53 Pull requests merged by 2 people

• Fix move constructor definition. My earlier change tried to define the move constructor explicitly to establish the correct invariants, but accidentally cleared the wrong container.

  #97541 merged Jul 25, 2025

• Fix GPU Zero Occupancy Stats

  #97453 merged Jul 24, 2025

• [XLA:GPU] Disable the change from PR #28184.

  #97535 merged Jul 24, 2025

• [XLA][Numerics][HLO Value Tracking] Define != operator for OriginalArray

  #96326 merged Jul 24, 2025

• [xla] Change Shape::element_type and Shape::array_or_buffer_element_type.

  #97436 merged Jul 24, 2025

• #sdy always lift meshes before deduping, since the dedup meshes pass ignores inlined meshes.

  #97510 merged Jul 24, 2025

• Add the skeleton for PropagateQsvPass in the tflite converter.

  #97386 merged Jul 24, 2025

• Update keras version and refresh lock files

  #97519 merged Jul 24, 2025

• Update the Keras and Tensorboard nightlies versions.

  #97522 merged Jul 24, 2025

• Don't export Shardy in MPMD before going to IFRT, since IFRT can support (and export) a module with StableHLO + Shardy.

  #97503 merged Jul 24, 2025

• PR #26445: [NFC] Clean up StreamKind and StreamId from GpuCliqueKey

  #97528 merged Jul 24, 2025

• PR #27904: [XLA:GPU][oneAPI] Enable Clang compiler as the host compiler

  #97333 merged Jul 24, 2025

• Remove nvshmem_configure.bzl file from XLA OSS and update rules_ml_toolchain in XLA dependencies to the version that doesn't have nvshmem_configure.bzl file.

  #97529 merged Jul 24, 2025

• Add a new constructor for HloRunnerAgnosticTestBase to accept an options struct.

  #97447 merged Jul 24, 2025

• Fix a postsubmit build failure

  #97493 merged Jul 24, 2025

• Conditions the import of cudnn_graph in cuda_dnn.cc only for v9_0_0 and above.

  #97532 merged Jul 24, 2025

• PR #29178: [XLA:GPU] Explicitly set shared memory config for kernels

  #97524 merged Jul 24, 2025

• RaggedDot in HloEvaluator now tolerates inputs with different precision, like conv.

  #97439 merged Jul 24, 2025

• Expose the variables interface for tpu_embedding_for_serving.

  #97464 merged Jul 24, 2025

• [IFRT] Support XLA GPU flag overrides.

  #97156 merged Jul 24, 2025

• [XLA][host offloading] Correctly handle queueing tasks on stream.

  #97353 merged Jul 24, 2025

• Define move constructor consistent with the invariants required by the destructor.

  #97517 merged Jul 24, 2025

• [XLA] Don't assume loops are canonicalzed

  #97477 merged Jul 24, 2025

• Migrate uses of XLA_TEST_BACKEND macros to use utilities in xla_test_backend_predicates.h

  #97468 merged Jul 24, 2025

• [xla:cpu] Add Thunk::async_resume() property to decide if thunk execution should be resumed using the TaskRunner or the caller thread

  #97473 merged Jul 24, 2025

• Also create reduce-scatters when SPMD-partitioning is disabled.

  #97211 merged Jul 24, 2025

• [xla] Change Shape::IsArray and Shape::IsArrayOrBuffer.

  #97389 merged Jul 24, 2025

• [XLA:GPU] Add tiling for variadic reduce.

  #97507 merged Jul 24, 2025

• Automated Code Change

  #97249 merged Jul 24, 2025

• [XLA:GPU] Adjust indexing cost model heuristic for register usage.

  #97434 merged Jul 24, 2025

• [XLA:GPU] Support different unroll factors than just 2.

  #97422 merged Jul 24, 2025

• Remove redundant conversion to std::string

  #97441 merged Jul 24, 2025

• Add bug id in the warning message for involuntary full rematerialization.

