@@ -276,31 +276,31 @@ VkFFTResult sample_11_precision_VkFFT_single(VkGPU* vkGPU, uint64_t file_output,
276276						uint64_t  loc_j = j;
277277						uint64_t  loc_l = l;
278278
279- 						// if (file_output) fprintf(output, "%f %f  - %f %f  \n", output_FFTW[i + j * dims[0] + l * dims[0] * dims[1]][0] / N, output_FFTW[i + j * dims[0] + l * dims[0] * dims[1]][1] / N, output_VkFFT[(loc_i + loc_j * dims[0] + loc_l * dims[0] * dims[1])][0], output_VkFFT[(loc_i + loc_j * dims[0] + loc_l * dims[0] * dims[1])][1]);
279+ 						// if (file_output) fprintf(output, "%.2e %.2e  - %.2e %.2e  \n", output_FFTW[i + j * dims[0] + l * dims[0] * dims[1]][0] / N, output_FFTW[i + j * dims[0] + l * dims[0] * dims[1]][1] / N, output_VkFFT[(loc_i + loc_j * dims[0] + loc_l * dims[0] * dims[1])][0], output_VkFFT[(loc_i + loc_j * dims[0] + loc_l * dims[0] * dims[1])][1]);
280280
281- 						// printf("%f %f  - %f %f  \n", output_FFTW[i + j * dims[0] + l * dims[0] * dims[1]][0], output_FFTW[i + j * dims[0] + l * dims[0] * dims[1]][1], output_VkFFT[(loc_i + loc_j * dims[0] + loc_l * dims[0] * dims[1])][0], output_VkFFT[(loc_i + loc_j * dims[0] + loc_l * dims[0] * dims[1])][1]);
281+ 						// printf("%.2e %.2e  - %.2e %.2e  \n", output_FFTW[i + j * dims[0] + l * dims[0] * dims[1]][0], output_FFTW[i + j * dims[0] + l * dims[0] * dims[1]][1], output_VkFFT[(loc_i + loc_j * dims[0] + loc_l * dims[0] * dims[1])][0], output_VkFFT[(loc_i + loc_j * dims[0] + loc_l * dims[0] * dims[1])][1]);
282282						double  current_data_norm = sqrt (output_FFTW[i + j * dims[0 ] + l * dims[0 ] * dims[1 ]][0 ] * output_FFTW[i + j * dims[0 ] + l * dims[0 ] * dims[1 ]][0 ] + output_FFTW[i + j * dims[0 ] + l * dims[0 ] * dims[1 ]][1 ] * output_FFTW[i + j * dims[0 ] + l * dims[0 ] * dims[1 ]][1 ]);
283283#if  defined(USE_cuFFT) || defined(USE_rocFFT)
284284						double  current_diff_x_extFFT = (output_extFFT[loc_i + loc_j * dims[0 ] + loc_l * dims[0 ] * dims[1 ]][0 ] - output_FFTW[i + j * dims[0 ] + l * dims[0 ] * dims[1 ]][0 ]);
285285						double  current_diff_y_extFFT = (output_extFFT[loc_i + loc_j * dims[0 ] + loc_l * dims[0 ] * dims[1 ]][1 ] - output_FFTW[i + j * dims[0 ] + l * dims[0 ] * dims[1 ]][1 ]);
286286						double  current_diff_norm_extFFT = sqrt (current_diff_x_extFFT * current_diff_x_extFFT + current_diff_y_extFFT * current_diff_y_extFFT);
287287						if  (current_diff_norm_extFFT > max_difference[0 ]) max_difference[0 ] = current_diff_norm_extFFT;
288288						avg_difference[0 ] += current_diff_norm_extFFT;
289- 						if  ((current_diff_norm_extFFT / current_data_norm > max_eps[0 ]) && (current_data_norm >  1e-10 ) ) {
289+ 						if  ((current_diff_norm_extFFT / current_data_norm > max_eps[0 ])) {
290290							max_eps[0 ] = current_diff_norm_extFFT / current_data_norm;
291291						}
292- 						avg_eps[0 ] += (current_data_norm >  1e-10 ) ?  current_diff_norm_extFFT / current_data_norm :  0 ;
292+ 						avg_eps[0 ] += current_diff_norm_extFFT / current_data_norm;
293293#endif 
294294
295295						double  current_diff_x_VkFFT = (output_VkFFT[loc_i + loc_j * dims[0 ] + loc_l * dims[0 ] * dims[1 ]][0 ] - output_FFTW[i + j * dims[0 ] + l * dims[0 ] * dims[1 ]][0 ]);
296296						double  current_diff_y_VkFFT = (output_VkFFT[loc_i + loc_j * dims[0 ] + loc_l * dims[0 ] * dims[1 ]][1 ] - output_FFTW[i + j * dims[0 ] + l * dims[0 ] * dims[1 ]][1 ]);
297297						double  current_diff_norm_VkFFT = sqrt (current_diff_x_VkFFT * current_diff_x_VkFFT + current_diff_y_VkFFT * current_diff_y_VkFFT);
298298						if  (current_diff_norm_VkFFT > max_difference[1 ]) max_difference[1 ] = current_diff_norm_VkFFT;
299299						avg_difference[1 ] += current_diff_norm_VkFFT;
300- 						if  ((current_diff_norm_VkFFT / current_data_norm > max_eps[1 ]) && (current_data_norm >  1e-10 ) ) {
300+ 						if  ((current_diff_norm_VkFFT / current_data_norm > max_eps[1 ])) {
