How to use the function ‘cblas_dgemm_compute’ of mkl?

lianchen — Mon, 26 Sep 2022 13:44:07 GMT

Are there any examples showing how to use those functions: cblas_dgemm_pack_get_size(),

cblas_dgemm_pack(), cblas_dgemm_compute() ? I would like to realize a specialized GEMM with a packed matrix B. Thanks.

This is my code. I have two questions：

①The function cblas_dgemm_pack_get_size() returns a very large number 7767808 when I want to pack the matrix B whose element type is double and dimension equals to 256 * 256. I think the buffer size needed to store the packed B will not bigger than 256 * 256 * 8 * 2 Bytes.

②The difference between C1 and C2 is larger than 1.0e-6. So I think there is a wrong.

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

{

// matrix parameters

int M, N, K;

int LDA, LDB, LDC;

printf("[INPUT] input M N K\n");

if(scanf("%d %d %d", &M, &N, &K) == 3){

printf("[TRUE] true parameters for scanf\n");

}

else{

printf("[FALSE] false parameters for scanf\n");

exit(EXIT_FAILURE);

}

// matrix buffer, column major

LDA = M, LDB = K, LDC = M;

double *A = NULL,

*B = NULL, *B_PACK = NULL,

*C1 = NULL, *C2 = NULL;

double alpha = 0.000001, beta = 0.000001;

A = (double *) malloc (sizeof(double) * M * K);

B = (double *) malloc (sizeof(double) * K * N);

C1 = (double *) malloc (sizeof(double) * M * N);

C2 = (double *) malloc (sizeof(double) * M * N);

gen_matrix(A, M, K), gen_matrix(B, K, N), gen_matrix(C1, M, N); // initialize matrix A、B、C1

matrix_copy(C1, M, N, C2); // copy the value from C1 to C2

B_PACK = (double *) malloc (cblas_dgemm_pack_get_size(CblasBMatrix, M, N, K));

cblas_dgemm_pack(CblasColMajor, CblasBMatrix, CblasNoTrans, M, N, K, alpha, B, LDB, B_PACK);

cblas_dgemm(CblasColMajor, CblasNoTrans, CblasNoTrans, M, N, K, alpha, A, LDA, B, LDB, beta, C1, LDC);

cblas_dgemm_compute(CblasColMajor, CblasNoTrans, CblasNoTrans, M, N, K, A, LDA, B_PACK , LDB, beta, C2, LDC);

double diff = max_abs_diff(M, N, C1, LDC, C2, LDC);

printf("diff = %lf\n", diff);

return 0;

}

topic How to use the function ‘cblas_dgemm_compute’ of mkl? in Intel® oneAPI Math Kernel Library

How to use the function ‘cblas_dgemm_compute’ of mkl?