229 Pull requests merged by 5 people

• Extend the Profiler Interface in Autotuner.

  #97437 merged Jul 25, 2025

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

  #97515 merged Jul 25, 2025

• Automated Code Change

  #97420 merged Jul 25, 2025

• Automated Code Change

  #97326 merged Jul 25, 2025

• Automated Code Change

  #97421 merged Jul 25, 2025

• Automated Code Change

  #97254 merged Jul 25, 2025

• Automated Code Change

  #97265 merged Jul 25, 2025

• Automated Code Change

  #97497 merged Jul 25, 2025

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

  #97545 merged Jul 25, 2025

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

  #97368 merged Jul 25, 2025

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

  #97336 merged Jul 25, 2025

• 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

• Optimize HasCombinableReplicaGroup and xla::CheckReplicaGroups.

  #97155 merged Jul 24, 2025

• Fixed dependency cleaner support for tf_kernel_library so that it will update both deps and gpu_deps.

  #97319 merged Jul 23, 2025

• [tf] Use template parameter to pass async callback

  #97440 merged Jul 23, 2025

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

  #97306 merged Jul 23, 2025

• PR #28540: [XLA:GPU] Add support for pre-padded scales for block scaled dot custom call

  #97347 merged Jul 23, 2025

• Parse and ignore 'mode' attribute on collectives.

  #97392 merged Jul 23, 2025

• Remove a redundant condition in mlir_hlo_to_hlo.cc.

  #97405 merged Jul 23, 2025

• Update keras-nightly and typing-extensions versions in the lock files.

  #97445 merged Jul 23, 2025

• [xla:cpu] Make dot thunk support int8 x int8 -> int32 matmul.

  #97435 merged Jul 23, 2025

• Avoid registering definition events until after the allocation succeeds

  #97418 merged Jul 23, 2025

• Add Hermetic C++ Toolchains for Linux x86_64 builds.

  #96820 merged Jul 23, 2025

• Unbreak HloPassPipeline silent/noop change debugging.

  #97431 merged Jul 23, 2025

• [xla] Optimize ShapeTree construction time for arrays

  #97394 merged Jul 23, 2025

• Correct the required gpu to test_core_h100

  #97352 merged Jul 23, 2025

• Use mdformat on the XNNPack delegate readme.

  #97182 merged Jul 23, 2025

• This is an automatic update to a device compatibility allowlist.

  #97423 merged Jul 23, 2025

• [tfcompile] Prefix xla generated symbols with tfcompile_xla_generated.

  #97427 merged Jul 23, 2025

• [XLA] Add stack trace breakdown to HloLiveRange::ToString for peak memory usage

  #94954 merged Jul 23, 2025

• [XLA:GPU] Do not fail when profiling executables, instead ignore failing configurations.

  #97344 merged Jul 23, 2025

• Reverts e4e87da4c3ed8a529f67bb1e3485101c37a4e1f9

  #97424 merged Jul 23, 2025

• Guard against possible null pointer dereferences.

  #97426 merged Jul 23, 2025

• [XLA:Autotuner] Allow gpu_profiler to collect output buffers.

  #97355 merged Jul 23, 2025

• Automated Code Change

  #97401 merged Jul 23, 2025

• [XLA:GPU][host offloading] Implement host offloading thunks.

  #97059 merged Jul 23, 2025

• PR #28735: [XLA:GPU] Enabling cuda graph concurrent mode by default

  #97045 merged Jul 23, 2025

• [XLA:GPU] Early return if there are no elements to compare with.

  #97402 merged Jul 23, 2025

• [XLA:CPU] Run CSE after inlining in fusion compiler.

  #97114 merged Jul 23, 2025

• Reverts 1b7ced5a66c8948c8b6dfa984ddea56e99c3cba3

  #97404 merged Jul 23, 2025

• [triton] Avoid propagating slice layouts across broadcast.

  #97350 merged Jul 23, 2025

• Prevent XLA from crashing when a Literal is too big to fit in memory.

  #97400 merged Jul 23, 2025

• Refactor shardy_xla_pass_test with IsCustomCall. No behavior change.

  #97388 merged Jul 23, 2025

• [XLA][Numerics][HLO Value Tracking] Support original values during HLO and stableHLO round trip

  #97148 merged Jul 23, 2025

• Automated Code Change

  #97342 merged Jul 23, 2025

• [xla] ShapeUtil: optimize shape traversal

  #97381 merged Jul 23, 2025

• [Hlo Diff] Use full constant values when generating fingerpints and simplify the ExactSubgraphMatcher.

