devs/libstrophe-gh-mirror
2
0
Fork 0
mirror of https://github.com/strophe/libstrophe.git synced 2026-07-17 19:36:22 +00:00
Code Issues Packages Projects Releases Wiki Activity

add option to use clock_gettime() in perf example

Browse Source 
This is done to support architectures where `rdtsc()` is not implemented.
In case it is available use `clock_gettime()`. Otherwise still fall back
to use `clock()`.

The first approach was to always use `clock_gettime()` if it is available
but the overhead it brings compared to `rdtsc()` is high, so I chose to
only try to use it if `rdtsc()` is not implemented.

Signed-off-by: Steffen Jaeckel <jaeckel-floss@eyet-services.de>
This commit is contained in:
Steffen Jaeckel
2022-08-05 12:35:04 +02:00
parent e29f422511
commit 6881a2e770
2 changed files with 38 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
configure.ac
View File

@@ -86,6 +86,7 @@ AX_VALGRIND_DFLT([helgrind], [off])
AX_VALGRIND_DFLT([sgcheck], [off])
AX_VALGRIND_CHECK
AC_SEARCH_LIBS([clock_gettime], [rt], AC_DEFINE(HAVE_CLOCK_GETTIME,1))
AC_SEARCH_LIBS([socket], [network socket])
AC_CHECK_FUNCS([snprintf vsnprintf], [], [have_snprintf=no])

49
examples/perf.c
View File

@@ -20,16 +20,20 @@
#include <stdio.h>
#include <stdint.h>
#include <inttypes.h>
#include <time.h>
static void init_timer(void);
static void t_start(void);
static uint64_t t_read(void);
static const char *perf_time_unit;
#define NUM_SAMPLES 1000u
static void perf_rand(xmpp_ctx_t *ctx)
{
xmpp_rand_t *rng = xmpp_rand_new(ctx);
uint64_t t1, t2;
uint64_t t1, t2, t3 = 0;
unsigned int n;
const size_t alloc_sz = 0x1000u;
size_t sz;
@@ -37,12 +41,12 @@ static void perf_rand(xmpp_ctx_t *ctx)
/* pre-heat */
for (n = 1; n < 4; ++n) {
xmpp_rand_bytes(rng, buf, n * 10);
xmpp_rand_bytes(rng, buf, alloc_sz / n);
}
for (sz = 2; sz <= alloc_sz; sz <<= 1) {
t2 = 0;
for (n = 0; n < 1000u; ++n) {
for (n = 0; n < NUM_SAMPLES; ++n) {
t_start();
t1 = t_read();
@@ -52,10 +56,15 @@ static void perf_rand(xmpp_ctx_t *ctx)
t1 = t_read() - t1;
t2 += t1;
}
t2 /= 1000;
fprintf(stderr,
"Reading %6zu bytes from PRNG took %8" PRIu64 " cycles\n", sz,
t2);
t2 /= NUM_SAMPLES;
fprintf(stderr, "Reading %6zu bytes from PRNG took %8" PRIu64 " %s\n",
sz, t2, perf_time_unit);
if (t3) {
double d3 = (double)t3, d2 = (double)t2;
fprintf(stderr, " +%.2f%%\n",
(d2 - d3) / ((d2 + d3) * 0.5) * 100.);
}
t3 = t2;
}
free(buf);
xmpp_rand_free(ctx, rng);
@@ -90,7 +99,7 @@ static uint64_t rdtsc(void)
unsigned hi, lo;
__asm__ __volatile__("rdtsc" : "=a"(lo), "=d"(hi));
return ((uint64_t)lo) | (((uint64_t)hi) << 32);
#elif defined(LTC_PPC32) || defined(TFM_PPC32)
#elif defined(__POWERPC__)
unsigned long a, b;
__asm__ __volatile__("mftbu %1 \nmftb %0\n" : "=r"(a), "=r"(b));
return (((uint64_t)b) << 32ULL) | ((uint64_t)a);
@@ -116,25 +125,41 @@ static uint64_t rdtsc(void)
uint64_t CNTVCT_EL0;
__asm__ __volatile__("mrs %0, cntvct_el0" : "=r"(CNTVCT_EL0));
return CNTVCT_EL0;
#elif defined HAVE_CLOCK_GETTIME
#define USE_CLOCK_GETTIME
struct timespec result;
clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &result);
return result.tv_sec * 1000000000 + result.tv_nsec;
#else
return clock();
#endif
#endif /* __GNUC__ */
/* Microsoft and Intel Windows compilers */
#elif defined _M_IX86 && !defined(LTC_NO_ASM)
#elif defined _M_IX86
__asm rdtsc
#elif defined _M_AMD64 && !defined(LTC_NO_ASM)
#elif defined _M_AMD64
return __rdtsc();
#elif defined _M_IA64 && !defined(LTC_NO_ASM)
#elif defined _M_IA64
#if defined __INTEL_COMPILER
#include <ia64intrin.h>
#endif
return __getReg(3116);
#elif defined HAVE_CLOCK_GETTIME
#define USE_CLOCK_GETTIME
struct timespec result;
clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &result);
return result.tv_sec * 1000000000 + result.tv_nsec;
#else
return clock();
#endif
}
#if defined USE_CLOCK_GETTIME
static const char *perf_time_unit = "ns";
#else
static const char *perf_time_unit = "cycles";
#endif
static void t_start(void)
{
timer = rdtsc();