All pull requests

Pull requests

• [DO NOT MERGE] Test jax-ml/jax#42141

  #1379·psanal35 opened Oct 30, 2026·

  ·

• [ROCm profiler] Put peer-copy rows on the source GPU (PR3)

  #1372·cj401-amd opened Oct 28, 2026Collaborator·2/2·

  ·
  1

• [XLA:GPU][Triton] Route eligible convolutions to Triton

  #1370·mmakevic-amd opened Oct 28, 2026Collaborator·6/6·

  ·
  6

• [XLA:GPU][Triton] Autotune Triton convolution fusions

  #1369·mmakevic-amd opened Oct 28, 2026Collaborator·5/6·

  ·
  7

• [XLA:GPU][Triton] Add TritonConvKey and a convolution autotuning search space

  #1368·mmakevic-amd opened Oct 28, 2026Collaborator·4/6·

  ·
  10

• [XLA:GPU][Triton] Add support gate for convolution

  #1367·mmakevic-amd opened Oct 28, 2026Collaborator·3/6·

  ·
  5

• [XLA:GPU][Triton] Emit convolution as an implicit GEMM

  #1366·mmakevic-amd opened Oct 28, 2026Collaborator·2/6·

  ·
  9

• [XLA:GPU][Triton] Add convolution support to tiling framework

  #1365·mmakevic-amd opened Oct 28, 2026Collaborator·1/6·

  ·
  4

• Enable rbe for multigpu tests on bb cluster

  #1364·alekstheod opened Oct 28, 2026Collaborator·

  ·

• [WIP][XLA:GPU][ROCm] Use the device warp size in the concat cost penalty

  #1361·nurmukhametov opened Oct 27, 2026Member·1 task done·

  ·
  20

• [ROCm] RCCL symmetric memory windows, ncclAlltoAll, and no-packing collective allocation

  #1358·mminutoli opened Oct 26, 2026Collaborator·

  ·
  10

• [WIP][ROCm] Add kpack to the Triton GEMM config

  #1357·nurmukhametov opened Oct 26, 2026Member·1 task done·

  ·
  3

• Update xla-101-advanced training

  #1356·eddieliao opened Oct 26, 2026·

  ·
  

• Update xla-101 training

  #1355·eddieliao opened Oct 26, 2026·

  ·
  

• [ROCm] Add LLVM patch to restrict DPP combine from performing bad tra…

  #1349·zoranjovanovic-ns opened Oct 25, 2026Collaborator·

  ·

• [ROCm profiler] Number trace rows by HIP device (PR2)

  #1345·cj401-amd opened Oct 23, 2026Collaborator·1/2·

  ·
  3

• [ROCm profiler] Add rocm_device_map: match SDK agents to HIP devices (PR1)

  #1344·cj401-amd opened Oct 23, 2026Collaborator·

  ·
  4

• Skip re-recording unchanged collectives in command buffers

  #1342·magaonka-amd opened Oct 23, 2026Collaborator·

  ·
  12

• [ROCm] Request exportable handles for VMM allocations

  #1341·mfrancepillois opened Oct 23, 2026Collaborator·

  ·
  19

• [ROCm] Add a temp buffer pool for --xla_gpu_temp_buffer_use_separate_color

  #1339·magaonka-amd opened Oct 23, 2026Collaborator·

  ·
  5

• Keep recorded collectives out of output allocations under VA remapping

  #1338·magaonka-amd opened Oct 22, 2026Collaborator·

  ·
  8

• [Do not merge] Full XLA GPU suite on rocmless with a patched 10.0 tarball

  #1337·psanal35 opened Oct 21, 2026·

  ·

• [WIP] hipblaslt small cleanup

  #1333·nurmukhametov opened Oct 20, 2026Member·1 task done·

  ·
  1

• Test

  #1325·alekstheod opened Oct 19, 2026Collaborator·1 task·

  ·

• [XLA:GPU] Encode ROCm float total-order sort keys before rocprim.

  #1311·linchen1-robot opened Oct 16, 2026Collaborator·

  ·
  16