| 361 | } |
| 362 | |
| 363 | bool test2(void) |
| 364 | { |
| 365 | float *h_Data, *h_Kernel, *h_ResultCPU, *h_ResultGPU; |
| 366 | |
| 367 | float *d_Data, *d_Kernel, *d_PaddedData, *d_PaddedKernel; |
| 368 | |
| 369 | fComplex *d_DataSpectrum0, *d_KernelSpectrum0; |
| 370 | |
| 371 | cufftHandle fftPlan; |
| 372 | |
| 373 | bool bRetVal; |
| 374 | StopWatchInterface *hTimer = NULL; |
| 375 | sdkCreateTimer(&hTimer); |
| 376 | |
| 377 | printf("Testing updated custom R2C / C2R FFT-based convolution\n"); |
| 378 | const int kernelH = 7; |
| 379 | const int kernelW = 6; |
| 380 | const int kernelY = 3; |
| 381 | const int kernelX = 4; |
| 382 | const int dataH = 2000; |
| 383 | const int dataW = 2000; |
| 384 | const int fftH = snapTransformSize(dataH + kernelH - 1); |
| 385 | const int fftW = snapTransformSize(dataW + kernelW - 1); |
| 386 | |
| 387 | printf("...allocating memory\n"); |
| 388 | h_Data = (float *)malloc(dataH * dataW * sizeof(float)); |
| 389 | h_Kernel = (float *)malloc(kernelH * kernelW * sizeof(float)); |
| 390 | h_ResultCPU = (float *)malloc(dataH * dataW * sizeof(float)); |
| 391 | h_ResultGPU = (float *)malloc(fftH * fftW * sizeof(float)); |
| 392 | |
| 393 | checkCudaErrors(cudaMalloc((void **)&d_Data, dataH * dataW * sizeof(float))); |
| 394 | checkCudaErrors(cudaMalloc((void **)&d_Kernel, kernelH * kernelW * sizeof(float))); |
| 395 | |
| 396 | checkCudaErrors(cudaMalloc((void **)&d_PaddedData, fftH * fftW * sizeof(float))); |
| 397 | checkCudaErrors(cudaMalloc((void **)&d_PaddedKernel, fftH * fftW * sizeof(float))); |
| 398 | |
| 399 | checkCudaErrors(cudaMalloc((void **)&d_DataSpectrum0, fftH * (fftW / 2) * sizeof(fComplex))); |
| 400 | checkCudaErrors(cudaMalloc((void **)&d_KernelSpectrum0, fftH * (fftW / 2) * sizeof(fComplex))); |
| 401 | |
| 402 | printf("...generating random input data\n"); |
| 403 | srand(2010); |
| 404 | |
| 405 | for (int i = 0; i < dataH * dataW; i++) { |
| 406 | h_Data[i] = getRand(); |
| 407 | } |
| 408 | |
| 409 | for (int i = 0; i < kernelH * kernelW; i++) { |
| 410 | h_Kernel[i] = getRand(); |
| 411 | } |
| 412 | |
| 413 | printf("...creating C2C FFT plan for %i x %i\n", fftH, fftW / 2); |
| 414 | checkCudaErrors(cufftPlan2d(&fftPlan, fftH, fftW / 2, CUFFT_C2C)); |
| 415 | |
| 416 | printf("...uploading to GPU and padding convolution kernel and input data\n"); |
| 417 | checkCudaErrors(cudaMemcpy(d_Data, h_Data, dataH * dataW * sizeof(float), cudaMemcpyHostToDevice)); |
| 418 | checkCudaErrors(cudaMemcpy(d_Kernel, h_Kernel, kernelH * kernelW * sizeof(float), cudaMemcpyHostToDevice)); |
| 419 | checkCudaErrors(cudaMemset(d_PaddedData, 0, fftH * fftW * sizeof(float))); |
| 420 | checkCudaErrors(cudaMemset(d_PaddedKernel, 0, fftH * fftW * sizeof(float))); |
no test coverage detected