gemm_comm.cu

#include <stdio.h>
#include <sys/time.h>

double get_walltime()
{
    struct timeval tp;
    gettimeofday(&tp, NULL);
    return (double)(tp.tv_sec + tp.tv_usec * 1e-6);
}

void matrixSerial(float *hostA, float *hostB, float *hostC, int M, int K, int N)
{
    for (int i = 0; i < M; i++)
    {
        for (int j = 0; j < N; j++)
        {
            float tmp = 0;
            for (int s = 0; s < K; s++)
            {
                tmp += hostA[i * K + s] * hostB[s * N + j];
            }
            hostC[i * N + j] = tmp;
        }
    }
}

void compare(float *hostC, float *serialC, int M, int N)
{
    float error = 0;
    bool tmp = true;
    for (int i = 0; i < M * N; i++)
    {
        error = fmax(error, fabs(hostC[i] - serialC[i]));
        if (error > 1e-2)
        {
            tmp = false;
            printf("error:hostC[%d] = %.3f, serialC[%d] = %.3f\n", i, hostC[i], i, serialC[i]);
            break;
        }
    }
    if (tmp)
    {
        printf("GPU output all right\n");
    }
}