Im2win: Memory Efficient Convolution On SIMD Architectures

Convolution is the most expensive operation among neural network operations, thus its performance is critical to the overall performance of neural networks. Commonly used convolution approaches, including general matrix multiplication (GEMM)-based convolution and direct convolution, rely on im2col f...

Ful tanımlama

Kaydedildi:
Detaylı Bibliyografya
Asıl Yazarlar: Lu, Shuai, Chu, Jun, Liu, Xu T.
Materyal Türü: Makale
Baskı/Yayın Bilgisi: 2023
Konular:
Online Erişim:https://arxiv.org/abs/2306.14320
http://arxiv.org/abs/2306.14320
Etiketler: Etiketle
Etiket eklenmemiş, İlk siz ekleyin!