All pull requests

Pull requests

• Apply per-context LLVM library options in LLVMCommandLineOptionsLock.

  #51173·copybara-service[bot] opened Oct 19, 2026Bot·

  ·

• Move UpdateLayoutForDynamicShapes and GetHloCostAnalysis to PjRtTopologyDescription, add a config for include_raw_buffer_in_ready_event, and remove the virtualness of SupportsPredeterminedError.

  #51172·copybara-service[bot] opened Oct 19, 2026Bot·

  ·
  

• Support parsing LLVM options into LLVMContext in LLVMCommandLineOptionsLock.

  #51171·copybara-service[bot] opened Oct 19, 2026Bot·

  ·

• [XLA:GPU][oneAPI] Align XPU Triton hook with upstream Triton

  #51170·akhilgoe opened Oct 19, 2026Contributor·

  ·

• Expose PjRtBuffer::CopyToMemorySpace with donated destination buffer to IFRT as Client::CopyArraysWithDonatedDst.

  #51169·copybara-service[bot] opened Oct 19, 2026Bot·

  ·
  

• Add intel iGPU definitions to oneapi_compute_capability

  #51168·Nyaa97 opened Oct 19, 2026·

  ·
  1

• Fix unused-variable warnings in tensorflow/compiler/xla

  #51167·copybara-service[bot] opened Oct 19, 2026Bot·

  ·

• [XLA:CPU] Round reference std::expm1(double) through float in ExhaustiveF16AndBF16ZeroUlpViaF32.

  #51165·copybara-service[bot] opened Oct 19, 2026Bot·

  ·

• Expand XLA:GPU AOT collective compatibility test coverage

  #51164·copybara-service[bot] opened Oct 19, 2026Bot·

  ·
  

• [ROCm] Hide GCC 8 std::filesystem symbols.

  #51162·zoranjovanovic-ns opened Oct 19, 2026Contributor·

  ·

• Synchronize external reference count tracking in the PJRT C API.

  #51161·copybara-service[bot] opened Oct 19, 2026Bot·

  ·

• [XLA:GPU] Mark fusion_emitter_int4_device_test as flaky on AMD GPUs.

  #51159·copybara-service[bot] opened Oct 19, 2026Bot·

  ·
  

• N/A

  #51158·copybara-service[bot] opened Oct 19, 2026Bot·

  ·
  

• N/A

  #51157·copybara-service[bot] opened Oct 19, 2026Bot·

  ·
  

• Update Eigen to commit:1da510b8155f8fa81924616dcc517ecc4202e4d3

  #51156·copybara-service[bot] opened Oct 19, 2026Bot·

  ·

• Refine backend layout constraints in a single LayoutAssignment::RunImpl pass

  #51154·copybara-service[bot] opened Oct 19, 2026Bot·

  ·

• [XLA] Add call filter predicate to UnflattenCallGraph.

  #51153·copybara-service[bot] opened Oct 19, 2026Bot·

  ·
  

• [XLA:CPU] Restore piecewise expm1 fallback in ElementalIrEmitter when C++ bitcode intrinsics are unavailable.

  #51152·copybara-service[bot] opened Oct 19, 2026Bot·

  ·
  

• Add Hlo config module to the custom pass interface

  #51151·dlcompilers-infra-bot opened Oct 19, 2026Contributor·

  ·
  

• Add ability to override threadpool implementation via Env

  #51150·copybara-service[bot] opened Oct 19, 2026Bot·

  ·
  

• Add stablehlo.fma op to StableHLO and VHLO 1.24.0.

  #51145·copybara-service[bot] opened Oct 19, 2026Bot·

  ·
  

• [XLA:GPU] Reject degenerate collectives for kernel codegen.

  #51144·copybara-service[bot] opened Oct 19, 2026Bot·

  ·
  

• Avoid pointless GitHub CI runs on Copybara merges

  #51143·copybara-service[bot] opened Oct 19, 2026Bot·

  ·
  

• [XLA] Ignore metadata and original_value when deduplicating computations in UnflattenCallGraph.

  #51140·copybara-service[bot] opened Oct 19, 2026Bot·

  ·
  

• [XLA:GPU] Do not fuse collectives into GEMM fusions in GemmFusionV2.

  #51139·copybara-service[bot] opened Oct 19, 2026Bot·

  ·
  1