| 17 | } |
| 18 | |
| 19 | int main(void) { |
| 20 | cudaError_t cudaStat; |
| 21 | cublasStatus_t stat; |
| 22 | cublasHandle_t handle; |
| 23 | int i, j; |
| 24 | float* devPtrA; |
| 25 | float* a = 0; |
| 26 | a = (float*)malloc(M * N * sizeof(*a)); |
| 27 | if (!a) { |
| 28 | printf("host memory allocation failed"); |
| 29 | return EXIT_FAILURE; |
| 30 | } |
| 31 | for (j = 0; j < N; j++) { |
| 32 | for (i = 0; i < M; i++) { |
| 33 | a[IDX2C(i, j, M)] = (float)(i * N + j + 1); |
| 34 | } |
| 35 | } |
| 36 | cudaStat = cudaMalloc((void**)&devPtrA, M * N * sizeof(*a)); |
| 37 | if (cudaStat != cudaSuccess) { |
| 38 | printf("device memory allocation failed"); |
| 39 | free(a); |
| 40 | return EXIT_FAILURE; |
| 41 | } |
| 42 | stat = cublasCreate(&handle); |
| 43 | if (stat != CUBLAS_STATUS_SUCCESS) { |
| 44 | printf("CUBLAS initialization failed\n"); |
| 45 | free(a); |
| 46 | cudaFree(devPtrA); |
| 47 | return EXIT_FAILURE; |
| 48 | } |
| 49 | stat = cublasSetMatrix(M, N, sizeof(*a), a, M, devPtrA, M); |
| 50 | if (stat != CUBLAS_STATUS_SUCCESS) { |
| 51 | printf("data download failed"); |
| 52 | free(a); |
| 53 | cudaFree(devPtrA); |
| 54 | cublasDestroy(handle); |
| 55 | return EXIT_FAILURE; |
| 56 | } |
| 57 | modify(handle, devPtrA, M, N, 1, 2, 16.0f, 12.0f); |
| 58 | stat = cublasGetMatrix(M, N, sizeof(*a), devPtrA, M, a, M); |
| 59 | if (stat != CUBLAS_STATUS_SUCCESS) { |
| 60 | printf("data upload failed"); |
| 61 | free(a); |
| 62 | cudaFree(devPtrA); |
| 63 | cublasDestroy(handle); |
| 64 | return EXIT_FAILURE; |
| 65 | } |
| 66 | cudaFree(devPtrA); |
| 67 | cublasDestroy(handle); |
| 68 | for (j = 0; j < N; j++) { |
| 69 | for (i = 0; i < M; i++) { |
| 70 | printf("%7.0f", a[IDX2C(i, j, M)]); |
| 71 | } |
| 72 | printf("\n"); |
| 73 | } |
| 74 | free(a); |
| 75 | return EXIT_SUCCESS; |
| 76 | } |