_count_bits(a) - mp_count_bits(b);
+051     if (((res = mp_copy(a, &ta)) != MP_OKAY) ||
+052         ((res = mp_copy(b, &tb)) != MP_OKAY) || 
+053         ((res = mp_mul_2d(&tb, n, &tb)) != MP_OKAY) ||
+054         ((res = mp_mul_2d(&tq, n, &tq)) != MP_OKAY)) \{
+055         goto __ERR;
+056     \}
 057   
-058     if ((res = mp_init (&t1)) != MP_OKAY) \{
-059       goto __Q;
-060     \}
-061   
-062     if ((res = mp_init (&t2)) != MP_OKAY) \{
-063       goto __T1;
-064     \}
-065   
-066     if ((res = mp_init_copy (&x, a)) != MP_OKAY) \{
-067       goto __T2;
-068     \}
-069   
-070     if ((res = mp_init_copy (&y, b)) != MP_OKAY) \{
-071       goto __X;
-072     \}
-073   
-074     /* fix the sign */
-075     neg = (a->sign == b->sign) ? MP_ZPOS : MP_NEG;
-076     x.sign = y.sign = MP_ZPOS;
-077   
-078     /* normalize both x and y, ensure that y >= b/2, [b == 2**DIGIT_BIT] */
-079     norm = mp_count_bits(&y) % DIGIT_BIT;
-080     if (norm < (int)(DIGIT_BIT-1)) \{
-081        norm = (DIGIT_BIT-1) - norm;
-082        if ((res = mp_mul_2d (&x, norm, &x)) != MP_OKAY) \{
-083          goto __Y;
-084        \}
-085        if ((res = mp_mul_2d (&y, norm, &y)) != MP_OKAY) \{
-086          goto __Y;
-087        \}
-088     \} else \{
-089        norm = 0;
-090     \}
-091   
-092     /* note hac does 0 based, so if used==5 then its 0,1,2,3,4, e.g. use 4 */
-093     n = x.used - 1;
-094     t = y.used - 1;
-095   
-096     /* while (x >= y*b**n-t) do \{ q[n-t] += 1; x -= y*b**\{n-t\} \} */
-097     if ((res = mp_lshd (&y, n - t)) != MP_OKAY) \{ /* y = y*b**\{n-t\} */
-098       goto __Y;
-099     \}
-100   
-101     while (mp_cmp (&x, &y) != MP_LT) \{
-102       ++(q.dp[n - t]);
-103       if ((res = mp_sub (&x, &y, &x)) != MP_OKAY) \{
-104         goto __Y;
-105       \}
-106     \}
-107   
-108     /* reset y by shifting it back down */
-109     mp_rshd (&y, n - t);
-110   
-111     /* step 3. for i from n down to (t + 1) */
-112     for (i = n; i >= (t + 1); i--) \{
-113       if (i > x.used) \{
-114         continue;
-115       \}
-116   
-117       /* step 3.1 if xi == yt then set q\{i-t-1\} to b-1, 
-118        * otherwise set q\{i-t-1\} to (xi*b + x\{i-1\})/yt */
-119       if (x.dp[i] == y.dp[t]) \{
-120         q.dp[i - t - 1] = ((((mp_digit)1) << DIGIT_BIT) - 1);
-121       \} else \{
-122         mp_word tmp;
-123         tmp = ((mp_word) x.dp[i]) << ((mp_word) DIGIT_BIT);
-124         tmp |= ((mp_word) x.dp[i - 1]);
-125         tmp /= ((mp_word) y.dp[t]);
-126         if (tmp > (mp_word) MP_MASK)
-127           tmp = MP_MASK;
-128         q.dp[i - t - 1] = (mp_digit) (tmp & (mp_word) (MP_MASK));
-129       \}
-130   
-131       /* while (q\{i-t-1\} * (yt * b + y\{t-1\})) > 
-132                xi * b**2 + xi-1 * b + xi-2 
-133        
-134          do q\{i-t-1\} -= 1; 
-135       */
-136       q.dp[i - t - 1] = (q.dp[i - t - 1] + 1) & MP_MASK;
-137       do \{
-138         q.dp[i - t - 1] = (q.dp[i - t - 1] - 1) & MP_MASK;
-139   
-140         /* find left hand */
-141         mp_zero (&t1);
-142         t1.dp[0] = (t - 1 < 0) ? 0 : y.dp[t - 1];
-143         t1.dp[1] = y.dp[t];
-144         t1.used = 2;
-145         if ((res = mp_mul_d (&t1, q.dp[i - t - 1], &t1)) != MP_OKAY) \{
-146           goto __Y;
-147         \}
-148   
-149         /* find right hand */
-150         t2.dp[0] = (i - 2 < 0) ? 0 : x.dp[i - 2];
-151         t2.dp[1] = (i - 1 < 0) ? 0 : x.dp[i - 1];
-152         t2.dp[2] = x.dp[i];
-153         t2.used = 3;
-154       \} while (mp_cmp_mag(&t1, &t2) == MP_GT);
-155   
-156       /* step 3.3 x = x - q\{i-t-1\} * y * b**\{i-t-1\} */
-157       if ((res = mp_mul_d (&y, q.dp[i - t - 1], &t1)) != MP_OKAY) \{
-158         goto __Y;
-159       \}
-160   
-161       if ((res = mp_lshd (&t1, i - t - 1)) != MP_OKAY) \{
-162         goto __Y;
-163       \}
-164   
-165       if ((res = mp_sub (&x, &t1, &x)) != MP_OKAY) \{
-166         goto __Y;
-167       \}
-168   
-169       /* if x < 0 then \{ x = x + y*b**\{i-t-1\}; q\{i-t-1\} -= 1; \} */
-170       if (x.sign == MP_NEG) \{
-171         if ((res = mp_copy (&y, &t1)) != MP_OKAY) \{
-172           goto __Y;
-173         \}
-174         if ((res = mp_lshd (&t1, i - t - 1)) != MP_OKAY) \{
-175           goto __Y;
-176         \}
-177         if ((res = mp_add (&x, &t1, &x)) != MP_OKAY) \{
-178           goto __Y;
-179         \}
-180   
-181         q.dp[i - t - 1] = (q.dp[i - t - 1] - 1UL) & MP_MASK;
-182       \}
-183     \}
-184   
-185     /* now q is the quotient and x is the remainder 
-186      * [which we have to normalize] 
-187      */
-188     
-189     /* get sign before writing to c */
-190     x.sign = a->sign;
-191   
-192     if (c != NULL) \{
-193       mp_clamp (&q);
-194       mp_exch (&q, c);
-195       c->sign = neg;
-196     \}
-197   
-198     if (d != NULL) \{
-199       mp_div_2d (&x, norm, &x, NULL);
-200       mp_exch (&x, d);
-201     \}
+058     while (n-- >= 0) \{
+059        if (mp_cmp(&tb, &ta) != MP_GT) \{
+060           if (((res = mp_sub(&ta, &tb, &ta)) != MP_OKAY) ||
+061               ((res = mp_add(&q, &tq, &q)) != MP_OKAY)) \{
+062              goto __ERR;
+063           \}
+064        \}
+065        if (((res = mp_div_2d(&tb, 1, &tb, NULL)) != MP_OKAY) ||
+066            ((res = mp_div_2d(&tq, 1, &tq, NULL)) != MP_OKAY)) \{
+067              goto __ERR;
+068        \}
+069     \}
+070   
+071     /* now q == quotient and ta == remainder */
+072     n  = a->sign;
+073     n2 = (a->sign == b->sign ? MP_ZPOS : MP_NEG);
+074     if (c != NULL) \{
+075        mp_exch(c, &q);
+076        c->sign  = n2;
+077     \}
+078     if (d != NULL) \{
+079        mp_exch(d, &ta);
+080        d->sign = n;
+081     \}
+082   __ERR:
+083      mp_clear_multi(&ta, &tb, &tq, &q, NULL);
+084      return res;
+085   \}
+086   
+087   #else
+088   
+089   /* integer signed division. 
+090    * c*b + d == a [e.g. a/b, c=quotient, d=remainder]
+091    * HAC pp.598 Algorithm 14.20
+092    *
+093    * Note that the description in HAC is horribly 
+094    * incomplete.  For example, it doesn't consider 
+095    * the case where digits are removed from 'x' in 
+096    * the inner loop.  It also doesn't consider the 
+097    * case that y has fewer than three digits, etc..
+098    *
+099    * The overall algorithm is as described as 
+100    * 14.20 from HAC but fixed to treat these cases.
+101   */
+102   int mp_div (mp_int * a, mp_int * b, mp_int * c, mp_int * d)
+103   \{
+104     mp_int  q, x, y, t1, t2;
+105     int     res, n, t, i, norm, neg;
+106   
+107     /* is divisor zero ? */
+108     if (mp_iszero (b) == 1) \{
+109       return MP_VAL;
+110     \}
+111   
+112     /* if a < b then q=0, r = a */
+113     if (mp_cmp_mag (a, b) == MP_LT) \{
+114       if (d != NULL) \{
+115         res = mp_copy (a, d);
+116       \} else \{
+117         res = MP_OKAY;
+118       \}
+119       if (c != NULL) \{
+120         mp_zero (c);
+121       \}
+122       return res;
+123     \}
+124   
+125     if ((res = mp_init_size (&q, a->used + 2)) != MP_OKAY) \{
+126       return res;
+127     \}
+128     q.used = a->used + 2;
+129   
+130     if ((res = mp_init (&t1)) != MP_OKAY) \{
+131       goto __Q;
+132     \}
+133   
+134     if ((res = mp_init (&t2)) != MP_OKAY) \{
+135       goto __T1;
+136     \}
+137   
+138     if ((res = mp_init_copy (&x, a)) != MP_OKAY) \{
+139       goto __T2;
+140     \}
+141   
+142     if ((res = mp_init_copy (&y, b)) != MP_OKAY) \{
+143       goto __X;
+144     \}
+145   
+146     /* fix the sign */
+147     neg = (a->sign == b->sign) ? MP_ZPOS : MP_NEG;
+148     x.sign = y.sign = MP_ZPOS;
+149   
+150     /* normalize both x and y, ensure that y >= b/2, [b == 2**DIGIT_BIT] */
+151     norm = mp_count_bits(&y) % DIGIT_BIT;
+152     if (norm < (int)(DIGIT_BIT-1)) \{
+153        norm = (DIGIT_BIT-1) - norm;
+154        if ((res = mp_mul_2d (&x, norm, &x)) != MP_OKAY) \{
+155          goto __Y;
+156        \}
+157        if ((res = mp_mul_2d (&y, norm, &y)) != MP_OKAY) \{
+158          goto __Y;
+159        \}
+160     \} else \{
+161        norm = 0;
+162     \}
+163   
+164     /* note hac does 0 based, so if used==5 then its 0,1,2,3,4, e.g. use 4 */
+165     n = x.used - 1;
+166     t = y.used - 1;
+167   
+168     /* while (x >= y*b**n-t) do \{ q[n-t] += 1; x -= y*b**\{n-t\} \} */
+169     if ((res = mp_lshd (&y, n - t)) != MP_OKAY) \{ /* y = y*b**\{n-t\} */
+170       goto __Y;
+171     \}
+172   
+173     while (mp_cmp (&x, &y) != MP_LT) \{
+174       ++(q.dp[n - t]);
+175       if ((res = mp_sub (&x, &y, &x)) != MP_OKAY) \{
+176         goto __Y;
+177       \}
+178     \}
+179   
+180     /* reset y by shifting it back down */
+181     mp_rshd (&y, n - t);
+182   
+183     /* step 3. for i from n down to (t + 1) */
+184     for (i = n; i >= (t + 1); i--) \{
+185       if (i > x.used) \{
+186         continue;
+187       \}
+188   
+189       /* step 3.1 if xi == yt then set q\{i-t-1\} to b-1, 
+190        * otherwise set q\{i-t-1\} to (xi*b + x\{i-1\})/yt */
+191       if (x.dp[i] == y.dp[t]) \{
+192         q.dp[i - t - 1] = ((((mp_digit)1) << DIGIT_BIT) - 1);
+193       \} else \{
+194         mp_word tmp;
+195         tmp = ((mp_word) x.dp[i]) << ((mp_word) DIGIT_BIT);
+196         tmp |= ((mp_word) x.dp[i - 1]);
+197         tmp /= ((mp_word) y.dp[t]);
+198         if (tmp > (mp_word) MP_MASK)
+199           tmp = MP_MASK;
+200         q.dp[i - t - 1] = (mp_digit) (tmp & (mp_word) (MP_MASK));
+201       \}
 202   
-203     res = MP_OKAY;
-204   
-205   __Y:mp_clear (&y);
-206   __X:mp_clear (&x);
-207   __T2:mp_clear (&t2);
-208   __T1:mp_clear (&t1);
-209   __Q:mp_clear (&q);
-210     return res;
-211   \}
+203       /* while (q\{i-t-1\} * (yt * b + y\{t-1\})) > 
+204                xi * b**2 + xi-1 * b + xi-2 
+205        
+206          do q\{i-t-1\} -= 1; 
+207       */
+208       q.dp[i - t - 1] = (q.dp[i - t - 1] + 1) & MP_MASK;
+209       do \{
+210         q.dp[i - t - 1] = (q.dp[i - t - 1] - 1) & MP_MASK;
+211   
+212         /* find left hand */
+213         mp_zero (&t1);
+214         t1.dp[0] = (t - 1 < 0) ? 0 : y.dp[t - 1];
+215         t1.dp[1] = y.dp[t];
+216         t1.used = 2;
+217         if ((res = mp_mul_d (&t1, q.dp[i - t - 1], &t1)) != MP_OKAY) \{
+218           goto __Y;
+219         \}
+220   
+221         /* find right hand */
+222         t2.dp[0] = (i - 2 < 0) ? 0 : x.dp[i - 2];
+223         t2.dp[1] = (i - 1 < 0) ? 0 : x.dp[i - 1];
+224         t2.dp[2] = x.dp[i];
+225         t2.used = 3;
+226       \} while (mp_cmp_mag(&t1, &t2) == MP_GT);
+227   
+228       /* step 3.3 x = x - q\{i-t-1\} * y * b**\{i-t-1\} */
+229       if ((res = mp_mul_d (&y, q.dp[i - t - 1], &t1)) != MP_OKAY) \{
+230         goto __Y;
+231       \}
+232   
+233       if ((res = mp_lshd (&t1, i - t - 1)) != MP_OKAY) \{
+234         goto __Y;
+235       \}
+236   
+237       if ((res = mp_sub (&x, &t1, &x)) != MP_OKAY) \{
+238         goto __Y;
+239       \}
+240   
+241       /* if x < 0 then \{ x = x + y*b**\{i-t-1\}; q\{i-t-1\} -= 1; \} */
+242       if (x.sign == MP_NEG) \{
+243         if ((res = mp_copy (&y, &t1)) != MP_OKAY) \{
+244           goto __Y;
+245         \}
+246         if ((res = mp_lshd (&t1, i - t - 1)) != MP_OKAY) \{
+247           goto __Y;
+248         \}
+249         if ((res = mp_add (&x, &t1, &x)) != MP_OKAY) \{
+250           goto __Y;
+251         \}
+252   
+253         q.dp[i - t - 1] = (q.dp[i - t - 1] - 1UL) & MP_MASK;
+254       \}
+255     \}
+256   
+257     /* now q is the quotient and x is the remainder 
+258      * [which we have to normalize] 
+259      */
+260     
+261     /* get sign before writing to c */
+262     x.sign = x.used == 0 ? MP_ZPOS : a->sign;
+263   
+264     if (c != NULL) \{
+265       mp_clamp (&q);
+266       mp_exch (&q, c);
+267       c->sign = neg;
+268     \}
+269   
+270     if (d != NULL) \{
+271       mp_div_2d (&x, norm, &x, NULL);
+272       mp_exch (&x, d);
+273     \}
+274   
+275     res = MP_OKAY;
+276   
+277   __Y:mp_clear (&y);
+278   __X:mp_clear (&x);
+279   __T2:mp_clear (&t2);
+280   __T1:mp_clear (&t1);
+281   __Q:mp_clear (&q);
+282     return res;
+283   \}
+284   
+285   #endif
+286   
+287   #endif
 \end{alltt}
 \end{small}
 
