Skip to content

Update CPU W-Tiling

Bram Veenboer requested to merge cpu-wtiling into master

Major update to CPU W-Tiling code:

  • Combine the forward and backward code in a single source file
  • Apply various optimizations
  • Split functions into smaller sub-functions
  • Expose the wtiles_to_grid and wtiles_from_grid kernels in InstanceCPU

These changes make the code both faster and more readable. Furthermore, it allows for a better comparison against the CUDA W-Tiling code. These two codes now look more similar and share a number of optimizations.

Edited by Bram Veenboer

Merge request reports