  #96312 merged Jul 23, 2025

• Fix a bug where doing a CPU jax.Array -> TPU jax.Array transfer was not setting the layout correctly on the output TPU array.

  #97393 merged Jul 23, 2025

• Change PjRtClient::LazyToLiteral to take a generator that returns a future of the literal

  #97162 merged Jul 23, 2025

• [HLO Diff] Fix bug in instruction users matching to use fingerprints instead of pointers.

  #97292 merged Jul 23, 2025

• Migrate away from ArrayRef(std::nullopt_t)

  #97382 merged Jul 23, 2025

• [tf] Migrate from std::map to absl::flat_hash_map

  #97376 merged Jul 23, 2025

• [IFRT] Define user_context() in Value and LoadedExecutable

  #97153 merged Jul 23, 2025

• [HLO Diff] Fix the order of operands in ?: operator in TopDownMatcher.

  #97291 merged Jul 23, 2025

• [HLO Diff] Fix existence check in TopDownMatcher.

  #97290 merged Jul 23, 2025

• XLA: Fix method ambiguity

  #97359 merged Jul 23, 2025

• [xla] ShapeUtil: avoid WithStatus callbacks on a hot path

  #97380 merged Jul 23, 2025

• Fix resource calculation for annotation groups that do not contain full pairs of start-done ops.

  #97366 merged Jul 23, 2025

• WithIndex: an iterator wrapper that adds iteration index to yielded value.

  #97370 merged Jul 23, 2025

• [IFRT] Remove the error message mentioning an invalid output target

  #97390 merged Jul 23, 2025

• [xla:tsl] Add Eigen contraction kernel template specialization for int8 LHS x int8 RHS -> int32 result.

  #97383 merged Jul 23, 2025

• Refactor StreamExecutorGpuTopologyDescription::DeviceDescriptions to share SetupDeviceDescription

  #97378 merged Jul 23, 2025

• Increase max stack depth of GPU compile callstack.

  #97384 merged Jul 22, 2025

• Update zlib/zstd patch after

  #97369 merged Jul 22, 2025

• Reverts c197694417be40d121b8b8ba9bb50efd81c4f227

  #97375 merged Jul 22, 2025

• #sdy update hasGspmdAttrsOrOps to look at the body of the main function.

  #97358 merged Jul 22, 2025

• [tf] Do not keep XlaComputation alive after it's compiled to Xla executable

  #97364 merged Jul 22, 2025

• Integrate LLVM at llvm/llvm-project@0a343098b0ea

  #97327 merged Jul 22, 2025

• [xla:cpu] Do not serialize HloProto if user didn't ask for it

  #97363 merged Jul 22, 2025

• Add reshard cache when the sharding of reshape is tile maximal (replicated or maximal).

  #97317 merged Jul 22, 2025

• Removed optimized batch matmul implementation from tflite in favor of XNNPACK. Also performed tiny refactoring along the way.

  #97184 merged Jul 22, 2025

• Internal change only.

  #97195 merged Jul 22, 2025

• [XLA:GPU] enable emitting dots through the generic emitter

  #96720 merged Jul 22, 2025

• Enable N-dimensional sparse tensor in tpu_embedding_v3.py

  #97151 merged Jul 22, 2025

• PR #29115: Bump github/codeql-action from 3.29.2 to 3.29.3

  #97337 merged Jul 22, 2025

• [XLA][host offloading] Open source annotate host compute offload.

  #97362 merged Jul 22, 2025

• [XLA:GPU] Replace "agent" memory sync scope with "device" for nVidia backend.

  #97360 merged Jul 22, 2025

• [XLA:CPU][XLA:GPU] Move concat fusion emitter to shared directory

  #97050 merged Jul 22, 2025

• Automated Code Change

  #97339 merged Jul 22, 2025

• This is an automatic update to a device compatibility allowlist.

  #97346 merged Jul 22, 2025

• [XLA:GPU] Remove horizontal fusion.

  #96241 merged Jul 22, 2025

• Automated Code Change

  #97247 merged Jul 22, 2025

• [XLA:GPU] Choose a tile size according to hardware constraints.