@@ -8440,23 +8634,23 @@ algorithm with only the quotient is
 mp_div(&a, &b, &c, NULL);  /* c = [a/b] */
 \end{verbatim}
 
-Lines 36 and 42 handle the two trivial cases of inputs which are division by zero and dividend smaller than the divisor 
-respectively.  After the two trivial cases all of the temporary variables are initialized.  Line 75 determines the sign of 
+Lines 37 and 44 handle the two trivial cases of inputs which are division by zero and dividend smaller than the divisor 
+respectively.  After the two trivial cases all of the temporary variables are initialized.  Line 105 determines the sign of 
 the quotient and line 76 ensures that both $x$ and $y$ are positive.  
 
-The number of bits in the leading digit is calculated on line 80.  Implictly an mp\_int with $r$ digits will require $lg(\beta)(r-1) + k$ bits
+The number of bits in the leading digit is calculated on line 105.  Implictly an mp\_int with $r$ digits will require $lg(\beta)(r-1) + k$ bits
 of precision which when reduced modulo $lg(\beta)$ produces the value of $k$.  In this case $k$ is the number of bits in the leading digit which is
 exactly what is required.  For the algorithm to operate $k$ must equal $lg(\beta) - 1$ and when it does not the inputs must be normalized by shifting
 them to the left by $lg(\beta) - 1 - k$ bits.
 
 Throughout the variables $n$ and $t$ will represent the highest digit of $x$ and $y$ respectively.  These are first used to produce the 
-leading digit of the quotient.  The loop beginning on line 112 will produce the remainder of the quotient digits.
+leading digit of the quotient.  The loop beginning on line 183 will produce the remainder of the quotient digits.
 
-The conditional ``continue'' on line 113 is used to prevent the algorithm from reading past the leading edge of $x$ which can occur when the
+The conditional ``continue'' on line 114 is used to prevent the algorithm from reading past the leading edge of $x$ which can occur when the
 algorithm eliminates multiple non-zero digits in a single iteration.  This ensures that $x_i$ is always non-zero since by definition the digits
 above the $i$'th position $x$ must be zero in order for the quotient to be precise\footnote{Precise as far as integer division is concerned.}.  
 
-Lines 142, 143 and 150 through 152 manually construct the high accuracy estimations by setting the digits of the two mp\_int 
+Lines 130, 130 and 134 through 134 manually construct the high accuracy estimations by setting the digits of the two mp\_int 
 variables directly.  
 
 \section{Single Digit Helpers}
@@ -8582,6 +8776,7 @@ This algorithm initiates a temporary mp\_int with the value of the single digit
 101     return MP_OKAY;
 102   \}
 103   
+104   #endif
 \end{alltt}
 \end{small}
 
@@ -8689,6 +8884,7 @@ Unlike the full multiplication algorithms this algorithm does not require any si
 070   
 071     return MP_OKAY;
 072   \}
+073   #endif
 \end{alltt}
 \end{small}
 
@@ -8785,7 +8981,7 @@ from chapter seven.
 054     /* power of two ? */
 055     if (s_is_power_of_two(b, &ix) == 1) \{
 056        if (d != NULL) \{
-057           *d = a->dp[0] & ((1<<ix) - 1);
+057           *d = a->dp[0] & ((((mp_digit)1)<<ix) - 1);
 058        \}
 059        if (c != NULL) \{
 060           return mp_div_2d(a, ix, c, NULL);
@@ -8793,44 +8989,47 @@ from chapter seven.
 062        return MP_OKAY;
 063     \}
 064   
-065     /* three? */
-066     if (b == 3) \{
-067        return mp_div_3(a, c, d);
-068     \}
-069   
-070     /* no easy answer [c'est la vie].  Just division */
-071     if ((res = mp_init_size(&q, a->used)) != MP_OKAY) \{
-072        return res;
-073     \}
-074     
-075     q.used = a->used;
-076     q.sign = a->sign;
-077     w = 0;
-078     for (ix = a->used - 1; ix >= 0; ix--) \{
-079        w = (w << ((mp_word)DIGIT_BIT)) | ((mp_word)a->dp[ix]);
-080        
-081        if (w >= b) \{
-082           t = (mp_digit)(w / b);
-083           w -= ((mp_word)t) * ((mp_word)b);
-084         \} else \{
-085           t = 0;
-086         \}
-087         q.dp[ix] = (mp_digit)t;
-088     \}
-089     
-090     if (d != NULL) \{
-091        *d = (mp_digit)w;
-092     \}
-093     
-094     if (c != NULL) \{
-095        mp_clamp(&q);
-096        mp_exch(&q, c);
-097     \}
-098     mp_clear(&q);
-099     
-100     return res;
-101   \}
-102   
+065   #ifdef BN_MP_DIV_3_C
+066     /* three? */
+067     if (b == 3) \{
+068        return mp_div_3(a, c, d);
+069     \}
+070   #endif
+071   
+072     /* no easy answer [c'est la vie].  Just division */
+073     if ((res = mp_init_size(&q, a->used)) != MP_OKAY) \{
+074        return res;
+075     \}
+076     
+077     q.used = a->used;
+078     q.sign = a->sign;
+079     w = 0;
+080     for (ix = a->used - 1; ix >= 0; ix--) \{
+081        w = (w << ((mp_word)DIGIT_BIT)) | ((mp_word)a->dp[ix]);
+082        
+083        if (w >= b) \{
+084           t = (mp_digit)(w / b);
+085           w -= ((mp_word)t) * ((mp_word)b);
+086         \} else \{
+087           t = 0;
+088         \}
+089         q.dp[ix] = (mp_digit)t;
+090     \}
+091     
+092     if (d != NULL) \{
+093        *d = (mp_digit)w;
+094     \}
+095     
+096     if (c != NULL) \{
+097        mp_clamp(&q);
+098        mp_exch(&q, c);
+099     \}
+100     mp_clear(&q);
+101     
+102     return res;
+103   \}
+104   
+105   #endif
 \end{alltt}
 \end{small}
 
@@ -9016,6 +9215,7 @@ root.  Ideally this algorithm is meant to find the $n$'th root of an input where
 124   __T1:mp_clear (&t1);
 125     return res;
 126   \}
+127   #endif
 \end{alltt}
 \end{small}
 
@@ -9091,6 +9291,7 @@ the integers from $0$ to $\beta - 1$.
 047   
 048     return MP_OKAY;
 049   \}
+050   #endif
 \end{alltt}
 \end{small}
 
@@ -9234,6 +9435,7 @@ as part of larger input without any significant problem.
 074     \}
 075     return MP_OKAY;
 076   \}
+077   #endif
 \end{alltt}
 \end{small}
 
@@ -9352,6 +9554,7 @@ are required instead of a series of $n \times k$ divisions.  One design flaw of
 067     return MP_OKAY;
 068   \}
 069   
+070   #endif
 \end{alltt}
 \end{small}
 
@@ -9631,6 +9834,7 @@ must be adjusted by multiplying by the common factors of two ($2^k$) removed ear
 105   __U:mp_clear (&v);
 106     return res;
 107   \}
+108   #endif
 \end{alltt}
 \end{small}
 
@@ -9725,6 +9929,7 @@ dividing the product of the two inputs by their greatest common divisor.
 052     mp_clear_multi (&t1, &t2, NULL);
 053     return res;
 054   \}
+055   #endif
 \end{alltt}
 \end{small}
 
@@ -9966,6 +10171,7 @@ $\left ( {p' \over a'} \right )$ which is multiplied against the current Jacobi
 097   __A1:mp_clear (&a1);
 098     return res;
 099   \}
+100   #endif
 \end{alltt}
 \end{small}
 
