Repository navigation

[XNNPACK] One polyphase pass for dense and depthwise transposed convs - #25041

Draft

JakeStevens wants to merge 20 commits into

pytorch:mainfrom

JakeStevens:xnnpack-unify-conv-transpose

Draft

[XNNPACK] One polyphase pass for dense and depthwise transposed convs#25041
JakeStevens wants to merge 20 commits into
pytorch:mainfrom
JakeStevens:xnnpack-unify-conv-transpose

Commits on Oct 23, 2026
Commits on Oct 26, 2026
- Remove the deleted spec with a slice assignment
  borys3kk
  committed
  743c690
Commits on Oct 27, 2026
- Use a Software Mansion copyright header on the new files
  borys3kk
  committed
  1c8a70d
- Drop the channels-last broadcast operand changes that moved out of #24113
  borys3kk
  committed
  445b1d5
Commits on Oct 29, 2026
- Key the phase weight placeholders by stride and padding as well
  borys3kk
  committed
  c22288b
Commits on Oct 30, 2026
- Merge remote-tracking branch 'origin/main' into xnnpack-depthwise-conv-transpose-main
  borys3kk
  committed
  12f2694
Commits on Nov 2, 2026
Commits on Nov 5, 2026
- Merge branch 'main' into xnnpack-depthwise-conv-transpose-main
  JakeStevens
  authored
  7ad83c0
Commits on Nov 12, 2026
- Add Phonon (Gradium 31M TTS) example with multi-backend export (#24810)
  
  psiddh
  authored and
  meta-codesync[bot]
  committed
  d7d0a0b
- Phonon: streaming output and per-backend weight quantization (#24929)
  
  psiddh
  authored and
  meta-codesync[bot]
  committed
  8ba6126
- Fix the linear_q4gsw reference implementation for output channels not a multiple of 4 (#24930)
  
  psiddh
  authored and
  meta-codesync[bot]
  committed
  423496d
- Do not delegate quantized linears that have no Vulkan kernel (#24971)
  
  psiddh
  authored and
  meta-codesync[bot]
  committed
  518777b
- Phonon: run the Mimi decoder's transposed convolutions as convolutions on XNNPACK
  
  psiddh
  authored and
  meta-codesync[bot]
  committed
  3495700
Commits on Nov 13, 2026