Created
January 19, 2015 22:36
-
-
Save pitrou/892219a7d4c6d37de201 to your computer and use it in GitHub Desktop.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
// gcc avx.c -mavx -std=c99 -O2 | |
// while true; do ./a.out; done > log | |
// analyze via np.genfromtxt, e.g. min, 10th percentile, median | |
#include <stdio.h> | |
#include <string.h> | |
#include <x86intrin.h> | |
#include <avxintrin.h> | |
#define N 5000 | |
/* Aligned AVX */ | |
void __attribute__((noinline)) add(double * a, int warmup) | |
{ | |
int dump; | |
long long ts = __rdtscp(&dump); | |
for (size_t i = 0; i < N; i+=4) { | |
__m256d av = _mm256_load_pd(&a[i]); | |
av = _mm256_add_pd(av, av); | |
_mm256_store_pd(&a[i], av); | |
} | |
if (!warmup) | |
printf("%lld ", __rdtscp(&dump) - ts); | |
} | |
/* Unaligned AVX */ | |
void __attribute__((noinline)) add2(double * a, int warmup) | |
{ | |
int dump; | |
long long ts = __rdtscp(&dump); | |
for (size_t i = 0; i < N; i+=4) { | |
__m256d av = _mm256_loadu_pd(&a[i]); | |
av = _mm256_add_pd(av, av); | |
_mm256_storeu_pd(&a[i], av); | |
} | |
if (!warmup) | |
printf("%lld ", __rdtscp(&dump) - ts); | |
} | |
/* Aligned SSE2 */ | |
void __attribute__((noinline)) add3(double * a, int warmup) | |
{ | |
int dump; | |
long long ts = __rdtscp(&dump); | |
for (size_t i = 0; i < N; i+=2) { | |
__m128d av = _mm_load_pd(&a[i]); | |
av = _mm_add_pd(av, av); | |
_mm_store_pd(&a[i], av); | |
} | |
if (!warmup) | |
printf("%lld ", __rdtscp(&dump) - ts); | |
} | |
/* Scalar */ | |
void __attribute__((noinline)) add9(double * a, int warmup) | |
{ | |
int dump; | |
long long ts = __rdtscp(&dump); | |
for (size_t i = 0; i < N; i+=1) { | |
a[i] = a[i] + a[i]; | |
} | |
if (!warmup) | |
printf("%lld ", __rdtscp(&dump) - ts); | |
} | |
int main(int argc, const char *argv[]) | |
{ | |
int i; | |
double * a = _mm_malloc(N *8 + 10, 32); | |
memset(a, 0, N * 8 + 10); | |
for (i=0; i<10; i++) | |
add(a, 1); | |
add(a, 0); | |
add2(a + 2, 1); | |
add2(a + 2, 0); | |
add2(a + 1, 1); | |
add2(a + 1, 0); | |
add3(a, 1); | |
add3(a, 0); | |
add9(a, 1); | |
add9(a, 0); | |
puts(""); | |
return 0; | |
} |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment