commit 95cec6a5ecfdc2a75ba041a6e9f9cd568d53a6fc
parent 37740e679273dbf76593a8da5d44772b94f07b28
Author: Mattias Andrée <maandree@kth.se>
Date: Wed, 27 Apr 2016 21:39:19 +0200
More accurate benchmarking
Signed-off-by: Mattias Andrée <maandree@kth.se>
Diffstat:
5 files changed, 158 insertions(+), 80 deletions(-)
diff --git a/Makefile b/Makefile
@@ -130,15 +130,15 @@ test-random.c: test-generate.py
test: test.c libzahl.a test-random.c
$(CC) $(LDFLAGS) $(CFLAGS_WITHOUT_O) -O0 $(CPPFLAGS) -o $@ test.c libzahl.a
-benchmark: bench/benchmark.c $(BENCHMARK_DEP_$(BENCHMARK_LIB))
+benchmark: bench/benchmark.c bench/benchmark.h $(BENCHMARK_DEP_$(BENCHMARK_LIB))
$(CC) $(LDFLAGS) $(CFLAGS) $(CPPFLAGS) -o $@ bench/benchmark.c \
$(BENCHMARK_LIB_$(BENCHMARK_LIB)) $(BENCHMARK_C_$(BENCHMARK_LIB))
-benchmark-func: bench/benchmark-func.c $(BENCHMARK_DEP_$(BENCHMARK_LIB))
+benchmark-func: bench/benchmark-func.c bench/benchmark.h $(BENCHMARK_DEP_$(BENCHMARK_LIB))
$(CC) $(LDFLAGS) $(CFLAGS) $(CPPFLAGS) -o $@ bench/benchmark-func.c \
$(BENCHMARK_LIB_$(BENCHMARK_LIB)) $(BENCHMARK_C_$(BENCHMARK_LIB))
-benchmark-zrand: bench/benchmark-zrand.c libzahl.a
+benchmark-zrand: bench/benchmark-zrand.c bench/benchmark.h libzahl.a
$(CC) $(LDFLAGS) $(CFLAGS) $(CPPFLAGS) -o $@ $^
check: test
diff --git a/bench/benchmark-func.c b/bench/benchmark-func.c
@@ -1,16 +1,7 @@
+#include "benchmark.h"
+
#include <limits.h>
-#include <stdio.h>
-#include <stdlib.h>
-#include <string.h>
-#include <time.h>
-#include <unistd.h>
-#ifdef BENCHMARK_LIB
-# include BENCHMARK_LIB
-#else
-# include "../zahl.h"
-# define BIGINT_LIBRARY "libzahl"
-#endif
enum {
HIGH_ONLY,
@@ -34,24 +25,10 @@ struct function {
size_t measurements;
};
-#if defined(__x86_64__)
-# undef clock_t
-# define clock_t unsigned long long int
-static inline clock_t rdtsc(void)
-{
- unsigned int low, high;
- __asm__ __volatile__ ("rdtsc" : "=a"(low), "=d"(high));
- return (clock_t)low | (((clock_t)high) << 32);
-}
-#else
-# define rdtsc clock
-#endif
-
#define M_MAX 200
static char buf[1000];
static z_t temp, temp2;
-static clock_t start, end;
static unsigned long long int measurements[M_MAX];
#if 1
@@ -124,12 +101,12 @@ gettime(size_t m)
INSTRUCTION;\
INSTRUCTION;\
j = f->runs;\
- start = rdtsc();\
+ TIC;\
while (j--) {\
INSTRUCTION;\
}\
- end = rdtsc();\
- measurements[k] = (unsigned long long int)(end - start);\
+ TOC;\
+ measurements[k] = TICKS;\
}\
printf("%llu\n", gettime(f->measurements));\
a++;\
@@ -301,6 +278,8 @@ main(int argc, char *argv[])
return 2;
}
+ benchmark_init();
+
if (setjmp(jmp)) {
zperror(argv[0]);
return 1;
diff --git a/bench/benchmark-zrand.c b/bench/benchmark-zrand.c
@@ -1,43 +1,28 @@
-#include <time.h>
-#include <stdio.h>
-
-#include "../zahl.h"
-
-
-#ifndef CLOCK_MONOTONIC_RAW
-# define CLOCK_MONOTONIC_RAW CLOCK_MONOTONIC
-#endif
+#include "benchmark.h"
#define BENCHMARK(INSTRUCTION, FAST)\
do {\
i = FAST ? 1000000L : 1000L;\
- clock_gettime(CLOCK_MONOTONIC_RAW, &start);\
+ TIC;\
while (i--) {\
INSTRUCTION;\
}\
- clock_gettime(CLOCK_MONOTONIC_RAW, &end);\
- end.tv_sec -= start.tv_sec;\
- end.tv_nsec -= start.tv_nsec;\
- if (end.tv_nsec < 0) {\
- end.tv_nsec += 1000000000L;\
- end.tv_sec -= 1;\
- }\
- printf("%s: %lli.%09li %s\n",\
- #INSTRUCTION,\
- (long long int)(end.tv_sec), end.tv_nsec,\
- FAST ? "µs" : "ms");\
+ TOC;\
+ printf("%s: %s %s\n",\
+ #INSTRUCTION, STIME, FAST ? "µs" : "ms");\
} while (0)
int
main(int argc, char *argv[])
{
- struct timespec start, end;
z_t r, n;
jmp_buf jmp;
size_t i;
+ benchmark_init();
+
if (setjmp(jmp)) {
zperror(argv[0]);
return 1;
diff --git a/bench/benchmark.c b/bench/benchmark.c
@@ -1,49 +1,29 @@
-#include <time.h>
-#include <stdio.h>
-
-#ifdef BENCHMARK_LIB
-# include BENCHMARK_LIB
-#else
-# include "../zahl.h"
-# define BIGINT_LIBRARY "libzahl"
-#endif
-
-
-#ifndef CLOCK_MONOTONIC_RAW
-# define CLOCK_MONOTONIC_RAW CLOCK_MONOTONIC
-#endif
+#include "benchmark.h"
#define BENCHMARK(INSTRUCTION, FAST)\
do {\
i = FAST ? 1000000L : 1000L;\
- clock_gettime(CLOCK_MONOTONIC_RAW, &start);\
+ TIC;\
while (i--) {\
(void)INSTRUCTION;\
}\
- clock_gettime(CLOCK_MONOTONIC_RAW, &end);\
- end.tv_sec -= start.tv_sec;\
- end.tv_nsec -= start.tv_nsec;\
- if (end.tv_nsec < 0) {\
- end.tv_nsec += 1000000000L;\
- end.tv_sec -= 1;\
- }\
- printf("%s: %lli.%09li %s (152 bits)\n",\
- #INSTRUCTION,\
- (long long int)(end.tv_sec), end.tv_nsec,\
- FAST ? "µs" : "ms");\
+ TOC;\
+ printf("%s: %s %s (152 bits)\n",\
+ #INSTRUCTION, STIME, FAST ? "µs" : "ms");\
} while (0)
int
main(int argc, char *argv[])
{
- struct timespec start, end;
char buf[1000];
z_t a, b, c, d, tiny;
jmp_buf jmp;
size_t i;
+ benchmark_init();
+
if (setjmp(jmp)) {
zperror(argv[0]);
return 1;
diff --git a/bench/benchmark.h b/bench/benchmark.h
@@ -0,0 +1,134 @@
+#if defined(__linux__)
+# define _GNU_SOURCE
+# include <sched.h>
+#endif
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <time.h>
+#include <unistd.h>
+
+
+#ifdef BENCHMARK_LIB
+# include BENCHMARK_LIB
+#else
+# include "../zahl.h"
+# define BIGINT_LIBRARY "libzahl"
+#endif
+
+
+#ifndef CLOCK_MONOTONIC_RAW
+# define CLOCK_MONOTONIC_RAW CLOCK_MONOTONIC
+#endif
+
+
+#ifdef __x86_64__
