Im2win: An Efficient Convolution Paradigm on GPU

TL;DR

This paper introduces im2win, a GPU convolution paradigm that reduces memory usage and improves performance by ensuring continuous memory access, outperforming existing methods in deep neural network computations.

Contribution

The paper proposes the im2win convolution method on GPU, combining memory efficiency with high performance, and demonstrates its superiority over existing approaches through extensive benchmarks.

Findings

Reduces memory footprint by up to 32.8%.

Achieves up to 3.5× TFLOPS compared to cuBLAS.

Outperforms cuDNN and direct convolution in speed and memory efficiency.

Abstract

Convolution is the most time-consuming operation in deep neural network operations, so its performance is critical to the overall performance of the neural network. The commonly used methods for convolution on GPU include the general matrix multiplication (GEMM)-based convolution and the direct convolution. GEMM-based convolution relies on the im2col algorithm, which results in a large memory footprint and reduced performance. Direct convolution does not have the large memory footprint problem, but the performance is not on par with GEMM-based approach because of the discontinuous memory access. This paper proposes a window-order-based convolution paradigm on GPU, called im2win, which not only reduces memory footprint but also offers continuous memory accesses, resulting in improved performance. Furthermore, we apply a range of optimization techniques on the convolution CUDA kernel, including shared memory, tiling, micro-kernel, double buffer, and prefetching. We compare our implementation with the direct convolution, and PyTorch's GEMM-based convolution with cuBLAS and six cuDNN-based convolution implementations, with twelve state-of-the-art DNN benchmarks. The experimental results show that our implementation 1) uses less memory footprint by 23.1% and achieves 3.5$\times$ TFLOPS compared with cuBLAS, 2) uses less memory footprint by 32.8% and achieves up to 1.8$\times$ TFLOPS compared with the best performant convolutions in cuDNN, and 3) achieves up to 155$\times$ TFLOPS compared with the direct convolution. We further perform an ablation study on the applied optimization techniques and find that the micro-kernel has the greatest positive impact on performance.