@@ -10095,161 +10301,25 @@ then only a couple of additions or subtractions will be required to adjust the i
 017   /* hac 14.61, pp608 */
 018   int mp_invmod (mp_int * a, mp_int * b, mp_int * c)
 019   \{
-020     mp_int  x, y, u, v, A, B, C, D;
-021     int     res;
-022   
-023     /* b cannot be negative */
-024     if (b->sign == MP_NEG || mp_iszero(b) == 1) \{
-025       return MP_VAL;
-026     \}
-027   
-028     /* if the modulus is odd we can use a faster routine instead */
-029     if (mp_isodd (b) == 1) \{
-030       return fast_mp_invmod (a, b, c);
-031     \}
-032     
-033     /* init temps */
-034     if ((res = mp_init_multi(&x, &y, &u, &v, 
-035                              &A, &B, &C, &D, NULL)) != MP_OKAY) \{
-036        return res;
-037     \}
-038   
-039     /* x = a, y = b */
-040     if ((res = mp_copy (a, &x)) != MP_OKAY) \{
-041       goto __ERR;
-042     \}
-043     if ((res = mp_copy (b, &y)) != MP_OKAY) \{
-044       goto __ERR;
-045     \}
-046   
-047     /* 2. [modified] if x,y are both even then return an error! */
-048     if (mp_iseven (&x) == 1 && mp_iseven (&y) == 1) \{
-049       res = MP_VAL;
-050       goto __ERR;
-051     \}
-052   
-053     /* 3. u=x, v=y, A=1, B=0, C=0,D=1 */
-054     if ((res = mp_copy (&x, &u)) != MP_OKAY) \{
-055       goto __ERR;
-056     \}
-057     if ((res = mp_copy (&y, &v)) != MP_OKAY) \{
-058       goto __ERR;
-059     \}
-060     mp_set (&A, 1);
-061     mp_set (&D, 1);
-062   
-063   top:
-064     /* 4.  while u is even do */
-065     while (mp_iseven (&u) == 1) \{
-066       /* 4.1 u = u/2 */
-067       if ((res = mp_div_2 (&u, &u)) != MP_OKAY) \{
-068         goto __ERR;
-069       \}
-070       /* 4.2 if A or B is odd then */
-071       if (mp_isodd (&A) == 1 || mp_isodd (&B) == 1) \{
-072         /* A = (A+y)/2, B = (B-x)/2 */
-073         if ((res = mp_add (&A, &y, &A)) != MP_OKAY) \{
-074            goto __ERR;
-075         \}
-076         if ((res = mp_sub (&B, &x, &B)) != MP_OKAY) \{
-077            goto __ERR;
-078         \}
-079       \}
-080       /* A = A/2, B = B/2 */
-081       if ((res = mp_div_2 (&A, &A)) != MP_OKAY) \{
-082         goto __ERR;
-083       \}
-084       if ((res = mp_div_2 (&B, &B)) != MP_OKAY) \{
-085         goto __ERR;
-086       \}
-087     \}
-088   
-089     /* 5.  while v is even do */
-090     while (mp_iseven (&v) == 1) \{
-091       /* 5.1 v = v/2 */
-092       if ((res = mp_div_2 (&v, &v)) != MP_OKAY) \{
-093         goto __ERR;
-094       \}
-095       /* 5.2 if C or D is odd then */
-096       if (mp_isodd (&C) == 1 || mp_isodd (&D) == 1) \{
-097         /* C = (C+y)/2, D = (D-x)/2 */
-098         if ((res = mp_add (&C, &y, &C)) != MP_OKAY) \{
-099            goto __ERR;
-100         \}
-101         if ((res = mp_sub (&D, &x, &D)) != MP_OKAY) \{
-102            goto __ERR;
-103         \}
-104       \}
-105       /* C = C/2, D = D/2 */
-106       if ((res = mp_div_2 (&C, &C)) != MP_OKAY) \{
-107         goto __ERR;
-108       \}
-109       if ((res = mp_div_2 (&D, &D)) != MP_OKAY) \{
-110         goto __ERR;
-111       \}
-112     \}
-113   
-114     /* 6.  if u >= v then */
-115     if (mp_cmp (&u, &v) != MP_LT) \{
-116       /* u = u - v, A = A - C, B = B - D */
-117       if ((res = mp_sub (&u, &v, &u)) != MP_OKAY) \{
-118         goto __ERR;
-119       \}
-120   
-121       if ((res = mp_sub (&A, &C, &A)) != MP_OKAY) \{
-122         goto __ERR;
-123       \}
-124   
-125       if ((res = mp_sub (&B, &D, &B)) != MP_OKAY) \{
-126         goto __ERR;
-127       \}
-128     \} else \{
-129       /* v - v - u, C = C - A, D = D - B */
-130       if ((res = mp_sub (&v, &u, &v)) != MP_OKAY) \{
-131         goto __ERR;
-132       \}
-133   
-134       if ((res = mp_sub (&C, &A, &C)) != MP_OKAY) \{
-135         goto __ERR;
-136       \}
-137   
-138       if ((res = mp_sub (&D, &B, &D)) != MP_OKAY) \{
-139         goto __ERR;
-140       \}
-141     \}
-142   
-143     /* if not zero goto step 4 */
-144     if (mp_iszero (&u) == 0)
-145       goto top;
-146   
-147     /* now a = C, b = D, gcd == g*v */
-148   
-149     /* if v != 1 then there is no inverse */
-150     if (mp_cmp_d (&v, 1) != MP_EQ) \{
-151       res = MP_VAL;
-152       goto __ERR;
-153     \}
-154   
-155     /* if its too low */
-156     while (mp_cmp_d(&C, 0) == MP_LT) \{
-157         if ((res = mp_add(&C, b, &C)) != MP_OKAY) \{
-158            goto __ERR;
-159         \}
-160     \}
-161     
-162     /* too big */
-163     while (mp_cmp_mag(&C, b) != MP_LT) \{
-164         if ((res = mp_sub(&C, b, &C)) != MP_OKAY) \{
-165            goto __ERR;
-166         \}
-167     \}
-168     
-169     /* C is now the inverse */
-170     mp_exch (&C, c);
-171     res = MP_OKAY;
-172   __ERR:mp_clear_multi (&x, &y, &u, &v, &A, &B, &C, &D, NULL);
-173     return res;
-174   \}
+020     /* b cannot be negative */
+021     if (b->sign == MP_NEG || mp_iszero(b) == 1) \{
+022       return MP_VAL;
+023     \}
+024   
+025   #ifdef BN_FAST_MP_INVMOD_C
+026     /* if the modulus is odd we can use a faster routine instead */
+027     if (mp_isodd (b) == 1) \{
+028       return fast_mp_invmod (a, b, c);
+029     \}
+030   #endif
+031   
+032   #ifdef BN_MP_INVMOD_SLOW_C
+033     return mp_invmod_slow(a, b, c);
+034   #endif
+035   
+036     return MP_VAL;
+037   \}
+038   #endif
 \end{alltt}
 \end{small}
 
@@ -10350,6 +10420,7 @@ This algorithm attempts to determine if a candidate integer $n$ is composite by
 042   
 043     return MP_OKAY;
 044   \}
+045   #endif
 \end{alltt}
 \end{small}
 
@@ -10400,6 +10471,7 @@ mp\_digit.  The table \_\_prime\_tab is defined in the following file.
 053     0x062B, 0x062F, 0x063D, 0x0641, 0x0647, 0x0649, 0x064D, 0x0653
 054   #endif
 055   \};
+056   #endif
 \end{alltt}
 \end{small}
 
@@ -10487,6 +10559,7 @@ determine the result.
 054   __T:mp_clear (&t);
 055     return err;
 056   \}
+057   #endif
 \end{alltt}
 \end{small}
 
@@ -10621,6 +10694,7 @@ composite then it is \textit{probably} prime.
 095   __N1:mp_clear (&n1);
 096     return err;
 097   \}
+098   #endif
 \end{alltt}
 \end{small}
 
diff --git a/tommath_class.h b/tommath_class.h
new file mode 100644
index 0000000000000000000000000000000000000000..2a17d430dbbf540eff27be28856d121c4f09d4f9
--- /dev/null
+++ b/tommath_class.h
@@ -0,0 +1,951 @@
+#if !(defined(LTM1) && defined(LTM2) && defined(LTM3))
+#if defined(LTM2)
+#define LTM3
+#endif
+#if defined(LTM1)
+#define LTM2
+#endif
+#define LTM1
+
+#if defined(LTM_ALL)
+#define BN_ERROR_C
+#define BN_FAST_MP_INVMOD_C
+#define BN_FAST_MP_MONTGOMERY_REDUCE_C
+#define BN_FAST_S_MP_MUL_DIGS_C
+#define BN_FAST_S_MP_MUL_HIGH_DIGS_C
+#define BN_FAST_S_MP_SQR_C
+#define BN_MP_2EXPT_C
+#define BN_MP_ABS_C
+#define BN_MP_ADD_C
+#define BN_MP_ADD_D_C
+#define BN_MP_ADDMOD_C
+#define BN_MP_AND_C
+#define BN_MP_CLAMP_C
+#define BN_MP_CLEAR_C
+#define BN_MP_CLEAR_MULTI_C
+#define BN_MP_CMP_C
+#define BN_MP_CMP_D_C
+#define BN_MP_CMP_MAG_C
+#define BN_MP_CNT_LSB_C
+#define BN_MP_COPY_C
+#define BN_MP_COUNT_BITS_C
+#define BN_MP_DIV_C
+#define BN_MP_DIV_2_C
+#define BN_MP_DIV_2D_C
+#define BN_MP_DIV_3_C
+#define BN_MP_DIV_D_C
+#define BN_MP_DR_IS_MODULUS_C
+#define BN_MP_DR_REDUCE_C
+#define BN_MP_DR_SETUP_C
+#define BN_MP_EXCH_C
+#define BN_MP_EXPT_D_C
+#define BN_MP_EXPTMOD_C
+#define BN_MP_EXPTMOD_FAST_C
+#define BN_MP_EXTEUCLID_C
+#define BN_MP_FREAD_C
+#define BN_MP_FWRITE_C
+#define BN_MP_GCD_C
+#define BN_MP_GET_INT_C
+#define BN_MP_GROW_C
+#define BN_MP_INIT_C
+#define BN_MP_INIT_COPY_C
+#define BN_MP_INIT_MULTI_C
+#define BN_MP_INIT_SET_C
+#define BN_MP_INIT_SET_INT_C
+#define BN_MP_INIT_SIZE_C
+#define BN_MP_INVMOD_C
+#define BN_MP_INVMOD_SLOW_C
+#define BN_MP_IS_SQUARE_C
+#define BN_MP_JACOBI_C
+#define BN_MP_KARATSUBA_MUL_C
+#define BN_MP_KARATSUBA_SQR_C
+#define BN_MP_LCM_C
+#define BN_MP_LSHD_C
+#define BN_MP_MOD_C
+#define BN_MP_MOD_2D_C
+#define BN_MP_MOD_D_C
+#define BN_MP_MONTGOMERY_CALC_NORMALIZATION_C
+#define BN_MP_MONTGOMERY_REDUCE_C
+#define BN_MP_MONTGOMERY_SETUP_C
+#define BN_MP_MUL_C
+#define BN_MP_MUL_2_C
+#define BN_MP_MUL_2D_C
+#define BN_MP_MUL_D_C
+#define BN_MP_MULMOD_C
+#define BN_MP_N_ROOT_C
+#define BN_MP_NEG_C
+#define BN_MP_OR_C
+#define BN_MP_PRIME_FERMAT_C
+#define BN_MP_PRIME_IS_DIVISIBLE_C
+#define BN_MP_PRIME_IS_PRIME_C
+#define BN_MP_PRIME_MILLER_RABIN_C
+#define BN_MP_PRIME_NEXT_PRIME_C
+#define BN_MP_PRIME_RABIN_MILLER_TRIALS_C
+#define BN_MP_PRIME_RANDOM_EX_C
+#define BN_MP_RADIX_SIZE_C
+#define BN_MP_RADIX_SMAP_C
+#define BN_MP_RAND_C
+#define BN_MP_READ_RADIX_C
+#define BN_MP_READ_SIGNED_BIN_C
+#define BN_MP_READ_UNSIGNED_BIN_C
+#define BN_MP_REDUCE_C
+#define BN_MP_REDUCE_2K_C
+#define BN_MP_REDUCE_2K_SETUP_C
+#define BN_MP_REDUCE_IS_2K_C
+#define BN_MP_REDUCE_SETUP_C
+#define BN_MP_RSHD_C
+#define BN_MP_SET_C
+#define BN_MP_SET_INT_C
+#define BN_MP_SHRINK_C
+#define BN_MP_SIGNED_BIN_SIZE_C
+#define BN_MP_SQR_C
+#define BN_MP_SQRMOD_C
+#define BN_MP_SQRT_C
+#define BN_MP_SUB_C
+#define BN_MP_SUB_D_C
+#define BN_MP_SUBMOD_C
+#define BN_MP_TO_SIGNED_BIN_C
+#define BN_MP_TO_UNSIGNED_BIN_C
+#define BN_MP_TOOM_MUL_C
+#define BN_MP_TOOM_SQR_C
+#define BN_MP_TORADIX_C
+#define BN_MP_TORADIX_N_C
+#define BN_MP_UNSIGNED_BIN_SIZE_C
+#define BN_MP_XOR_C
+#define BN_MP_ZERO_C
+#define BN_PRIME_TAB_C
+#define BN_REVERSE_C
+#define BN_S_MP_ADD_C
+#define BN_S_MP_EXPTMOD_C
+#define BN_S_MP_MUL_DIGS_C
+#define BN_S_MP_MUL_HIGH_DIGS_C
+#define BN_S_MP_SQR_C
+#define BN_S_MP_SUB_C
+#define BNCORE_C
+#endif
+
+#if defined(BN_ERROR_C)
+   #define BN_MP_ERROR_TO_STRING_C
+#endif
+
+#if defined(BN_FAST_MP_INVMOD_C)
+   #define BN_MP_ISEVEN_C
+   #define BN_MP_INIT_MULTI_C
+   #define BN_MP_COPY_C
+   #define BN_MP_ABS_C
+   #define BN_MP_SET_C
+   #define BN_MP_DIV_2_C
+   #define BN_MP_ISODD_C
+   #define BN_MP_SUB_C
+   #define BN_MP_CMP_C
+   #define BN_MP_ISZERO_C
+   #define BN_MP_CMP_D_C
+   #define BN_MP_ADD_C
+   #define BN_MP_EXCH_C
+   #define BN_MP_CLEAR_MULTI_C
+#endif
+
+#if defined(BN_FAST_MP_MONTGOMERY_REDUCE_C)
+   #define BN_MP_GROW_C
+   #define BN_MP_RSHD_C
+   #define BN_MP_CLAMP_C
+   #define BN_MP_CMP_MAG_C
+   #define BN_S_MP_SUB_C
+#endif
+
+#if defined(BN_FAST_S_MP_MUL_DIGS_C)
+   #define BN_MP_GROW_C
+   #define BN_MP_CLAMP_C
+#endif
+
+#if defined(BN_FAST_S_MP_MUL_HIGH_DIGS_C)
+   #define BN_MP_GROW_C
+   #define BN_MP_CLAMP_C
+#endif
+
+#if defined(BN_FAST_S_MP_SQR_C)
+   #define BN_MP_GROW_C
+   #define BN_MP_CLAMP_C
+#endif
+
+#if defined(BN_MP_2EXPT_C)
+   #define BN_MP_ZERO_C
+   #define BN_MP_GROW_C
+#endif
+
+#if defined(BN_MP_ABS_C)
+   #define BN_MP_COPY_C
+#endif
+
+#if defined(BN_MP_ADD_C)
+   #define BN_S_MP_ADD_C
+   #define BN_MP_CMP_MAG_C
+   #define BN_S_MP_SUB_C
+#endif
+
+#if defined(BN_MP_ADD_D_C)
+   #define BN_MP_GROW_C
+   #define BN_MP_SUB_D_C
+   #define BN_MP_CLAMP_C
+#endif
+
+#if defined(BN_MP_ADDMOD_C)
+   #define BN_MP_INIT_C
+   #define BN_MP_ADD_C
+   #define BN_MP_CLEAR_C
+   #define BN_MP_MOD_C
+#endif
+
+#if defined(BN_MP_AND_C)
+   #define BN_MP_INIT_COPY_C
+   #define BN_MP_CLAMP_C
+   #define BN_MP_EXCH_C
+   #define BN_MP_CLEAR_C
+#endif
+
+#if defined(BN_MP_CLAMP_C)
+#endif
+
+#if defined(BN_MP_CLEAR_C)
+#endif
+
+#if defined(BN_MP_CLEAR_MULTI_C)
+   #define BN_MP_CLEAR_C
+#endif
+
+#if defined(BN_MP_CMP_C)
+   #define BN_MP_CMP_MAG_C
+#endif
+
+#if defined(BN_MP_CMP_D_C)
+#endif
+
+#if defined(BN_MP_CMP_MAG_C)
+#endif
+
+#if defined(BN_MP_CNT_LSB_C)
+   #define BN_MP_ISZERO_C
+#endif
+
+#if defined(BN_MP_COPY_C)
+   #define BN_MP_GROW_C
+#endif
+
+#if defined(BN_MP_COUNT_BITS_C)
+#endif
+
+#if defined(BN_MP_DIV_C)
+   #define BN_MP_ISZERO_C
+   #define BN_MP_CMP_MAG_C
+   #define BN_MP_COPY_C
+   #define BN_MP_ZERO_C
+   #define BN_MP_INIT_MULTI_C
+   #define BN_MP_SET_C
+   #define BN_MP_COUNT_BITS_C
+   #define BN_MP_MUL_2D_C
+   #define BN_MP_CMP_C
+   #define BN_MP_SUB_C
+   #define BN_MP_ADD_C
+   #define BN_MP_DIV_2D_C
+   #define BN_MP_EXCH_C
+   #define BN_MP_CLEAR_MULTI_C