301301							max_eps[1 ] = current_diff_norm_VkFFT / current_data_norm;
302302						}
303- 						avg_eps[1 ] += (current_data_norm >  1e-10 ) ?  current_diff_norm_VkFFT / current_data_norm :  0 ;
303+ 						avg_eps[1 ] += current_diff_norm_VkFFT / current_data_norm;
304304					}
305305				}
306306			}
@@ -310,17 +310,17 @@ VkFFTResult sample_11_precision_VkFFT_single(VkGPU* vkGPU, uint64_t file_output,
310310			avg_eps[1 ] /= (dims[0 ] * dims[1 ] * dims[2 ]);
311311#ifdef  USE_cuFFT
312312			if  (file_output)
313- 				fprintf (output, " cuFFT System: %"   PRIu64 " x%"   PRIu64 " x%"   PRIu64 "  avg_difference: %f  max_difference: %f  avg_eps: %f  max_eps: %f \n "  , dims[0 ], dims[1 ], dims[2 ], avg_difference[0 ], max_difference[0 ], avg_eps[0 ], max_eps[0 ]);
314- 			printf (" cuFFT System: %"   PRIu64 " x%"   PRIu64 " x%"   PRIu64 "  avg_difference: %f  max_difference: %f  avg_eps: %f  max_eps: %f \n "  , dims[0 ], dims[1 ], dims[2 ], avg_difference[0 ], max_difference[0 ], avg_eps[0 ], max_eps[0 ]);
313+ 				fprintf (output, " cuFFT System: %"   PRIu64 " x%"   PRIu64 " x%"   PRIu64 "  avg_difference: %.2e  max_difference: %.2e  avg_eps: %.2e  max_eps: %.2e \n "  , dims[0 ], dims[1 ], dims[2 ], avg_difference[0 ], max_difference[0 ], avg_eps[0 ], max_eps[0 ]);
314+ 			printf (" cuFFT System: %"   PRIu64 " x%"   PRIu64 " x%"   PRIu64 "  avg_difference: %.2e  max_difference: %.2e  avg_eps: %.2e  max_eps: %.2e \n "  , dims[0 ], dims[1 ], dims[2 ], avg_difference[0 ], max_difference[0 ], avg_eps[0 ], max_eps[0 ]);
315315#endif 
316316#ifdef  USE_rocFFT
317317			if  (file_output)
318- 				fprintf (output, " rocFFT System: %"   PRIu64 " x%"   PRIu64 " x%"   PRIu64 "  avg_difference: %f  max_difference: %f  avg_eps: %f  max_eps: %f \n "  , dims[0 ], dims[1 ], dims[2 ], avg_difference[0 ], max_difference[0 ], avg_eps[0 ], max_eps[0 ]);
319- 			printf (" rocFFT System: %"   PRIu64 " x%"   PRIu64 " x%"   PRIu64 "  avg_difference: %f  max_difference: %f  avg_eps: %f  max_eps: %f \n "  , dims[0 ], dims[1 ], dims[2 ], avg_difference[0 ], max_difference[0 ], avg_eps[0 ], max_eps[0 ]);
318+ 				fprintf (output, " rocFFT System: %"   PRIu64 " x%"   PRIu64 " x%"   PRIu64 "  avg_difference: %.2e  max_difference: %.2e  avg_eps: %.2e  max_eps: %.2e \n "  , dims[0 ], dims[1 ], dims[2 ], avg_difference[0 ], max_difference[0 ], avg_eps[0 ], max_eps[0 ]);
319+ 			printf (" rocFFT System: %"   PRIu64 " x%"   PRIu64 " x%"   PRIu64 "  avg_difference: %.2e  max_difference: %.2e  avg_eps: %.2e  max_eps: %.2e \n "  , dims[0 ], dims[1 ], dims[2 ], avg_difference[0 ], max_difference[0 ], avg_eps[0 ], max_eps[0 ]);
320320#endif 
321321			if  (file_output)
322- 				fprintf (output, " VkFFT System: %"   PRIu64 " x%"   PRIu64 " x%"   PRIu64 "  avg_difference: %f  max_difference: %f  avg_eps: %f  max_eps: %f \n "  , dims[0 ], dims[1 ], dims[2 ], avg_difference[1 ], max_difference[1 ], avg_eps[1 ], max_eps[1 ]);
323- 			printf (" VkFFT System: %"   PRIu64 " x%"   PRIu64 " x%"   PRIu64 "  avg_difference: %f  max_difference: %f  avg_eps: %f  max_eps: %f \n "  , dims[0 ], dims[1 ], dims[2 ], avg_difference[1 ], max_difference[1 ], avg_eps[1 ], max_eps[1 ]);
322+ 				fprintf (output, " VkFFT System: %"   PRIu64 " x%"   PRIu64 " x%"   PRIu64 "  avg_difference: %.2e  max_difference: %.2e  avg_eps: %.2e  max_eps: %.2e \n "  , dims[0 ], dims[1 ], dims[2 ], avg_difference[1 ], max_difference[1 ], avg_eps[1 ], max_eps[1 ]);
323+ 			printf (" VkFFT System: %"   PRIu64 " x%"   PRIu64 " x%"   PRIu64 "  avg_difference: %.2e  max_difference: %.2e  avg_eps: %.2e  max_eps: %.2e \n "  , dims[0 ], dims[1 ], dims[2 ], avg_difference[1 ], max_difference[1 ], avg_eps[1 ], max_eps[1 ]);
324324			free (output_VkFFT);
325325			for  (uint64_t  i = 0 ; i < numBuf; i++) {
326326#if (VKFFT_BACKEND==0)
0 commit comments