@lvc would you mind adding tomsfastmath to your tool please, including develop?

ifndef SIGCLD
define SIGCLD SIGCHLD
endif
solved

2.src/mont/fp_montgomery_reduce.c:514:15: error: unknown register name '%cc' in
asm
INNERMUL8;

We should also have a TFM implementation of the sqrtmod_prime() function as done for LTM in libtom/libtommath#33

I'm trying to package Tomsfastmath for Homebrew, but the latest release (0.13.1) was published in April 2017, and requires patching in the Homebrew formula to build. There have also been a lot of changes since Tomsfastmath 0.13.1, in addition to a license change. Therefore, it would be great if you would publish a new release.

One thing we found productive is TFM can be fairly easily ported to be variable precision. This in turn saves a lot of memory at a small (trivial) cost in performance.

I can't contribute code (right now) but it's something that is worth looking into. It would also make FP_MAX_SIZE obsolete.

If neither ENDIAN_BIG nor ENDIAN_LITTLE is defined, the following code is executed in fp_read_unsigned_bin

#else
  /* read the bytes in */
  for (; c > 0; c--) {
     fp_mul_2d (a, 8, a);
     a->dp[0] |= *b++;
     a->used += 1;
  }
#endif

This will overflow a->dp array if c is large enough. I implemented a local change by having fp_read_unsigned_bin return a failure status in such case but I see now that the code above seems to truncate the input.

LibTomCrypt and LibTomMath come with mingw makefiles, thanks for that. Please add this makefile also for TomsFastMath.

Hi Tom,
I am trying to perform left and right shift x digits via the functions fp_lshd and fp_rshd:

  fp_int a;

  fp_set(&a, 60); fp_print("", "\n", a, 10);
  fp_lshd(&a, 1); fp_print("", "\n", a, 10);

  fp_set(&a, 60); fp_print("", "\n", a, 10);
  fp_rshd(&a, 1); fp_print("", "\n", a, 10);

  printf("%d\n", (60 << 1));
  printf("%d\n", (60 >> 1));

The output is:

60
1106804644422573096960
60
0
120
30

fp_print defined as follows:

void fp_print(const char *start, const char *end, const fp_int u, int base)
{
  int size_u;
  fp_radix_size((fp_int *)&u, base, &size_u);
  char *buf_u = malloc(size_u);
  fp_toradix_n((fp_int *)&u, buf_u, base, size_u);
  printf("%s%s%s", start, buf_u, end);
}

Am I wrong using it?

Obviously my use case is for larger numbers

Hi Tom

fp_iseven() appears a little incorrect to me, and it should be the following IMO:

diff --git a/src/headers/tfm.h b/src/headers/tfm.h
index 0d1dbf1..6f00c08 100644
--- a/src/headers/tfm.h
+++ b/src/headers/tfm.h
@@ -335,7 +335,7 @@ const char *fp_ident(void);
 
 /* zero/even/odd ? */
 #define fp_iszero(a) (((a)->used == 0) ? FP_YES : FP_NO)
-#define fp_iseven(a) (((a)->used >= 0 && (((a)->dp[0] & 1) == 0)) ? FP_YES : FP_NO)
+#define fp_iseven(a) (((a)->used == 0 || (((a)->dp[0] & 1) == 0)) ? FP_YES : FP_NO)
 #define fp_isodd(a)  (((a)->used > 0  && (((a)->dp[0] & 1) == 1)) ? FP_YES : FP_NO)
 
 /* set to a small digit */

Setting M to "2 ^ 4423 - 1", then "exptmod(3,M,M)" should give '3'.

The current code returns 1 from fp_exptmod, presumably because the modulus is so large.

#include "stdio.h"
#include "tfm.h"

void s_mp_sqr(fp_int *x)
{
	fp_int tmp;
	fp_init(&tmp);
	fp_sqr(x,&tmp);
	fp_copy(&tmp, x);
}

void fp_expt(fp_int *a, fp_int *b, fp_int *c)
{
	fp_int   s, x;
	fp_digit d;
	int      dig, bit;


	if (b->sign == FP_NEG)
		return ;

	fp_init(&s);

	fp_set(&s, 1);

	fp_init_copy(&x, a);

	/* Loop over low-order digits in ascending order */
	for(dig = 0; dig < (b->used - 1); dig++) {
		d = b->dp[dig];

		/* Loop over bits of each non-maximal digit */
		fp_int tmp;
		for(bit = 0; bit < DIGIT_BIT; bit++) {
			if(d & 1) {
				s.sign = FP_ZPOS;
				x.sign = FP_ZPOS;
				fp_init(&tmp);
				fp_mul(&s, &x, &tmp);
				fp_copy(&tmp, &s);
			}

			d >>= 1;

			s_mp_sqr(&x);
		}
	}

	/* Consider now the last digit... */
	d = b->dp[dig];

	fp_int tmp;
	while(d) {
		if(d & 1) {
			s.sign = x.sign = FP_ZPOS;
			fp_init(&tmp);
			fp_mul(&s, &x, &tmp);
			fp_copy(&tmp, &s);
		}

		d >>= 1;

		s_mp_sqr(&x);
	}

	if(fp_iseven(b))
		s.sign = FP_ZPOS;
	else
		s.sign = a->sign;

	fp_copy(&s, c);

} /* end fp_expt() */

int main(int argc, char ** argv )
{
	fp_int a,b,c,d;
	fp_init(&a);
	fp_init(&b);
	fp_init(&c);
	fp_init(&d);

	fp_set(&a, 3);
	fp_set(&b, 2);
	fp_set(&c, 4423);
	fp_expt(&b, &c, &d);
	fp_copy(&d, &b);

	fp_init(&d);
	int rc = fp_exptmod(&a, &b, &b, &d);

	int size=0;
	fp_radix_size(&d, 10, &size);
	char * outstr = (char *) malloc(size+1);
	fp_toradix(&d, outstr, 10);
	outstr[size] = '\0';
	printf("rc = %d, result (should be 3): %s\n", rc, outstr);
	return 0;
}

fp_sqr does not handle case when parameter a = 00....00

When a=00.....00 the result b is not updated.

Adding the following code after y = A->used; seems to resolve this
if (y==0) {
B->used = 0;
goto clean;
}

imo everybody intending to do any real-life work with tomsfastmath will require this function to import his binary private keys etc. it should be mentioned right after fp_init() and fp_set(). i spent about half an hour to search for this functionality in both tfm pdf and ltm pdf without success, and only found out by digging through the ltc rsa code.