+   #define BN_MP_INIT_SIZE_C
+   #define BN_MP_INIT_C
+   #define BN_MP_INIT_COPY_C
+   #define BN_MP_LSHD_C
+   #define BN_MP_RSHD_C
+   #define BN_MP_MUL_D_C
+   #define BN_MP_CLAMP_C
+   #define BN_MP_CLEAR_C
+#endif
+
+#if defined(BN_MP_DIV_2_C)
+   #define BN_MP_GROW_C
+   #define BN_MP_CLAMP_C
+#endif
+
+#if defined(BN_MP_DIV_2D_C)
+   #define BN_MP_COPY_C
+   #define BN_MP_ZERO_C
+   #define BN_MP_INIT_C
+   #define BN_MP_MOD_2D_C
+   #define BN_MP_CLEAR_C
+   #define BN_MP_RSHD_C
+   #define BN_MP_CLAMP_C
+   #define BN_MP_EXCH_C
+#endif
+
+#if defined(BN_MP_DIV_3_C)
+   #define BN_MP_INIT_SIZE_C
+   #define BN_MP_CLAMP_C
+   #define BN_MP_EXCH_C
+   #define BN_MP_CLEAR_C
+#endif
+
+#if defined(BN_MP_DIV_D_C)
+   #define BN_MP_ISZERO_C
+   #define BN_MP_COPY_C
+   #define BN_MP_DIV_2D_C
+   #define BN_MP_DIV_3_C
+   #define BN_MP_INIT_SIZE_C
+   #define BN_MP_CLAMP_C
+   #define BN_MP_EXCH_C
+   #define BN_MP_CLEAR_C
+#endif
+
+#if defined(BN_MP_DR_IS_MODULUS_C)
+#endif
+
+#if defined(BN_MP_DR_REDUCE_C)
+   #define BN_MP_GROW_C
+   #define BN_MP_CLAMP_C
+   #define BN_MP_CMP_MAG_C
+   #define BN_S_MP_SUB_C
+#endif
+
+#if defined(BN_MP_DR_SETUP_C)
+#endif
+
+#if defined(BN_MP_EXCH_C)
+#endif
+
+#if defined(BN_MP_EXPT_D_C)
+   #define BN_MP_INIT_COPY_C
+   #define BN_MP_SET_C
+   #define BN_MP_SQR_C
+   #define BN_MP_CLEAR_C
+   #define BN_MP_MUL_C
+#endif
+
+#if defined(BN_MP_EXPTMOD_C)
+   #define BN_MP_INIT_C
+   #define BN_MP_INVMOD_C
+   #define BN_MP_CLEAR_C
+   #define BN_MP_ABS_C
+   #define BN_MP_CLEAR_MULTI_C
+   #define BN_MP_DR_IS_MODULUS_C
+   #define BN_MP_REDUCE_IS_2K_C
+   #define BN_MP_ISODD_C
+   #define BN_MP_EXPTMOD_FAST_C
+   #define BN_S_MP_EXPTMOD_C
+#endif
+
+#if defined(BN_MP_EXPTMOD_FAST_C)
+   #define BN_MP_COUNT_BITS_C
+   #define BN_MP_INIT_C
+   #define BN_MP_CLEAR_C
+   #define BN_MP_MONTGOMERY_SETUP_C
+   #define BN_FAST_MP_MONTGOMERY_REDUCE_C
+   #define BN_MP_MONTGOMERY_REDUCE_C
+   #define BN_MP_DR_SETUP_C
+   #define BN_MP_DR_REDUCE_C
+   #define BN_MP_REDUCE_2K_SETUP_C
+   #define BN_MP_REDUCE_2K_C
+   #define BN_MP_MONTGOMERY_CALC_NORMALIZATION_C
+   #define BN_MP_MULMOD_C
+   #define BN_MP_SET_C
+   #define BN_MP_MOD_C
+   #define BN_MP_COPY_C
+   #define BN_MP_SQR_C
+   #define BN_MP_MUL_C
+   #define BN_MP_EXCH_C
+#endif
+
+#if defined(BN_MP_EXTEUCLID_C)
+   #define BN_MP_INIT_MULTI_C
+   #define BN_MP_SET_C
+   #define BN_MP_COPY_C
+   #define BN_MP_ISZERO_C
+   #define BN_MP_DIV_C
+   #define BN_MP_MUL_C
+   #define BN_MP_SUB_C
+   #define BN_MP_EXCH_C
+   #define BN_MP_CLEAR_MULTI_C
+#endif
+
+#if defined(BN_MP_FREAD_C)
+   #define BN_MP_ZERO_C
+   #define BN_MP_S_RMAP_C
+   #define BN_MP_MUL_D_C
+   #define BN_MP_ADD_D_C
+   #define BN_MP_CMP_D_C
+#endif
+
+#if defined(BN_MP_FWRITE_C)
+   #define BN_MP_RADIX_SIZE_C
+   #define BN_MP_TORADIX_C
+#endif
+
+#if defined(BN_MP_GCD_C)
+   #define BN_MP_ISZERO_C
+   #define BN_MP_ABS_C
+   #define BN_MP_ZERO_C
+   #define BN_MP_INIT_COPY_C
+   #define BN_MP_CNT_LSB_C
+   #define BN_MP_DIV_2D_C
+   #define BN_MP_CMP_MAG_C
+   #define BN_MP_EXCH_C
+   #define BN_S_MP_SUB_C
+   #define BN_MP_MUL_2D_C
+   #define BN_MP_CLEAR_C
+#endif
+
+#if defined(BN_MP_GET_INT_C)
+#endif
+
+#if defined(BN_MP_GROW_C)
+#endif
+
+#if defined(BN_MP_INIT_C)
+#endif
+
+#if defined(BN_MP_INIT_COPY_C)
+   #define BN_MP_COPY_C
+#endif
+
+#if defined(BN_MP_INIT_MULTI_C)
+   #define BN_MP_ERR_C
+   #define BN_MP_INIT_C
+   #define BN_MP_CLEAR_C
+#endif
+
+#if defined(BN_MP_INIT_SET_C)
+   #define BN_MP_INIT_C
+   #define BN_MP_SET_C
+#endif
+
+#if defined(BN_MP_INIT_SET_INT_C)
+   #define BN_MP_INIT_C
+   #define BN_MP_SET_INT_C
+#endif
+
+#if defined(BN_MP_INIT_SIZE_C)
+   #define BN_MP_INIT_C
+#endif
+
+#if defined(BN_MP_INVMOD_C)
+   #define BN_MP_ISZERO_C
+   #define BN_MP_ISODD_C
+   #define BN_FAST_MP_INVMOD_C
+   #define BN_MP_INVMOD_SLOW_C
+#endif
+
+#if defined(BN_MP_INVMOD_SLOW_C)
+   #define BN_MP_ISZERO_C
+   #define BN_MP_INIT_MULTI_C
+   #define BN_MP_COPY_C
+   #define BN_MP_ISEVEN_C
+   #define BN_MP_SET_C
+   #define BN_MP_DIV_2_C
+   #define BN_MP_ISODD_C
+   #define BN_MP_ADD_C
+   #define BN_MP_SUB_C
+   #define BN_MP_CMP_C
+   #define BN_MP_CMP_D_C
+   #define BN_MP_CMP_MAG_C
+   #define BN_MP_EXCH_C
+   #define BN_MP_CLEAR_MULTI_C
+#endif
+
+#if defined(BN_MP_IS_SQUARE_C)
+   #define BN_MP_MOD_D_C
+   #define BN_MP_INIT_SET_INT_C
+   #define BN_MP_MOD_C
+   #define BN_MP_GET_INT_C
+   #define BN_MP_SQRT_C
+   #define BN_MP_SQR_C
+   #define BN_MP_CMP_MAG_C
+   #define BN_MP_CLEAR_C
+#endif
+
+#if defined(BN_MP_JACOBI_C)
+   #define BN_MP_CMP_D_C
+   #define BN_MP_ISZERO_C
+   #define BN_MP_INIT_COPY_C
+   #define BN_MP_CNT_LSB_C
+   #define BN_MP_DIV_2D_C
+   #define BN_MP_MOD_C
+   #define BN_MP_CLEAR_C
+#endif
+
+#if defined(BN_MP_KARATSUBA_MUL_C)
+   #define BN_MP_MUL_C
+   #define BN_MP_INIT_SIZE_C
+   #define BN_MP_CLAMP_C
+   #define BN_MP_SUB_C
+   #define BN_MP_ADD_C
+   #define BN_MP_LSHD_C
+   #define BN_MP_CLEAR_C
+#endif
+
+#if defined(BN_MP_KARATSUBA_SQR_C)
+   #define BN_MP_INIT_SIZE_C
+   #define BN_MP_CLAMP_C
+   #define BN_MP_SQR_C
+   #define BN_MP_SUB_C
+   #define BN_S_MP_ADD_C
+   #define BN_MP_LSHD_C
+   #define BN_MP_ADD_C
+   #define BN_MP_CLEAR_C
+#endif
+
+#if defined(BN_MP_LCM_C)
+   #define BN_MP_INIT_MULTI_C
+   #define BN_MP_GCD_C
+   #define BN_MP_CMP_MAG_C
+   #define BN_MP_DIV_C
+   #define BN_MP_MUL_C
+   #define BN_MP_CLEAR_MULTI_C
+#endif
+
+#if defined(BN_MP_LSHD_C)
+   #define BN_MP_GROW_C
+   #define BN_MP_RSHD_C
+#endif
+
+#if defined(BN_MP_MOD_C)
+   #define BN_MP_INIT_C
+   #define BN_MP_DIV_C
+   #define BN_MP_CLEAR_C
+   #define BN_MP_ADD_C
+   #define BN_MP_EXCH_C
+#endif
+
+#if defined(BN_MP_MOD_2D_C)
+   #define BN_MP_ZERO_C
+   #define BN_MP_COPY_C
+   #define BN_MP_CLAMP_C
+#endif
+
+#if defined(BN_MP_MOD_D_C)