  #97335 merged Jul 22, 2025

• Use CONCURRENT_KERNEL tracing in KernelNameTracer

  #97280 merged Jul 22, 2025

• Fix bf16 propagation for host memory space

  #97310 merged Jul 22, 2025

• Avoid crash in RequestActivityBuffer

  #97279 merged Jul 22, 2025

• Automated Code Change

  #97241 merged Jul 22, 2025

• Add TPU v4i to GetDeviceMemoryInBytes()

  #97316 merged Jul 22, 2025

• [xla] ShapedBuffer: Don't compute on_host_shape if it's the same as device shape

  #97323 merged Jul 22, 2025

• Automated Code Change

  #97248 merged Jul 22, 2025

• Automated Code Change

  #97025 merged Jul 22, 2025

• Fixes minor bug with low memory limits in Peak Priority remat allowing them to be zero.

  #97298 merged Jul 22, 2025

• [xla:cpu] Refactor a few helper functions DotLibraryRewriter for readablility.

  #97324 merged Jul 22, 2025

• Remove AbstractCpuBuffer. All subclasses can be replaced with CommonPjRtBufferImpl and removed.

  #97194 merged Jul 22, 2025

• [xla:cpu] Add useful TraceMe annotations to CpuExecutable

  #97318 merged Jul 22, 2025

• [xla:cpu] Pass --xla_cpu_experimental_onednn_fusion_types to OneDnnMatcher.

  #97320 merged Jul 22, 2025

• [xla] Annotate likely branches with ABSL_PREDICT_TRUE

  #97311 merged Jul 22, 2025

• Add missing includes.

  #96385 merged Jul 22, 2025

• Add a warning log in xla::ShardingPropagation that it will be deprecated and replaced with Shardy.

  #97299 merged Jul 22, 2025

• Introduce AllGatherDynamicSliceShuffledOffsetSimplifier, a new HLO pass that collapse dynamic-slice(all-gather) with shuffled offset into collective-permute.

  #97287 merged Jul 22, 2025

• Updating necessary visibilities ahead of sa-train-google in-place migrations.

  #97304 merged Jul 22, 2025

• Introduce PickUnusedPort() that doesn't terminate when failing to find a port. The new function returns -1 when failing to find an unused port. The original PickUnusedPortOrDie() still terminates when a port can't be found.

  #97308 merged Jul 21, 2025

• [xla] Make Shape::Equal faster

  #97296 merged Jul 21, 2025

• lite: Add config option to enable benchmark_model

  #96999 merged Jul 21, 2025

• Add a RequiresQuantizedBiasInterface

  #96580 merged Jul 21, 2025

• Remove unused API

  #97303 merged Jul 21, 2025

• Update BUILD files with dependencies

  #97305 merged Jul 21, 2025

• Optimize BatchMatmul to Fully Connected when RHS is reshaped after dequantization.

  #97205 merged Jul 21, 2025

• Add SparseCore documentation

  #97200 merged Jul 21, 2025

• Pin python dependencies and update lock files

  #97302 merged Jul 21, 2025

• Update version numbers for TensorFlow 2.20.0-rc0

  #97295 merged Jul 21, 2025

• Use stablehlo precision config conversion for stablehlo ops

  #97233 merged Jul 21, 2025

• Replace implementation of TSL status matchers with Abseil.

  #97283 merged Jul 21, 2025

• [xla:cpu] Make se::HostStream fully synchronous

  #97288 merged Jul 21, 2025

• Skip flaky test on Windows.

  #97286 merged Jul 21, 2025

• Added WatchJobState RPC to coordination service.

  #97077 merged Jul 21, 2025

• Add individual test targets for tensorflow/core:legacy_lib_core_tests.

  #97285 merged Jul 21, 2025

• Updating XNNPACK readme to include the correct weight cache file path field.

  #97289 merged Jul 21, 2025

• Add 10 Maxtext-derived HLO-based benchmarks

  #97132 merged Jul 21, 2025

• [xla] Add benchmarks for Shape::Equal

  #97293 merged Jul 21, 2025

• Avoid heap allocation for the sub buffer address

  #97230 merged Jul 21, 2025

• Avoid checking captured_tensors' usage when deciding if

  #97141 merged Jul 21, 2025

• LatencyHidingScheduler: Only recalculate when we've touched an already-scheduled computation, and use computation-specific peak rather than module peak in statistics.

  #97192 merged Jul 21, 2025

• Reverts 0c0a40e9e8dd2f04ab9a66643ca628acdef5840d

  #97281 merged Jul 21, 2025

• #sdy delete round_trip_import dir since JAX export can now export a model with StableHLO + Shardy dialects.

