All pull requests

Pull requests

• [NFC] Migrate the Intel backend off deprecated op builders

  #8654·lowpolyneko opened Nov 4, 2026Contributor·

  ·

• [Pipeliner] Cost model for dropping a gapped dot-operand prefetch

  run-benchmarks
  #8653·etiotto opened Nov 4, 2026Contributor·

  ·
  1

  

• [TritonIntelGPU] Lower row-limit masks on 2d_block_load_from_ptr to base_height

  #8651·anmyachev opened Nov 4, 2026Contributor·

  ·

• [CI][Benchmarks] Use distinct artifact names for Triton, vLLM and SGLang benchmark reports

  run-benchmarksrun-sglang-benchmarksrun-vllm-benchmarks
  #8637·quinnlp opened Nov 3, 2026Contributor·

  ·

• [Intel] Raise OutOfResources when num_warps * warp_size > wg size

  #8632·dudilester opened Nov 3, 2026Contributor·

  ·
  1

  1

• [RemoveLayoutConversions] Rematerialize in-loop converts up to their anchors

  run-benchmarks
  #8631·wdziurdz opened Nov 3, 2026Contributor·

  ·
  1

  3

• [StageLargeFMADotsViaSLM] Stage fp8 dot operands decoded

  #8628·dudilester opened Nov 3, 2026Contributor·

  ·
  1

• [Analysis] Stop under-approximating loop-carried ranges in IntegerRangeAnalysis

  run-benchmarks
  #8621·etiotto opened Nov 2, 2026Contributor·

  ·
  1

  1

  

• [TritonIntelGPU] Enable 2D block IO for runtime row-bound masks

  run-benchmarks
  #8615·anmyachev opened Nov 2, 2026Contributor·

  ·
  1

  28

• [Benchmarks] Route GEMM shapes with a mostly empty last wave to stream-K

  run-benchmarks
  #8611·januszjah opened Nov 2, 2026Contributor·

  ·
  1

  1

• [AxisInfo][StrideInfo] Fix unsound deductions for signed division and remainder

  #8610·whitneywhtsang opened Nov 1, 2026Contributor·

  ·
  1

  10

• [vLLM][CI] Warmup cache for benchmarks

  #8608·anmyachev opened Nov 1, 2026Contributor·

  ·

• [RemoveLayoutConversions] Fall back to convert_layout when a result can't be retyped in place

  #8600·anmyachev opened Oct 30, 2026Contributor·

  ·
  4

• Update Triton version: 3.9.0 -> 3.9.1

  #8598·anmyachev opened Oct 30, 2026Contributor·

  ·
  2

• Use SYCL for large attention references

  #8595·sarah12121212 opened Oct 29, 2026Contributor·

  ·

• [Intel] Correct rank-reducing 2D block stores and fold unit-dim-adding reshapes into them

  run-benchmarks
  #8593·etiotto opened Oct 29, 2026Contributor·

  ·
  1

  7

  

• [E2E] Testing

  #8587·anmyachev opened Oct 29, 2026Contributor·

  ·

• [SGLang] Add an XPU default and a B60 config for the fp8 fused MoE

  #8585·wdziurdz opened Oct 29, 2026Contributor·

  ·

• Make the autotune benchmarker's L2 flush GRF-mode neutral

  #8582·dev-tomek opened Oct 29, 2026Contributor·

  ·
  1

• Let the backend's Python files load outside triton.backends.intel

  #8554·mieshkiwrk opened Oct 27, 2026Contributor·

  ·
  14

  

• [XPU] Add an opt-in large-GRF-first build for predicted rebuilds

  #8550·etiotto opened Oct 26, 2026Contributor·

  ·

• [CI] Run TritonKernelRunner tests in build-test with its Triton wheel

  #8524·kwasd opened Oct 22, 2026Contributor·

  ·
  6

  

• Add generic Triton compile-cache warmup script for triton_kernels_benchmark

  #8513·anmyachev opened Oct 22, 2026Contributor·

  ·

• [Benchmarks] Add autotune config pinning to triton-benchmarks

  #8503·mieshkiwrk opened Oct 21, 2026Contributor·

  ·
  15

  

• [CI] Raise Windows tests timeout to 900 minutes

  #8493·kwasd opened Oct 20, 2026Contributor·

  ·
  2