| COMMIT |
0.70 |
PR #39621: Add test coverage for AllReduceSimplifier unteste |
|
Explicit AI attribution: 'Generated with |
2026-03-24 |
| PR |
0.50 |
tools/docs: normalize eager Tensor repr numpy fields |
|
Contains Cursor tool attribution; brief |
2026-03-25 |
| PR |
0.10 |
Fix DeleteMultiDeviceIterator crash (Issue #107121) |
|
Concise technical summary; no AI-style p |
2026-03-25 |
| PR |
0.10 |
Fix ResourceGather dtype mismatch CHECK failure (Issue #1130 |
|
Straightforward technical description; l |
2026-03-25 |
| COMMIT |
0.00 |
Migrate while_test to use PjRt runtime. |
|
Brief, domain-specific commit with Piper |
2026-03-25 |
| COMMIT |
0.00 |
Fix deserialization of empty tuples in OriginalValue |
|
Technical description with domain jargon |
2026-03-25 |
| COMMIT |
0.00 |
Check memory space color in `CommonPjRtLoadedExecutable::Che |
|
Concise technical change, no AI indicato |
2026-03-25 |
| COMMIT |
0.00 |
Fix forward on [XLA] Disalbe LICM on scalar and support Rang |
|
Forward revert with informal tone; lacks |
2026-03-25 |
| COMMIT |
0.00 |
Reverts e950a4b7a4c64fca41b9b040c8ffb5e2fd5e1041 |
|
Pure revert; no free-text content beyond |
2026-03-25 |
| COMMIT |
0.00 |
Enable global scheduler for priority-aware batching in all c |
|
Straightforward technical commit; no AI |
2026-03-25 |
| COMMIT |
0.00 |
Remove unused device info related code from coordination ser |
|
Short, technical, minimal commit with do |
2026-03-25 |
| COMMIT |
0.00 |
[PjRt-IFRT] Remove AdjustExecutableDevicesForPmap now that j |
|
Technical, domain-specific phrasing; hum |
2026-03-25 |
| COMMIT |
0.00 |
Update comments to reflect the new name of the CSE pass. |
|
Specific and detailed comment update; la |
2026-03-25 |
| COMMIT |
0.00 |
Inline addCommonPreImportPasses. |
|
Terse, direct technical commit; no AI si |
2026-03-25 |
| COMMIT |
0.00 |
Use utils to resolve conflicts on func and call results duri |
|
Terse technical wording with domain refe |
2026-03-25 |
| COMMIT |
0.00 |
[CollectiveCombiner] Deduplicate redundant buffers sent |
|
Concise, domain-specific phrasing, no AI |
2026-03-25 |
| COMMIT |
0.00 |
Migrate dynamic_reshape_test to PjRt runtime. |
|
Direct migration mention, no formal AI-s |
2026-03-25 |
| COMMIT |
0.00 |
Simplify utils for inserting reshards to resolve conflicts o |
|
Brief, technical explanation; slightly f |
2026-03-25 |
| COMMIT |
0.00 |
[SPMD] Add support for a multi pad operation |
|
Domain jargon, terse title, no AI signal |
2026-03-25 |
| COMMIT |
0.00 |
Add host target machine options to GpuTopology. |
|
Technical summary with platform referenc |
2026-03-25 |
| COMMIT |
0.00 |
Add overload of IsCompatibleWithTargetTopology that accepts |
|
Direct technical phrasing, no AI traits |
2026-03-25 |
| COMMIT |
0.00 |
[XLA:GPU] Plumb through fast interconnect slice size overrid |
|
Informal, domain-specific explanation, l |
2026-03-25 |
| COMMIT |
0.00 |
[XLA:GPU] Default dot precision to BF16_BF16_F32 for F32 arg |
|
Technical, domain-specific summary with |
2026-03-25 |
| COMMIT |
0.00 |
[XLA:GPU]: Use error callbacks in the runtime to notify of a |
|
Concise technical update, informal and j |