  #96423 merged Jul 21, 2025

• #sdy use new API to pass dump-index and use MeshAttr::getMaximal

  #97275 merged Jul 21, 2025

• Avoid recomputation of pjrt_buffer->memory_space() in MakeMemoryKindFromPjRtBuffer.

  #97068 merged Jul 21, 2025

• [XLA:GPU] Use correct unroll factor in indexing map

  #97276 merged Jul 21, 2025

• Add missing dependencies.

  #97261 merged Jul 21, 2025

• [XLA] NFC: remove duplication in diag message

  #97274 merged Jul 21, 2025

• [XLA:GPU] Add Get method to GpuPerformanceModelOwning.

  #97277 merged Jul 21, 2025

• Reverts b96ee234c76e3f76341d0acec46e2e7d067cd5a5

  #97273 merged Jul 21, 2025

• [XLA:GPU][host offloading] Implement gpu host offloading allocator.

  #97051 merged Jul 21, 2025

• [XLA:GPU] Refactor tests of IndexingMap

  #97138 merged Jul 21, 2025

• Fix cost analysis on for output byte accessed when result is tuple

  #96927 merged Jul 21, 2025

• Remove LLVM dependency from KernelThunk

  #97178 merged Jul 21, 2025

• Fix a problem with a variable being used out of scope.

  #97272 merged Jul 21, 2025

• [XLA:GPU][Tiling] Use SmallVector<OneDimTile> to store tiling info.

  #97229 merged Jul 21, 2025

• Automated Code Change

  #97268 merged Jul 21, 2025

• Give better error in run_hlo_module if HLO has collectives.

  #97208 merged Jul 21, 2025

• Introduces a new utility function, MatchPermutedSliceAndPartitionOffset, to detect a pattern where a DynamicSlice consumes the output of an AllGather with a permuted set of offsets. This pattern is equivalent to a CollectivePermute and can be optimized accordingly.

  #97189 merged Jul 20, 2025

• Automated Code Change

  #97259 merged Jul 20, 2025

• Add sdy shardings in frontend_attributes alongside hlo shardings for extra wrapper main added in tf2xla bridge.

  #96919 merged Jul 19, 2025

• [XLA:GPU] Support control flow thunks in command buffer conversion pass. We only convert kWhile and kConditional thunks if all thunks in all brunches are convertible.

  #97183 merged Jul 19, 2025

• #sdy Fix forward of making XLA C++ changes so we can fall back to GSPMD in JAX export if the loaded module was lowered for GSPMD.

  #96368 merged Jul 19, 2025

• [XLA] Add helper function GetIndicesSpecForDynamicSlice to get indices spec for dynamic slice fed by all-gather, the spec includes the mapping from slice offsets to corresponding partition IDs(flattened-id).

  #96805 merged Jul 19, 2025

• Internal, visibility only changes to public code.

  #97207 merged Jul 19, 2025

• Add visibility to hlo_input_output_format

  #96758 merged Jul 19, 2025

• Use a literal sentinel value for kernel init failure

  #97193 merged Jul 19, 2025

• Reduce redundancy between StringTo* enum functions.

  #97201 merged Jul 19, 2025

• [XLA:CPU] Refactor Intrinsic and use it in all math intrinsics.

  #97000 merged Jul 19, 2025

• Integrate LLVM at llvm/llvm-project@06ae0c2a1086

  #97154 merged Jul 19, 2025

• Update nccl_archive BUILD file to fix TF GPU wheel build.

  #97206 merged Jul 19, 2025

• [XLA:GPU] Add a verifier to the GPU compiler before post-scheduling pipeline.

  #97150 merged Jul 19, 2025

• Use host callback in the CopyToHostFuture method in Async PjRt.

  #97203 merged Jul 18, 2025

• Add function ExtractDynamicSliceFromCollectiveUser to extract a dynamic slice user from a collective.

  #96802 merged Jul 18, 2025

• no external change

  #97100 merged Jul 18, 2025

• Reverts 849435a30d0487e415126507953575358ed3c4eb

  #97190 merged Jul 18, 2025

• Reverts 2a45c5b0c326e20eafe833df055326b39edadcf2

  #97071 merged Jul 18, 2025

• Bump sqlite to 3.50.3

  #97191 merged Jul 18, 2025

• Typo fix "perferred" -> "preferred".

  #97198 merged Jul 18, 2025

• PR #28257: [XLA:GPU] Update ONEAPI crosstool compiler wrapper

  #97149 merged Jul 18, 2025

• Use ASSERT_THAT to check pass.Run() result

  #97164 merged Jul 18, 2025

• Update the XNNPack delegate README.

  #97181 merged Jul 18, 2025

• Annotate some XLA:GPU flags as stable i.e. they should provide 6 month deprecation notice.

  #97134 merged Jul 18, 2025

• [XLA:GPU] Add a test for DotForInt4vsIdentityBF16ReturnsCorrectResult.

  #97064 merged Jul 18, 2025

• PR #28985: [XLA:GPU] Add shared_memory_per_block_optin device info member

  #97140 merged Jul 18, 2025

• Update README.md

  #96902 merged Jul 18, 2025

• Update dependencies to XNNPACK.

  #97177 merged Jul 18, 2025

• [XLA:GPU] Move Dot strength reduction out of algebraic simplifier

  #97166 merged Jul 18, 2025

• [XLA:GPU] Remove CHECK-CSE since it is not used.

  #97129 merged Jul 18, 2025

• #sdy improve the error messaging when importing and exporting sharding custom calls.

  #97041 merged Jul 18, 2025

• Introduce new helper function that produces device lists for iota tile assignment. Apply it in xla_sharding_util.cc.

  #97176 merged Jul 18, 2025

• Introduce stable flags and associated deprecation policy for XLA debug options.

  #97049 merged Jul 18, 2025

• Use GetInPlaceInputOutputPairs from AliasInfo instead of HloDataflowAnalysis.

  #97170 merged Jul 18, 2025

• Remove ifdef from ir_emitter_unnested and fix various clang-tidy warnings

  #97127 merged Jul 18, 2025

• Add TmaMetadata serialization support

  #97103 merged Jul 18, 2025

140 Pull requests opened by 8 people

• Remove HLO and Autotuner dependency from CublasLtMatmulThunk

  #97180 opened Jul 18, 2025

• [XLA] Use sort instead of btree in MakeFreeChunks.

  #97185 opened Jul 18, 2025

• Automated Code Change

  #97188 opened Jul 18, 2025

• IFRT proxy logging fix: Do not log error when Executable is destroyed before its metadata is queried by the server (and sent over to the client).

  #97197 opened Jul 18, 2025

• Handle negative permutations in IsTransposeTrivial.

  #97199 opened Jul 18, 2025

• [xla:gpu][triton] In squeeze-dims pass, keep at least two dimensions.

  #97223 opened Jul 19, 2025

• Add a scratch implemention of muon

  #97231 opened Jul 19, 2025

• Automated Code Change

  #97236 opened Jul 19, 2025

• Automated Code Change

  #97237 opened Jul 19, 2025

• Automated Code Change

  #97238 opened Jul 19, 2025

• Add HloInstruction extractor for custom-call kernel metadata.

  #97239 opened Jul 19, 2025

• Automated Code Change

  #97240 opened Jul 19, 2025

• Fix stride overflow, and solution to issue number #97165

  #97260 opened Jul 20, 2025

• Have HloModule::Clone copy layout_canonicalization_callback.

  #97262 opened Jul 20, 2025

• Automated Code Change

  #97267 opened Jul 21, 2025

• Integrate LLVM at llvm/llvm-project@13f7786f72d1

  #97271 opened Jul 21, 2025

• Add an xla flag to lower ragged dots through the ragged dot fusion emitter

  #97278 opened Jul 21, 2025

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

  #97282 opened Jul 21, 2025

• Integrate LLVM at llvm/llvm-project@13f7786f72d1

  #97284 opened Jul 21, 2025

• [HLO Diff] Use full constant values when generating fingerpints.

  #97294 opened Jul 21, 2025

• Update release notes for TensorFlow 2.19.1

  #97297 opened Jul 21, 2025

• [XLA:GPU] Remove horizontal fusion.

  #97301 opened Jul 21, 2025

• PR #28919: [NVIDIA GPU] [XLA_GPU_MS_COLLECTIVE] Round-robin stream assignment for async communications

  #97307 opened Jul 21, 2025

• Adding Metal delegate kernel

  #97312 opened Jul 21, 2025

• Integrate LLVM at llvm/llvm-project@c384ec431dd7

  #97314 opened Jul 21, 2025

• Add gemm_config for xnnpack gemms.

  #97315 opened Jul 22, 2025

• [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 opened Jul 22, 2025

• Integrate LLVM at llvm/llvm-project@c384ec431dd7

  #97322 opened Jul 22, 2025

• Automated Code Change

  #97330 opened Jul 22, 2025

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

  #97340 opened Jul 22, 2025

• Add `_XlaShardingV2` in `TPUPartitionedOps` and use if present for all op's as opposed to just `XlaSharding` op.

  #97348 opened Jul 22, 2025

• Improvements to KernelNameTracer

  #97351 opened Jul 22, 2025

• [XLA:GPU] Enable strength reduction for s32xs32->s32 dots

  #97356 opened Jul 22, 2025

• Internal change only.

  #97365 opened Jul 22, 2025

• [ #HLODiff ]Fix a bug in hlo_diff_summary.cc

  #97367 opened Jul 22, 2025

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

  #97371 opened Jul 22, 2025

• Add float data generation within given range to test utils

  #97372 opened Jul 22, 2025

• Add int8/int16 support for SQRT op to AEQ

  #97373 opened Jul 22, 2025

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

  #97374 opened Jul 22, 2025

• avoid removing some copy kernels.

  #97377 opened Jul 22, 2025

• changes to the build process for litert

  #97385 opened Jul 23, 2025

• Integrate LLVM at llvm/llvm-project@13f7786f72d1

  #97395 opened Jul 23, 2025

• Add an optimization pattern that tranform const<[a, 1]> @ <[1, b]> to <[1, b]> * const<[a, 1]>.

  #97396 opened Jul 23, 2025

• PR #29204: [XLA:GPU] Remove Stream Id from command buffer

  #97397 opened Jul 23, 2025

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

  #97398 opened Jul 23, 2025

• Automated Code Change

  #97399 opened Jul 23, 2025

• PR #26187: Simplify HloComputation::IsFusionComputation semantics.

  #97406 opened Jul 23, 2025

• Replace lite/tools/versioning:versioning with compiler/mlir/lite/tools/versioning:versioning

  #97407 opened Jul 23, 2025

• Automated Code Change

  #97410 opened Jul 23, 2025

• Simplify `get-tuple-element(tuple(A))` as `copy(A)` in mhlo->hlo conversion.

  #97413 opened Jul 23, 2025

• Automated Code Change

  #97416 opened Jul 23, 2025

• [XLA] Improve hlo_live_range_test to use HLO textual form

  #97425 opened Jul 23, 2025

• Expose GPU tracing knobs for 3P

  #97428 opened Jul 23, 2025

• Make CustomCallThunk own the associated HloComputation

  #97429 opened Jul 23, 2025

• [XLA:GPU] Enable deviceless AutotunerPass.

  #97430 opened Jul 23, 2025

• Use file mapping on Windows for the XNNPack weight cache.

  #97433 opened Jul 23, 2025

• Add Multi output reduce test

  #97438 opened Jul 23, 2025

• Implement PjRtClient::MakeCrossHostReceiveBuffers and PjRtBuffer:CopyToRemoteDevice in the PjRt C API.

  #97444 opened Jul 23, 2025

• Add an option to swallow errors in HloRunnerAgnosticTestBase and friends.

  #97446 opened Jul 23, 2025

• Allow MHLO passes to operate on `mlir::FunctionOpInterface` instead of `func::FuncOp`.

  #97449 opened Jul 23, 2025

• [XLA][Numerics][HLO Value Tracking] Support MLIR dumps from Shardy passes in get_hlo_original_value_coverage.sh

  #97452 opened Jul 23, 2025

• MPMD XLA bump

  #97457 opened Jul 23, 2025

• Add JAX tests for deadlock verifier

  #97459 opened Jul 23, 2025

• Add option to enable coordination service client to recover.

  #97460 opened Jul 23, 2025

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

  #97461 opened Jul 23, 2025

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

  #97462 opened Jul 23, 2025

• Add option to enable coordination service client to recover.

  #97463 opened Jul 23, 2025

• Update the `coords` of `PjRtStreamExecutorDeviceDescription`.

  #97465 opened Jul 23, 2025

• Add Pass `AllGatherPadDsSimplifier` to replaces a DynamicSlice(Pad(AllGather)) pattern with a more efficient sequence of HLOs, primarily using CollectivePermute and Concatenate.

  #97466 opened Jul 23, 2025

• Add declarative quantization optimization patterns.

  #97467 opened Jul 23, 2025

• [StableHLO] infra fix

  #97469 opened Jul 23, 2025

• Add quantization utility functions for PropagateQSVPass.

  #97470 opened Jul 23, 2025

• [IFRT Proxy] Bump the minimum version to `kSerDesVersioning`.

  #97472 opened Jul 24, 2025

• 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

  #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

• 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

• Avoid calling `VisibleDeviceCount()` in `BaseGPUDeviceFactory::CreateDevices` if `num_gpus_to_use == 0` so that the underlying platform does not get initialized unnecessarily. Calling `stream_executor::gpu::CudaPlatform::VisibleDeviceCount` initializes CUDA which is then held onto in memory.

  #97546 opened Jul 25, 2025

• Automated Code Change

  #97547 opened Jul 25, 2025

• Automated Code Change

  #97548 opened Jul 25, 2025

• Automated Code Change

  #97549 opened Jul 25, 2025

• Automated Code Change

  #97550 opened Jul 25, 2025

• Automated Code Change

  #97551 opened Jul 25, 2025

• Automated Code Change

  #97552 opened Jul 25, 2025

• Automated Code Change

  #97553 opened Jul 25, 2025

• Explicitly include the header for absl::string_view

  #97554 opened Jul 25, 2025

• Automated Code Change

  #97555 opened Jul 25, 2025

• Automated Code Change

  #97556 opened Jul 25, 2025

• Support bfloat16 type when serializing to numpy

  #97557 opened Jul 25, 2025

• Automated Code Change

  #97558 opened Jul 25, 2025

• Automated Code Change

  #97559 opened Jul 25, 2025

• Automated Code Change

  #97560 opened Jul 25, 2025

• Automated Code Change

  #97561 opened Jul 25, 2025

• Automated Code Change

  #97562 opened Jul 25, 2025

• Automated Code Change

  #97563 opened Jul 25, 2025

• [CUDA] Add new headers to CUDA cupti headers

  #97564 opened Jul 25, 2025

• Automated Code Change

  #97565 opened Jul 25, 2025

• Automated Code Change

  #97566 opened Jul 25, 2025

7 Issues closed by 3 people

• 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

• tensorflow load failed

  #97432 closed Jul 23, 2025

• how to use libtensorflowlite_c.so C API and delegate gpu opencl correctly?

  #95795 closed Jul 23, 2025

• Error in loading Tensorflow in python

  #97345 closed Jul 22, 2025

• Inconsistent NotEqual broadcasting behavior between CPU and GPU (CPU fails silently, GPU raises error)

  #97227 closed Jul 19, 2025

6 Issues opened by 6 people

• tf.no_op DAG cycle issue

  #97512 opened Jul 24, 2025

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

  #97403 opened Jul 23, 2025

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

  #97387 opened Jul 23, 2025

• Request for Python 3.13 support in TensorFlow

  #97361 opened Jul 22, 2025

• Inconsistent behavior for `tf.raw_ops.NotEqual` between CPU and GPU with non-broadcastable shapes

  #97204 opened Jul 18, 2025

• could you add support of the new optimizer: Muon

  #97187 opened Jul 18, 2025

48 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.

• Add a Reflection Map to `emitc` class

  #96263 commented on Jul 25, 2025 • 11 new comments

• Core Dump When Training

  #97016 commented on Jul 25, 2025 • 0 new comments

• lib new version not support 16kb pages in android

  #96602 commented on Jul 25, 2025 • 0 new comments

• Fix compile error in tensorflow/python/tfcompile_wrapper.cc on s390x

  #87676 commented on Jul 21, 2025 • 0 new comments

• Move duplicate CUDA/XLA registration logs from INFO to VLOG

  #89808 commented on Jul 21, 2025 • 0 new comments

• Fix comparison functions and add unit tests

  #94484 commented on Jul 21, 2025 • 0 new comments

• Update Protobuf to 6.31.1

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

• Remove LiteRT modules from TF python deps.

  #95991 commented on Jul 22, 2025 • 0 new comments

• [XLA:SCHEDULING] Defer scheduling of allocate buffer custom calls

  #96110 commented on Jul 25, 2025 • 0 new comments

• Construct XLA GPU client with coordination service client.

