208 Pull requests merged by 5 people

• Change AsyncRunTask to be move-only.

  #95365 merged Jun 14, 2025

• [IFRT] Use SerDes versioning in most IFRT and PjRt-IFRT types

  #94890 merged Jun 14, 2025

• Return error when non-zero checks fail for div.

  #95290 merged Jun 14, 2025

• #HLODiff Add a new function that group instructions by opcode and then call bipartite MatchSameTypeInstructions for each type.

  #95218 merged Jun 14, 2025

• Use blocking_thread_pool for H2D transfer

  #95358 merged Jun 14, 2025

• Reverts 75aa15587bec5fd8b24c22b21f8d6c0ed6726f86

  #95363 merged Jun 13, 2025

• Record subphase action information in HloRunnerPjRt.

  #94700 merged Jun 13, 2025

• Add precise resource calculation for scheduling groups with "keep_original_sequence_order_in_group" attribute.

  #95357 merged Jun 13, 2025

• Update dependencies for XNNPACK and KleidiAI.

  #95259 merged Jun 13, 2025

• Update visibility for python/data/experimental/ops/global_shuffle_op.

  #95291 merged Jun 13, 2025

• Fix scheduling of instructions simplified by tuple simplifier in the presence of control predecessors by assigning replacement instructions to be rescheduled.

  #95298 merged Jun 13, 2025

• [IFRT] Change DeviceTest to DeviceTestFixture

  #95353 merged Jun 13, 2025

• [XLA:CPU] Emit xla.exp from legacy pipeline.

  #95090 merged Jun 13, 2025

• [XLA:benchmarks] Create an HLO folder for benchmarking and add a sample hlo

  #95355 merged Jun 13, 2025

• Reverts 0acc54bd1bb16186eb04b4681395b58935da52d0

  #95350 merged Jun 13, 2025

• Delete all uses of test_macros.h as this file is now empty.

  #95284 merged Jun 13, 2025

• Rename options to options_ptr in tflite::xnnpack::Delegate::Delegate().

  #94957 merged Jun 13, 2025

• Added heartbeat_timeout argument.

  #95292 merged Jun 13, 2025

• Add the FusedConv2DBiasActivation op tf dialect.

  #95147 merged Jun 13, 2025

• Remove lite/quantization/ir:QuantOps from tf_quant_opt.cc

  #95285 merged Jun 13, 2025

• Change SerDesVersion's enum to SerDesVersionNumber

  #95288 merged Jun 13, 2025

• [XLA conditional code motion] Pass accept_different_shape = true in call to set_root_instruction to handle case where output of conditional isn't a tuple.

  #95344 merged Jun 13, 2025

• Update dynamic shape support to be based on raw buffers.

  #95224 merged Jun 13, 2025

• [pjrt] Improved CUDA detection in the Triton PjRt extension

  #95338 merged Jun 13, 2025

• [XLA:GPU] Add the Triton emitter support for pad in nested fusions.

  #95267 merged Jun 13, 2025

• PR #27481: [ROCm] enable hidden unit tests on rocm-1

  #95337 merged Jun 13, 2025

• PR #27266: [ROCm] analytical latency estimator support for rocm

  #95187 merged Jun 13, 2025

• [XLA:GPU] Fix an issue with not resetting preferred_consumer_

  #95334 merged Jun 13, 2025

• [XLA:GPU] Add scoped timer to HLO execution in functional HLO runner.

  #95279 merged Jun 13, 2025

• Cache invalidation for XProf

  #93080 merged Jun 13, 2025

• [XLA:GPU] Add separate expected precision tolerance for BLAS.

  #95330 merged Jun 13, 2025

• Reduce severity of logging in StreamExecutorGpuClient::gpu_run_option.

  #95331 merged Jun 13, 2025

• PR #26948: SPMD Partial Windowed Einsums

  #95191 merged Jun 13, 2025

• [XLA:GPU] Improve dot algorithm precision tests with reference computation

  #95328 merged Jun 13, 2025

• PR #27737: [ROCm] fixed gpu_hlo_unoptimized_llvm.hlo.test on rocm

  #95287 merged Jun 13, 2025

• PR #26632: Add support for NVSHMEM collective_permute

  #95263 merged Jun 13, 2025

• Move Shardy mesh inlining to right after deserialization.

  #95281 merged Jun 13, 2025

• [XLA:GPU] Return empty vector instead of InvalidArgument error for GetSupportedConfig if the instruction is not supported.

  #95273 merged Jun 13, 2025

• Replace outdated select() on --cpu in tensorflow/BUILD and related files with platform API equivalent.

  #94745 merged Jun 13, 2025

• Extract HloOperandIndex into its own target (NFC).

  #95323 merged Jun 13, 2025

• Automated Code Change

  #95229 merged Jun 13, 2025

• strings_ops: Test "invalid" unicode_encode/unicode_decode inputs

  #95321 merged Jun 13, 2025

• Replace deprecated tsl::errors::Internal with absl::InternalError

  #95225 merged Jun 13, 2025

• Add missing rocm_headers dependency in rocm_library macro

  #95276 merged Jun 13, 2025

• [IFRT] Add empty method to base DeviceList.

  #95289 merged Jun 13, 2025

• [DOC] Update Docker image in build_from_source and dev_guide docs.

  #95220 merged Jun 13, 2025

• Refactor subprocess_compilation target.

  #95063 merged Jun 13, 2025

• Refactor CopyRawDeviceToHost into CopyRawDeviceToHostAndReturnEvent

  #95201 merged Jun 13, 2025

• Always use staging buffer for ToLiteral.

  #95198 merged Jun 13, 2025

• Don't reshape resource tensors in XNNPACK

  #95283 merged Jun 13, 2025

• Use incarnation ids in GPU clique keys.

  #93271 merged Jun 12, 2025

• cleanup for attrs_and_constraints target

  #95129 merged Jun 12, 2025

• Replace DISABLED_ON_TPU with test::DeviceTypeIs(test::kTpu) in xla

  #95221 merged Jun 12, 2025

• Set --gunit_fail_if_no_test_selected for XLA.

  #95065 merged Jun 12, 2025

• #sdy Move unreduced frontend attribute to xla constants and clean up.

  #95202 merged Jun 12, 2025

• Reverts 693e65e93e8f547d42ed01816bf5f56175b191bc

  #95227 merged Jun 12, 2025

• Replace NVSHMEM stub with hermetic library in the target dependencies when --@local_config_nvshmem//:include_nvshmem_libs=True.

  #95150 merged Jun 12, 2025

• Update xprof to 2.20.0

  #95179 merged Jun 12, 2025

• Don't record callstack for XLA:CPU in OSS since streamz doesn't work there

  #95207 merged Jun 12, 2025

• LiteralUtil: add method to get literal pointers

  #95258 merged Jun 12, 2025

• [XLA:GPU] Make ScopedLoggingTimer also logs the elapsed time in micros.

  #95270 merged Jun 12, 2025

• [xla:gpu] Add blas backends to the precision tests

  #95272 merged Jun 12, 2025

• PR #27188: [XLA:GPU] Hermetic Build oneAPI

  #95199 merged Jun 12, 2025

• [xla:cpu] Do not fuse too many reductions together

  #95223 merged Jun 12, 2025

• [XLA:CPU] Add reduce-window and reduce over outer dimension microbenchmarks.

  #95266 merged Jun 12, 2025

• [XLA:GPU]: Use Store/Load with a counter instead of CAS for all-reduce

  #95189 merged Jun 12, 2025

• Bump the version of StableHLO and VHLO indicating that mixed serialization is supported.

  #95256 merged Jun 12, 2025

• Clean up dependencies of ROCm targets

  #95252 merged Jun 12, 2025

• Create separate methods for creating kernel from ptx and cubin

  #95251 merged Jun 12, 2025

• [XLA:GPU] print histogram of the relative errors in the Triton dot algorithm test.

  #95257 merged Jun 12, 2025

• PR #27700: [debug-options-dump] Fix dumping for repeated string fields

  #95197 merged Jun 12, 2025

• PR #27721: Always evaluate while init in loop analysis.

  #95250 merged Jun 12, 2025

• Automated Code Change

  #95238 merged Jun 12, 2025

• PR #23688: [ROCm] Triton performance fixes

  #95255 merged Jun 12, 2025

• Remove unused ForwardsValue hook (NFC).

  #95246 merged Jun 12, 2025

• Internal change only

  #95254 merged Jun 12, 2025

• PR #27718: Adds latency hiding scheduling support for DUS s5

  #95226 merged Jun 12, 2025

• PR #27128: Make copy_fusion respect the budget limit

  #95186 merged Jun 12, 2025

• PR #27680: Build nvshmem on arm64 machines

  #95237 merged Jun 12, 2025

• Work around a compiler issue in GCC 8

  #95233 merged Jun 12, 2025

• Automated Code Change

  #95161 merged Jun 12, 2025

• Add a shared, SchedulingContext class to pass around the commonly-used objects for scheduling, to help clean up code as well as avoid computation associated with re-creation on demand.

  #95062 merged Jun 12, 2025

• Bump requests from 2.32.3 to 2.32.4

  #95112 merged Jun 12, 2025

• Automated Code Change

  #95167 merged Jun 12, 2025

• Add retry_on_oom paramater to AllocateRawBuffer to allow

  #95088 merged Jun 12, 2025

• Integrate LLVM at llvm/llvm-project@842377882a3f

  #95174 merged Jun 12, 2025

• [XLA] Disable conditional code motion with outfeed instruction that tend to be used for summaries that extend live ranges.

  #95206 merged Jun 12, 2025

• remove old copy of quantization_lib and rename new one

  #95085 merged Jun 12, 2025

• Remove all uses of XLA_TYPED_TEST

  #95211 merged Jun 11, 2025

• Support QAT aware conversion with dynamic shape models

  #94830 merged Jun 11, 2025

• Revamp StableHLO folder patterns to use MLIR folding infra, to not expand splats.

  #95154 merged Jun 11, 2025

• #sdy Skip all-reduce along sharding dimension on ops with unreduced frontend attribute.

  #94903 merged Jun 11, 2025

• ifdef guard recording GPU compilation stacks to Google

  #95204 merged Jun 11, 2025

• Add DeviceTypeIs to xla_test_backend_predicates to replace XLA_TEST_BACKEND_TPU and similar

  #95209 merged Jun 11, 2025

• [IFRT IR] Add method for fingerprinting ModuleOp.

  #95196 merged Jun 11, 2025

• Add hlo_module_name parameter to Compiler::CreateMetricsHook to make available the hlo module name of the recorded hlo program in the logging infrastructure.

  #95194 merged Jun 11, 2025

• #sdy Mark unreduced ops with frontend attributes.

  #94979 merged Jun 11, 2025

• Implement Shardy to HLO transformation

  #95200 merged Jun 11, 2025

• [XLA] Early reject for Reduce Window Rewriter to eliminate cases faster

  #95145 merged Jun 11, 2025

• [GPU] Replace CudnnLegacyFusedConvRunner with CudnnExecutionPlanRunner

  #95078 merged Jun 11, 2025

• Optimize rotation loop during Eval

  #95159 merged Jun 11, 2025

• Extract hlo input / output format enums to separate files outside of functional_hlo_runner.

  #94136 merged Jun 11, 2025

• Add user-agent for GCS C-API

  #95131 merged Jun 11, 2025

• [XLA] Allow ReshapeMover to perform chain moves.

  #95148 merged Jun 11, 2025

• Fix xnnpack-delegate performance regression

  #94944 merged Jun 11, 2025

• [XLA:GPU]: Use double buffering for the one-shot all-reduce kernel

  #95004 merged Jun 11, 2025

• [XLA:GPU] NFC sort ops in support test

  #95190 merged Jun 11, 2025

• [XLA:GPU]: Add tests for all-reduce within a while-loop

  #94934 merged Jun 11, 2025

• [xla:gpu] NestGemmFusion: also hoist bitcasts downwards across transposes/broadcasts.

  #94866 merged Jun 11, 2025

• [XLA:GPU] update symbolic tile analysis to propagate rutime variables

  #95123 merged Jun 11, 2025

• [XLA:GPU][Emitters] Avoid using tiled transpose when transposing the last dimension if one of the dimensions is too small

  #95180 merged Jun 11, 2025

• [XLA:GPU] Convert RendezvousArg pair to a struct.

  #95182 merged Jun 11, 2025

• PR #27635: [ROCm] enable hidden unit tests on rocm-2

  #95181 merged Jun 11, 2025

• [XLA:CPU] Create memory mappers named after LLVM modules.

  #95055 merged Jun 11, 2025

• Make GpuTestKernels use the GpuKernelRegistry

  #95109 merged Jun 11, 2025

• Enable cublas backend. Skip codegen backends which do not support an instruction.

  #95116 merged Jun 11, 2025

• [XLA:CPU] Save obj file names in the executable proto

  #95057 merged Jun 11, 2025

• PR #27498: [ROCm] Add new hip_runtime bazel target

  #95111 merged Jun 11, 2025

• [XLA:CPU] Block fusions of subcomputations if the parent can be fused.

  #95117 merged Jun 11, 2025

• fix HloToStablehlo copy of backend_config into frontend_attributes

  #95139 merged Jun 11, 2025

• [XLA:GPU] Return empty vector instead of throwing and error in the Triton autotuner backend.

  #95178 merged Jun 11, 2025

• PR #27371: PJRT_Executable_DeserializeAndLoad: plumb compile options

  #95175 merged Jun 11, 2025

• [XLA:CPU][autotuning] LLVM kernel autotuner implementation

  #94880 merged Jun 11, 2025

• [XLA:CPU] Fix order of lowered work item id

  #94941 merged Jun 11, 2025

• Automated Code Change

  #95165 merged Jun 11, 2025

• Remove unused multi PTX feature from CudaPtxInMemory

  #95125 merged Jun 11, 2025

• Automated Code Change

  #95169 merged Jun 11, 2025

• Make IFRT Proxy create device list using the underlying IFRT Client's MakeDeviceList

  #95172 merged Jun 11, 2025

• Allow out of bounds strided slice ops to delegate

  #95144 merged Jun 11, 2025

• [XLA:Collective] Fix reduction computation type bug in while_loop_all_reduce_code_motion_setup.

  #95082 merged Jun 11, 2025

• Reverts a854d8322fa93c9b644e3474fe0e4866d769669b

  #94713 merged Jun 11, 2025

• Reverts 667a7af1e172dec7fd24718c15dc9e961d990384

  #95155 merged Jun 11, 2025

• [XLA] Make EraseElementFromVector return a bool instead of a status.

  #95153 merged Jun 11, 2025

• remove lite quant deps from tfr-opt and use tf quant deps instead

  #95143 merged Jun 10, 2025

• cleanup for lift_as_function_call

  #95126 merged Jun 10, 2025

• Bump googletest revision to pick up new fail_if_no_test_selected flag.

  #95146 merged Jun 10, 2025

• Elide gpu_ prefix from GPU backends in xla_test

  #94752 merged Jun 10, 2025

• [XLA:GPU] Make functional runner respect disabled SPMD partitioning

  #95067 merged Jun 10, 2025

• Add num_sparsecores_per_device attribute to custom combiner BWD ops.

  #95093 merged Jun 10, 2025

• [XLA] Refactoring for Reduce Window Rewriter

  #95073 merged Jun 10, 2025

• [XLA] Make reshape-mover a bit less dependent on algsimp.

  #95134 merged Jun 10, 2025

• Reverts cc47e60b44b55fab0a7731b5e1f411b67aef57b4

  #94948 merged Jun 10, 2025

• [XLA] Remove XLA_TEST_P and XLA_TEST_F

  #95135 merged Jun 10, 2025

• Use NVSHMEM tar files in RBE CUDA builds for XLA and Tensorflow.

  #95140 merged Jun 10, 2025

• Add helper for stripping out dynamic shape metadata for a raw buffer.

  #95091 merged Jun 10, 2025

• Account for control dependencies in HloSchedule::Update().

  #95074 merged Jun 10, 2025

• Fix variable ops in the flex delegate

  #95009 merged Jun 10, 2025

• Move memory space clearance to right before host offloader work

  #93890 merged Jun 10, 2025

• Add last_checkpoint_step parameter to CheckpointManager init.

  #95136 merged Jun 10, 2025

• Fix ops attempting to handle filter and bias of mismatched floating point types

  #95054 merged Jun 10, 2025

• #HLODiff Extends HloComputationGraphMatcher to match leaf nodes other than constants and parameters.

  #94760 merged Jun 10, 2025

• Extend tf2xla to support Shardy.

  #94489 merged Jun 10, 2025

• Add python extension for registering AutoSharding to XLA pipeline.

  #95133 merged Jun 10, 2025

• Improve device assignment string format and add AbslStringify implementation.

  #94882 merged Jun 10, 2025

• Reverts 9882facbb94c0b0e5a07bc38e445cd177440c9e9

  #95128 merged Jun 10, 2025

• When subgraph reshaping is enabled, we need to clear all the externals when reshaping, not just the inputs.

  #95053 merged Jun 10, 2025

• Reverts fc61865027fbe24c6755f71e979b78b2474db1e9

  #94980 merged Jun 10, 2025

• Build python3.14.0b1 from source and install it in the linux docker images

  #95068 merged Jun 10, 2025

• cleanup for fake_quant_utils

  #95124 merged Jun 10, 2025

• [XLA] Replace XLA_TEST_P with TEST_P

  #94972 merged Jun 10, 2025

• [XLA] Replace XLA_TEST_P with TEST_P

  #94975 merged Jun 10, 2025

• [XLA] Replace XLA_TEST_F with TEST_F

  #94968 merged Jun 10, 2025

• [XLA] Replace XLA_TEST_F with TEST_F

  #94962 merged Jun 10, 2025

• Remove unused LlvmHostKernel

  #95122 merged Jun 10, 2025

• Fix CopyToMemorySpace typo bug.

  #95076 merged Jun 10, 2025

• [XLA:CPU] Add ExpF64Avx512 benchmark.

  #95081 merged Jun 10, 2025

• Move REVERSE_V2 implementation to separate header file.

  #94856 merged Jun 10, 2025

• [XLA:CPU] Fix xla math lib to scan for vectorized function names.

  #95080 merged Jun 10, 2025

• [XLA:GPU] Add proto serialization for PartitionIdThunk

  #94659 merged Jun 10, 2025

• [XLA:GPU] indexing map serialization: print undefined + tests

  #94809 merged Jun 10, 2025

• [XLA] Remove merging pad into reduce-window - because of creating unsual patterns

  #94924 merged Jun 10, 2025

• Sync patches to unbreak windows build

  #95118 merged Jun 10, 2025

• PR #27537: [run-hlo-module] Fix the debug options test

  #95070 merged Jun 10, 2025

• PR #27596: Bump github/codeql-action from 3.28.10 to 3.28.19

  #95110 merged Jun 10, 2025

• [XLA:GPU] Run SoL cost model for -On.

  #94861 merged Jun 10, 2025

• PR #26268: [ROCm] Introduce xla_gpu_use_inprocess_lld to invoke ldd as a library

  #94923 merged Jun 10, 2025

• Better error messages for thunk deserialization errors

  #94867 merged Jun 10, 2025

• #sdy Replace manual axes that are bound to a duplicate mesh in SdyRoundTripDedupMeshesPass

  #94936 merged Jun 10, 2025

• Add serialization support for DeviceToDeviceCopyThunk

  #94862 merged Jun 10, 2025

• Adds interface to update target specific states in latency hiding scheduler.

  #94763 merged Jun 10, 2025

• Remove unused Bazel macro build_cub_sort_kernels

  #95108 merged Jun 10, 2025

• Make the kv store timeout configurable for cross-host device transfers.

  #95069 merged Jun 10, 2025

• Remove tensorflow-intel from Linux wheel metadata

  #94854 merged Jun 10, 2025

• Improve error handling in transmission of buffer metadata for experimental cross-host device transfers.

  #94685 merged Jun 10, 2025

• Integrate LLVM at llvm/llvm-project@649020c68016

  #95064 merged Jun 10, 2025

• * Support buffer coloring by updating MSA behavior:

  #94483 merged Jun 10, 2025

• Move trace event after we select the device

  #95084 merged Jun 10, 2025

• [XLA] Better logs for cycle detector for scheduling groups

  #95083 merged Jun 10, 2025

• [XLA] Enable more ops for conditional code motion and only deal with the cost model when moving reduction opeations

  #94901 merged Jun 10, 2025

• Fix buffer overflow bug in HloLexer::LexInt64Impl and add regression tests.

  #94982 merged Jun 10, 2025

• [XLA] Handle more cases in IotaTileAssignment::Transpose.

  #94947 merged Jun 10, 2025

• Integrate hermetic nvshmem repository in XLA and TF projects.

  #94894 merged Jun 9, 2025

• Remove redundant build dep.

  #95077 merged Jun 9, 2025

• Uses the correct SparseCore count for custom combiner BWD op in AoT compilation.

  #95071 merged Jun 9, 2025

• remove old copy of uniform_op_quant_spec, tf_to_uniform_attribute_utils, tf_op_quant and rename new one

  #94754 merged Jun 9, 2025

• remove old copy of fuse_convolution_pass and rename new one

  #94321 merged Jun 9, 2025

• Fix TF nightly auditwheel repair due to pywrap

  #94921 merged Jun 9, 2025

• Update rules_python patch file to get python 3.14.0b1

  #95066 merged Jun 9, 2025

• remove old copy of constant_fold and rename new one

  #94813 merged Jun 9, 2025

• Remove all uses of DISABLED_ON_DEBUG and delete it. We can check if NDEBUG is set directly

  #94978 merged Jun 9, 2025

• [tosa] Support variable bias for convolutional ops

  #94438 merged Jun 9, 2025

• Remove all uses of DISABLED_ON_INTERPRETER_TSAN

  #94976 merged Jun 9, 2025

• Add a test case in collective pipeliner that runs backward & forward passes back-to-back.

  #94974 merged Jun 9, 2025

• Remove TestPlatform from test_macros.{cc,h} as it is unused

  #94981 merged Jun 9, 2025

• #sdy don't remove size 1 axes from manual axes, so a fully manual computation remains fully manual after this pass.

  #95056 merged Jun 9, 2025

• Add support for converting sdy.reduce_scatter into (1 or more) stablehlo.reduce_scatter.

  #94938 merged Jun 9, 2025

• Files should be open with O_BINARY on Windows.

  #95010 merged Jun 8, 2025

• Introduces an IOPDDL-based implementation of the heuristic solver(s).

  #94942 merged Jun 7, 2025

102 Pull requests opened by 5 people

• Automated Code Change

  #95026 opened Jun 8, 2025

• Automated Code Change

  #95028 opened Jun 8, 2025

• Fix: Safely Capture and Store TF Operation Stack Trace at Creation to Prevent Dangling Reference Errors

  #95034 opened Jun 8, 2025

• Automated Code Change

  #95044 opened Jun 9, 2025

• Automated Code Change

  #95045 opened Jun 9, 2025

• Automated Code Change

  #95046 opened Jun 9, 2025

• Cleanup: rename from tensorflow_stats to framework_op_stats

  #95072 opened Jun 9, 2025

• [Phase Compilation] Part-1: PJRT extensions to implement phase compilation.

  #95079 opened Jun 9, 2025

• Add an API to overwrite the current execution_stream_id and respect it in XLA CPU dispatch.

  #95086 opened Jun 10, 2025

• [Phase Compilation] Part-2: Introduces xla::PjRtPhaseCompiler

  #95087 opened Jun 10, 2025

• [Phase Compilation] Part-3: Add C++ layers to test and interact with C PJRT API.

  #95089 opened Jun 10, 2025

• [xla:gpu] Reimplement FindBlockLevelParameters()

  #95107 opened Jun 10, 2025

• PR #27481: [ROCm] enable hidden unit tests on rocm-1

  #95115 opened Jun 10, 2025

• Adjust passes that deal with aliasing logic to have a callback.

  #95119 opened Jun 10, 2025

• Replace outdated select() on --cpu in tensorflow/BUILD and related files with platform API equivalent.

  #95120 opened Jun 10, 2025

• [XLA:benchmarks] Add a filter to skip blocking performance presubmit

  #95132 opened Jun 10, 2025

• [XLA][Numerics][HLO Value Tracking] Uses std::unique_ptr for reference to the original value in HloInstruction

  #95138 opened Jun 10, 2025

• This change updates XLA and sets the default CUDA and cuDNN versions to 12.8.0 and 9.8.0, respectively, in all configurations.

  #95141 opened Jun 10, 2025

• Reverts 18d3864ae7b5951d10389fcffd15780b544f6cd1

  #95156 opened Jun 10, 2025

• Enhance hadamard_rotation_test before optimizing rotation algorithm.

  #95157 opened Jun 10, 2025

• Register ChloDialect in tf_tfl_translate.cc

  #95158 opened Jun 10, 2025

• Enable subgraph reshaping by default in XNNPACK delegate

  #95173 opened Jun 11, 2025

• Integrate Triton up to [0a4aa696](https://github.com/openai/triton/commits/0a4aa6960599b17290eb942d71d14afc1775b175)

  #95177 opened Jun 11, 2025

• Add AliasHints class that will replace the separate alias hint hooks.

  #95183 opened Jun 11, 2025

• [XLA:GPU] Legalize dot precision into casts+algorithm.

  #95188 opened Jun 11, 2025

• Commented out the legacy test files. They are not buildable.

  #95192 opened Jun 11, 2025

• Changing the evaluator to add support for a custom call

  #95195 opened Jun 11, 2025

• Allow TFL to TOSA pipeline to load TFL dialect

  #95205 opened Jun 11, 2025

• Refactor `custom_call` to use common `FindCudaExecutable` method from XLA repository to find CUDA binaries.

  #95208 opened Jun 11, 2025

• Port recent MHLO changes to StableHLO optimization path.

  #95210 opened Jun 11, 2025

• Break sparse tensors to find users

  #95213 opened Jun 11, 2025

• [XLA:benchmarks] Test onboard a new hlo from repo path

  #95215 opened Jun 11, 2025

• add dynamic registration helper

  #95216 opened Jun 11, 2025

• Reserve the last 100 custom XPlane IDs for NCCL Net Plugin.

  #95217 opened Jun 11, 2025

• #HLODiff Add a BipartiteTopDownMatcher after strict GreedyTopDownMatcher

  #95219 opened Jun 11, 2025

• Integrate LLVM at llvm/llvm-project@02550da93291

  #95234 opened Jun 12, 2025

• Simplify MultiKernelLoaderSpec

  #95260 opened Jun 12, 2025

• [PROTOTYPE] Cleanup TFL dependencies in tosa

  #95262 opened Jun 12, 2025

• Add the method to Autotune HloModule.

  #95264 opened Jun 12, 2025

• Split `ImplicitArithOpBuilder` into its own target.

  #95265 opened Jun 12, 2025

• [XLA:CPU] Add pass to rewrite f32 <-> bf16 conversions.

  #95268 opened Jun 12, 2025

• Introduce repo environment variable CUDA_EXTRA_COPTS

  #95269 opened Jun 12, 2025

• [XLA:CPU] Disable loop unrolling for certain reduce operations.

  #95271 opened Jun 12, 2025

• [XLA:benchmarks] Add README guide for onboarding new benchmarks to OpenXLA.

  #95275 opened Jun 12, 2025

• Implement HLO to Shardy transformation.

  #95277 opened Jun 12, 2025

• Use TensorShape instead of RuntimeShape in TF.

  #95278 opened Jun 12, 2025

• [XLA:benchmarks] Upload performance regression in presubmit to GCS buckets for tracking

  #95280 opened Jun 12, 2025

• [xla:gpu] Nest gemm fusion: only hoist bitcasts upwards.

  #95282 opened Jun 12, 2025

• Correct the int type of `output_id` in NanoRT IFRT Client.

  #95286 opened Jun 12, 2025

• Remove old heartbeat flags and arguments.

  #95293 opened Jun 12, 2025

• Remove old heartbeat options.

  #95294 opened Jun 12, 2025

• Set heartbeat_timeout argument and flag.

  #95295 opened Jun 12, 2025

• Use heartbeat_timeout argument.

  #95296 opened Jun 12, 2025

• Test fusion model for an operand not belonging to fusion op.

  #95297 opened Jun 12, 2025

• Upgrade TF_SYSROOT for rbe_linux_cpu to /dt10

  #95299 opened Jun 13, 2025

• Change ownership of the file descriptor from the weight cache builder to the provider.

  #95300 opened Jun 13, 2025

• Test CPU build

  #95301 opened Jun 13, 2025

• [XLA] Refactor if-else chain into a switch.

  #95302 opened Jun 13, 2025

• Add Hermetic C++ Toolchains for XLA project.

  #95304 opened Jun 13, 2025

• Experiment!

  #95305 opened Jun 13, 2025

• Support for nested while loops in while_loop_unroller.

  #95306 opened Jun 13, 2025

• Add AcquireScopedRawBuffer(...) to CommonPjRtBuffer which

  #95307 opened Jun 13, 2025

• Integrate LLVM at llvm/llvm-project@8890706db673

  #95319 opened Jun 13, 2025

• Automated Code Change

  #95320 opened Jun 13, 2025

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

  #95322 opened Jun 13, 2025

• Automated Code Change

  #95325 opened Jun 13, 2025

• PR #27784: Use 256 byte alignment to avoid breakages starting with cublas 12.9.1.4.

  #95329 opened Jun 13, 2025

• [XLA:CPU] Naming module memory regions after emitters that produced the modules.

  #95333 opened Jun 13, 2025

• Fix subprocess.check_output decoding issue in pip_smoke_test.py to handle byte output safely

  #95335 opened Jun 13, 2025

• [XLA:GPU] NFC optional separator when printing tiled HLO instruction and tiling

  #95336 opened Jun 13, 2025

• [XLA:GPU] Flip LHS SoL by default on Hopper and supported HLOs.

  #95339 opened Jun 13, 2025

• [XLA:GPU] Extract dynamic slicing related utils into a separate file.

  #95340 opened Jun 13, 2025

• Migrate MultiKernelLoaderSpec users to the new APIs

  #95341 opened Jun 13, 2025

• [stablehlo] Update StablehloToLinalgRandom to use stablehlo.reshape instead of getReassociationIndicesForCollapse

  #95342 opened Jun 13, 2025

• Remove the tfl dialect dependency from tf-tfrt-opt.

  #95343 opened Jun 13, 2025

• #sdy When exporting `sdy.sharding_constraint` to custom call, mark it as side effecting so it won't get CSEd.

  #95345 opened Jun 13, 2025

• [XLA:CPU] Ensure that the work splitting is done on the outer dimension

  #95346 opened Jun 13, 2025

• [XLA:CPU] Use legacy fusion for dot fusion.

  #95347 opened Jun 13, 2025

• [XLA:CPU][XLA:GPU] Add nuw to add/mul when affine map is lowered

  #95348 opened Jun 13, 2025

• Split ImplicitArithOpBuilder into its own target.

  #95349 opened Jun 13, 2025

• Extend the CUDA root candidates and add `FindNvdisasmExecutable` to `subprocess_compilation.cc`.

  #95351 opened Jun 13, 2025

• Delete `test_macros.h` and final remaining uses

  #95352 opened Jun 13, 2025

• Adds flags for NCCL non-blocking communicators and async execution.

  #95354 opened Jun 13, 2025

• [xla:cpu] Deprecate API_VERSION_STATUS_RETURNING_UNIFIED custom calls

  #95356 opened Jun 13, 2025

• Only abort collectives on failure.

  #95359 opened Jun 13, 2025

• Load PyInfo from rules_python (Attempt 2)

  #95360 opened Jun 13, 2025

• Add `ShouldWarmupAllBatchSizes` overload that accepts name/version directly

  #95361 opened Jun 13, 2025

• na na na

  #95362 opened Jun 13, 2025

• Introduce the flag `--@xla//xla/tsl/platform/default:cuda_rpath` that controls whether we set rpath linker flags for use with nvidia wheel-packaged libs.

  #95364 opened Jun 13, 2025

• [XLA][Numerics][HLO Value Tracking] Deduplicate original values in an HLO module

  #95366 opened Jun 14, 2025

• Integrate LLVM at llvm/llvm-project@2c440232e261

  #95367 opened Jun 14, 2025

• [PjRt] Block on external references when `CommonPjRtBuffer` is destroyed

  #95368 opened Jun 14, 2025

• Automated Code Change

  #95373 opened Jun 14, 2025

• Automated Code Change

  #95374 opened Jun 14, 2025

• Automated Code Change

  #95376 opened Jun 14, 2025

• Automated Code Change

  #95378 opened Jun 14, 2025

• Automated Code Change

  #95379 opened Jun 14, 2025

• Automated Code Change

  #95380 opened Jun 14, 2025

• Automated Code Change

  #95382 opened Jun 14, 2025

• Automated Code Change

  #95383 opened Jun 14, 2025

• Automated Code Change

  #95384 opened Jun 14, 2025

• Automated Code Change

  #95385 opened Jun 14, 2025

6 Issues closed by 5 people

• Looking for the the reasoning speed comparison of different reaasoning frameworks

  #94534 closed Jun 12, 2025

• Unexpected UnicodeDecodeError: invalid continuation byte when reading lines from a file

  #27537 closed Jun 10, 2025

• tensorflow.python.framework.errors_impl.FailedPreconditionError: Could not find variable bn_conv1/moving_mean

  #93739 closed Jun 10, 2025

• Are there any training acceleration solutions for operations like embedding in a search and recommendation model trained on a CPU using TensorFlow?

  #93798 closed Jun 10, 2025

• Bug: Chained HashedCrossing in TF 2.16.2 results in (None, D) vs (batch_size, D) input shapes

  #93830 closed Jun 10, 2025

• Build error in tensorflow lite minimal example

  #70730 closed Jun 9, 2025

4 Issues opened by 4 people

• Deeplabcut issue

  #95274 opened Jun 12, 2025

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

  #95235 opened Jun 12, 2025

• how to build libtensorflowlite_c.so with Address Sanitizer

  #95222 opened Jun 12, 2025

• TensorFlow disables SwiftUI Previews

  #95106 opened Jun 10, 2025

60 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.

• build(aarch64): Update to oneDNN-3.7 + ACL-24.12 (fix)

  #93951 commented on Jun 12, 2025 • 5 new comments

• Fix comparison functions and add unit tests

  #94484 commented on Jun 12, 2025 • 1 new comment

• Fix: Ensure boolean_mask_v2() only accepts boolean dtype for mask

  #89370 commented on Jun 11, 2025 • 0 new comments

• [ONEDNN] upgrading onednn 3.7

  #90388 commented on Jun 11, 2025 • 0 new comments

• Optimize `xla::HloSharding::PartialTile` by using vectorized highway sort `hwy::VQSort`.

  #92165 commented on Jun 11, 2025 • 0 new comments

• Major deps update:

  #92241 commented on Jun 14, 2025 • 0 new comments

• TfLite elementwise_ops add type support (#104)

  #92706 commented on Jun 11, 2025 • 0 new comments

• Only abort collectives with failed tasks.

  #93273 commented on Jun 12, 2025 • 0 new comments

• Allow Xprof frontend to use the query parameter `use_saved_result=False` which will skip the intermediate analysis and regenerate the tool data from XSpace

  #93411 commented on Jun 13, 2025 • 0 new comments

• Stable delegate python api

  #93850 commented on Jun 12, 2025 • 0 new comments

• [NCCL] Upgrade TF NCCL version to 2.26.5

  #94053 commented on Jun 10, 2025 • 0 new comments

• This change introduces a new mechanism for tracking CUDA graph events within the profiling system.

  #94057 commented on Jun 11, 2025 • 0 new comments

• Enable Stablehlo -> HLO lowering by default.

  #94296 commented on Jun 12, 2025 • 0 new comments

• Add source_target_pairs to send/recv ops in StableHLO

  #94495 commented on Jun 14, 2025 • 0 new comments

• Add hlo_module_name parameter to Compiler::CreateMetricsHook to make available the hlo module name of the recorded hlo program in the logging infrastructure.

  #94649 commented on Jun 10, 2025 • 0 new comments

• Use Hermetic C++ toolchain for Linux x86_64 builds.

  #94704 commented on Jun 13, 2025 • 0 new comments

• Added strongly typed int for `IncarnationId`.

  #94761 commented on Jun 13, 2025 • 0 new comments

• [XLA] Flatten nested tuple return shape of async computations in tuple simplifier.

  #94765 commented on Jun 9, 2025 • 0 new comments

• [XLA:CPU][tfcompile] Enable thunk runtime by default.

