| 81 | float getRand(void) { return (float)(rand() % 16); } |
| 82 | |
| 83 | bool test0(void) |
| 84 | { |
| 85 | float *h_Data, *h_Kernel, *h_ResultCPU, *h_ResultGPU; |
| 86 | |
| 87 | float *d_Data, *d_PaddedData, *d_Kernel, *d_PaddedKernel; |
| 88 | |
| 89 | fComplex *d_DataSpectrum, *d_KernelSpectrum; |
| 90 | |
| 91 | cufftHandle fftPlanFwd, fftPlanInv; |
| 92 | |
| 93 | bool bRetVal; |
| 94 | StopWatchInterface *hTimer = NULL; |
| 95 | sdkCreateTimer(&hTimer); |
| 96 | |
| 97 | printf("Testing built-in R2C / C2R FFT-based convolution\n"); |
| 98 | const int kernelH = 7; |
| 99 | const int kernelW = 6; |
| 100 | const int kernelY = 3; |
| 101 | const int kernelX = 4; |
| 102 | const int dataH = 2000; |
| 103 | const int dataW = 2000; |
| 104 | const int fftH = snapTransformSize(dataH + kernelH - 1); |
| 105 | const int fftW = snapTransformSize(dataW + kernelW - 1); |
| 106 | |
| 107 | printf("...allocating memory\n"); |
| 108 | h_Data = (float *)malloc(dataH * dataW * sizeof(float)); |
| 109 | h_Kernel = (float *)malloc(kernelH * kernelW * sizeof(float)); |
| 110 | h_ResultCPU = (float *)malloc(dataH * dataW * sizeof(float)); |
| 111 | h_ResultGPU = (float *)malloc(fftH * fftW * sizeof(float)); |
| 112 | |
| 113 | checkCudaErrors(cudaMalloc((void **)&d_Data, dataH * dataW * sizeof(float))); |
| 114 | checkCudaErrors(cudaMalloc((void **)&d_Kernel, kernelH * kernelW * sizeof(float))); |
| 115 | |
| 116 | checkCudaErrors(cudaMalloc((void **)&d_PaddedData, fftH * fftW * sizeof(float))); |
| 117 | checkCudaErrors(cudaMalloc((void **)&d_PaddedKernel, fftH * fftW * sizeof(float))); |
| 118 | |
| 119 | checkCudaErrors(cudaMalloc((void **)&d_DataSpectrum, fftH * (fftW / 2 + 1) * sizeof(fComplex))); |
| 120 | checkCudaErrors(cudaMalloc((void **)&d_KernelSpectrum, fftH * (fftW / 2 + 1) * sizeof(fComplex))); |
| 121 | checkCudaErrors(cudaMemset(d_KernelSpectrum, 0, fftH * (fftW / 2 + 1) * sizeof(fComplex))); |
| 122 | |
| 123 | printf("...generating random input data\n"); |
| 124 | srand(2010); |
| 125 | |
| 126 | for (int i = 0; i < dataH * dataW; i++) { |
| 127 | h_Data[i] = getRand(); |
| 128 | } |
| 129 | |
| 130 | for (int i = 0; i < kernelH * kernelW; i++) { |
| 131 | h_Kernel[i] = getRand(); |
| 132 | } |
| 133 | |
| 134 | printf("...creating R2C & C2R FFT plans for %i x %i\n", fftH, fftW); |
| 135 | checkCudaErrors(cufftPlan2d(&fftPlanFwd, fftH, fftW, CUFFT_R2C)); |
| 136 | checkCudaErrors(cufftPlan2d(&fftPlanInv, fftH, fftW, CUFFT_C2R)); |
| 137 | |
| 138 | printf("...uploading to GPU and padding convolution kernel and input data\n"); |
| 139 | checkCudaErrors(cudaMemcpy(d_Kernel, h_Kernel, kernelH * kernelW * sizeof(float), cudaMemcpyHostToDevice)); |
| 140 | checkCudaErrors(cudaMemcpy(d_Data, h_Data, dataH * dataW * sizeof(float), cudaMemcpyHostToDevice)); |
no test coverage detected