All pull requests

Pull requests

• [BugFix][Reduce] Fit reduce batch to the target's per-thread output count

  bugdocumentation
  #3684·saiyambharara opened Oct 26, 2026Contributor·

  ·
  1

  2

• [BugFix][Transform] Fix divergent and missing barriers in thread_storage_sync

  arithbugUX
  #3682·stealth-rklopper opened Oct 26, 2026·

  ·
  1

  2

• [BugFix][CUDA] Pack a float32 broadcast to 8 lanes without the address of a temporary

  bug
  #3681·stealth-rklopper opened Oct 26, 2026·

  ·
  1

  2

• [Ascend] Add Manual Scheduled FlashMLA kernels

  #3678·Denverjin opened Oct 26, 2026Contributor·

  ·
  4

• [BugFix][Language] Validate transpose destination extents in the operator

  bugcpuUX
  #3677·173787247 opened Oct 26, 2026Contributor·

  ·
  5

• [Ascend]: Support opt-in batch scheduling

  ascendbugenhancement
  #3675·TZX789 opened Oct 26, 2026·

  ·
  2

• [BugFix] Region-based kill reorder for shared memory liveness

  bug
  #3670·Guan-jeans opened Oct 26, 2026Contributor·

  ·
  5

• [Ascend] Interleaved strided UB->GM stores from AIV cores inside T.PerCoreTask()

  arithascendbugdocumentationenhancement
  #3668·LLMZhangYC opened Oct 26, 2026Contributor·

  ·
  5

• [Refactor] Rebase TileLang onto unmodified standalone TIRx

  #3666·SiriusNEO opened Oct 25, 2026Collaborator·

  ·
  2

• [BugFix][ROCm] Fix float-to-FP4 casts and add the E8M0 type in the HIP templates

  #3665·xiaohong42 opened Oct 25, 2026·

  ·
  2

• [BugFix][Examples] Fit attention tiles to shared-memory limits

  bugpython
  #3662·Donkeyzhenbang opened Oct 25, 2026·

  ·
  4

• [BugFix][CUDA] Publish synchronous stores in warp-specialized producers

  bug
  #3661·Donkeyzhenbang opened Oct 25, 2026·

  ·
  1

  2

• [Bugfix][Autotuner] Serialize cache repair and publication across processes 🤖

  bugdependenciespython
  #3660·yampolskiyy opened Oct 25, 2026·

  ·
  1

  2

• [Bugfix] Preserve print options during manual profiler checks 🤖

  bugpythontoolsUX
  #3658·yampolskiyy opened Oct 25, 2026·

  ·
  2

• [Bugfix] Fall back when CPU affinity cannot be queried 🤖

  bugpython
  #3657·yampolskiyy opened Oct 25, 2026·

  ·
  2

• [Bugfix] Correct bool comparisons and uint8 error diagnostics 🤖

  bugpythonUX
  #3656·yampolskiyy opened Oct 25, 2026·

  ·
  2

• [Ascend] Reject on-chip memory overflow in MergeUBAllocations

  arithascendbugUX
  #3655·CeleNewYear opened Oct 25, 2026Collaborator·

  ·
  6

• [Bugfix] Handle non-contiguous tensors in mismatch diagnostics 🤖

  bugpythonUX
  #3654·yampolskiyy opened Oct 25, 2026·

  ·
  2

• [BugFix] Respect buffer alignment when vectorizing memory accesses

  arithbug
  #3652·sepcnt opened Oct 25, 2026Contributor·

  ·
  1

  2

• [BugFix][Codegen] Wrap scalar sub-word int arithmetic to dtype width

  bugcpu
  #3644·arcusbuilds opened Oct 24, 2026Contributor·

  ·
  1

  6

• [BugFix][CuTeDSL] Repair the CuTeDSL backend and make its CI job run CuTeDSL

  arithbugdependenciesenhancementgithub_actionspythonUX
  #3640·lucifer1004 opened Oct 24, 2026Collaborator·

  ·
  38

• [Ascend] Support E4M3 ND/NZ UB-to-L1 copies

  ascenddocumentationenhancement
  #3639·mmyy1217 opened Oct 24, 2026·

  ·
  12

• [BugFix][JIT] Preserve identity of same-named dynamic symbols

  #3638·erhsh opened Oct 24, 2026Contributor·

  ·
  1

  2

• Add Intel AMX BF16 and INT8 CPU backend

  cpudocumentationenhancementpythontoolsUX
  #3637·blossomin opened Oct 24, 2026·

  ·
  7

• [BugFix][JIT] Evaluate dynamic output shape expressions

  arithascendbugcpudocumentationpythonUX
  #3636·erhsh opened Oct 24, 2026Contributor·

  ·
  1

  4