I recently switched from armhf to armv7 and noticed that memcpy on armv7 is much slower than on armhf.

Here is a simple test:

#include <time.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>

#define N 131072

int main()
{
  int i;
  clock_t time[2];
  void *src = malloc(N);
  void *dst = malloc(N);

  time[0] = clock();
  for(i = 0; i < N; ++i)
  {
    memcpy(dst, src, N);
  }
  time[1] = clock();

  printf("time: %ld\n", time[1] - time[0]);

  return 0;
}

and the results:

armhf:~# time ./test
time: 7310658
real    0m 7.31s
user    0m 7.31s
sys     0m 0.00s

armv7:~# time ./test
time: 12973756
real    0m 12.97s
user    0m 12.97s
sys     0m 0.00s

I suspect the source of the problem is the --with-mode=thumb flag on line 235 in gcc/APKBUILD.

Would it be possible to replace --with-mode=thumb with --with-mode=arm?