All pull requests

Pull requests

• Add IndexDomain::Intersect to compute the intersection of two IndexDomains.

  #51001·copybara-service[bot] opened Oct 17, 2026Bot·

  ·

• Serialize parameter dtype and shape in PjRt-IFRT executable metadata

  #51000·copybara-service[bot] opened Oct 17, 2026Bot·

  ·

• Add DeserializeLoadedExecutable RPC to IFRT Proxy

  #50999·copybara-service[bot] opened Oct 17, 2026Bot·

  ·

• Add an option to prevent donated buffers for being resused for temporary allocations.

  #50998·copybara-service[bot] opened Oct 16, 2026Bot·

  ·
  

• [XLA] Move buffer_assignment and heap_simulator to service/buffer_assignment

  #50997·copybara-service[bot] opened Oct 16, 2026Bot·

  ·
  

• Roll forward of "[Mosaic] Fix tpu.memref_squeeze..." with relaxed verifier

  #50996·copybara-service[bot] opened Oct 16, 2026Bot·

  ·

• Support async compile in CommonPjRtClient.

  #50995·copybara-service[bot] opened Oct 16, 2026Bot·

  ·
  

• Update XNNPACK and slinky in TensorFlow/XLA.

  #50994·copybara-service[bot] opened Oct 16, 2026Bot·

  ·

• Allow kAfterAll and kAddDependency to be identical to themselves.

  #50993·copybara-service[bot] opened Oct 16, 2026Bot·

  ·

• [XLA] Honor mixed precision in the scan verifier when it ignores layouts

  #50992·copybara-service[bot] opened Oct 16, 2026Bot·

  ·
  

• [XLA] Emit buffer aliases once per HloBuffer in BufferAssignment::ToProto

  #50991·copybara-service[bot] opened Oct 16, 2026Bot·

  ·
  

• [XLA:CPU] Add vectorized f64 log and log2 intrinsics.

  #50989·copybara-service[bot] opened Oct 16, 2026Bot·

  ·
  

• [XLA:GPU] Splat scalar operands in Triton atomics lowering

  #50988·janbernloehr opened Oct 16, 2026·

  ·
  1

• Use axis-indexed device_id dict in XLA chiplet-to-chiplet DMA benchmark.

  #50987·copybara-service[bot] opened Oct 16, 2026Bot·

  ·

• PR #50924: Fix FFI record handler registration in the Python HLO runner

  #50984·copybara-service[bot] opened Oct 16, 2026Bot·

  ·

• Record original value recovery for no-op transpose removal

  #50983·copybara-service[bot] opened Oct 16, 2026Bot·

  ·
  

• PR #50858: [PJRT:GPU] Align topology and client platform versions

  #50981·copybara-service[bot] opened Oct 16, 2026Bot·

  ·

• [XLA:CPU] Add parameter buffer slices to ComputationKernelArguments.

  #50979·copybara-service[bot] opened Oct 16, 2026Bot·

  ·
  

• Copy host<->device transfers directly under GPU Confidential Computing.

  #50978·copybara-service[bot] opened Oct 16, 2026Bot·

  ·

• Relax tuple all-to-all layout verification before layout assignment.

  #50976·copybara-service[bot] opened Oct 16, 2026Bot·

  ·
  

• Populate parameter shape and dtype in PjRt-IFRT serialized metadata.

  #50975·copybara-service[bot] opened Oct 16, 2026Bot·

  ·

• Report estimated custom-call costs in XProf op metrics.

  #50974·copybara-service[bot] opened Oct 16, 2026Bot·

  ·

• Enable LLO-based cost estimation for Pallas kernels

  #50973·copybara-service[bot] opened Oct 16, 2026Bot·

  ·

• Add estimated FLOPs and bytes accessed stat types to the XPlane schema.

  #50972·copybara-service[bot] opened Oct 16, 2026Bot·

  ·

• PR #50785: AsyncCollectiveCustomCallRewriter: Propagate frontend attributes for rewritten explicit async collectives

  #50967·copybara-service[bot] opened Oct 16, 2026Bot·

  ·