All pull requests

Pull requests

• [LLVMGPU] Align logical reduction axes before FMA-chain flattening

  #25157·agourakis82 opened Oct 29, 2026·

  ·
  3

• [GlobalOptimization] Pack constant QDQ weights into i4 storage

  #25145·roberto-laudani opened Oct 29, 2026Contributor·

  ·

• [Stream] Add storage policy to async transfers

  #25143·purnima-nlp opened Oct 29, 2026Contributor·

  ·
  18

• [Runtime] Fix shift-count-overflow in next_power_of_two on 32-bit targets

  #25142·riestmo-nxp opened Oct 29, 2026·

  ·
  1

  3

• [Codegen][GPU] Add NVIDIA sparse MMA layout descriptor

  #25141·weimin023 opened Oct 29, 2026Contributor·

  ·

• [DO NOT MERGE] Test LLVM Python dialect stub generation

  #25140·PragmaTwice opened Oct 29, 2026Member·

  ·

• [Torch] Convert integral float zero points of PT2E ops

  #25136·roberto-laudani opened Oct 28, 2026Contributor·

  ·
  

• [Codegen][CPU] Select inner_tiled distribution tiles with the mmt4d cost model

  #25130·shuangjuliu opened Oct 27, 2026·

  ·
  1

• [Codegen][CPU] Honor --iree-llvmcpu-skip-intermediate-roundings=false for RVV Zvfhmin f16 inner_tiled MMA

  #25125·liuyalong opened Oct 26, 2026Contributor·

  ·
  1

  8

• [VMVX] Prioritize dynamic loops in default distribution

  #25123·agourakis82 opened Oct 25, 2026·

  ·
  1

  2

• [BugFix] online attention tiling type error

  #25122·ppppqp opened Oct 24, 2026·

  ·
  1

  2

• [Codegen][CPU] Add RISC-V V bf16 vfwmaccbf16 inner_tiled MMA.

  #25120·TMahlatini opened Oct 23, 2026Contributor·

  ·

• [LLVMCPU] Write embedded executable object data before linking

  #25117·rpqt opened Oct 23, 2026·

  ·
  15

• [GlobalOpt] Fold broadcasts into batch_matmul indexing maps

  #25114·Manewing opened Oct 23, 2026Contributor·2/2·

  ·
  3

• [GlobalOptimization] Do not demote contraction inputs with a tiny reduction extent

  #25105·tonykuttai opened Oct 22, 2026·

  ·
  1

  8

• [Torch] Convert torchao affine quantization to LinalgExt

  #25104·roberto-laudani opened Oct 21, 2026Contributor·

  ·
  3

  

• [CodeGen] Fix linkExecutablesInto creating incorrect executable references

  #25102·devtbi opened Oct 21, 2026Contributor·

  ·
  1

• fix(compiler/GlobalOptimization): change pass ordering so img2col convolutions in quantized conv models will result in integer math

  #25099·Manewing opened Oct 21, 2026Contributor·

  ·
  5

• [Stream] Avoid repeated transfers for concrete await results

  #25098·zhangpan2001 opened Oct 21, 2026Contributor·

  ·
  1

• [Runtime] reject bytecode modules with more than 65535 function descriptors

  #25097·yl120064227 opened Oct 20, 2026·

  ·
  6

• [Runtime] validate gguf alignment and guard the tensor data offset against underflow

  #25096·yl120064227 opened Oct 20, 2026·

  ·
  2

• [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

  1

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

  #25079·anutosh491 opened Oct 19, 2026Member·

  ·