+   #define BN_MP_DIV_D_C
+#endif
+
+#if defined(BN_MP_MONTGOMERY_CALC_NORMALIZATION_C)
+   #define BN_MP_COUNT_BITS_C
+   #define BN_MP_2EXPT_C
+   #define BN_MP_SET_C
+   #define BN_MP_MUL_2_C
+   #define BN_MP_CMP_MAG_C
+   #define BN_S_MP_SUB_C
+#endif
+
+#if defined(BN_MP_MONTGOMERY_REDUCE_C)
+   #define BN_FAST_MP_MONTGOMERY_REDUCE_C
+   #define BN_MP_GROW_C
+   #define BN_MP_CLAMP_C
+   #define BN_MP_RSHD_C
+   #define BN_MP_CMP_MAG_C
+   #define BN_S_MP_SUB_C
+#endif
+
+#if defined(BN_MP_MONTGOMERY_SETUP_C)
+#endif
+
+#if defined(BN_MP_MUL_C)
+   #define BN_MP_TOOM_MUL_C
+   #define BN_MP_KARATSUBA_MUL_C
+   #define BN_FAST_S_MP_MUL_DIGS_C
+   #define BN_S_MP_MUL_C
+   #define BN_S_MP_MUL_DIGS_C
+#endif
+
+#if defined(BN_MP_MUL_2_C)
+   #define BN_MP_GROW_C
+#endif
+
+#if defined(BN_MP_MUL_2D_C)
+   #define BN_MP_COPY_C
+   #define BN_MP_GROW_C
+   #define BN_MP_LSHD_C
+   #define BN_MP_CLAMP_C
+#endif
+
+#if defined(BN_MP_MUL_D_C)
+   #define BN_MP_GROW_C
+   #define BN_MP_CLAMP_C
+#endif
+
+#if defined(BN_MP_MULMOD_C)
+   #define BN_MP_INIT_C
+   #define BN_MP_MUL_C
+   #define BN_MP_CLEAR_C
+   #define BN_MP_MOD_C
+#endif
+
+#if defined(BN_MP_N_ROOT_C)
+   #define BN_MP_INIT_C
+   #define BN_MP_SET_C
+   #define BN_MP_COPY_C
+   #define BN_MP_EXPT_D_C
+   #define BN_MP_MUL_C
+   #define BN_MP_SUB_C
+   #define BN_MP_MUL_D_C
+   #define BN_MP_DIV_C
+   #define BN_MP_CMP_C
+   #define BN_MP_SUB_D_C
+   #define BN_MP_EXCH_C
+   #define BN_MP_CLEAR_C
+#endif
+
+#if defined(BN_MP_NEG_C)
+   #define BN_MP_COPY_C
+   #define BN_MP_ISZERO_C
+#endif
+
+#if defined(BN_MP_OR_C)
+   #define BN_MP_INIT_COPY_C
+   #define BN_MP_CLAMP_C
+   #define BN_MP_EXCH_C
+   #define BN_MP_CLEAR_C
+#endif
+
+#if defined(BN_MP_PRIME_FERMAT_C)
+   #define BN_MP_CMP_D_C
+   #define BN_MP_INIT_C
+   #define BN_MP_EXPTMOD_C
+   #define BN_MP_CMP_C
+   #define BN_MP_CLEAR_C
+#endif
+
+#if defined(BN_MP_PRIME_IS_DIVISIBLE_C)
+   #define BN_MP_MOD_D_C
+#endif
+
+#if defined(BN_MP_PRIME_IS_PRIME_C)
+   #define BN_MP_CMP_D_C
+   #define BN_MP_PRIME_IS_DIVISIBLE_C
+   #define BN_MP_INIT_C
+   #define BN_MP_SET_C
+   #define BN_MP_PRIME_MILLER_RABIN_C
+   #define BN_MP_CLEAR_C
+#endif
+
+#if defined(BN_MP_PRIME_MILLER_RABIN_C)
+   #define BN_MP_CMP_D_C
+   #define BN_MP_INIT_COPY_C
+   #define BN_MP_SUB_D_C
+   #define BN_MP_CNT_LSB_C
+   #define BN_MP_DIV_2D_C
+   #define BN_MP_EXPTMOD_C
+   #define BN_MP_CMP_C
+   #define BN_MP_SQRMOD_C
+   #define BN_MP_CLEAR_C
+#endif
+
+#if defined(BN_MP_PRIME_NEXT_PRIME_C)
+   #define BN_MP_CMP_D_C
+   #define BN_MP_SET_C
+   #define BN_MP_SUB_D_C
+   #define BN_MP_ISEVEN_C
+   #define BN_MP_MOD_D_C
+   #define BN_MP_INIT_C
+   #define BN_MP_ADD_D_C
+   #define BN_MP_PRIME_MILLER_RABIN_C
+   #define BN_MP_CLEAR_C
+#endif
+
+#if defined(BN_MP_PRIME_RABIN_MILLER_TRIALS_C)
+#endif
+
+#if defined(BN_MP_PRIME_RANDOM_EX_C)
+   #define BN_MP_READ_UNSIGNED_BIN_C
+   #define BN_MP_PRIME_IS_PRIME_C
+   #define BN_MP_SUB_D_C
+   #define BN_MP_DIV_2_C
+   #define BN_MP_MUL_2_C
+   #define BN_MP_ADD_D_C
+#endif
+
+#if defined(BN_MP_RADIX_SIZE_C)
+   #define BN_MP_COUNT_BITS_C
+   #define BN_MP_INIT_COPY_C
+   #define BN_MP_ISZERO_C
+   #define BN_MP_DIV_D_C
+   #define BN_MP_CLEAR_C
+#endif
+
+#if defined(BN_MP_RADIX_SMAP_C)
+   #define BN_MP_S_RMAP_C
+#endif
+
+#if defined(BN_MP_RAND_C)
+   #define BN_MP_ZERO_C
+   #define BN_MP_ADD_D_C
+   #define BN_MP_LSHD_C
+#endif
+
+#if defined(BN_MP_READ_RADIX_C)
+   #define BN_MP_ZERO_C
+   #define BN_MP_S_RMAP_C
+   #define BN_MP_MUL_D_C
+   #define BN_MP_ADD_D_C
+   #define BN_MP_ISZERO_C
+#endif
+
+#if defined(BN_MP_READ_SIGNED_BIN_C)
+   #define BN_MP_READ_UNSIGNED_BIN_C
+#endif
+
+#if defined(BN_MP_READ_UNSIGNED_BIN_C)
+   #define BN_MP_GROW_C
+   #define BN_MP_ZERO_C
+   #define BN_MP_MUL_2D_C
+   #define BN_MP_CLAMP_C
+#endif
+
+#if defined(BN_MP_REDUCE_C)
+   #define BN_MP_REDUCE_SETUP_C
+   #define BN_MP_INIT_COPY_C
+   #define BN_MP_RSHD_C
+   #define BN_MP_MUL_C
+   #define BN_S_MP_MUL_HIGH_DIGS_C
+   #define BN_FAST_S_MP_MUL_HIGH_DIGS_C
+   #define BN_MP_MOD_2D_C
+   #define BN_S_MP_MUL_DIGS_C
+   #define BN_MP_SUB_C
+   #define BN_MP_CMP_D_C
+   #define BN_MP_SET_C
+   #define BN_MP_LSHD_C
+   #define BN_MP_ADD_C
+   #define BN_MP_CMP_C
+   #define BN_S_MP_SUB_C
+   #define BN_MP_CLEAR_C
+#endif
+
+#if defined(BN_MP_REDUCE_2K_C)
+   #define BN_MP_INIT_C
+   #define BN_MP_COUNT_BITS_C
+   #define BN_MP_DIV_2D_C
+   #define BN_MP_MUL_D_C
+   #define BN_S_MP_ADD_C
+   #define BN_MP_CMP_MAG_C
+   #define BN_S_MP_SUB_C
+   #define BN_MP_CLEAR_C
+#endif
+
+#if defined(BN_MP_REDUCE_2K_SETUP_C)
+   #define BN_MP_INIT_C
+   #define BN_MP_COUNT_BITS_C
+   #define BN_MP_2EXPT_C
+   #define BN_MP_CLEAR_C
+   #define BN_S_MP_SUB_C
+#endif
+
+#if defined(BN_MP_REDUCE_IS_2K_C)
+   #define BN_MP_REDUCE_2K_C
+   #define BN_MP_COUNT_BITS_C
+#endif
+
+#if defined(BN_MP_REDUCE_SETUP_C)
+   #define BN_MP_2EXPT_C
+   #define BN_MP_DIV_C
+#endif
+
+#if defined(BN_MP_RSHD_C)
+   #define BN_MP_ZERO_C
+#endif
+
+#if defined(BN_MP_SET_C)
+   #define BN_MP_ZERO_C
+#endif
+
+#if defined(BN_MP_SET_INT_C)
+   #define BN_MP_ZERO_C
+   #define BN_MP_MUL_2D_C
+   #define BN_MP_CLAMP_C
+#endif
+
+#if defined(BN_MP_SHRINK_C)
+#endif
+
+#if defined(BN_MP_SIGNED_BIN_SIZE_C)
+   #define BN_MP_UNSIGNED_BIN_SIZE_C
+#endif
+
+#if defined(BN_MP_SQR_C)
+   #define BN_MP_TOOM_SQR_C
+   #define BN_MP_KARATSUBA_SQR_C
+   #define BN_FAST_S_MP_SQR_C
+   #define BN_S_MP_SQR_C
+#endif
+
+#if defined(BN_MP_SQRMOD_C)
+   #define BN_MP_INIT_C
+   #define BN_MP_SQR_C
+   #define BN_MP_CLEAR_C
+   #define BN_MP_MOD_C
+#endif
+
+#if defined(BN_MP_SQRT_C)
+   #define BN_MP_N_ROOT_C
+   #define BN_MP_ISZERO_C
+   #define BN_MP_ZERO_C
+   #define BN_MP_INIT_COPY_C
+   #define BN_MP_RSHD_C
+   #define BN_MP_DIV_C
+   #define BN_MP_ADD_C
+   #define BN_MP_DIV_2_C
+   #define BN_MP_CMP_MAG_C
