[Bug c/43052] New: Inline memcmp is *much* slower than glibc's

[bjorn at haxx dot se]

memcmp-intensive code becomes up to 6 times slower if compiled with the -O3
option than with the -g or -O0 option. The reason for this is that the inline
memcmp function is *much* slower than the glibc memcmp.

Here's a simple test case:

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/times.h>

void* list[1024 * 1024];

int main(void)
{
    int count = sizeof(list) / sizeof(char*);
    int i;
    for (i=0; i < count; i++)
        list[i] = calloc(1024, 1);

    int dupes = 0;
    int start = times(NULL);
    for (i=0; i<count-1; i++)
        if (!memcmp(list[i], list[i+1], 1024))
            dupes++;

    int ticks = times(NULL) - start;
    printf("Time: %d ticks (%d memcmp/tick)\n", ticks, dupes/ticks);

    return 0;
}


# gcc -O3 -o test test.c
# ./test
Time: 188 ticks (5577 memcmp/tick)
# gcc -O0 -o test test.c
# ./test
Time: 30 ticks (34952 memcmp/tick)

System is Debian testing with libc package 2.10.1-7.


-- 
           Summary: Inline memcmp is *much* slower than glibc's
           Product: gcc
           Version: 4.4.3
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: bjorn at haxx dot se
 GCC build triplet: i486-linux-gnu
  GCC host triplet: i486-linux-gnu
GCC target triplet: i486-linux-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43052