All pull requests

Pull requests

• Initial removal of AITER prebuilt mechanism

  ci-level 1P2
  #782·Micky774 opened Oct 28, 2026Contributor·13 tasks·

  ·

• Support gfx1250-strict

  ci-level 1
  #781·ipanfilo opened Oct 27, 2026Collaborator·1 of 13 tasks·

  ·
  1

• Enforce using of source CI image

  ci-level 1
  #780·ipanfilo opened Oct 24, 2026Collaborator·13 tasks·

  ·

• [gfx1250] Use TDM for MXFP8 and FP8 quantization

  ci-level 1
  #778·matthiasdiener opened Oct 22, 2026Contributor·1 of 13 tasks·

  ·
  

• XCD work stealing improvements for HK Grouped GEMMs

  #777·alextmagro opened Oct 22, 2026Contributor·

  ·
  11

• fix examples, wire them into CI summary, fix dispatch when multiple labels are added simultaneously

  ci-level 1P2
  #776·matthiasdiener opened Oct 21, 2026Contributor·1 of 13 tasks·

  ·
  2

  

• split-K and LDS offset buffer for HipKittens GEMMs

  ci-level 3
  #772·alextmagro opened Oct 20, 2026Contributor·

  ·
  7

• Ifu dev 20261019 v2.21

  ci-level 3P1
  #771·VeeraRajasekhar opened Oct 20, 2026Contributor·

  ·
  22

  

• Dynamic WS sizing

  ci-level 1P2
  #770·matthiasdiener opened Oct 20, 2026Contributor·1 of 13 tasks·

  ·
  2

  

• gfx1250: use hipblaslt device alpha for NVFP4 GEMM

  ci-level 1P2
  #766·matthiasdiener opened Oct 16, 2026Contributor·1 of 13 tasks·

  ·
  4

  

• Ipanfilo/security scaning

  ci-level 1P2
  #762·ipanfilo opened Oct 16, 2026Collaborator·13 tasks·

  ·
  12

• Release 2.20.2

  ci-level 3P1
  #760·Micky774 opened Oct 15, 2026Contributor·13 tasks·

  ·
  6

• speed up multi-stream execution

  ci-level 1P3
  #759·matthiasdiener opened Oct 13, 2026Contributor·1 of 13 tasks·

  ·
  

• [TE] CK fused attention: return max_logit

  P1
  #755·AllenFarcas opened Oct 7, 2026Contributor·3 of 13 tasks·

  ·
  19

  

• FlyDSL persistent work-stealing grouped GEMM enablement (gfx950)

  ci-level 1P2
  #746·aris134 opened Sep 29, 2026Contributor·13 tasks·

  ·
  70

  

• extend MXFP4 recipe to GroupedLinear and a8w4 (Triton grouped GEMM kernels)

  ci-level 1P2
  #745·matthiasdiener opened Sep 28, 2026Contributor·1 of 13 tasks·

  ·
  46

  

• chore: add security scanning workflows

  P2
  #743·haribabug opened Sep 24, 2026·

  ·
  4

• microbenchmarks: use pytest as execution backend

  P2
  #726·matthiasdiener opened Aug 31, 2026Contributor·1 of 13 tasks·

  ·
  28

  

• [proof-of-concept] comm overlap microbenchmarks

  P2
  #724·matthiasdiener opened Aug 31, 2026Contributor·1 of 15 tasks·

  ·
  2

  

• Add gfx950 MXFP8 CK grouped GEMM

  P2
  #718·aris134 opened Aug 26, 2026Contributor·13 tasks·

  ·
  

• Integrate Grouped Gemm v2 on ROCm

  P2
  #717·VeeraRajasekhar opened Aug 26, 2026Contributor·13 tasks·

  ·
  

• CI: upload Python coverage.json from pytest (non-blocking)

  community-contributionP2
  #712·jiagaoxiang opened Aug 21, 2026·4 tasks·

  ·
  1

• [ROCm] Route dense CP softmax-LSE correction through a native kernel

  community-contributionP2
  #708·zjin-lcf opened Aug 19, 2026·

  ·
  1

• [proof-of-concept] Kernel autotuning in TE

  P3
  #705·matthiasdiener opened Aug 18, 2026Contributor·1 of 13 tasks·

  ·
  

• [gfx1250] detect when building on gfx1250, add to build archs

  ci-level 1P2
  #700·matthiasdiener opened Aug 11, 2026Contributor·1 of 13 tasks·

  ·
  3