All pull requests

Pull requests

• test(tensilelite): add coverage for doTailLoopOpt and calcEpilogueTurns

  ci:hipsparselt-fastorganization: ROCmproject: hipblasltproject: hipsparselt
  #15128·pdhirajkumarprasad opened Nov 5, 2026Contributor·1 task·

  ·
  2

• feat(miopen): extend ConvDirectDepthwiseFwd2D to 9x9 and 1xK (up to 31) depthwise filters

  documentationorganization: ROCmproject: miopen
  #15127·klin2024 opened Nov 5, 2026·1 task done·

  ·
  2

• test(tensilelite): add coverage for generateSrcStrForMFMA and getMaxRegsForOccupancy

  ci:hipsparselt-fastorganization: ROCmproject: hipblasltproject: hipsparselt
  #15126·pdhirajkumarprasad opened Nov 5, 2026Contributor·1 task·

  ·
  2

• fix(hipdnn): reject negative tensor strides in frontend validation (ALMIOPEN-3114)

  organization: ROCmproject: hipdnn
  #15125·CMiservaAMD opened Nov 5, 2026Contributor·3 of 4 tasks·

  ·
  2

• test: Add coverage for untested gfx950 Python functions

  ci:hipsparselt-fastorganization: ROCmproject: hipblasltproject: hipsparselt
  #15124·pdhirajkumarprasad opened Nov 5, 2026Contributor·1 task·

  ·
  2

• revert(hipblaslt): remove the gfx1250v0 subtree and the asicRevision check

  ci:hipsparselt-fastdocumentationNot ready to Revieworganization: ROCmproject: hipblasltproject: hipsparselt
  #15123·geotseng-amd opened Nov 5, 2026Contributor·1 task·

  ·
  2

  

• perf(ck_tile/fmha): [gfx1250 opt 5] optimize the qr_tdm_pingpong pipeline and extend it to d64

  ck: attention-moeorganization: ROCmproject: composablekernel
  #15122·Yanxing-Shi opened Nov 5, 2026Contributor·

  ·
  2

• Addresses static coverage gaps identified by automated agent analysis:

  ci:hipsparselt-fastorganization: ROCmproject: hipblasltproject: hipsparselt
  #15121·pdhirajkumarprasad opened Nov 5, 2026Contributor·1 task·

  ·
  2

• perf(hipblaslt): use dot2 WaveSplitK for gfx1151 bf16->fp32 TN decode

  organization: ROCmproject: hipblaslt
  #15119·masteryi-0018 opened Nov 5, 2026·1 task done·

  ·
  2

• feat(ck-tile): enable gfx1250 Stream-K on the comp_tdm_v3 pipeline

  documentationorganization: ROCmproject: composablekernel
  #15118·ozturkosu opened Nov 5, 2026Contributor·2/2·

  ·
  2

• test(tensilelite): add unit tests for static coverage gaps by Quality loop #93, #97 and #101

  ci:hipsparselt-fastorganization: ROCmproject: hipblasltproject: hipsparselt
  #15117·pdhirajkumarprasad opened Nov 5, 2026Contributor·1 task·

  ·
  2

• feat(hipblaslt): replace gfx1250 MXFP4 F4BS origami library with pruned BoK kernels

  Not ready to Revieworganization: ROCmproject: hipblaslt
  #15116·pkamd opened Nov 5, 2026Contributor·1 of 4 tasks·

  ·
  2

• feat(ck-tile): adaptive kernel search for the gfx1250 TDM pipelines

  documentationorganization: ROCmproject: composablekernel
  #15115·ozturkosu opened Nov 5, 2026Contributor·

  ·
  2

• feat(hip-kernel-provider): gfx950 unified SDPA rankers and LSE-on kernels

  documentationorganization: ROCmproject: hip-kernel-providerproject: hipdnnproject: hipdnn-integration-testsrocke: platformrocke-lib: attention
  #15113·ThruptiRajLakshmanaGowda opened Nov 5, 2026Contributor·2 of 4 tasks·

  ·
  2

• ALMIOPEN-2597: Batch kernel binary inserts in bounded SQLite transactions

  documentationorganization: ROCmproject: miopen
  #15112·iq136boy opened Nov 5, 2026Contributor·4 of 6 tasks·

  ·
  2

• perf(hipblaslt): split K for gfx1151 fp32 one-row TN GEMMs

  organization: ROCmproject: hipblaslt
  #15111·masteryi-0018 opened Nov 5, 2026·1 task done·

  ·
  2

• perf(hipblaslt): merge gfx950 BBS and SS Equality winners from Isabel…

  organization: ROCmproject: hipblaslt
  #15109·mathialakan opened Nov 5, 2026Contributor·4 tasks·

  ·
  2

• fix(tensilelite): allow gfx1250 MX StreamK with USO

  ci:hipsparselt-fastorganization: ROCmproject: hipblasltproject: hipsparselt
  #15108·ThanHenderson opened Nov 5, 2026Contributor·1 task·

  ·
  3

• fix(tensilelite): query the device for WorkgroupMappingXCCCheck on first use

  ci:hipsparselt-fastorganization: ROCmproject: hipblasltproject: hipsparselt
  #15107·archana-ramalingam opened Nov 5, 2026Contributor·1 task done·

  ·
  3

• fix(rocfft): stop concurrent plan items from sharing work memory

  ci:multi-gpudocumentationorganization: ROCmproject: rocfft
  #15106·regan-amd opened Nov 5, 2026Contributor·1 task done·

  ·
  3

• feat(hipdnn): rocke conv gfx950 wgrad bf16/fp16 engine and ranker

  documentationNot ready to Revieworganization: ROCmproject: hip-kernel-providerproject: hipdnnrocke: platform
  #15105·bartekxk opened Nov 5, 2026Contributor·1 task·

  ·
  2

• feat(miopen): add the lightweight miopen-review skill

  ci:docs-onlydocumentationorganization: ROCmproject: miopen
  #15104·JonathanLichtnerAMD opened Nov 5, 2026Contributor·

  ·
  2

• fix(tensilelite): synchronize persistent TDM LDS reuse

  ci:hipsparselt-fastorganization: ROCmproject: hipblasltproject: hipsparselt
  #15103·ThanHenderson opened Nov 4, 2026Contributor·1 task·

  ·
  4

• feat(hipdnn): ship the trained gfx942 unified 3D kernel ranker (UHD)

  documentationorganization: ROCmproject: hip-kernel-providerproject: hipdnn
  #15102·AviralGoelAMD opened Nov 4, 2026Contributor·

  ·
  

• feat(hipdnn): IngestorGenerator emits per-arch heuristic bindings (engine.heuristics)

  documentationorganization: ROCmproject: hipdnn
  #15101·AviralGoelAMD opened Nov 4, 2026Contributor·

  ·
  3