import torch from fused_ssim import fused_ssim from pytorch_msssim import SSIM import matplotlib.pyplot as plt import numpy as np import time import os plt.style.use('ggplot') # GPU Device Detection and Configuration # This script supports benchmarking on multiple GPU backends: # - CUDA: For Nvidia GPUs and AMD GPUs (via ROCm) # - MPS: For Apple Silicon (M1, M2, M3, etc.) # - XPU: For Intel GPUs (via SYCL/oneAPI) # The appropriate device is automatically detected and configured below. if torch.cuda.is_available(): # CUDA backend for Nvidia GPUs and AMD GPUs (with ROCm) gpu = torch.cuda.get_device_name() fused_ssim_device = "cuda" fused_ssim_module = torch.cuda elif torch.mps.is_available(): # MPS (Metal Performance Shaders) backend for Apple Silicon gpu = "Apple Silicon (MPS)" fused_ssim_device = "mps" fused_ssim_module = torch.mps elif hasattr(torch, 'xpu') and torch.xpu.is_available(): # XPU backend for Intel GPUs (via SYCL/oneAPI) gpu = torch.xpu.get_device_name(0) fused_ssim_device = "xpu" fused_ssim_module = torch.xpu if __name__ == "__main__": # Benchmark Configuration torch.manual_seed(0) B, CH = 5, 1 dimensions = list(range(50, 1550, 50)) iterations = 50 data = { "pytorch_mssim": [], "fused-ssim": [] } pm_ssim = SSIM(data_range=1.0, channel=CH) # Training Benchmark: Measure forward and backward pass performance for d in dimensions: with torch.no_grad(): img1_og = torch.rand([B, CH, d, d], device=fused_ssim_device) img2_og = torch.rand([B, CH, d, d], device=fused_ssim_device) img1_mine_same = torch.nn.Parameter(img1_og.clone()) img2_mine_same = img2_og.clone() img1_pm = torch.nn.Parameter(img1_og.clone()) img2_pm = img2_og.clone() begin = time.time() for _ in range(iterations): pm_ssim_val = pm_ssim(img1_pm, img2_pm) pm_ssim_val.backward() fused_ssim_module.synchronize() # Ensure GPU operations complete before timing end = time.time() data["pytorch_mssim"].append((end - begin) / iterations * 1000) begin = time.time() for _ in range(iterations): mine_ssim_val_same = fused_ssim(img1_mine_same, img2_mine_same) mine_ssim_val_same.backward() fused_ssim_module.synchronize() # Ensure GPU operations complete before timing end = time.time() data["fused-ssim"].append((end - begin) / iterations * 1000) num_pixels = (B * np.array(dimensions) ** 2) / 1e6 plt.plot(num_pixels, data["pytorch_mssim"], label="pytorch_mssim") plt.plot(num_pixels, data["fused-ssim"], label="fused-ssim") plt.legend() plt.xlabel("Number of pixels (in millions).") plt.ylabel("Time for one training iteration (ms).") plt.title(f"Training Benchmark on {gpu}.") plt.savefig(os.path.join("..", "images", f"training_time-{gpu.lower().replace(' ', '-')}.png"), dpi=300) data = { "pytorch_mssim": [], "fused-ssim": [] } plt.clf() # Inference Benchmark: Measure forward pass only (no gradients) for d in dimensions: with torch.no_grad(): img1_og = torch.rand([B, CH, d, d], device=fused_ssim_device) img2_og = torch.rand([B, CH, d, d], device=fused_ssim_device) img1_mine_same = torch.nn.Parameter(img1_og.clone()) img2_mine_same = img2_og.clone() img1_pm = torch.nn.Parameter(img1_og.clone()) img2_pm = img2_og.clone() begin = time.time() for _ in range(iterations): pm_ssim_val = pm_ssim(img1_pm, img2_pm) fused_ssim_module.synchronize() # Ensure GPU operations complete before timing end = time.time() data["pytorch_mssim"].append((end - begin) / iterations * 1000) begin = time.time() for _ in range(iterations): mine_ssim_val_same = fused_ssim(img1_mine_same, img2_mine_same, train=False) fused_ssim_module.synchronize() # Ensure GPU operations complete before timing end = time.time() data["fused-ssim"].append((end - begin) / iterations * 1000) num_pixels = (B * np.array(dimensions) ** 2) / 1e6 plt.plot(num_pixels, data["pytorch_mssim"], label="pytorch_mssim") plt.plot(num_pixels, data["fused-ssim"], label="fused-ssim") plt.legend() plt.xlabel("Number of pixels (in millions).") plt.ylabel("Time for one inference iteration (ms).") plt.title(f"Inference Benchmark on {gpu}.") plt.savefig(os.path.join("..", "images", f"inference_time-{gpu.lower().replace(' ', '-')}.png"), dpi=300)