  #96210 commented on Jul 23, 2025 • 0 new comments

• Add validation for label probability distribution in softmax_cross_entropy_with_logits

  #96387 commented on Jul 21, 2025 • 0 new comments

• Call `UpdateGlobalProcessInfo` from PjRt IFRT client.

  #96494 commented on Jul 23, 2025 • 0 new comments

• Fixes L2Pool implementation to not average pooling region squares

  #96599 commented on Jul 24, 2025 • 0 new comments

• PR #19067: [XLA:CPU][oneDNN] Move simplification pass before oneDNN pass

  #96617 commented on Jul 19, 2025 • 0 new comments

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

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

• [TFLite] Support 16KB page sizes alignment for libtensorflowlite_gpu_gl.so

  #96702 commented on Jul 21, 2025 • 0 new comments

• Automated Code Change

  #96929 commented on Jul 23, 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

• [ #HLODiff ] Add support for manual node matching.

  #97060 commented on Jul 22, 2025 • 0 new comments

• PR #28883: [XLA:CPU][oneDNN] Add build flag to enable asynchronous support in oneDNN

  #97115 commented on Jul 19, 2025 • 0 new comments

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

  #97118 commented on Jul 25, 2025 • 0 new comments

• Add Metal LiteRt Tensor Buffer support

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

• [XLA:MSA] Reduce available memory bandwidth for instruction that are overlapped with bandwidth limiting asynchronous instructions.

  #97163 commented on Jul 21, 2025 • 0 new comments

• [Compatibility][Upgrade] TensorFlow 2.x to 2.15.0: Dependency Conflict and Version Downgrade Issue

  #96694 commented on Jul 19, 2025 • 0 new comments

• Build Error While Compiling TensorFlow Lite Using CMake

  #96654 commented on Jul 19, 2025 • 0 new comments

• Memory leak in tf.data when iterating over Dataset.from_generator

  #65675 commented on Jul 19, 2025 • 0 new comments

• TensorFlow on RTX 5090

  #89272 commented on Jul 19, 2025 • 0 new comments

• GPU Not Detected by TensorFlow Despite Proper System Setup

  #96707 commented on Jul 22, 2025 • 0 new comments

• Some sorting related ops produce results inconsistent with NumPy when tensor contains NaN

  #95235 commented on Jul 22, 2025 • 0 new comments

• tf.data.experimental.prefetch_to_device has no effect inside tf.distribute.Strategy.distribute_datasets_from_function.

  #94735 commented on Jul 22, 2025 • 0 new comments

• `tf.linalg.solve` behaves inconsistently on GPU for singular matrices depending on shape

  #94657 commented on Jul 22, 2025 • 0 new comments

• Convolution: CPU memory increase with growing number of different sequence lengths

  #62441 commented on Jul 23, 2025 • 0 new comments

• Remove or update zh-cn translation from installation instructions

  #62245 commented on Jul 23, 2025 • 0 new comments

• tf.strings.to_number cannot convert positive integers prefixed with "+" when out_type is tf.int32 or tf.int64

  #62191 commented on Jul 23, 2025 • 0 new comments

• failed to build branch r2.13

  #60716 commented on Jul 23, 2025 • 0 new comments

• Incorrect gradient in divide_no_nan and reciprocal_no_nan when divide by 0

  #60715 commented on Jul 23, 2025 • 0 new comments

• `tf.split` or `tf.transpose` cause errors for quantize-aware training with `quantize_apply`

  #60714 commented on Jul 23, 2025 • 0 new comments

• Dataset.ragged_batch does not produce correct specs with tf.py_function and tf.numpy_function

  #60710 commented on Jul 23, 2025 • 0 new comments

• `tf.math.argmax` throws `InvalidArgumentError` with valid `axis` of `int16` dtype

  #97096 commented on Jul 23, 2025 • 0 new comments

• Will TF supprot triton at future

  #96876 commented on Jul 23, 2025 • 0 new comments

• `tf.linalg.matrix_rank` produces inconsistent output on CPU vs GPU with `tol=6`

  #97102 commented on Jul 23, 2025 • 0 new comments

• `tf.nn.local_response_normalization` returns incorrect output

  #97105 commented on Jul 23, 2025 • 0 new comments

• `tf.pow` returns inconsistent value on CPU vs GPU

  #97125 commented on Jul 23, 2025 • 0 new comments

• `tf.nn.depthwise_conv2d` crashes with large `strides` values when ONEDNN is enabled

  #97165 commented on Jul 23, 2025 • 0 new comments

• 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