2026-03-25 |
| COMMIT |
0.00 |
[XLA:GPU] Exchange pointer is RaggedAllToAllThunk thunk with |
|
Domain-specific, terse, uses jargon, inf |
2026-03-25 |
| COMMIT |
0.00 |
[XLA:GPU]: Use the host callback registry for cuda stream. |
|
Jargon-heavy, brief, lacks AI-style boil |
2026-03-25 |
| COMMIT |
0.00 |
[XLA:GPU]: Unify the compute capability b/w codegen and runt |
|
Uses abbreviations, terse, technical, no |
2026-03-25 |
| COMMIT |
0.00 |
Reverts d2e00f2c722d2fa1d18e55354b16561b998cdf5c |
|
Classic revert message, human, no AI sig |
2026-03-25 |
| COMMIT |
0.00 |
Automated Code Change |
|
Generic phrase, but likely human due to |
2026-03-25 |
| COMMIT |
0.00 |
Add host CPU target machine options to GpuTopology. |
|
Technical, structured, slight formality |
2026-03-25 |
| COMMIT |
0.00 |
Simplify reduce_window_test setup. |
|
Casual, domain-aware, lacks AI signals. |
2026-03-25 |
| COMMIT |
0.00 |
Remove some comments |
|
Very terse, classic human commit style. |
2026-03-25 |
| COMMIT |
0.00 |
Automated Code Change |
|
Brief, automated code change; lacks AI s |
2026-03-25 |
| COMMIT |
0.00 |
Automated Code Change |
|
Brief, automated code change; lacks AI s |
2026-03-25 |
| COMMIT |
0.00 |
Integrate LLVM at llvm/llvm-project@31b17c4789ab |
|
Brief changelog and standard integration |
2026-03-25 |
| COMMIT |
0.00 |
Enable lazy imports for //third_party/tensorflow/python/data |
|
Uses domain-specific notation and concis |
2026-03-25 |
| COMMIT |
0.00 |
Guard `CreatePropagateStaticShapesPass` with an `IfrtModelCo |
|
Human style, terse, technical; no AI hal |
2026-03-25 |
| COMMIT |
0.00 |
Make absl::string_view version of TSL `ReadTextProto`. |
|
Concise, technical update; domain-specif |
2026-03-25 |
| COMMIT |
0.00 |
Fix flaky test in memory_usage_monitor_test.cc |
|
Slightly more verbose but still technica |
2026-03-24 |
| COMMIT |
0.00 |
Remove dependency of "tensorflow/core:all_kernels" when buil |
|
Direct description, domain jargon; no AI |
2026-03-24 |
| COMMIT |
0.00 |
Integrate LLVM at llvm/llvm-project@17d28a5b53b0 |
|
Structured, technical, matches human com |
2026-03-24 |
| COMMIT |
0.00 |
Add visibility of `pywrap_saved_model` |
|
Minimal, domain-specific free text; typi |
2026-03-24 |
| COMMIT |
0.00 |
Add a "metadata" command to interactive_graphviz. |
|
Brief explanation, domain vocab, no AI t |
2026-03-24 |
| COMMIT |
0.00 |
Move over to .mm as opposed to .cc so we don't need special |
|
Clear, informal explanation of technical |
2026-03-24 |
| COMMIT |
0.00 |
[XLA:GPU] Add specific deprecated overloads for tsl::errors: |
|
Technical phrasing and domain-specific c |
2026-03-24 |
| COMMIT |
0.00 |
Delete now unused stablehlo shard map import pass. |
|
Concise, domain-specific commit; human s |
2026-03-24 |
| COMMIT |
0.00 |
Add CUDA compute capability 103 to build tools. |
|
Brief, technical update; no AI-like lang |
2026-03-24 |
| COMMIT |
0.00 |
Remove unused coordination service protos. |
|
Terse, technical removal; human commit s |
2026-03-24 |
| COMMIT |
0.00 |
make constants emitted for BitonicSort's comparator internal |
|
Informal tone, domain jargon; not AI phr |
