Add missing command line option registrations#1

Draft

IMVVVVVIP wants to merge 10000 commits intoIMVVVVVIP:masterfrom

tensorflow:master

This pull request is big! We're only showing the most recent 250 commits

Commits on Mar 30, 2026

PR #39497: [XLA:GPU][oneAPI] Register base oneCCL collective support for XLA

nhatleSummer22
authored and
tensorflower-gardener
committed
[XLA:GPU] Add command buffer scheduling mode `CONCURRENT_REGIONS`.

thomasjoerg
authored and
tensorflower-gardener
committed
Automated Code Change
tensorflower-gardener
committed
PR #39025: [xla] Fix potentially expensive spin in ObjectPool

ezhulenev
authored and
tensorflower-gardener
committed
Automated Code Change
tensorflower-gardener
committed
PR #39956: [xla:gpu] Log thunk progress in chronological order

ezhulenev
authored and
tensorflower-gardener
committed
PR #39417: [ROCm] Implement empty graph node support and safeguards for HIP command buffer

phambinhfin
authored and
tensorflower-gardener
committed
PR #39508: [XLA:GPU][oneAPI] Remove hardcoded spirv-binary in SYCL tests

bhavani-subramanian
authored and
tensorflower-gardener
committed
PR #39931: Add test coverage for AllGatherRemoveDegenerateDims untested code paths

kredd2506
authored and
tensorflower-gardener
committed
Automated Code Change
tensorflower-gardener
committed
[XLA:GPU] Add support for constants via new tiling.

pifon2a
authored and
tensorflower-gardener
committed
[XLA:GPU] Fix ReorderFilterAndBiasHloTest.TestCudnnReorderFilterAndBias

Marcin Radomski
authored and
tensorflower-gardener
committed
Disable unflattener for Shardy to GSPMD fallback.

ekayaaslan
authored and
tensorflower-gardener
committed
PR #39950: [ROCm] Fix hipblasLt Int8 GEMM support and autotuner output comparison

cj401-amd
authored and
tensorflower-gardener
committed
[XLA:GPU] Use copied device info to init members of GpuCostModelStatsCollection

nputikhin
authored and
tensorflower-gardener
committed
[XLA:GPU] Add transposes via the new tiling.

pifon2a
authored and
tensorflower-gardener
committed
[xla] Add FloatNormalizationExcessPrecisionTest

penpornk
authored and
tensorflower-gardener
committed
[XLA:GPU] Move ScatterDeterminismExpander pass to backends/gpu/transforms.

akuegel
authored and
tensorflower-gardener
committed
[XLA:GPU] Reroute dots to a detailed cost model in the indexing model

nputikhin
authored and
tensorflower-gardener
committed
[XLA:GPU][NFC] Pad the tile sizes before the tiling propagation.

pifon2a
authored and
tensorflower-gardener
committed
Make CPU devices report as cpu:0 instead of TFRT_CPU_0.

hawkinsp
authored and
tensorflower-gardener
committed
[XLA:GPU] Preserve point dimension for concat if it occurs in concatenate dimension
tensorflower-gardener
committed
Add dummy execute handler for CubScanUnassignedScratchSizeTarget FFI.

chsigg
authored and
tensorflower-gardener
committed
Rename `XNNPACK_CACHE_NO_MMAP_FOR_TEST` to `XNNPACK_CACHE_NO_FILE_MAPPING_FOR_DEBUG`.

qukhan
authored and
tensorflower-gardener
committed
disable flaky test

vwbaker
authored and
tensorflower-gardener
committed
Add reshards for when func has no result shardings on unflattener.

ekayaaslan
authored and
tensorflower-gardener
committed
Migrate XLA GPU emitters and HWIR to use SymbolicMap.
tensorflower-gardener
committed
[xla] Make some FloatNormalization tests more readable.

penpornk
authored and
tensorflower-gardener
committed
Support extracting multiple consumers and use that in autotuner's compile.
tensorflower-gardener
committed
Automated Code Change

QEDady
authored and
tensorflower-gardener
committed
Log configs in coordination service `client_server_test.cc`

mwhittaker
authored and
tensorflower-gardener
committed
Remove timeout long for slice_test.

akuegel
authored and
tensorflower-gardener
committed
[SC Serving] Bypass `sample_ids` scatter for univalent features in XLA:TPU:SC compiler.
tensorflower-gardener
committed
Add reshards for when func has no result shardings on shardy outliner.

ekayaaslan
authored and
tensorflower-gardener
committed
PR #40017: [xla:gpu] Correctly track thunk progress in presence of loops

ezhulenev
authored and
tensorflower-gardener
committed
Support running IFRT through TFRT session on tf serving model server

SiqiaoWu1993
authored and
tensorflower-gardener
committed
Add a `device_memory_limit_bytes` attribute to PjRt GPU devices

junwhanahn
authored and
tensorflower-gardener
committed
Clean up coordination service `client_server_test.cc`.

mwhittaker
authored and
tensorflower-gardener
committed
Reverts a0a1f5b376ab08c50c6559c4d43804e9d3634e27
tensorflower-gardener
committed
Remove TensorBoard dependency from TensorFlow
tensorflower-gardener
committed
[XLA:GPU] Add pad via the new tiling.

pifon2a
authored and
tensorflower-gardener
committed
Automated Code Change
tensorflower-gardener
committed
Update XNNPACK in tensorflow/xla

dsharletg
authored and
tensorflower-gardener
committed
Integrate Triton up to [565c08520d](https://github.com/triton-lang/triton/commits/565c08520dcdee3a949476395fbceb340e57b4ba)

vwbaker
authored and
tensorflower-gardener
committed
PR #40053: [xla:ifrt] Allow passing PjRt client for autotuning during cross-compile

ezhulenev
authored and
tensorflower-gardener
committed
PR #40112: [xla:gpu] Correctly report thunk execution index

ezhulenev
authored and
tensorflower-gardener
committed
[StreamExecutor] RAII-compliant Resource attachment via Pimpl

apivovarov
authored and
tensorflower-gardener
committed
Reverts e117560c18eb072d8f41622f625acb18d8e4d48d

hyeontaek
authored and
tensorflower-gardener
committed
Refactor `CommonPjRtClient::DefineBuffer` to take `shared_ptr<const Shape>`

junwhanahn
authored and
tensorflower-gardener
committed
Introduce DeviceInterconnectResource to bridge Global Topology to Backend Thunks

apivovarov
authored and
tensorflower-gardener
committed
Update DynamicUpdateSlice for safer bytes size in memcpy.
tensorflower-gardener
committed
Translate async collectives between StableHLO and HLO.

mwhittaker
authored and
tensorflower-gardener
committed
Fix building TF CUDA wheel by adding dependency on stub alias.

ybaturina
authored and
tensorflower-gardener
committed
Add HLO constraint propagation for test argument generation

bhatuzdaname
authored and
tensorflower-gardener
committed
[ReplicaGroupV3] Cleanup replica group parsing in HLO parser.

Varcho
authored and
tensorflower-gardener
committed

Commits on Mar 31, 2026

PR #40132: [xla:gpu] Allow cross-compilation via PjRtCompiler::Compile

ezhulenev
authored and
tensorflower-gardener
committed
[PjRt] Implement `PjRt(Loaded)Executable::GetParameterMemoryKinds()` for various PjRt clients

hyeontaek
authored and
tensorflower-gardener
committed
[XLA:CPU] Set PjRt compiler thread stack size to 2 MiB

seantalts
authored and
tensorflower-gardener
committed
Automated Code Change
tensorflower-gardener
committed
Automated Code Change
tensorflower-gardener
committed
Automated Code Change
tensorflower-gardener
committed
Automated Code Change
tensorflower-gardener
committed
1. Fixed Missing Profiles: Enabled modern telemetry to capture HBM metrics (Memory Profiles).

bhupendradubey-cmd
authored and
tensorflower-gardener
committed
Reverts fc307c1a508ec80cbbb6a91dccf3f14345091704

ermilovmaxim
authored and
tensorflower-gardener
committed
Automated Code Change
tensorflower-gardener
committed
[IFRT IR] Add ifrt_ir_compile_options to compile_mpmd()

ICGog
authored and
tensorflower-gardener
committed
Fix copyright year

wsmoses
authored and
tensorflower-gardener
committed
Automated Code Change
tensorflower-gardener
committed
Allow dots with multiple batch dimensions through GemmFusion - the new emitter can handle this, so there's no reason to restrict this

vwbaker
authored and
tensorflower-gardener
committed
Automated Code Change
tensorflower-gardener
committed
[XLA Presubmit] Add SM 12.0 to the list of known SM architecture

beckerhe
authored and
tensorflower-gardener
committed
PR #39693: Fix tracked_gpu_device_buffer_test in OSS by adding cuda_platform dep

and-ivanov
authored and
tensorflower-gardener
committed
PR #39393: [xla:gpu] Add VA remapping for command buffer thunks

shawnwang18
authored and
tensorflower-gardener
committed
PR #39695: Fix xla aot compile gpu test

and-ivanov
authored and
tensorflower-gardener
committed
Rollback of Migrate XLA GPU emitters and HWIR to use SymbolicMap.
tensorflower-gardener
committed
[xla:cpu] Make small_while_loop_hoisting_pass_test use the same byte threshold as cpu_compiler.

penpornk
authored and
tensorflower-gardener
committed
PR #40141: [xla:gpu] Fix a bug when emitting degenerate collective permute

ezhulenev
authored and
tensorflower-gardener
committed
PR #40108: Update protobuf dependency version to 32.1

ezhulenev
authored and
tensorflower-gardener
committed
PR #39871: [ROCm] Fix bf16 upcast handling for libdevice calls.

zoranjovanovic-ns
authored and
tensorflower-gardener
committed
PR #39309: [ROCm] Add scope_range_id support to ROCm profiler

magaonka-amd
authored and
tensorflower-gardener
committed
PR #39991: Add test coverage for BatchNormExpander untested code paths

kredd2506
authored and
tensorflower-gardener
committed
PR #40111: Bump pygments from 2.18.0 to 2.20.0 in /xla/backends/cpu/benchmarks/e2e/gemma2/flax_2b

dependabot[bot]
authored and
tensorflower-gardener
committed
PR #40118: Fix tensor memory size check to use tcgen05 capability instead of IsBlackwell()

and-ivanov
authored and
tensorflower-gardener
committed
[XLA:GPU] Move triton emitter c++ tests closer to the lit tests.

pifon2a
authored and
tensorflower-gardener
committed
[XLA:GPU] Add slice and iota emission via the new tiling.

pifon2a
authored and
tensorflower-gardener
committed
[XLA:GPU] Simplify xla_aot_compile_gpu_test by using HloPjRtTestBase.

akuegel
authored and
tensorflower-gardener
committed
Fully serialize stream_executor::DeviceDescription to/from proto.

beckerhe
authored and
tensorflower-gardener
committed
[xla:cpu] Roll-back small_while_loop_hoisting_pass threshold refactor.

penpornk
authored and
tensorflower-gardener
committed
Remove gen_gpu_hlo_compile_tests bazel rule.

akuegel
authored and
tensorflower-gardener
committed
PR #39744: [ROCm] Porting CUB sort FFI handler consolidation to ROCm.

tsrw2048
authored and
tensorflower-gardener
committed
Automated Code Change
tensorflower-gardener
committed
PR #39537: [xla:gpu] Add hang watchdog to PJRT GPU client

ezhulenev
authored and
tensorflower-gardener
committed
Simplify partial replication check in SPMD partitioner.

Kingston Mandisodza
authored and
tensorflower-gardener
committed
Rename test case in SpmdPartitionerTest.

Kingston Mandisodza
authored and
tensorflower-gardener
committed
Check that cached buffers offset and size do not point out of the flatbuffer.

qukhan
authored and
tensorflower-gardener
committed
Run shardy/google/integrate_latest.sh to integrate.

ekayaaslan
authored and
tensorflower-gardener
committed
Remove legacy autotuner compile util.
tensorflower-gardener
committed
Handle mixed V2/V3 shardings in PartialReplicateReshardCompatibleSharding.

Kingston Mandisodza
authored and
tensorflower-gardener
committed
PR #39929: [ROCm] Fix parse empty rocm distro links

alekstheod
authored and
tensorflower-gardener
committed
Add a specialized kernel for scans of many short rows.

chsigg
authored and
tensorflower-gardener
committed
Add `stablehlo.future` to C API.

mwhittaker
authored and
tensorflower-gardener
committed
Remove `kLeaveBarriersOnRecoverableAgentRestart` from coordination service.

mwhittaker
authored and
tensorflower-gardener
committed
Allow non-kPadUpPolicy for priority-aware batching when max_execution_batch_size is 1.
tensorflower-gardener
committed
Fix RunHandler::ScheduleIntraOpClosure to call the correct impl method.
tensorflower-gardener
committed
Use absl::HashOf for key_hash in LocalRendezvous.

ezbr
authored and
tensorflower-gardener
committed
port away from TF_ASSIGN_OR_RETURN

ermilovmaxim
authored and
tensorflower-gardener
committed
PR #39567: [ROCm] Fix fp8 on ROCm when xla_gpu_enable_cublaslt=false

Eetusjo
authored and
tensorflower-gardener
committed
[XLA:GPU] Emit broadcast via the new tiling.

pifon2a
authored and
tensorflower-gardener
committed
Add PRNG seed to PjRt ExecuteOptions and add runtime support where possible.

nvgrw
authored and
tensorflower-gardener
committed
Fix an integer overflow in symbolic_tile_analysis.

hawkinsp
authored and
tensorflower-gardener
committed
Add GetOrCreateRuntimeError to PJRT Megascale Extension

hhb
authored and
tensorflower-gardener
committed
Rename TSL Monitoring's `MakeStatic` factory functions to `NoDestructor`

mrguenther
authored and
tensorflower-gardener
committed
Migrate prng_test to use PjRt.

nvgrw
authored and
tensorflower-gardener
committed
Support the XLA GPU compilation flags in Orbax
tensorflower-gardener
committed
Correct strided slice output shape calculation.

lynne0326
authored and
tensorflower-gardener
committed
Add a test for dataset batching using `dtypes.int64.max` batch size.

yangustc07
authored and
tensorflower-gardener
committed
Migrate qr_test to PjRt runtime.

nvgrw
authored and
tensorflower-gardener
committed
Fix WinDefFileParse on Linux in pywrap_library by disabling evaluation of generated rules

cantonios
authored and
tensorflower-gardener
committed
Fix use of deprecated `xnn_define_static_constant_pad`

dsharletg
authored and
tensorflower-gardener
committed
Fix aligned_alloc usage for older Android APIs
tensorflower-gardener
committed
Add missing bf16 reduce benchmarks similarly to f32

alexander-shaposhnikov
authored and
tensorflower-gardener
committed
Migrate quantize_test to PjRt runtime.

nvgrw
authored and
tensorflower-gardener
committed
Add support for reshape and reduce fusions.

alexander-shaposhnikov
authored and
tensorflower-gardener
committed
Test elements shape and dtype in `testLargeBatchSize`.

yangustc07
authored and
tensorflower-gardener
committed

Commits on Apr 1, 2026

Remove friend classing in coordination service.

mwhittaker
authored and
tensorflower-gardener
committed
Migrate xla::ifrt::GetDeviceMemoryInBytes to `device_memory_bytes_limit` device attributes

junwhanahn
authored and
tensorflower-gardener
committed
Automated Code Change
tensorflower-gardener
committed
Automated Code Change
tensorflower-gardener
committed
Remove test timeout long from grouped_convolution_test.

akuegel
authored and
tensorflower-gardener
committed
Remove ptxas regression test.

akuegel
authored and
tensorflower-gardener
committed
Make GpuTargetConfig device-agnostic

beckerhe
authored and
tensorflower-gardener
committed
Automated Code Change
tensorflower-gardener
committed
Always populate target config in GpuTopology created in PjRt client

beckerhe
authored and
tensorflower-gardener
committed
Automated Code Change
tensorflower-gardener
committed
[XLA:GPU] Prevent collective memory from being destructed until the module execution on devices is completed.

PatriosTheGreat
authored and
tensorflower-gardener
committed
PR #40180: [XLA:GPU][oneAPI] Remove context activation and update SYCL memfill/memset tests

bhavani-subramanian
authored and
tensorflower-gardener
committed
Move gpu_device_info_test to target_config directory.

beckerhe
authored and
tensorflower-gardener
committed
Fix scan rewriter to handle broadcasted initial values and correct layout calculation.

chsigg
authored and
tensorflower-gardener
committed
PR #33269: Addin a knob to control the limitation of async-compute resource.

mingxu1067
authored and
tensorflower-gardener
committed
Automated Code Change
tensorflower-gardener
committed
PR #39601: [xla:gpu] Add support for dynamically sized packed kernel args

ezhulenev
authored and
tensorflower-gardener
committed
Automated Code Change
tensorflower-gardener
committed
Move utility for in/outliner into shardy. part 2.

ekayaaslan
authored and
tensorflower-gardener
committed
PR #39854: [ROCm] Use hermetic clang for rocm

alekstheod
authored and
tensorflower-gardener
committed
[XLA:GPU]: Unify collective codegen checks in one place.

sohaibiftikhar
authored and
tensorflower-gardener
committed
[XLA:GPU] Actually use `ConcurrentRegionsHloOrdering` for command buffer scheduling mode `CONCURRENT_REGIONS`.

thomasjoerg
authored and
tensorflower-gardener
committed
[XLA:GPU] Sort the tiled instructions taking RT variables into account.

pifon2a
authored and
tensorflower-gardener
committed
Run shardy/google/integrate_latest.sh to integrate.

ekayaaslan
authored and
tensorflower-gardener
committed
[XLA:GPU] Remove NCCL from dependencies if no_nccl definition is provided.

PatriosTheGreat
authored and
tensorflower-gardener
committed
Use utility to walk on call graphs.

ekayaaslan
authored and
tensorflower-gardener
committed
PR #39725: [xla] Add global per-process hang watchdog

ezhulenev
authored and
tensorflower-gardener
committed
[XLA:GPU] Mark NVSHMEM thunks as deprecated.

PatriosTheGreat
authored and
tensorflower-gardener
committed
Automated Code Change
tensorflower-gardener
committed
Wrap XNNPack delegate check macros in `do { ... } while(0)`.

qukhan
authored and
tensorflower-gardener
committed
[XLA:CPU] Add benchmarks for Cbrt

akuegel
authored and
tensorflower-gardener
committed
Enforce consistent simplification at creation time in SymbolicExpr
tensorflower-gardener
committed
Handle one size axis in `IsReplicated` checks

KanishAnand
authored and
tensorflower-gardener
committed
Allow creating a GPU PjRtClient without devices

beckerhe
authored and
tensorflower-gardener
committed
Add reshards for when func has no argument shardings but the call has.

ekayaaslan
authored and
tensorflower-gardener
committed
Run shardy/google/integrate_latest.sh to integrate.

ekayaaslan
authored and
tensorflower-gardener
committed
Add if_onednn_async to tf_copts.

akuegel
authored and
tensorflower-gardener
committed
[XLA:GPU] Fix Triton loads dropping the mask and causing OOB with aligned reads that read pass the end of the original shape.

dimitar-asenov
authored and
tensorflower-gardener
committed
Fix use-after-free in `SymbolicMap::GetResults()`
tensorflower-gardener
committed
Add original func names on shardy outliner.

ekayaaslan
authored and
tensorflower-gardener
committed
Handle V3 named sharding in windowed input resharding.

Kingston Mandisodza
authored and
tensorflower-gardener
committed
[XLA:GPU] Simplify and retune the splitk heuristic.

mooskagh
authored and
tensorflower-gardener
committed
[XLA:CPU] Lower Cbrt to Libm call.

akuegel
authored and
tensorflower-gardener
committed
Drop func and call sharding conflict resolution on outliner.

ekayaaslan
authored and
tensorflower-gardener
committed
Fix dependency issues.
tensorflower-gardener
committed
Add SymbolicExpr::ReplaceDims(absl::Span<const SymbolicExpr> replacements)
tensorflower-gardener
committed
Drop fully deduplicating on outliner and leave it to unflattenner.

ekayaaslan
authored and
tensorflower-gardener
committed
Internal visibility change.
tensorflower-gardener
committed
Add a test for `jit(lambda x: x, out_shardings=unreduced)` and use that in device_put.

yashk2810
authored and
tensorflower-gardener
committed
Grant users of XLA access to gpu_topology_util

beckerhe
authored and
tensorflower-gardener
committed
[XLA] Remove obsolete shmap & sdy manual computation inliner handling.

Varcho
authored and
tensorflower-gardener
committed
Update Shardy to the latest commit.

yashk2810
authored and
tensorflower-gardener
committed
[tfrt] Introduce DeviceInterconnectResource to bridge Global Topology to Backend Thunks

apivovarov
authored and
tensorflower-gardener
committed
Benchmark for traceme_encode.
tensorflower-gardener
committed
[XLA] Make `ObjectPool::Entry::next` atomic

hyeontaek
authored and
tensorflower-gardener
committed
Add PID metadata to XPlanes and TraceMeConsumers to enable producer/consumer differentiation across processes profiled together.

bmass02
authored and
tensorflower-gardener
committed
[ReplicaGroupV3][Refactor][17/n] migrate functions in spmd/ to use collectivedevicelist base instead of old representations.

Varcho
authored and
tensorflower-gardener
committed
Make an explicit ref type for PjRtDeviceEvent to allow easily changing this type to a c-api compatible type later on.

pschuh
authored and
tensorflower-gardener
committed
Remove `allow_new_incarnation_to_reconnect` config from coordination service.

mwhittaker
authored and
tensorflower-gardener
committed
[XLA] Relax memory ordering of `entry->next` access in `~ObjectPool`

hyeontaek
authored and
tensorflower-gardener
committed
[ReplicaGroupV3][Refactor][18/n] migrate more XLA functions to use collectivedevicelist instead of older replica group representations.

Varcho
authored and
tensorflower-gardener
committed
Reverts f6b594dc036bed5a4d59ee60e69698e8219501b2
tensorflower-gardener
committed
[GPU] Use CheckFabricHomogeneity in RaggedAllToAllThunk

apivovarov
authored and
tensorflower-gardener
committed
Mark coordination service movable but not copyable.

mwhittaker
authored and
tensorflower-gardener
committed

Commits on Apr 2, 2026