Develop upstream sync 251224#3170

Merged

mmakevic-amd merged 765 commits intodevelop-upstreamfrom

develop-upstream-sync-251224

Jan 14, 2026

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

Commits on Dec 18, 2025

Automated Code Change
tensorflower-gardener
committed
[XLA:GPU] Add method for printing unsatisfied Constraints for ConstraintExpression.

golechwierowicz
authored and
tensorflower-gardener
committed
Automated Code Change
tensorflower-gardener
committed
Automated Code Change
tensorflower-gardener
committed
[XLA:GPU]Disable TransposeDimensionGrouper pass and replace it with OTF normalization in emitters

thcmbs
authored and
tensorflower-gardener
committed
Refactor: Dynamically register custom call targets in custom_call_test.cc

beckerhe
authored and
tensorflower-gardener
committed
Add a function to check for empty/non existing files.

qukhan
authored and
tensorflower-gardener
committed
Update XNNPack version

qukhan
authored and
tensorflower-gardener
committed
Automated Code Change
tensorflower-gardener
committed
Automated Code Change
tensorflower-gardener
committed
When opening a file, check that the file path is not null.

qukhan
authored and
tensorflower-gardener
committed
Automated Code Change
tensorflower-gardener
committed
Remove forgotten ROCM version checks from NcclCollectives

beckerhe
authored and
tensorflower-gardener
committed
Automated Code Change
tensorflower-gardener
committed
Automated Code Change
tensorflower-gardener
committed
Automated Code Change
tensorflower-gardener
committed
[XLA:GPU] Support partitioned across replicas modules
tensorflower-gardener
committed
Apply llvm-use-new-mlir-op-builder fixes
tensorflower-gardener
committed
Refactor: Extract AotCompilationResult to CompiledModule.

beckerhe
authored and
tensorflower-gardener
committed
[PJRT] Change the two optimizations in Transpose to operate on Loop nests, rather than on the original dimensions.

hawkinsp
authored and
tensorflower-gardener
committed
Reverts 408bf09796590bc66233afff288bf926e2736a9d
tensorflower-gardener
committed
PR #35510: [ROCm] Initialze collectives to nullptr to force its allocation later

hsharsha
authored and
tensorflower-gardener
committed
Apply llvm-use-new-mlir-op-builder fixes
tensorflower-gardener
committed
Remove unnecessary local_defines and add missing includes in gpu_executor_test.

beckerhe
authored and
tensorflower-gardener
committed
PR #35482: Correctly generate compile_commands.json

ezhulenev
authored and
tensorflower-gardener
committed
In `FileDescriptor` tests, improve temporary file path generation.

qukhan
authored and
tensorflower-gardener
committed
PR #35463: [xla:gpu] Support ncclAlltoall directly for contiguous send/recv buffers

ezhulenev
authored and
tensorflower-gardener
committed
[xla:cpu] Do not expand convolution feature group if the convolution is supported by libraries.

penpornk
authored and
tensorflower-gardener
committed
[XLA:GPU] Use StreamState as rendezvous value.

olegshyshkov
authored and
tensorflower-gardener
committed
[XLA:GPU] don't stop traversal when sinking bitcasts

metaflow
authored and
tensorflower-gardener
committed
Add basic replicated execution support to ClientLibraryTestRunnerMixin<T>.

nvgrw
authored and
tensorflower-gardener
committed
Expose execution_options in client_library_test_runner_mixin.

nvgrw
authored and
tensorflower-gardener
committed
Remove unneeded `CoordinationServiceAgent::ShutdownInternal` method.

mwhittaker
authored and
tensorflower-gardener
committed
Merge pull request #106273 from tensorflow:dependabot/pip/ci/official/requirements_updater/numpy1_requirements/werkzeug-3.1.4
tensorflower-gardener
committed
Add proto serialization for AllToAllStartThunk

ermilovmaxim
authored and
tensorflower-gardener
committed
Merge pull request #106221 from tensorflow:fix-bicubic-doc
tensorflower-gardener
committed
[XLA:CPU] Enable i8xi8->i32 grouped convolutions with ynnpack.

vksnk
authored and
tensorflower-gardener
committed
[PJRT C API] Implement `xla::PjRtCrossHostSendCancelNotifier`.

emilyfertig
authored and
tensorflower-gardener
committed
Expose XProf Collection Walltime

Becker-ZH
authored and
tensorflower-gardener
committed
Enabling indexing analysis directly on StableHLO.

jpienaar
authored and
tensorflower-gardener
committed
Update bazelisk, buildifier, buildozer and bats-core to the latest version in ml_build Dockerfile

quoctruong
authored and
tensorflower-gardener
committed
[XLA] Add DeadDynamicUpdateSliceElimination pass.

congliuthu
authored and
tensorflower-gardener
committed
add low_priority queue batch attr to mlrt batch_kernel
tensorflower-gardener
committed
Clean up device assignment logic in HloRunnerPjRt.

nvgrw
authored and
tensorflower-gardener
committed
Add ScopedAsyncTrackingEvent to PJRT C API.

hhb
authored and
tensorflower-gardener
committed
Use a deterministic device assignment for split compilation.

nvgrw
authored and
tensorflower-gardener
committed
[PJRT] Change transpose code to chunk nodes on the Loop nest representation, not the Node representation.

hawkinsp
authored and
tensorflower-gardener
committed

Commits on Dec 19, 2025

[XLA:GPU] Split SortRewriter into two passes.

pifon2a
authored and
tensorflower-gardener
committed
switch from deprecated TF_CHECK_OK

ermilovmaxim
authored and
tensorflower-gardener
committed
Simplify `CoordinationServiceAgent` creation.

mwhittaker
authored and
tensorflower-gardener
committed
(4/N) Add support for `NamedSharding` in existing `HloShardingUtil` methods. Remaining methods will be updated in follow up cl's.

KanishAnand
authored and
tensorflower-gardener
committed
Integrate StableHLO at openxla/stablehlo@d496423c

abhigunj
authored and
tensorflower-gardener
committed
Remove unused `CoordinationService::Config::recoverable_jobs` field.

mwhittaker
authored and
tensorflower-gardener
committed
[JAX] Refresh a custom layout if a buffer is copied across clients or memories

hyeontaek
authored and
tensorflower-gardener
committed
Add `testonly` parameter to `py_import` macros to allow testonly dependencies.

ybaturina
authored and
tensorflower-gardener
committed
[PJRT] Remove the .start field on transpose plan Nodes.

hawkinsp
authored and
tensorflower-gardener
committed
Automated Code Change
tensorflower-gardener
committed
Automated Code Change
tensorflower-gardener
committed
Automated Code Change
tensorflower-gardener
committed
Fix struct size check in PJRT_Executable_GetCompiledMemoryStats.

hhb
authored and
tensorflower-gardener
committed
Update symbols in nccl stub for NCCL 2.28.9.

ybaturina
authored and
tensorflower-gardener
committed
Automated Code Change
tensorflower-gardener
committed
Remove unused dependencies and conditional compilation from XLA GPU emitter transforms.

beckerhe
authored and
tensorflower-gardener
committed
Automated Code Change
tensorflower-gardener
committed
Add GetCompilationOptions to PjRt C API

hhb
authored and
tensorflower-gardener
committed
Automated Code Change
tensorflower-gardener
committed
[XLA:GPU] Add xla_gpu_default_to_alg_dot_bf16_bf16_f32 flag to XLA.

thomasjoerg
authored and
tensorflower-gardener
committed
compat: Update forward compatibility horizon to 2025-12-19
tensorflower-gardener
committed
Update GraphDef version to 2446.
tensorflower-gardener
committed
Refactor: Pass GpuComputeCapability to GetBlasComputationType

beckerhe
authored and
tensorflower-gardener
committed
Add more logs to the CuDNN autotuner backend.

derdrdirk
authored and
tensorflower-gardener
committed
[XLA:GPU] Fix bugs when sorting two elements with a fused iota.

akuegel
authored and
tensorflower-gardener
committed
Make filecheck prefixes a parameter.

beckerhe
authored and
tensorflower-gardener
committed
Automated Code Change
tensorflower-gardener
committed
Automated Code Change
tensorflower-gardener
committed
PR #35354: [ROCM] bug-fixing SortRewriter and fixing self_adjoint_test on ROCM

pemeliya
authored and
tensorflower-gardener
committed
[XLA:GPU] Only mark HoistFusedBitcasts as changed when modifications occur.

thcmbs
authored and
tensorflower-gardener
committed
[PJRT] Add a ExecuteChunk method that allows the user to execute a single chunk of the transpose only.

hawkinsp
authored and
tensorflower-gardener
committed
Add LINT.IfChange to keep batch function op registrations in sync.

deqiangc
authored and
tensorflower-gardener
committed
Enable using custom hermetic NCCL version.

ybaturina
authored and
tensorflower-gardener
committed
Allow serialization for sync collectives

ermilovmaxim
authored and
tensorflower-gardener
committed
Enable using custom hermetic NCCL version.

ybaturina
authored and
tensorflower-gardener
committed
[StableHLO] Update CHLO broadcast ops to use StableHLO bounded-dynamism-aware broadcasting for type inference.

GleasonK
authored and
tensorflower-gardener
committed
[XLA][Numerics][HLO Value Tracking] Support HLO original values in ConditionalSimplifier pass

jcai19
authored and
tensorflower-gardener
committed
Simplify TrackedCpuDeviceBuffer.

pschuh
authored and
tensorflower-gardener
committed
[ReplicaGroupV3][Refactor][3/n] Use CollectiveDeviceListBase for polymorphic device list support.

Varcho
authored and
tensorflower-gardener
committed
Internal changes only.

fengwuyao
authored and
tensorflower-gardener
committed
[XLA] Continue trying to unroll pipelined loops after failure

vsytch
authored and
tensorflower-gardener
committed
[XLA:GPU] Allow to fuse iotas into sort.

akuegel
authored and
tensorflower-gardener
committed
Apply llvm-use-new-mlir-op-builder fixes
tensorflower-gardener
committed
Integrate LLVM at llvm/llvm-project@7d381f2a5634
tensorflower-gardener
committed
[ReplicaGroupV3][Refactor][4/n] Update replica group classes to hide implementation details (via private/protected) to external callers.

Varcho
authored and
tensorflower-gardener
committed
Apply llvm-use-new-mlir-op-builder fixes
tensorflower-gardener
committed
Apply llvm-use-new-mlir-op-builder fixes
tensorflower-gardener
committed
Migrate reshape_motion_test to PjRt.

nvgrw
authored and
tensorflower-gardener
committed

Commits on Dec 20, 2025

Commits on Dec 21, 2025

Commits on Dec 22, 2025

Commits on Dec 23, 2025

Commits on Dec 24, 2025

Commits on Dec 25, 2025

Remove leftover diff symbols
mmakevic-amd
committed

Commits on Dec 30, 2025

Commits on Jan 12, 2026

Remove remaining cuda-only tags and move failing subtests to test scripts
mmakevic-amd
committed

Commits on Jan 13, 2026