gemmManagedOutOfCore.cu

#include <stdlib.h>
#include <stdio.h>
#include <cuda_runtime.h>
#include <cublas.h>
#include "common.hh"


int
main(int argc, char *argv[])
{
    size_t N = 16000;
    clock_t start, end; 
    cublasHandle_t handle;
    float *a, *b, *c;
    const float alpha = 1;
    const float beta = 0;
    size_t count, nn;

    if (argc == 2) {
        N = checked_strtosize(argv[1]);
    }
    nn = checked_mul(N, N);
    count = checked_mul(nn, sizeof(float));

    check(cublasCreate(&handle));

    start = clock();
    check(cudaMallocManaged(&a, count));
    check(cudaMallocManaged(&b, count));
    check(cudaMallocManaged(&c, count));

    for (size_t i = 0; i < N*N; i++) {
        a[i] = i / 37.0;
        b[i] = i / 101.0;
    }
    end = clock();
    log("host: MallocManaged+init", start, end);

    start = clock();
    check(cublasSgemm(handle, CUBLAS_OP_N, CUBLAS_OP_N,
                      N, N, N,
                      &alpha,
                      a, N,
                      b, N,
                      &beta,
                      c, N));
    check(cudaDeviceSynchronize());
    end = clock();
    log("cublasSgemm", start, end);

    return 0;
}