All pull requests

Pull requests

• test(hipblaslt): add an on-demand fast_check SDC-hunt recipe

  documentationproject: hipblaslt
  #13025·tony-davis opened Oct 3, 2026Contributor·8/8·1 task done·

  ·

• fix(rocke/dispatch): unbreak the gfx950 dense-attention ingestor bridge after the dispatcher refactor

  Not ready to Revieworganization: ROCmproject: hip-kernel-providerproject: hipdnnrocke-lib: attention
  #13024·ThruptiRajLakshmanaGowda opened Oct 3, 2026Contributor·3 tasks done·

  ·
  2

• fix(rocfft): fix multi-device plans with Bluestein lengths

  ci:multi-gpudocumentationorganization: ROCmproject: rocfft
  #13023·regan-amd opened Oct 2, 2026Contributor·4/4·1 task done·

  ·
  2

• test(hipblaslt): run fast_check on edge, bookkeeping and per-kernel configurations

  organization: ROCmproject: hipblaslt
  #13022·tony-davis opened Oct 2, 2026Contributor·7/8·1 task done·

  ·

• test(hipblaslt): sweep fast_check across 2^31, 2^32 and launch-size thresholds

  documentationorganization: ROCmproject: hipblaslt
  #13021·tony-davis opened Oct 2, 2026Contributor·6/8·1 task done·

  ·

• fix(rocfft): fix GPU faults of fused Bluestein with unit batch distance

  documentationorganization: ROCmproject: rocfft
  #13020·regan-amd opened Oct 2, 2026Contributor·3/4·1 task done·

  ·
  3

• feat(stinkytofu): Support barrier signal/wait scheduling

  ci:hipsparselt-fastdocumentationorganization: ROCmproject: hipblasltproject: hipsparseltshared: stinkytofu
  #13019·hcman2 opened Oct 2, 2026Contributor·1 task·

  ·
  2

• fix(rocfft): fix plan creation failures for fused Bluestein with padded odist

  documentationorganization: ROCmproject: rocfft
  #13017·regan-amd opened Oct 2, 2026Contributor·2/4·1 task done·

  ·
  1

• [MIOpen] Sync hipconv v0.3.2

  documentationorganization: ROCmproject: miopen
  #13016·amd-alavin opened Oct 2, 2026Contributor·

  ·
  3

• test(hipblaslt): check fast_check side operands (scales, E, amaxD, bias gradients)

  documentationorganization: ROCmproject: hipblaslt
  #13015·tony-davis opened Oct 2, 2026Contributor·5/8·1 task done·

  ·
  1

• fix(rocke): bound gfx1250 GEMM loads and preserve persistent CTA counts

  documentationgithub actionsorganization: ROCmproject: hip-kernel-providerproject: nonerocke: platform
  #13014·tenpercent opened Oct 2, 2026Contributor·6 of 8 tasks·

  ·

• test(hipblaslt): keep fast_check exact at large K

  documentationorganization: ROCmproject: hipblaslt
  #13013·tony-davis opened Oct 2, 2026Contributor·4/8·1 task done·

  ·
  1

• Save per-iteration benchmark data during GA optimization

  ci:hipsparselt-fastNot ready to Revieworganization: ROCmproject: hipblasltproject: hipsparselt
  #13012·pkamd opened Oct 2, 2026Contributor·

  ·
  3

• fix(tensilelite): protect persistent vector epilogue LDS

  ci:hipsparselt-fastdocumentationNot ready to Revieworganization: ROCmproject: hipblasltproject: hipsparselt
  #13011·ThanHenderson opened Oct 2, 2026Contributor·

  ·
  3

• Save users/nguyen nhu di/rocprim kernel tuner find run length topk

  Not ready to Revieworganization: ROCm
  #13010·NguyenNhuDi opened Oct 2, 2026Contributor·1 task·

  ·
  2

• fix(rocprim,rocrand,hipcub,rocthrust): find amd_smi for the benchmarks (port of #12532 to hipccl3)

  organization: ROCm
  #13009·stanleytsang-amd opened Oct 2, 2026Contributor·4 of 5 tasks·

  ·
  2

• docs(rocthrust, rocrand, rocprim, hipcub) updated the docs to point to the external primbench docs (port of #11826 to hipccl3)

  ci:docs-onlydocumentationorganization: ROCm
  #13008·stanleytsang-amd opened Oct 2, 2026Contributor·

  ·
  2

• feat(rocprim): Implement Kernel Tuner for Various Algorithms (port of #12008 to hipccl3)

  organization: ROCm
  #13007·stanleytsang-amd opened Oct 2, 2026Contributor·1 task·

  ·
  2

• perf(rocprim): reduced branches in device_search and device_find_end (port of #11299 to hipccl3)

  documentationorganization: ROCm
  #13006·stanleytsang-amd opened Oct 2, 2026Contributor·1 task done·

  ·
  2

• test(hipblaslt): check every fast_check solution on every iteration

  documentationorganization: ROCmproject: hipblaslt
  #13005·tony-davis opened Oct 2, 2026Contributor·3/8·1 task done·

  ·
  1

• feat(rocprim): add tuning with kernel tuner for device_transform and device_transform_pointer (port of #12705 to hipccl3)

  organization: ROCm
  #13004·stanleytsang-amd opened Oct 2, 2026Contributor·1 task done·

  ·
  2

• feat(rocprim): kernel tuner cleanup and add colors to logs (port of #12838 to hipccl3)

  organization: ROCm
  #13003·stanleytsang-amd opened Oct 2, 2026Contributor·1 task done·

  ·
  2

• fix(rocthrust): harden HIPSTDPAR allocation interposition (port of #11498 to hipccl3)

  documentationorganization: ROCm
  #13002·stanleytsang-amd opened Oct 2, 2026Contributor·1 task done·

  ·
  2

• build(deps): bump pyjwt from 2.8.0 to 2.15.0 in /projects/hipccl/hipccl2/rocthrust/docs/sphinx

  ci:docs-onlydependenciesdocumentationexternal contributionpython
  #13001·dependabot[bot] opened Oct 2, 2026Bot·

  ·
  1

• build(deps): bump tornado from 6.5.5 to 6.5.9 in /projects/hipccl/hipccl2/rocthrust/docs/sphinx

  ci:docs-onlydependenciesdocumentationexternal contributionpython
  #13000·dependabot[bot] opened Oct 2, 2026Bot·

  ·
  1