| 216 | } |
| 217 | |
| 218 | bool test1(void) |
| 219 | { |
| 220 | float *h_Data, *h_Kernel, *h_ResultCPU, *h_ResultGPU; |
| 221 | |
| 222 | float *d_Data, *d_Kernel, *d_PaddedData, *d_PaddedKernel; |
| 223 | |
| 224 | fComplex *d_DataSpectrum0, *d_KernelSpectrum0, *d_DataSpectrum, *d_KernelSpectrum; |
| 225 | |
| 226 | cufftHandle fftPlan; |
| 227 | |
| 228 | bool bRetVal; |
| 229 | StopWatchInterface *hTimer = NULL; |
| 230 | sdkCreateTimer(&hTimer); |
| 231 | |
| 232 | printf("Testing custom R2C / C2R FFT-based convolution\n"); |
| 233 | const uint fftPadding = 16; |
| 234 | const int kernelH = 7; |
| 235 | const int kernelW = 6; |
| 236 | const int kernelY = 3; |
| 237 | const int kernelX = 4; |
| 238 | const int dataH = 2000; |
| 239 | const int dataW = 2000; |
| 240 | const int fftH = snapTransformSize(dataH + kernelH - 1); |
| 241 | const int fftW = snapTransformSize(dataW + kernelW - 1); |
| 242 | |
| 243 | printf("...allocating memory\n"); |
| 244 | h_Data = (float *)malloc(dataH * dataW * sizeof(float)); |
| 245 | h_Kernel = (float *)malloc(kernelH * kernelW * sizeof(float)); |
| 246 | h_ResultCPU = (float *)malloc(dataH * dataW * sizeof(float)); |
| 247 | h_ResultGPU = (float *)malloc(fftH * fftW * sizeof(float)); |
| 248 | |
| 249 | checkCudaErrors(cudaMalloc((void **)&d_Data, dataH * dataW * sizeof(float))); |
| 250 | checkCudaErrors(cudaMalloc((void **)&d_Kernel, kernelH * kernelW * sizeof(float))); |
| 251 | |
| 252 | checkCudaErrors(cudaMalloc((void **)&d_PaddedData, fftH * fftW * sizeof(float))); |
| 253 | checkCudaErrors(cudaMalloc((void **)&d_PaddedKernel, fftH * fftW * sizeof(float))); |
| 254 | |
| 255 | checkCudaErrors(cudaMalloc((void **)&d_DataSpectrum0, fftH * (fftW / 2) * sizeof(fComplex))); |
| 256 | checkCudaErrors(cudaMalloc((void **)&d_KernelSpectrum0, fftH * (fftW / 2) * sizeof(fComplex))); |
| 257 | checkCudaErrors(cudaMalloc((void **)&d_DataSpectrum, fftH * (fftW / 2 + fftPadding) * sizeof(fComplex))); |
| 258 | checkCudaErrors(cudaMalloc((void **)&d_KernelSpectrum, fftH * (fftW / 2 + fftPadding) * sizeof(fComplex))); |
| 259 | |
| 260 | printf("...generating random input data\n"); |
| 261 | srand(2010); |
| 262 | |
| 263 | for (int i = 0; i < dataH * dataW; i++) { |
| 264 | h_Data[i] = getRand(); |
| 265 | } |
| 266 | |
| 267 | for (int i = 0; i < kernelH * kernelW; i++) { |
| 268 | h_Kernel[i] = getRand(); |
| 269 | } |
| 270 | |
| 271 | printf("...creating C2C FFT plan for %i x %i\n", fftH, fftW / 2); |
| 272 | checkCudaErrors(cufftPlan2d(&fftPlan, fftH, fftW / 2, CUFFT_C2C)); |
| 273 | |
| 274 | printf("...uploading to GPU and padding convolution kernel and input data\n"); |
| 275 | checkCudaErrors(cudaMemcpy(d_Data, h_Data, dataH * dataW * sizeof(float), cudaMemcpyHostToDevice)); |
no test coverage detected