2026-03-24 |
| COMMIT |
0.00 |
[XLA:GPU] Add pointer exchange capability to MultiGpuBarrier |
|
Technical, domain language and concise d |
2026-03-24 |
| COMMIT |
0.00 |
[XLA:MSA] Block Prefetching - do not add uses that are not i |
|
Domain-specific, succinct explanation; n |
2026-03-24 |
| COMMIT |
0.00 |
[XLA:GPU] Implement round-nearest-even support and enable BF |
|
Technical, specific update; no AI phrasi |
2026-03-24 |
| COMMIT |
0.00 |
[XLA:GPU]: Unify the host callback registry across all strea |
|
Informal and specific, includes develope |
2026-03-24 |
| COMMIT |
0.00 |
[XLA:GPU] Introduce CombinedGpuPerformanceModel |
|
Informal tone, domain jargon, human comm |
2026-03-24 |
| COMMIT |
0.00 |
[XLA:GPU] Add an option to pass output pointer to RaggedAllT |
|
Technical jargon and detailed explanatio |
2026-03-24 |
| COMMIT |
0.00 |
[XLA:GPU] Make GpuPerformanceModelCache thread-safe |
|
Contains domain-specific terminology and |
2026-03-24 |
| COMMIT |
0.00 |
[XLA:GPU] Expand Triton autotuner search space and update te |
|
Casual tone, bug references, and test de |
2026-03-24 |
| COMMIT |
0.00 |
Automated Code Change |
|
Generic brief commit message; no AI sign |
2026-03-24 |
| COMMIT |
0.00 |
[XLA:CPU] Pass LLVM data layout information to the XLA:CPU C |
|
Short, technical description—clearly hum |
2026-03-24 |
| COMMIT |
0.00 |
[XLA:GPU] Another correction to the FuseGeluActivation logic |
|
Short, clear update with domain specific |
2026-03-24 |
| COMMIT |
0.00 |
PR #39690: Skip AutoShardingTest.MatMulWithAutosharding in O |
|
Long-form prose, but domain details and |
2026-03-24 |
| COMMIT |
0.00 |
Create configs with unroll factor for Loop fusion in NativeE |
|
Technical jargon and informal explanatio |
2026-03-24 |
| COMMIT |
0.00 |
Use utils for inserting copies/reshards to resolve conflicts |
|
Brief technical description, typical hum |
2026-03-24 |
| COMMIT |
0.00 |
PR #38909: [ROCm] Support hipblaslt group-gemm 3/5 |
|
Structured PR, but content is domain-spe |
2026-03-24 |
| COMMIT |
0.00 |
[XLA:GPU] PriorityFusion: remove block_level_parameters_cach |
|
Technical explanation with informal tone |
2026-03-24 |
| COMMIT |
0.00 |
[XLA:GPU/CPU] Add printing after each pass for the CPU emitt |
|
Concise commit, no AI signals |
2026-03-24 |
| COMMIT |
0.00 |
Improve SymbolicExpr printing by using operator precedence |
|
Technical description, some typos, casua |
2026-03-24 |
| COMMIT |
0.00 |
[XLA:GPU] Add an option to use MultiGpuBarrierWithNcclKernel |
|
Technical commit, no AI signals |
2026-03-24 |
| COMMIT |
0.00 |
Automated Code Change |
|
Standard 'Automated Code Change', no AI |
2026-03-24 |
| COMMIT |
0.00 |
Automated Code Change |
|
Standard 'Automated Code Change', no AI |
2026-03-24 |
| COMMIT |
0.00 |
Add stream output operators for SymbolicExpr and SymbolicMap |
|
Technical addition, no AI signals |
2026-03-24 |
| COMMIT |
0.00 |
[XLA:GPU] Improve num_blocks propagation in tiled cost model |
|
Technical commit, no AI signals |
2026-03-24 |
| COMMIT |
0.00 |
Add RTX 6000 Pro support |
|
Brief, domain-specific commit with no AI |
2026-03-24 |
| COMMIT |
0.00 |
[XLA:GPU] Handle concat in indexing cost model `FlopsPerElem |
|
Terese, domain-specific commit message w |
2026-03-24 |
| COMMIT |
0.00 |
Bumped nanobind version |
|
Brief, minimal version bump; no AI hallm |
2026-03-24 |
| COMMIT |
0.00 |
[Revert] Convert sharding in SpmdPartitioner::HandleReshape |
|
Standard revert with technical detail; h |
2026-03-24 |
| COMMIT |
0.00 |
[XLA:GPU] Fix FuseGeluActivation for the case with aux and b |
|
Informal explanation with technical cont |
2026-03-24 |
| COMMIT |
0.00 |
Rewrite HloScanInstruction to tree reduction in ReduceWindow |
|
Technical, detailed changelog; normal fo |
2026-03-24 |
| COMMIT |
0.00 |
[XLA:GPU] Add GB300 GPU target configuration to the library |
|
Brief technical commit, strong domain re |
2026-03-24 |
| COMMIT |
0.00 |
[IFRT IR] Return mlir::walk::interrupted instead of op->emit |
|
Minimal technical phrasing, direct and c |
2026-03-24 |
| COMMIT |
0.00 |
PR #39718: [ROCm] Disable keep_going for rocm pipelines |
|
Template PR sections, some formality but |
2026-03-24 |
| COMMIT |
0.00 |
[IFRT IR] Update verify sharding specified pass to only be a |
|
Domain-specific commit, no signs of AI p |
2026-03-24 |
| COMMIT |
0.00 |
[XLA:GPU] Update caches in MultiOutputFusion pass. |
|
Technical, brief changelog; no AI indica |
2026-03-24 |
| COMMIT |
0.00 |
GPU thunk tests require direct access to GPU executables and |
|
Casual domain-specific tone, no AI hallm |
2026-03-24 |
| COMMIT |
0.00 |
Add test for FullyConnected int16 versioning. |
|
Contains domain jargon and casual phrasi |
2026-03-24 |
| COMMIT |
0.00 |
Convert sharding in `SpmdPartitioner::HandleReshape` to supp |
|
Concise technical commit, no AI signals. |
2026-03-24 |
| COMMIT |
0.00 |
[IFRT] Fix `EquivalentLayouts` to use shard shapes to get co |
|
Technical description, natural tone, dom |
2026-03-24 |
| COMMIT |
0.00 |
Explicitly quiesce the async_dispatch threads when shutting |
|
Short, domain-specific, no AI traits. |
2026-03-24 |
| COMMIT |
0.00 |
Reverts 7e75a5a75f308ba630f734e69972e9e89c777e19 |
|
Revert message, typical terse human comm |
2026-03-24 |
| COMMIT |
0.00 |
Simplify coordination service config. |
|
Changelog style, no AI phrasing or trait |
2026-03-24 |
| COMMIT |
0.00 |
PR #38955: [XLA:GPU] include cuda VMM allocator to GPU clien |
|
Technical explanation, domain details, n |
2026-03-24 |
| COMMIT |
0.00 |
[XLA:MSA] Remove dead code - do_not_touch_instructions is ne |
|
Terse, domain-specific, very brief. |
2026-03-24 |
| COMMIT |
0.00 |
[SDY][Cleanup] Use mlir::sdy::getTensorRank in StableHLO imp |
|
Succinct, domain specific, no AI traits. |
2026-03-24 |
| COMMIT |
0.00 |
Remove Serialize() from Topology interface. |
|
Direct technical phrasing, no AI hallmar |
2026-03-23 |
| COMMIT |
0.00 |
This CL modifies the MLIR MemoryEffects for RecordEventMetri |
|
Technical jargon and detailed explanatio |
2026-03-23 |
| COMMIT |
0.00 |
Add int16 x int16 support to TFLite FC op |
|
Informal tone with domain-specific detai |
2026-03-23 |
| COMMIT |
0.00 |
PR #39428: [xla:gpu] Use generic AsynStart/Done thunks for h |
|
Structured changelog, domain terms, huma |
2026-03-23 |
| COMMIT |
