EDIT:
///////////////////////////////
never mind, found the type error, was inconsistency transferring from GPU to CPU and one or two others
///////////////////////////////
Probably a very simple mistake but I’ve just started with cublas so can’t see it. I’m getting the wrong results when I run this code:
#include
#include
#include “cublas_v2.h”
#include <cuda_runtime.h>
using namespace std;
template void printMatrix(int rowCount, int colCount, const T* matrix) {
for (int i=0;i<rowCount;i++) {
for (int j=0;j<colCount;j++) {
cout << matrix[j*colCount+i] << “\t”;
}
cout << endl;
}
}
void gpu_blas_mmul(const float *A, const float *B, float *C, const int m, const int k, const int n) {
int lda=m,ldb=k,ldc=m;
const float alf = 1;
const float bet = 0;
const float *alpha = &alf;
const float *beta = &bet;
cublasHandle_t handle;
cublasCreate(&handle);
cublasSgemm(handle, CUBLAS_OP_N, CUBLAS_OP_N, m, n, k, alpha, A, lda, B, ldb, beta, C, ldc);
cublasDestroy(handle);
}
void gpu_blas_mmul2(const double *A, const double *B, double *C, const int m, const int k, const int n) {
int lda=m,ldb=k,ldc=m;
const double alf = 1, bet = 0;
const double *beta = &bet, *alpha = &alf;
cublasHandle_t handle;
cublasCreate(&handle);
cublasDgemm(handle, CUBLAS_OP_N, CUBLAS_OP_N, m, n, k, alpha, A, lda, B, ldb, beta, C, ldc);
cublasDestroy(handle);
}
int main() {
// Allocate 3 arrays on CPU
int nr_rows_A, nr_cols_A, nr_rows_B, nr_cols_B, nr_rows_C, nr_cols_C;
// Using 2x2 square arrays
nr_rows_A = nr_cols_A = nr_rows_B = nr_cols_B = nr_rows_C = nr_cols_C = 2;
float h_C[4];
// Allocate 3 arrays on GPU
double *d_A, *d_B, *d_C;
cudaMalloc(&d_A,nr_rows_A * nr_cols_A * sizeof(float));
cudaMalloc(&d_B,nr_rows_B * nr_cols_B * sizeof(float));
cudaMalloc(&d_C,nr_rows_C * nr_cols_C * sizeof(float));
// Fill host arrays and move to device
//float h_A[] = {3,1,2,4};
//float h_B[] = {1,0,0,1};
double h_A[] = {3,1,2,4};
double h_B[] = {1,0,0,1};
cout << "Matrix A:\n";
printMatrix(2,2,h_A);
cout << "Matrix B:\n";
printMatrix(2,2,h_B);
cudaMemcpy(d_A,h_A,4*sizeof(float),cudaMemcpyHostToDevice);
cudaMemcpy(d_B,h_B,4*sizeof(float),cudaMemcpyHostToDevice);
// Multiply A and B on GPU
gpu_blas_mmul2(d_A, d_B, d_C, nr_rows_A, nr_cols_A, nr_cols_B);
// Copy (and print) the result on host memory
cudaMemcpy(h_C,d_C,nr_rows_C * nr_cols_C * sizeof(float),cudaMemcpyDeviceToHost);
cout << "Matrix C:\n";
printMatrix(nr_rows_C, nr_cols_C, h_C);
//Free GPU memory
cudaFree(d_A);
cudaFree(d_B);
cudaFree(d_C);
return 0;
}
I compile it with
“nvcc code.cu -lcublas”
and it compiles without error. When I run it, it prints out:
CUBLAS initialization success…
Matrix A:
3 2
1 4
Matrix B:
1 0
0 1
Matrix C:
-nan -nan
-nan -nan
Application completed, cleaning up and exiting…
which wasn’t what I was expecting. Can anyone see what is wrong? It should just be a basic matrix multiplication. If I alter the comments so the matrices are floats instead and use gpu_blas_mmul instead of gpu_blas_mmul2, I get the correct output so it must be something to do with a type error somewhere…
Any help appreciated. Thanks