+# define RDTSC_MAYBE_SUPPORTED
+#endif
+
+
+#if !defined(USE_RDTSC) && !defined(USE_CLOCK) && !defined(USE_GETTIME)
+# if 1 && defined(RDTSC_MAYBE_SUPPORTED) && defined(__linux__)
+# define USE_RDTSC
+# elif 1
+# define USE_CLOCK
+# else
+# define USE_GETTIME
+# endif
+#endif
+
+
+static struct timespec dur;
+static char timebuf[512];
+#if defined(USE_RDTSC)
+typedef unsigned long long int rdtsc_t;
+static unsigned int start_high, start_low, end_high, end_low;
+static unsigned long long int freq;
+#elif defined(USE_CLOCK)
+static clock_t start, end;
+#else
+static struct timespec start;
+#endif
+
+
+static void
+benchmark_init(void)
+{
+#if defined(__linux__)
+ cpu_set_t cpuset;
+# if defined(USE_RDTSC)
+ FILE *f;
+ char *line = 0;
+ size_t size = 0;
+# endif
+ CPU_ZERO(&cpuset);
+ CPU_SET(0, &cpuset);
+ sched_setaffinity(getpid(), sizeof(cpuset), &cpuset);
+# if defined(USE_RDTSC)
+ f = fopen("/sys/devices/system/cpu/cpu0/cpufreq/cpuinfo_max_freq", "r");
+ if (getline(&line, &size, f) < 0)
+ abort();
+ fclose(f);
+ freq = strtoull(line, 0, 10);
+ free(line);
+# endif
+#endif
+ (void) timebuf;
+}
+
+
+#if defined(USE_RDTSC) && defined(__x86_64__)
+static inline void
+rdtsc(unsigned int *low, unsigned int *high)
+{
+ __asm__ __volatile__ ("rdtsc" : "=a"(*low), "=d"(*high));
+}
+static inline rdtsc_t
+rdtsc_join(unsigned int low, unsigned int high)
+{
+ return (rdtsc_t)low | (((rdtsc_t)high) << 32);
+}
+#endif
+
+
+#if defined(USE_RDTSC)
+# define TIC (rdtsc(&start_low, &start_high))
+# define TOC\
+ do {\
+ rdtsc_t dur_cycles;\
+ double dur_seconds;\
+ rdtsc(&end_low, &end_high);\
+ dur_cycles = rdtsc_join(end_low, end_high);\
+ dur_cycles -= rdtsc_join(start_low, start_high);\
+ dur_seconds = dur_cycles;\
+ dur_seconds /= freq;\
+ dur_seconds /= 1000;\
+ dur_seconds -= dur.tv_sec = (int)dur_seconds;\
+ dur.tv_nsec = dur_seconds * 1000000000L;\
+ } while (0)
+#elif defined(USE_CLOCK)
+# define TIC (start = clock())
+# define TOC\
+ do {\
+ end = clock();\
+ dur.tv_sec = (end - start) / 1000000ULL;\
+ dur.tv_nsec = ((end - start) % 1000000ULL) * 1000;\
+ } while (0)
+#elif defined(USE_GETTIME)
+# define TIC clock_gettime(CLOCK_MONOTONIC_RAW, &start)
+# define TOC\
+ do {\
+ clock_gettime(CLOCK_MONOTONIC_RAW, &dur);\
+ dur.tv_sec -= start.tv_sec;\
+ dur.tv_nsec -= start.tv_nsec;\
+ if (dur.tv_nsec < 0) {\
+ dur.tv_nsec += 1000000000L;\
+ dur.tv_sec -= 1;\
+ }\
+ } while (0)
+#endif
+
+
+#define TICKS (dur.tv_sec * 1000000000ULL + dur.tv_nsec)
+#define STIME (sprintf(timebuf, "%lli.%09li", (long long)(dur.tv_sec), dur.tv_nsec), timebuf)
