Hi,
I started digging into the possibility of reducing the number of malloc/free
calls (by replacing small/quick fixed-size allocations with stack arrays),
and I think I found one that makes a substantial difference...
The call to mp_div_2d() in a loop inside the function
mp_to_unsigned_bin()...
If I create a version of mp_init() (call it mp_init_nomalloc() that does
not allocate
and then I put that small array on the stack in mp_div_2d(), I observed
a dramatic
decrease in the number of malloc/free calls (dropped from about 2100 to
about 550) for
a single session that simply connects and then exits.
This is tough to observe dynamically without a debuggable malloc (which
I have);
but its always possible that my diagnostics are off too; hence it would
be nice to get a second set of eyes looking at this to see if my
observation is
correct. If it is, then it would apply to any function that does both the
mp_init() and the matching mp_clear() in the same block of code.
New versions of libtomath/bn_mp_div_2d.c and bn_mp_init.c are attached
if someone has time to double check me.
Hopefully I won't realize my mistake immediately after I send this as I
did before and have to retract my statement!! :-(
Ed
#include <tommath.h>
#ifdef BN_MP_INIT_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, [email protected], http://math.libtomcrypt.com
*/
int mp_init_nomalloc (mp_int * a)
{
int i;
/* 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;
a->alloc = MP_PREC;
a->sign = MP_ZPOS;
return MP_OKAY;
}
/* 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) DB_MALLOC (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;
a->alloc = MP_PREC;
a->sign = MP_ZPOS;
return MP_OKAY;
}
#endif
/* $Source: /cvs/libtom/libtommath/bn_mp_init.c,v $ */
/* $Revision: 1.3 $ */
/* $Date: 2006/03/31 14:18:44 $ */
#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
* 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, [email protected], http://math.libtomcrypt.com
*/
//#define ELIMINATE_MALLOC
#ifdef ELIMINATE_MALLOC
extern int mp_init_nomalloc(mp_int *a);
#define mp_init(a) mp_init_nomalloc(a)
#define mp_clear(a)
#endif
/* 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)
{
mp_digit D, r, rr;
int x, res;
mp_int t;
#ifdef ELIMINATE_MALLOC
mp_digit mpdata[MP_PREC];
t.dp = mpdata;
#endif
/* if the shift count is <= 0 then we do no work */
if (b <= 0) {
res = mp_copy (a, c);
if (d != NULL) {
mp_zero (d);
}
return res;
}
if ((res = mp_init (&t)) != MP_OKAY) {
return res;
}
/* get the remainder */
if (d != NULL) {
if ((res = mp_mod_2d (a, b, &t)) != MP_OKAY) {
mp_clear (&t);
return res;
}
}
/* copy */
if ((res = mp_copy (a, c)) != MP_OKAY) {
mp_clear (&t);
return res;
}
/* shift by as many digits in the bit count */
if (b >= (int)DIGIT_BIT) {
mp_rshd (c, b / DIGIT_BIT);
}
/* shift any bit count < DIGIT_BIT */
D = (mp_digit) (b % DIGIT_BIT);
if (D != 0) {
register mp_digit *tmpc, mask, shift;
/* mask */
mask = (((mp_digit)1) << D) - 1;
/* shift for lsb */
shift = DIGIT_BIT - D;
/* alias */
tmpc = c->dp + (c->used - 1);
/* carry */
r = 0;
for (x = c->used - 1; x >= 0; x--) {
/* get the lower bits of this word in a temp */
rr = *tmpc & mask;
/* shift the current word and mix in the carry bits from the previous word */
*tmpc = (*tmpc >> D) | (r << shift);
--tmpc;
/* set the carry to the carry bits of the current word found above */
r = rr;
}
}
mp_clamp (c);
if (d != NULL) {
mp_exch (&t, d);
}
mp_clear (&t);
return MP_OKAY;
}
#endif
/* $Source: /cvs/libtom/libtommath/bn_mp_div_2d.c,v $ */
/* $Revision: 1.3 $ */
/* $Date: 2006/03/31 14:18:44 $ */