  #97448 merged Jul 24, 2025

• Automated Code Change

  #97494 merged Jul 24, 2025

• Fix reading mapped names in LayerNameMapper

  #94651 merged Jul 24, 2025

• [XLA:GPU] Add a set of tuning configs for Blackwell

  #97483 merged Jul 24, 2025

• Automated Code Change

  #97492 merged Jul 24, 2025

• Update deps:

  #97035 merged Jul 24, 2025

• [XLA:GPU] Fix a typo in triton support logic.

  #97488 merged Jul 24, 2025

• [xla] Optimize ShapeTree construction time for non-tuples

  #97454 merged Jul 24, 2025

• [StableHLO<->HLO] Remove special handling for AddOp to Xor for booleans

  #97451 merged Jul 24, 2025

• Use constexpr absl::string_view for hlo string instead of const char*

  #97458 merged Jul 24, 2025

• [xla:cpu] Make DotLibraryRewriter fuse in both up and down directions.

  #97481 merged Jul 24, 2025

• [StableHLO] Fix boolean converters

  #97443 merged Jul 24, 2025

• Enable dynamic-slice offset based on multiply for AllGatherPermutedDsSimplifierTest, more specifically,

  #97332 merged Jul 24, 2025

• [xla] Remove NOINLINE from ShapeIndex constructor

  #97455 merged Jul 24, 2025

• Remove local_config_nvshmem repository from XLA and Tensorflow WORKSPACE files.

  #97082 merged Jul 24, 2025

• [xla] Switch from std::set to absl::flat_hash_set

  #97450 merged Jul 24, 2025

• [XLA:MGPU][Experimental] HLO -> Pallas.

  #97357 merged Jul 24, 2025

• [xla] Switch from std::unordered_map to absl::flat_hash_map

  #97456 merged Jul 24, 2025

• [XLA] Avoid incidental changes in WhileLoopConstantSinking.

  #97471 merged Jul 24, 2025

• [XLA] Don't require flattened control-flow in InfeedTokenPropagation.

  #97442 merged Jul 24, 2025

• Add support for deserializing xplanes to Jaxlib

  #96282 merged Jul 24, 2025

49 Pull requests opened by 5 people

• Automated Code Change

  #97475 opened Jul 24, 2025

• Automated Code Change

  #97476 opened Jul 24, 2025

• [XLA] Allow tuple-like while body roots during unrolling

  #97478 opened Jul 24, 2025

• Add type-safe Go session config API, docs, and tests

  #97479 opened Jul 24, 2025

• Automated Code Change

  #97480 opened Jul 24, 2025

• Automated Code Change

  #97482 opened Jul 24, 2025

• Remove unused variable in ResizeBilinear888Uint8

  #97484 opened Jul 24, 2025

• Automated Code Change

  #97485 opened Jul 24, 2025

• Automated Code Change

  #97486 opened Jul 24, 2025

• Automated Code Change

  #97487 opened Jul 24, 2025

• Automated Code Change

  #97489 opened Jul 24, 2025

• Automated Code Change

  #97490 opened Jul 24, 2025

• Replace ElementsAreArray(<float>) to Pointwise(FloatingPointEq(), <float>) for StablehloCaseTestFloat

  #97491 opened Jul 24, 2025

• Automated Code Change

  #97495 opened Jul 24, 2025

• Automated Code Change

  #97496 opened Jul 24, 2025

• Automated Code Change

  #97497 opened Jul 24, 2025

• Automated Code Change

  #97498 opened Jul 24, 2025

• Automated Code Change

  #97499 opened Jul 24, 2025

• Automated Code Change

  #97501 opened Jul 24, 2025

• [XLA:GPU] Add a set of tuning configs for Blackwell

  #97502 opened Jul 24, 2025

• Share the logic to check if a module has an unapproved dialect for versioned serialization.

  #97504 opened Jul 24, 2025

• [XLA:GPU] Enable wgmma with block_n=8 in Triton and autotuner

  #97505 opened Jul 24, 2025

• #sdy set `PropagationOptions::enableAutoPartitioning` in shardy_xla_pass and cleanup legacy auto partitioner code, as it's no longer needed.

  #97506 opened Jul 24, 2025

• Fix: Remove dangling reference to ExperimentalTFLiteToTosaBytecode

  #97508 opened Jul 24, 2025

• [XLA:GPU] Update test to account for new MLIR lowering patterns.

  #97509 opened Jul 24, 2025

• [XLA:GPU] ignore configs in autotuner if the triton emitter does not support them

  #97511 opened Jul 24, 2025

• Fix: Clarify GPU compatibility check in TFLite Model Analyzer

  #97513 opened Jul 24, 2025

• [XLA:GPU] NFC move fusion block level rewriter logic to the pass

  #97515 opened Jul 24, 2025

• Reverts 3c1a8779ff9951a0797ff64ad1551eaf2e8c37eb

  #97516 opened Jul 24, 2025

• Add support for deserializing xplanes to Jaxlib

  #97518 opened Jul 24, 2025

• [XLA] Don't flatten the CFG inside WhileLoopUnroller.

  #97520 opened Jul 24, 2025

• Add int8 and int16x8 kernel support for sqrt

  #97521 opened Jul 24, 2025

• [xla:cpu] Add Thunk::async_resume() property to decide if thunk execution should be resumed using the TaskRunner or the caller thread

  #97523 opened Jul 24, 2025

• Attribute exporter absl

  #97525 opened Jul 24, 2025

• [XLA] WhileLoopUnroller should only inline newly created calls.

  #97526 opened Jul 24, 2025

• Fix nccl for libc++.

  #97527 opened Jul 24, 2025

• Enable libnvptxcompiler and libnvjitlink support in OSS by default.

  #97530 opened Jul 24, 2025

• use "xla_dump_to" to dump PjrtCompile inputs

  #97531 opened Jul 24, 2025

• Checks for zero steps before iterating through a range.

  #97533 opened Jul 24, 2025

• removed tflite deps from _pywrap_tensorflow and add it to tflite_pywrap_deps instead

  #97534 opened Jul 24, 2025

• Update Maxtext-derived HLOs with large constants

  #97536 opened Jul 24, 2025

• Add a heuristic to choose the quantization type for same-scale ops.

  #97537 opened Jul 24, 2025

• Migrate away from ListInit::getValues()

  #97538 opened Jul 24, 2025

• Migrate away from ArrayRef(std::nullopt_t)

  #97539 opened Jul 24, 2025

• Pipe incarnations through C plugin.

  #97540 opened Jul 24, 2025

• Pallas/Mosaic support host output.

  #97542 opened Jul 24, 2025

• OSS-ing the generic pjrtTopologyDescription proto

  #97543 opened Jul 25, 2025

• [SDY] Add pretty print methods for sdy attrs used with explicit collective

  #97544 opened Jul 25, 2025

• hold the usage event until the host callback is done when doing d2h

  #97545 opened Jul 25, 2025

3 Issues closed by 1 person

• Corrección código complejos

  #97474 closed Jul 24, 2025

• DLL load failed while importing _pywrap_tensorflow_internal: A dynamic link library (DLL) initialization routine failed.

  #97514 closed Jul 24, 2025

• Failed to load the native TensorFlow runtime.

  #97500 closed Jul 24, 2025

1 Issue opened by 1 person

• tf.no_op DAG cycle issue

  #97512 opened Jul 24, 2025

31 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• Getting AttributeError when running tf.ones_like

  #89250 commented on Jul 24, 2025 • 0 new comments

• The API documentation of tf.no_op specifies that tf.no_op() should return a TensorFlow Operation.

  #89248 commented on Jul 24, 2025 • 0 new comments

• [TOSA] NameError: name 'ExperimentalTFLiteToTosaBytecode' is not defined

  #88053 commented on Jul 24, 2025 • 0 new comments

• tf.keras.Model have not attribute : submodules but the document still use the error attribute

  #97403 commented on Jul 24, 2025 • 0 new comments

• TensorFlow was not built with CUDA kernel binaries compatible with compute capability 12.0 CUDA_ERROR_INVALID_HANDLE

  #97387 commented on Jul 25, 2025 • 0 new comments

• Update Protobuf to 6.31.1

  #95873 commented on Jul 24, 2025 • 0 new comments

• Use RPG's solution as a hint to CP-SAT

  #96674 commented on Jul 24, 2025 • 0 new comments

• Avoid crashing when LRU cache keys change.

  #96930 commented on Jul 24, 2025 • 0 new comments

• [XLA:GPU] Move the s4 unpacking sequence from llvm pass to int4->int8 pass

  #97047 commented on Jul 24, 2025 • 0 new comments

• Introduce --dump_tflite_model_dir to dump TFLite models in Delegate Test Suite (DTS)

  #97118 commented on Jul 24, 2025 • 0 new comments

• Add Metal LiteRt Tensor Buffer support

  #97142 commented on Jul 25, 2025 • 0 new comments

• Integrate Triton up to [6af1c4b5](https://github.com/openai/triton/commits/6af1c4b507dfddd0d62cdc8613839f46bdf6acb0)

  #97282 commented on Jul 24, 2025 • 0 new comments

• [SDY] dump source sharding info when output shardyDir is non-empty. This occurs when `xla_dump_to` flag is set and `xla_dump_hlo_pass_re` regex matches the shardy pass (see `getShardyDirIfShouldDump`).

  #97321 commented on Jul 24, 2025 • 0 new comments

• Add utility function `MatchDsPadAllGather` to match the pattern of `DynamicSlice(Pad(AllGather))`.

  #97336 commented on Jul 25, 2025 • 0 new comments

• [XLA:GPU] Migrate GemmAlgorithmPicker to new autotuner.

  #97340 commented on Jul 24, 2025 • 0 new comments

• Turn DCHECK into runtime error. Because literal is user provided, it should

  #97368 commented on Jul 25, 2025 • 0 new comments

• [XLA:CPU] Lower to xla.rsqrt for all rsqrt ops.

  #97371 commented on Jul 24, 2025 • 0 new comments

• [XLA:GPU] update deadlocks verifier behavior to get computations from the scheduler

  #97374 commented on Jul 24, 2025 • 0 new comments

• Integrate LLVM at llvm/llvm-project@22b083539051

  #97398 commented on Jul 24, 2025 • 0 new comments

• Automated Code Change

  #97420 commented on Jul 24, 2025 • 0 new comments

• Automated Code Change

  #97421 commented on Jul 24, 2025 • 0 new comments

• [XLA] Improve hlo_live_range_test to use HLO textual form

  #97425 commented on Jul 24, 2025 • 0 new comments

• Extend the Profiler Interface in Autotuner.

  #97437 commented on Jul 24, 2025 • 0 new comments

• Add Multi output reduce test

  #97438 commented on Jul 24, 2025 • 0 new comments

• MPMD XLA bump

  #97457 commented on Jul 24, 2025 • 0 new comments

• Add JAX tests for deadlock verifier

  #97459 commented on Jul 24, 2025 • 0 new comments

• Add option to enable coordination service client to recover.

  #97460 commented on Jul 24, 2025 • 0 new comments

• Add environment variable to enable TFRT GPU client in GPU plugin.

  #97461 commented on Jul 24, 2025 • 0 new comments

• Add option to enable TFRT GPU client in C API plugin.

  #97462 commented on Jul 24, 2025 • 0 new comments

• Add option to enable coordination service client to recover.

  #97463 commented on Jul 24, 2025 • 0 new comments

• Add quantization utility functions for PropagateQSVPass.

  #97470 commented on Jul 24, 2025 • 0 new comments