Skip to content

Instantly share code, notes, and snippets.

@mingfeima

mingfeima/pytorch_perf_optimization_cpu.md

Last active December 29, 2017 01:51
Show Gist options
  • Save mingfeima/1c0ac42822c75978f2129d4776f2eb8c to your computer and use it in GitHub Desktop.
Save mingfeima/1c0ac42822c75978f2129d4776f2eb8c to your computer and use it in GitHub Desktop.
Download ZIP
Raw
pytorch_perf_optimization_cpu.md

PyTorch Performance Optimization on CPU

  1. pytorch mkldnn integration prototype design
  • mkldnn conv integration
  • conv3d parallelization: vol2col, col2vol
  • LSTM optimization non-fused: tanh/sigmoid parallelization

  1. Create MKLDNN conda channel

  2. MKLDNN tensor type

  • create lib/THMKL?
  • create ATen backend?
  • move mkldnn_conv_xxx from torch/csrc to ATen
  • mkldnn global primitive cache
  • reorder definition
  1. RNN fused kernel on Xeon
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment