timing.c

Dropbear is an SSH 2 server, designed to be usable in small memory environments. It supports:

#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;
   }
}

/* RDTSC from Scott Duplichan */
static ulong64 TIMFUNC(void)
{
#if defined __GNUC__
#if defined(__i386__) || defined(__x86_64__)
   unsigned long long a;
   __asm__ __volatile__("rdtsc\nmovl %%eax,%0\nmovl %%edx,4+%0\n"::
			"m"(a):"%eax", "%edx");
   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
}

#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, *logd;
   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);
   goto exptmod;
   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 */
 multtest:
   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;