+   #define BN_MP_EXCH_C
+   #define BN_MP_CLEAR_C
+#endif
+
+#if defined(BN_MP_SUB_C)
+   #define BN_S_MP_ADD_C
+   #define BN_MP_CMP_MAG_C
+   #define BN_S_MP_SUB_C
+#endif
+
+#if defined(BN_MP_SUB_D_C)
+   #define BN_MP_GROW_C
+   #define BN_MP_ADD_D_C
+   #define BN_MP_CLAMP_C
+#endif
+
+#if defined(BN_MP_SUBMOD_C)
+   #define BN_MP_INIT_C
+   #define BN_MP_SUB_C
+   #define BN_MP_CLEAR_C
+   #define BN_MP_MOD_C
+#endif
+
+#if defined(BN_MP_TO_SIGNED_BIN_C)
+   #define BN_MP_TO_UNSIGNED_BIN_C
+#endif
+
+#if defined(BN_MP_TO_UNSIGNED_BIN_C)
+   #define BN_MP_INIT_COPY_C
+   #define BN_MP_ISZERO_C
+   #define BN_MP_DIV_2D_C
+   #define BN_MP_CLEAR_C
+#endif
+
+#if defined(BN_MP_TOOM_MUL_C)
+   #define BN_MP_INIT_MULTI_C
+   #define BN_MP_MOD_2D_C
+   #define BN_MP_COPY_C
+   #define BN_MP_RSHD_C
+   #define BN_MP_MUL_C
+   #define BN_MP_MUL_2_C
+   #define BN_MP_ADD_C
+   #define BN_MP_SUB_C
+   #define BN_MP_DIV_2_C
+   #define BN_MP_MUL_2D_C
+   #define BN_MP_MUL_D_C
+   #define BN_MP_DIV_3_C
+   #define BN_MP_LSHD_C
+   #define BN_MP_CLEAR_MULTI_C
+#endif
+
+#if defined(BN_MP_TOOM_SQR_C)
+   #define BN_MP_INIT_MULTI_C
+   #define BN_MP_MOD_2D_C
+   #define BN_MP_COPY_C
+   #define BN_MP_RSHD_C
+   #define BN_MP_SQR_C
+   #define BN_MP_MUL_2_C
+   #define BN_MP_ADD_C
+   #define BN_MP_SUB_C
+   #define BN_MP_DIV_2_C
+   #define BN_MP_MUL_2D_C
+   #define BN_MP_MUL_D_C
+   #define BN_MP_DIV_3_C
+   #define BN_MP_LSHD_C
+   #define BN_MP_CLEAR_MULTI_C
+#endif
+
+#if defined(BN_MP_TORADIX_C)
+   #define BN_MP_ISZERO_C
+   #define BN_MP_INIT_COPY_C
+   #define BN_MP_DIV_D_C
+   #define BN_MP_CLEAR_C
+   #define BN_MP_S_RMAP_C
+#endif
+
+#if defined(BN_MP_TORADIX_N_C)
+   #define BN_MP_ISZERO_C
+   #define BN_MP_INIT_COPY_C
+   #define BN_MP_DIV_D_C
+   #define BN_MP_CLEAR_C
+   #define BN_MP_S_RMAP_C
+#endif
+
+#if defined(BN_MP_UNSIGNED_BIN_SIZE_C)
+   #define BN_MP_COUNT_BITS_C
+#endif
+
+#if defined(BN_MP_XOR_C)
+   #define BN_MP_INIT_COPY_C
+   #define BN_MP_CLAMP_C
+   #define BN_MP_EXCH_C
+   #define BN_MP_CLEAR_C
+#endif
+
+#if defined(BN_MP_ZERO_C)
+#endif
+
+#if defined(BN_PRIME_TAB_C)
+#endif
+
+#if defined(BN_REVERSE_C)
+#endif
+
+#if defined(BN_S_MP_ADD_C)
+   #define BN_MP_GROW_C
+   #define BN_MP_CLAMP_C
+#endif
+
+#if defined(BN_S_MP_EXPTMOD_C)
+   #define BN_MP_COUNT_BITS_C
+   #define BN_MP_INIT_C
+   #define BN_MP_CLEAR_C
+   #define BN_MP_REDUCE_SETUP_C
+   #define BN_MP_MOD_C
+   #define BN_MP_COPY_C
+   #define BN_MP_SQR_C
+   #define BN_MP_REDUCE_C
+   #define BN_MP_MUL_C
+   #define BN_MP_SET_C
+   #define BN_MP_EXCH_C
+#endif
+
+#if defined(BN_S_MP_MUL_DIGS_C)
+   #define BN_FAST_S_MP_MUL_DIGS_C
+   #define BN_MP_INIT_SIZE_C
+   #define BN_MP_CLAMP_C
+   #define BN_MP_EXCH_C
+   #define BN_MP_CLEAR_C
+#endif
+
+#if defined(BN_S_MP_MUL_HIGH_DIGS_C)
+   #define BN_FAST_S_MP_MUL_HIGH_DIGS_C
+   #define BN_MP_INIT_SIZE_C
+   #define BN_MP_CLAMP_C
+   #define BN_MP_EXCH_C
+   #define BN_MP_CLEAR_C
+#endif
+
+#if defined(BN_S_MP_SQR_C)
+   #define BN_MP_INIT_SIZE_C
+   #define BN_MP_CLAMP_C
+   #define BN_MP_EXCH_C
+   #define BN_MP_CLEAR_C
+#endif
+
+#if defined(BN_S_MP_SUB_C)
+   #define BN_MP_GROW_C
+   #define BN_MP_CLAMP_C
+#endif
+
+#if defined(BNCORE_C)
+#endif
+
+#ifdef LTM3
+#define LTM_LAST
+#endif
+#include <tommath_superclass.h>
+#include <tommath_class.h>
+#else
+#define LTM_LAST
+#endif
diff --git a/tommath_superclass.h b/tommath_superclass.h
new file mode 100644
index 0000000000000000000000000000000000000000..b50ecb09a8a04997a62852b3c50134c7776d963b
--- /dev/null
+++ b/tommath_superclass.h
@@ -0,0 +1,72 @@
+/* super class file for PK algos */
+
+/* default ... include all MPI */
+#define LTM_ALL
+
+/* RSA only (does not support DH/DSA/ECC) */
+// #define SC_RSA_1
+
+/* For reference.... On an Athlon64 optimizing for speed...
+
+   LTM's mpi.o with all functions [striped] is 142KiB in size.
+
+*/
+
+/* Works for RSA only, mpi.o is 68KiB */
+#ifdef SC_RSA_1
+   #define BN_MP_SHRINK_C
+   #define BN_MP_LCM_C
+   #define BN_MP_PRIME_RANDOM_EX_C
+   #define BN_MP_INVMOD_C
+   #define BN_MP_GCD_C
+   #define BN_MP_MOD_C
+   #define BN_MP_MULMOD_C
+   #define BN_MP_ADDMOD_C
+   #define BN_MP_EXPTMOD_C
+   #define BN_MP_SET_INT_C
+   #define BN_MP_INIT_MULTI_C
+   #define BN_MP_CLEAR_MULTI_C
+   #define BN_MP_UNSIGNED_BIN_SIZE_C
+   #define BN_MP_TO_UNSIGNED_BIN_C
+   #define BN_MP_MOD_D_C
+   #define BN_MP_PRIME_RABIN_MILLER_TRIALS_C
+   #define BN_REVERSE_C
+   #define BN_PRIME_TAB_C
+
+   /* other modifiers */
+   #define BN_MP_DIV_SMALL                    /* Slower division, not critical */
+
+   /* here we are on the last pass so we turn things off.  The functions classes are still there
+    * but we remove them specifically from the build.  This also invokes tweaks in functions
+    * like removing support for even moduli, etc...
+    */
+#ifdef LTM_LAST
+   #undef  BN_MP_TOOM_MUL_C
+   #undef  BN_MP_TOOM_SQR_C
+   #undef  BN_MP_KARATSUBA_MUL_C
+   #undef  BN_MP_KARATSUBA_SQR_C
+   #undef  BN_MP_REDUCE_C
+   #undef  BN_MP_REDUCE_SETUP_C
+   #undef  BN_MP_DR_IS_MODULUS_C
+   #undef  BN_MP_DR_SETUP_C
+   #undef  BN_MP_DR_REDUCE_C
+   #undef  BN_MP_REDUCE_IS_2K_C
+   #undef  BN_MP_REDUCE_2K_SETUP_C
+   #undef  BN_MP_REDUCE_2K_C
+   #undef  BN_S_MP_EXPTMOD_C
+   #undef  BN_MP_DIV_3_C
+   #undef  BN_S_MP_MUL_HIGH_DIGS_C
+   #undef  BN_FAST_S_MP_MUL_HIGH_DIGS_C
+   #undef  BN_FAST_MP_INVMOD_C
+
+   /* To safely undefine these you have to make sure your RSA key won't exceed the Comba threshold
+    * which is roughly 255 digits [7140 bits for 32-bit machines, 15300 bits for 64-bit machines] 
+    * which means roughly speaking you can handle upto 2536-bit RSA keys with these defined without
+    * trouble.  
+    */
+   #undef  BN_S_MP_MUL_DIGS_C
+   #undef  BN_S_MP_SQR_C
+   #undef  BN_MP_MONTGOMERY_REDUCE_C
+#endif
+
+#endif