All pull requests

Pull requests

• [Stream] Honor parameter subrange offsets in split encoders

  #25091·crystarm opened Oct 20, 2026·

  ·
  1

  2

• [DispatchCreation] Skip split reduction candidates with unsupported combiners

  #25090·losewayy opened Oct 20, 2026·

  ·
  1

  2

• [Encoding] Replace packed_storage special cases with interface queries

  #25089·Prajit-Rahul opened Oct 19, 2026Contributor·

  ·
  1

• fix(vm/bytecode): connect ISA decoder validation macros and tighten register verification

  #25088·professor-moody opened Oct 19, 2026·

  ·
  2

• [GlobalOptimization] Expand QDQ contractions over collapsed operands

  #25086·ziereis opened Oct 19, 2026Contributor·

  ·
  1

• [WIP][CI] Run WebGPU HAL tests on Linux

  #25079·anutosh491 opened Oct 19, 2026Contributor·

  ·

• [Torchax] Introduce experimental stateless export and composite registration

  #25074·jschuhmacher opened Oct 19, 2026Contributor·2/2·

  ·

• [LLVMCPU] Generate bounded whole-head attention tiles

  #25073·jschuhmacher opened Oct 19, 2026Contributor·2/2·

  ·

• [StableHLO] Lower Torch SDPA composites with dynamic grouped-query attention

  #25072·jschuhmacher opened Oct 19, 2026Contributor·1/2·

  ·

• [LinalgExt] Preserve attention score coordinates through transformations

  #25071·jschuhmacher opened Oct 19, 2026Contributor·1/2·

  ·

• [WebGPU] Retain resources referenced by command buffers

  #25070·anutosh491 opened Oct 19, 2026Contributor·

  ·
  8

• [WebGPU] Fix staging buffer mapping lifetimes

  #25069·anutosh491 opened Oct 19, 2026Contributor·

  ·
  6

• [StableHLO] Add plugin registration for composite and custom-call rewrites

  #25067·jschuhmacher opened Oct 18, 2026Contributor·

  ·

• [StableHLO] Preserve symbolic shape checks and optimization facts

  #25066·jschuhmacher opened Oct 18, 2026Contributor·

  ·

• [StableHLO] Support runtime slice updates and concatenation

  #25065·jschuhmacher opened Oct 18, 2026Contributor·

  ·

• [Flow] Add a pass that generalizes executables differing only in constants

  #25059·jonathansong opened Oct 17, 2026Contributor·

  ·
  5

• Add big-endian MIPS64 (mips64) VMVX-only runtime cross-compilation

  #25056·andrea037 opened Oct 15, 2026·

  ·
  2

• Add MIPS64 (mips64el) VMVX-only runtime cross-compilation

  #25055·andrea037 opened Oct 15, 2026·

  ·
  2

• [TensorExt] Add ResolveSparseInterfaceOps pass and rewrite patterns

  #25052·MaheshRavishankar opened Oct 15, 2026Collaborator·

  ·

• [LLVMGPU] Tile outer reduction dimensions in contraction config

  #25044·purnima-nlp opened Oct 14, 2026Contributor·

  ·
  1

• [Torch] Register RestructureNonConstantAxes in the Torch input pipeline

  #25041·liuyalong opened Oct 14, 2026Contributor·

  ·
  2

• [Stream] Make "consolidate" mode of SplitParameterEncoding actually consolidate

  #25040·AGindinson opened Oct 14, 2026Contributor·

  ·
  1

  1

• [Stream] Copy resources stored into multiple mutable globals before in-place updates

  #25030·npitre opened Oct 12, 2026·

  ·
  1

  5

• [Stream] Move out all parameters in consolidate mode of SplitParameterEncoding

  #25027·AGindinson opened Oct 12, 2026Contributor·

  ·

• [Codegen] Bound static workgroup counts by target limits

  #25017·zhangpan2001 opened Oct 9, 2026Contributor·

  ·
  1

  2