All pull requests

Pull requests

• fix(rocalution): compute the ItILU0 residual norm for the stopping criteria and the history

  bugorganization: ROCmproject: rocalution
  #14190·ntrost57 opened Oct 22, 2026Contributor·2/2·

  ·
  2

• feat(tensilelite): generate W4A16 prefill kernels

  ci:hipsparselt-fastdocumentationorganization: ROCmproject: hipblasltproject: hipsparselt
  #14189·mgehre-amd opened Oct 22, 2026Contributor·1 task·

  ·

• feat(tensilelite): add host and client support for Int4 block-scaled A

  ci:hipsparselt-fastorganization: ROCmproject: hipblasltproject: hipsparselt
  #14188·mgehre-amd opened Oct 22, 2026Contributor·1 task·

  ·
  2

• fix(rocalution): move the inverse diagonal and redo the analysis when IC moves

  bugorganization: ROCmproject: rocalution
  #14187·ntrost57 opened Oct 22, 2026Contributor·2/2·

  ·
  2

• fix(rocalution): analyse the U part of the iterative triangular solve into its own info

  bugorganization: ROCmproject: rocalution
  #14186·ntrost57 opened Oct 22, 2026Contributor·

  ·
  3

• fix(rocalution): use a 0-based index in HIPAcceleratorVector::Amax

  bugorganization: ROCmproject: rocalution
  #14185·ntrost57 opened Oct 22, 2026Contributor·2/2·

  ·
  2

• fix(ck-tile): reject split-K for TdmEpilogue and fix TdmMultiDEpilogue LDS race

  ck: convolutionNot ready to Revieworganization: streamhpcproject: composablekernel
  #14183·chris-tsiaousis-hpc opened Oct 22, 2026Contributor·3 tasks done·

  ·
  2

• build(hipsparse): stop fetching build dependencies and test matrices by default

  BuildPackagingdocumentationorganization: ROCmproject: hipsparse
  #14182·ntrost57 opened Oct 22, 2026Contributor·

  ·
  2

• perf(ck_tile): reduce fp8 qr_tdm forward overhead on gfx1250

  ck: attention-moeNot ready to Revieworganization: ROCmproject: composablekernel
  #14181·steven112163 opened Oct 22, 2026Contributor·1 task done·

  ·
  2

• test(miopen): skip grouped-conv perf-config search test when CK has no instances

  organization: ROCmproject: miopen
  #14180·amd-ahyttine opened Oct 22, 2026Contributor·

  ·
  3

• perf(rocprim): block_histogram_atomic optimization by using __match_any intrinsic

  organization: streamhpcproject: hipcclproject: rocprim
  #14179·NB4444 opened Oct 22, 2026Contributor·1 task·

  ·
  2

  

• test(hipblaslt): make memory_shortfall_reclaims_idle_buffers deterministic

  organization: ROCmproject: hipblaslt
  #14177·archana-ramalingam opened Oct 22, 2026Contributor·1 task done·

  ·
  3

• feat(rocke): tile x knob registry for gfx950 GDN/KDA decode

  documentationorganization: ROCmproject: hip-kernel-providerrocke: platform
  #14176·AviralGoelAMD opened Oct 22, 2026Contributor·

  ·

• test(tensilelite): run spmm_gate on gfx1250-strict

  ci:hipsparselt-fastorganization: ROCmproject: hipblasltproject: hipsparselt
  #14175·archana-ramalingam opened Oct 22, 2026Contributor·1 task done·

  ·
  3

• test(tensilelite): mark two new gfx1250 configs ffm_fail

  ci:hipsparselt-fastorganization: ROCmproject: hipblasltproject: hipsparselt
  #14174·archana-ramalingam opened Oct 22, 2026Contributor·1 task done·

  ·
  3

• perf(hipblaslt): tune gfx1201 INT8 TN H3 GEMMs

  ci:gfx120X-allorganization: ROCmproject: hipblaslt
  #14173·zihaomu opened Oct 22, 2026Member·1 task done·

  ·
  2

• fix(rocke): use the compiler's VGPR allocation step for CDNA occupancy

  organization: ROCmproject: hip-kernel-providerrocke: platformrocke-lib: attention
  #14171·cgmillette opened Oct 22, 2026Contributor·1 task·

  ·
  3

• fix(rocSolver) SVD Batched gesvdj fails for batch size > 64 * 1024

  documentationorganization: ROCmproject: rocsolver
  #14170·EdDAzevedo opened Oct 22, 2026Contributor·1 task·

  ·
  9

• fix(miopen): order LSTM multi-stream fwd bias add before GEMM

  organization: ROCmproject: miopen
  #14169·JonathanLichtnerAMD opened Oct 22, 2026Contributor·

  ·
  2

• feat(tensilelite): add the A2A+GEMM shard-loop kernel (FusedA2A: A2AGemm)

  ci:hipsparselt-fastdocumentationorganization: ROCmproject: hipblasltproject: hipsparselt
  #14168·DDEle opened Oct 22, 2026Contributor·5/5·1 task done·

  ·
  3

• docs(hipblaslt): Updated changelog for 10.2.0

  ci:docs-onlydocumentationNot ready to Revieworganization: ROCmproject: hipblaslt
  #14165·pmoutsias-amd opened Oct 22, 2026Contributor·1 task done·

  ·
  5

• test(rocke): split pinned GPU references by operation, architecture, and variant

  documentationorganization: ROCmproject: hip-kernel-providerrocke: platformrocke-lib: attention
  #14164·brockhargreaves-amd opened Oct 22, 2026Contributor·1 task·

  ·

• fix(ck-tile): run misaligned Stream-K problems through the vector-width fallback (gfx1250)

  documentationorganization: ROCmproject: composablekernel
  #14163·ozturkosu opened Oct 22, 2026Contributor·

  ·
  2

  

• ci(hipblaslt): update rocjitsu and check gfx950 races

  github actionsorganization: ROCmproject: nonetest:hipblaslt
  #14162·newling opened Oct 21, 2026Contributor·1 task done·

  ·
  2

• test(rocke): move arch-domain reproduction to CTest

  documentationorganization: ROCmproject: hip-kernel-providerrocke: platform
  #14160·shumway opened Oct 21, 2026Contributor·9 tasks done·

  ·
  2