All pull requests

Pull requests

• [BugFix][Language] Normalize scalar numeric conditions in the eager frontend

  bugdocumentationpython
  #3563·kobecai opened Oct 18, 2026Contributor·

  ·
  1

  2

• [Perf][CUDA] Use NaN-propagating min/max for FP32 clamp on SM80+

  documentationenhancementpythontools
  #3562·JayceSu98 opened Oct 18, 2026Contributor·

  ·
  4

• [BugFix][Profiler] Exclude cache-flush time from early-stop decisions

  bugpythontools
  #3561·JayceSu98 opened Oct 18, 2026Contributor·

  ·
  2

• [BugFix][Profiler] Snapshot outputs before later calls overwrite them

  bugpythontools
  #3560·JayceSu98 opened Oct 18, 2026Contributor·

  ·
  6

• [BugFix][Profiler] Fix empty CUDA Graph captures across devices

  bugpythontools
  #3559·JayceSu98 opened Oct 18, 2026Contributor·

  ·
  2

• [BugFix] Preserve eager frontend Python source locations

  bugpythonUX
  #3558·Ethereal49 opened Oct 17, 2026Contributor·

  ·
  1

  4

• [Bugfix] Diagnose floating-point operands to integer div/mod operations

  bugpythonUX
  #3557·Edwardssss opened Oct 17, 2026·

  ·
  1

  2

• [Fix] Preserve operand precision for rounding narrow floating-point constants

  arithbugpython
  #3556·Edwardssss opened Oct 17, 2026·

  ·
  1

  6

• [CUDA] Validate MMA K alignment before layout inference

  bugpythonUX
  #3554·Edwardssss opened Oct 17, 2026·

  ·
  1

  2

• [Transform] Keep async WGMMA shared operands live until wait_wgmma

  bug
  #3553·LibertychaserUS opened Oct 17, 2026Contributor·

  ·
  1

  10

• [BugFix][Metal] Skip metal.cooperative_tensor in MergeSharedMemoryAllocations

  bugmetal
  #3552·emilianbold opened Oct 17, 2026·

  ·
  5

• [Bugfix][CuTeDSL] Preserve signed integer truncation

  bugpython
  #3551·SuperGoodGame opened Oct 17, 2026Contributor·

  ·
  1

  5

• [BugFix][TIR] Synchronize predicates with promoted indices

  arithbug
  #3550·penguin-wwy opened Oct 17, 2026Collaborator·

  ·
  2

• [ROCm] Support pip-SDK ROCm 10 (arch detection + Dockerfile)

  bugdependenciesenhancementpython
  #3549·zhangnju opened Oct 16, 2026Collaborator·

  ·
  6

• [BugFix][ROCm] Print fp4 (float4_e2m1fn) float constants in HIP codegen

  bug
  #3548·zhangnju opened Oct 16, 2026Collaborator·

  ·
  6

• [BugFix][Codegen] Handle non-constant cp_async wait depth

  bug
  #3547·zhangnju opened Oct 16, 2026Collaborator·

  ·
  2

• [BugFix][Autotuner] Keep set_autotune_inputs tensors on their capture device

  ascendbugpython
  #3546·liggest opened Oct 16, 2026Contributor·

  ·
  1

  4

• [Metal] Lower tirx.rsqrt to Metal's native rsqrt()

  bugmetal
  #3544·emilianbold opened Oct 16, 2026·

  ·
  2

• [BugFix][Ascend][OOB] Group GM-to-L1 copies and OOB padding into one task

  #3543·liguanglin opened Oct 16, 2026·

  ·
  2

• [BugFix][Metal] Print hoisted statements before a local.var assignment

  bugmetal
  #3542·emilianbold opened Oct 16, 2026·

  ·
  2

• [BugFix] Preserve NamedTuple types in macro returns

  bugpython
  #3540·sepcnt opened Oct 16, 2026Contributor·

  ·
  1

  2

• [BugFix][ThreadSync] Order shared predicate reads before branch writes

  bug
  #3538·Ethereal49 opened Oct 16, 2026Contributor·

  ·
  1

  2

• [BugFix] Fence generic-proxy stage reads before warp-specialized consumer releases

  bug
  #3537·alumkal opened Oct 16, 2026·

  ·
  1

  2

• [BugFix] Fix T.cumsum on shared buffers under warp specialization

  bug
  #3536·samridh10exe opened Oct 15, 2026·

  ·
  1

  4

• [BugFix][Ascend][PTO] Emit Ramp values as typed vectors

  ascendbug
  #3535·erhsh opened Oct 15, 2026Contributor·

  ·
  1

  4