All pull requests

Pull requests

• [TLX][AMD] Fix int32 offset overflow in gfx950 flash_attn_varlen (#4722)

  CLA Signedmeta-exported
  #4722·FindHao opened Nov 5, 2026Contributor·

  ·
  5

• [triton] Gluon TMA loads with multicast=True lower to a non-multicast copy (#4721)

  CLA Signedmeta-exported
  #4721·ifding opened Nov 5, 2026·

  ·
  1

• Optimize Muse dense BF16 GEMMs on gfx950

  CLA Signedmeta-exported
  #4720·wlei-llvm opened Nov 5, 2026Contributor·

  ·
  2

• [triton][beta] [Cherry-pick] '[AMD] Avoid false HIP runtime mismatch with TheRock (#11492)'

  CLA Signedmeta-exported
  #4719·agron911 opened Nov 4, 2026Contributor·

  ·
  1

• [TLX] [AMD] Improve gfx950 MM production-shape performance (#4718)

  CLA Signedmeta-exported
  #4718·wlei-llvm opened Nov 4, 2026Contributor·

  ·
  1

• [Cherry-pick] [AMD][BACKEND] Fix free-variable mask key to enable register deduplication in stores (#11465)

  CLA Signedmeta-exported
  #4717·ianbarber opened Nov 4, 2026Contributor·

  ·
  1

• [Triton][NPOT] Admit tt.histogram on Product layouts (#4716)

  CLA Signedmeta-exported
  #4716·ianbarber opened Nov 4, 2026Contributor·

  ·
  1

• [Triton][NPOT] Admit tt.atomic_rmw and tt.atomic_cas on Product layouts (#4715)

  CLA Signedmeta-exported
  #4715·ianbarber opened Nov 4, 2026Contributor·

  ·
  1

• [Triton][NPOT] Cover the cross-warp tail of a reduce with warps on an NPOT axis (#4714)

  CLA Signedmeta-exported
  #4714·ianbarber opened Nov 4, 2026Contributor·

  ·
  1

• [Triton][NPOT] Scan a Product layout in blocked register order (#4713)

  CLA Signedmeta-exported
  #4713·ianbarber opened Nov 4, 2026Contributor·

  ·
  1

• [triton][beta] compilation-pipeline tutorials: parse the current MLIR pass-dump header in _ir_utils

  CLA Signedmeta-exported
  #4710·zztbit opened Nov 4, 2026Contributor·

  ·
  1

• [Triton][NPOT] Admit tt.reduce on Product layouts and add the rejection hook (#4706)

  CLA Signedmeta-exported
  #4706·ianbarber opened Nov 4, 2026Contributor·

  ·
  1

• [Triton][NPOT] Keep input radices in the reduce layout helpers (#4705)

  CLA Signedmeta-exported
  #4705·ianbarber opened Nov 4, 2026Contributor·

  ·
  1

• [Triton][NPOT] Keep the shape of NPOT webs that lower a Product layout (#4704)

  CLA Signedmeta-exported
  #4704·ianbarber opened Nov 4, 2026Contributor·

  ·
  1

• [TLX] Skip the consumer-side proxy fence when the kernel already fenced its writes

  CLA Signedmeta-exported
  #4689·htyu opened Nov 4, 2026Contributor·

  ·
  1

• [AMD][gfx950][TLX] Production-shape MXFP8 attention forward/backward repro vs AITER

  CLA Signedmeta-exported
  #4683·wlei-llvm opened Nov 4, 2026Contributor·

  ·
  1

• Back out "[triton] Remove identical-layout conversions in RLC"

  CLA Signedmeta-exported
  #4679·karimelazzouni opened Nov 4, 2026·

  ·
  1

• [TLX] Pad blackwell_gemm_ws launch grid M tiles to the cluster width

  CLA Signed
  #4652·jananisriram opened Nov 3, 2026Contributor·

  ·

• [triton][ci] Add gfx1250 MI450 emulation lane via roc-jitsu

  CLA Signedmeta-exported
  #4641·njriasan opened Nov 3, 2026Contributor·

  ·
  3

• [Cherry-pick] Cherry-pick 55 upstream commits (plan-96f23fca786175ca)

  CLA Signed
  #4639·agron911 opened Nov 3, 2026Contributor·

  ·

• [triton][3.8] Remove unreachable break in Proton Graph.cpp

  CLA Signedmeta-exported
  #4635·r-barnes opened Nov 3, 2026Contributor·

  ·
  1

• [triton][3.8] Release unused input in test_reduce

  CLA Signedmeta-exported
  #4626·minjang opened Nov 3, 2026Contributor·

  ·
  1

• [TLX] Stage blackwell_gemm_ws TMA epilogue subtiles in the output dtype

  CLA Signed
  #4618·jananisriram opened Nov 2, 2026Contributor·

  ·

• [triton][beta] Remove identical-layout conversions in RLC (#4615)

  CLA Signedmeta-exported
  #4615·njriasan opened Nov 2, 2026Contributor·

  ·
  1

• [triton][tlx] Round-trip 2-CTA barriers in TTGIR-to-TLX

  CLA Signedmeta-exported
  #4614·jdonald opened Nov 2, 2026Contributor·

  ·
  1