Weird performance behaviour

[Marcel Keller <m dot keller at bristol dot ac dot uk>]

Hi,

I've noticed that the performance of mpn_addmul_1 from MPIR can depend 
considerably on whether I link against libpthread, which strikes me as 
very weird:

$ g++ -O3 Time-addmul_1.cpp addmul_1.o -o a.out

$ g++ -O3 Time-addmul_1.cpp addmul_1.o -o b.out  -lpthread

$ ./a.out
mpn_addmul_1: 0.506279

$ ./b.out
mpn_addmul_1: 0.682086

Disassembling the binaries shows that the mpn function in 
Time-addmul_1.cpp is compiled exactly the same way.

I'm running CentOS 7 and GCC 6.2 on a Haswell CPU, but the difference 
also appears on an Ivy Bridge. All inputs and outputs are attached.

Does anyone have any idea why this could be?

Best regards,
Marcel

Attachment: b.out
Description: Binary data

Attachment: a.out
Description: Binary data

#include <mpir.h>
#include <stdlib.h>
#include <time.h>

const int t = 5;
const int n = 1e8;

void mpn(mp_limb_t* zz, mp_limb_t* x, mp_limb_t y)
{
    struct timespec start, stop;
    clock_gettime(CLOCK_REALTIME, &start);
    for (int i = 0; i < n; i++)
        zz[t] = mpn_addmul_1(zz, x, t, y);
    clock_gettime(CLOCK_REALTIME, &stop);
    printf("mpn_addmul_1: %f\n", 1e-9 * (stop.tv_nsec - start.tv_nsec) +
            (stop.tv_sec - start.tv_sec));
}

int main()
{
    mp_limb_t x[t+1], y, z[t+1];
    mpn(z, x, y);
}

Attachment: addmul_1.o
Description: Binary data