finding executed time using PGI_ACC_TIME

Accelerated Computing HPC Compilers Legacy PGI Compilers
1

Hi, I’m new to open ACC.

For testing open ACC compiler well and how fast it makes a program,

I compiled with pcgcc -acc -o test03.exe test03.c -ta=tesla:cc1x -Minfo=accel, and executed it.

But when varying n from 1 to 10^6, the time measured by using PGI_ACC_TIME doesn’t satisfy me.

I heard when n is small number, there must be some overhead for sending and getting datas from host to device, and versus.

Approximately, the time varys 41 at 1 times, 41 at 10 times, 41 at 100 times, 82 at 1000 times, about 400 at 10000 times, about 4000 at 100000 times.

Any problems in my code? or any advices for me?

I really really appreciate to your very kind reply, in advance :).



#include <stdio.h>
#include <stdlib.h>
#include <assert.h>
#include "stdafx.h"
#include <windows.h>

int main( int argc, char* argv[] )
{

    int n;      /* size of the vector */
    float *a;  /* the vector */
    float *restrict r;  /* the results */
    float *e;  /* expected results */
    int i;

    if( argc > 1 )
        n = atoi( argv[1] );
    else 
        n = 1000;
    if( n <= 0 ) n = 1000;

    a = (float*)malloc(n*sizeof(float));
    r = (float*)malloc(n*sizeof(float));
    e = (float*)malloc(n*sizeof(float));
    /* initialize */
    for( i = 0; i < n; ++i ) a[i] = (float)(i+1);

	printf("start!!\n");

#pragma acc kernels loop
    for( i = 0; i < n; ++i ) r[i] = a[i]*2.0f;
    /* compute on the host to compare */
    for( i = 0; i < n; ++i ) e[i] = a[i]*2.0f;
    /* check the results */
    for( i = 0; i < n; ++i )
	{
        assert( r[i] == e[i] );
	
	}

	printf("end!!\n");
	printf( "%d iterations completed\n", n );

	
    return 0;
}
2

Hi Tae Hyeok, Jang,

Note that there’s very little compute here so it’s not a great example to show performance. There is some overhead but the times you’re seeing are having more to do with GPU saturation. It looks to take approximately 41ms per block so once you saturate the GPU, the remaining blocks still need to run, hence the 82 ms. If you graph the times, you’d most likely see a stair step.

  • Mat