Skip to content

GitLab

Switch branch/tag
  1. 16 Jul, 2025 2 commits
    • Thorsten Kurth's avatar
      cleanup with contiguous checks · 45fc2a46
      Thorsten Kurth authored
      45fc2a46
    • Mauro Bisson's avatar
      Optimized BWD kernel with the same changes for FWD from commit 8cb399ee: · 9a463332
      Mauro Bisson authored 
      * Replaced PyTorch's slow permutation.
* Split kernel into general and specialized versions (for num_channel <= 8192)
* Enabled float4-based vectorized memory access, when possible.
* Added runtime dispatch logic for kernel specialization.

Aligned attention_fwd_cuda.cu with attention_bwd_cuda.cu in terms of naming conventions and kernel parameters.

Extracted shared host/device functions and declarations into a separate module:
* attention_utils.cuh
* attention_utils.cu
      9a463332