All pull requests

Pull requests

• [AMD] Avoid GIL deadlock during HIP library lookup (cherry pick of #11265)

  #974·nemanjaudovic opened Sep 24, 2026·

  ·

• [release 3.8] Support Python 3.15

  #971·ethanwee1 opened Sep 21, 2026·

  ·
  2

• [release/internal/3.8.x] Remove git version suffix

  #957·ethanwee1 opened Jul 23, 2026·

  ·

• [AMD] Improve ld.lld detection for ROCm SDK / ROCM_PATH installs

  #950·adityas-amd opened Jul 9, 2026Member·

  ·
  1

• [release/internal/3.5.x] Fix null-ptr crash in TritonAMDGPUOptimizeDotOperands when tt.trans takes load result directly

  #929·mstankov-amd opened Mar 3, 2026·

  ·

• Cherry pick commits to fix mxfp4 perf

  #897·zhanglx13 opened Oct 29, 2025Collaborator·

  ·

• tune_gemm: use correct output option for rocprofv3

  #887·matthiasdiener opened Oct 3, 2025·

  ·
  

• [AMD] Option to use custom scheduler in LLVM backend

  #884·alefimov-amd opened Sep 25, 2025·

  ·
  1

• [DRAFT] Move quantized dot operands in LDS in pipeliner

  #882·alefimov-amd opened Sep 23, 2025·

  ·

• Pytorch/rocm7.1 internal testing hstu drop 1

  #881·scxiao opened Sep 23, 2025Collaborator·

  ·

• Adding more support for machine model cycles per dot/mfma

  #870·guacamoleo opened Aug 15, 2025·

  ·

• Add bf16 gemm pingpong with num_stages=3

  #818·jungpark-mlir opened May 30, 2025·

  ·
  6

• Bypass LDS for scale B operand for skinny gemms

  #817·plognjen opened May 29, 2025·

  ·
  5

• [DO NOT MERGE!!!] Introduce tilesPerWarp parameter, scale preshuffling and bypassingLDS

  #809·plognjen opened May 23, 2025·

  ·
  2

• [DRAFT] Shared/aggregate load

  #804·alefimov-amd opened May 21, 2025·

  ·

• [AMD] Improve Scheduling for Async BF16 GEMM

  #802·raikonenfnu opened May 21, 2025Member·

  ·
  1

• add predicate mask for atomic_rmw ops

  #799·scxiao opened May 19, 2025Collaborator·

  ·

• [AMD] added slicing ttg.async_copy_global_to_local

  #797·ravil-mobile opened May 15, 2025·

  ·
  31

• Enchanged rocm-triton-prof.py; regex, config-yaml file

  #793·ravil-mobile opened May 7, 2025·

  ·
  1

• [AMD] Added canonicalization pattern to propagate DotOp attrs

  #792·ravil-mobile opened May 7, 2025·

  ·
  4

• [AMD] Added Refine-Reschedule hint for rescheduling

  #791·ravil-mobile opened May 5, 2025·

  ·
  1

• Shaoclee/compare ck

  #788·k50112113 opened May 2, 2025·

  ·

• [WIP] [StreamK]

  #782·zhanglx13 opened Apr 28, 2025Collaborator·

  ·
  8

• [AMD] Added bufferOps refinement

  #776·ravil-mobile opened Apr 14, 2025·

  ·
  15

• update scale dot assertion in plot_layout.py

  #774·jtang10 opened Apr 10, 2025·

  ·