  #94787 commented on Jun 10, 2025 • 0 new comments

• Move converting a mixed SDY+GSPMD module down to a pure GSPMD targeting module to `MlirToXlaComputation`.

  #94798 commented on Jun 9, 2025 • 0 new comments

• derive all the lines of each device, instead of deriving events from only the line with most event.

  #94829 commented on Jun 14, 2025 • 0 new comments

• Move Mosaic into XLA

  #94877 commented on Jun 9, 2025 • 0 new comments

• PR #27412: Command buffer respect control dependency of HloInstruction when running with concurrent mode.

  #94883 commented on Jun 12, 2025 • 0 new comments

• Refactor CUPTI callback ID logic to cupti_tracer.

  #94898 commented on Jun 11, 2025 • 0 new comments

• Add test and refactor Device Assignment.

  #94904 commented on Jun 10, 2025 • 0 new comments

• Implementation of phase-compilation using PJRT Extensions

  #94922 commented on Jun 9, 2025 • 0 new comments

• Make `Thunk::ToProto()` return an error if serialization is not implemented

  #94930 commented on Jun 10, 2025 • 0 new comments

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

  #94937 commented on Jun 10, 2025 • 0 new comments

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

  #94954 commented on Jun 10, 2025 • 0 new comments

• Use raw string literals for regex patterns in kernel test assertions

  #95005 commented on Jun 12, 2025 • 0 new comments

• Adding TensorFlow Hub KerasLayer to Sequential Model Raises ValueError

  #63849 commented on Jun 8, 2025 • 0 new comments

• `../tensorflow/third_party/xla/third_party/tsl/tsl/platform/ml_dtypes.h:19:10: error: 'ml_dtypes/include/float8.h' file not found [clang-diagnostic-error]`

  #93130 commented on Jun 9, 2025 • 0 new comments

• tf.data.Dataset .map().batch() pattern is not matched to use fused implementation.

  #53572 commented on Jun 9, 2025 • 0 new comments

• Segmentation fault in tf.sets.size

  #94863 commented on Jun 9, 2025 • 0 new comments

• Pybind11 Exception

  #60534 commented on Jun 9, 2025 • 0 new comments

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

  #94735 commented on Jun 9, 2025 • 0 new comments

• TensorFlow Docker `tensorflow/tensorflow:latest-gpu` fails to detect GPU due to CUDA/cuDNN mismatch

  #94593 commented on Jun 10, 2025 • 0 new comments

• Crash in `tf.raw_ops.BiasAdd` when executing on GPU

  #94379 commented on Jun 10, 2025 • 0 new comments

• Muting Tensorflow Lite logs

  #92216 commented on Jun 10, 2025 • 0 new comments

• tf.transpose crashes with negative perm value: "Check failed: d >= 0 (0 vs. -1)"

  #94433 commented on Jun 11, 2025 • 0 new comments

• 16KB pagination support for TF Lite Select Ops

  #94048 commented on Jun 11, 2025 • 0 new comments

• Enhance Memory Optimizer with Dynamic Cost Model for Operation Recomputation

  #94653 commented on Jun 11, 2025 • 0 new comments

• Support/Feature Request: Pre-processing very large corpus text file as tokens to train GPT Models.

  #60539 commented on Jun 11, 2025 • 0 new comments

• tf.linalg.matrix_rank results has different results with or without @tf.function for numpy inputs under tensorflow-cpu

  #60547 commented on Jun 11, 2025 • 0 new comments

• TensorFlow DLL failed to load with newer version of TF

  #91656 commented on Jun 11, 2025 • 0 new comments

• crash when two model parallel inference in two instances using libtensorflowlite_c.so and run delegate gpu opencl

  #94274 commented on Jun 12, 2025 • 0 new comments

• AttributeError with Protobuf >= 6.30

  #94030 commented on Jun 12, 2025 • 0 new comments

• rejection_resample loses track of ragged tensors

  #60583 commented on Jun 12, 2025 • 0 new comments

• [TFLite] flatbuffer64 support for TFlite

  #60570 commented on Jun 12, 2025 • 0 new comments

• Weird memory usage of shuffling in `tf.data.Dataset`

  #60599 commented on Jun 12, 2025 • 0 new comments

• control_flow_ops_test unit test is flaky

  #60629 commented on Jun 12, 2025 • 0 new comments

• graph execution error bug with tfm.nlp.layers.MultiHeadRelativeAttention

  #94599 commented on Jun 13, 2025 • 0 new comments

• java.lang.IllegalArgumentException: Internal error: Error applying delegate:

  #93525 commented on Jun 13, 2025 • 0 new comments

• Mismatch Between Quantized TFLite Layer Outputs and Expected Mathematical Values When Using get_tensor()

  #93917 commented on Jun 13, 2025 • 0 new comments

• cuDNN, cuFFT, and cuBLAS Errors

  #62075 commented on Jun 13, 2025 • 0 new comments

• How to run Android demo which uses NPU to inference?

  #94853 commented on Jun 13, 2025 • 0 new comments

• Numpy and tf experimental Numpy differ in vander matrix creation case for N=0

  #60628 commented on Jun 13, 2025 • 0 new comments

• tensorflow-macos still required for version 2.16.1

  #63495 commented on Jun 14, 2025 • 0 new comments

• Tensorflow is aborting with CompositeTensorVariant already registered

  #94709 commented on Jun 14, 2025 • 0 new comments

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

  #87676 commented on Jun 11, 2025 • 0 new comments