Created
August 23, 2011 16:57
-
-
Save mmalecki/1165847 to your computer and use it in GitHub Desktop.
memset
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
#include <stdio.h> | |
#include <stdlib.h> | |
#include <time.h> | |
void* memset1(void* mem, int value, size_t num) { | |
char* m = mem; | |
char v = value; | |
while (num-- != 0) | |
*(m + num) = v; | |
return mem; | |
} | |
void* memset2(void* mem, int value, size_t num) { | |
int* m = mem; | |
char c; | |
char* mc; | |
for (; num >= 4; num -= 4) | |
*(m + num) = value; | |
if (num != 0) { | |
c = value; | |
mc = m + num; | |
while (num--) | |
*(mc + num) = 0; | |
} | |
return mem; | |
} | |
void* memset_sse(void* mem, int value, size_t num) { | |
value = value | (value << 8); | |
for (; num >= 16; num -= 16) { | |
asm( | |
"movd %0, %%xmm0 \n\t" | |
"movdqu %%xmm0, 0(%1) \n\t" | |
: : "r" (value), "r" (mem)); | |
} | |
} | |
void* memset_asm(void* ptr, int value, size_t num) { | |
asm("rep stosl;" : : "a"(value | (value << 8) | (value << 16) | (value << 24)), "D" ((size_t) ptr), "c" (num / 4)); | |
asm("rep stosb;" : : "a"(value), "D" (((size_t) ptr) + ((num / 4) * 4)), | |
"c" (num - ((num / 4) * 4))); | |
} | |
int main(int argc, char** argv) { | |
unsigned int i, e = 1024, sz = 1024 * 1024; | |
clock_t start, stop; | |
char* mem = (char*) malloc(sz); | |
start = clock(); | |
for (i = 0; i < e; i++) { | |
memset(mem, 64, sz); | |
} | |
stop = clock(); | |
printf("Standard memset: %f\n", ((double) stop - start) / CLOCKS_PER_SEC); | |
memset(mem, 0, sz); | |
start = clock(); | |
for (i = 0; i < e; i++) { | |
memset1(mem, 61, sz); | |
} | |
stop = clock(); | |
printf("Int memset: %f\n", ((double) stop - start) / CLOCKS_PER_SEC); | |
for (i = 0; i < sz; i++) { | |
if (*(mem + i) != 61) { | |
printf("Failed at %i!\n", i); | |
break; | |
} | |
} | |
memset(mem, 0, sz); | |
start = clock(); | |
for (i = 0; i < e; i++) { | |
memset_asm(mem, 61, sz); | |
} | |
stop = clock(); | |
printf("SSE memset: %f\n", ((double) stop - start) / CLOCKS_PER_SEC); | |
for (i = 0; i < sz; i++) { | |
if (*(mem + i) != 61) { | |
printf("Failed at %i (%x)!\n", i, *(mem + i)); | |
break; | |
} | |
} | |
return 0; | |
} |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment