yeppp sin() performance comparison with glibc 2.21

[Virendra Kumar Pathak]

Hi Yeppp Group,

I was comparing sin() performance time between yeppp 1.0.0 & glibc2.21 on x86_64 (haswell).
I saw yeppp was performing better than glibc2.21 even with vector size 1.

Is this expected ? I read somewhere on this group that yeppp's performance will be low if vector/array size is low.
Please shed some insight on this.

Please find below the code which I am using with glibc 2.21 & yeppp

Time for glibc 2.21 = 4.8155 second

Time for yeppp = 1.63625

glibc 2.21

void sin_func(void)
{
        double incr = 0.732;
        double result, count = 0.0;

        while (count < 105414350) {
                result = sin(count);
                count += incr;
        }
}

yeppp

void sin_func(void)
{
        double incr = 0.732; double result, count = 0.0;
        Yep64f x[1], y[1];
        YepSize sz =1;

        while (count < 105414350) {
                x[0] = count; y[0] = 0;
                if (yepMath_Sin_V64f_V64f(&x[0], &y[0], sz) != YepStatusOk) {
                        printf("Wait error \n");
                        return;
                }
                count += incr;
        }
}

Thanks.

with regards
Virendra Kumar Pathak

[Virendra Kumar Pathak]

Hi All,

Just to add, I am using libm from inside glibc2.21.

Thanks.