0.00 |
[ReplicaGroupV3] SPMD partitioner changes to support v3 in d |
|
Concise and domain-specific, no AI trait |
2026-03-23 |
| COMMIT |
0.00 |
Add utility for constructing a basic CommonPjRtLoadedExecuta |
|
Technical language, concise reasoning, h |
2026-03-23 |
| COMMIT |
0.00 |
Bump rules_ml_toolchain to version with reverted SYCL (OneAP |
|
Brief, domain term usage, no AI signs. |
2026-03-23 |
| COMMIT |
0.00 |
Update Megascale debugging playbook. |
|
Terese description, template likely; cle |
2026-03-23 |
| COMMIT |
0.00 |
A partial and manual rollback for factor out. |
|
Short, cryptic, and informal; human writ |
2026-03-23 |
| COMMIT |
0.00 |
Do not call `HostShapeToDeviceShape` on device shapes |
|
Direct technical content, informal tone, |
2026-03-23 |
| PR |
0.00 |
Fix `HandleReshape` using old `operand_sharding` where `oper |
|
— |
2026-03-25 |
| PR |
0.00 |
Remove legacy event-based synchronization from RaggedAllToAl |
|
— |
2026-03-23 |
| PR |
0.00 |
PR #39382: [xla:gpu] Migrate collective thunks to generic As |
|
— |
2026-03-25 |
| PR |
0.00 |
Add aggregation mode to cupti collector. |
|
— |
2026-03-18 |
| PR |
0.00 |
Fix NCLL use rules_python py_binary to accommodate rules_pyt |
|
— |
2026-03-25 |
| PR |
0.00 |
Roll back defaulting dot precision to BF16_BF16_F32 for F32 |
|
— |
2026-03-25 |
| PR |
0.00 |
[XLA:CPU] Add Softmax regression test for AArch64 (Pass for |
|
— |
2026-03-25 |
| PR |
0.00 |
Migrate grouped_convolution_test to PjRt runtime. |
|
— |
2026-03-25 |
| PR |
0.00 |
Rewrite matching python_repo requirement entries with local |
|
— |
2026-03-25 |
| PR |
0.00 |
[XLA:CPU]: Make OneAPI builds blocking |
|
— |
2026-03-24 |
| PR |
0.00 |
Migrate while_test to use PjRt runtime. |
|
— |
2026-03-25 |
| PR |
0.00 |
add new visibility. |
|
— |
2026-03-25 |
| PR |
0.00 |
Fix deserialization of empty tuples in OriginalValue |
|
— |
2026-03-20 |
| PR |
0.00 |
Check memory space color in `CommonPjRtLoadedExecutable::Che |
|
— |
2026-03-21 |
| PR |
0.00 |
Internal build file cleanup. |
|
— |
2026-03-25 |
| PR |
0.00 |
Add recognize reduce window |
|
— |
2026-03-25 |
| PR |
0.00 |
Disable compatibility test for Python 3.14 since google-past |
|
— |
2026-03-18 |
| PR |
0.00 |
Fix forward on [XLA] Disalbe LICM on scalar and support Rang |
|
— |
2026-03-24 |
| PR |
0.00 |
Remove PjRtTopologyDescription::Serialize method. |
|
— |
2026-03-24 |
| PR |
0.00 |
Refactor vhlo::TensorV1Attr to store DenseElementsAttr |
|
— |
2026-03-23 |
| PR |
0.00 |
[IFRT IR] Add alias for IfrtShardingParamAttr to reduce size |
|
— |
2026-03-21 |
| PR |
0.00 |
utilize PreOptimizationHook and avoid code duplication |
|
— |
2026-03-24 |
| PR |
0.00 |
Fix HloDCE pruning root multi-output fusions with side-effec |
|
— |
2026-03-19 |
| PR |
0.00 |
Reverts e950a4b7a4c64fca41b9b040c8ffb5e2fd5e1041 |
|
— |
2026-03-25 |
| PR |
0.00 |
Move tests from service/gpu/tests to backends/gpu/tests |
|
— |
2026-03-25 |
| PR |
0.00 |
Integrate LLVM at llvm/llvm-project@6ac9ab87dac3 |
|
— |
2026-03-25 |
| PR |
0.00 |
Allow mixing V2 and V3 shardings in utility functions. |
|
— |
2026-03-24 |
| PR |
0.00 |
Enable global scheduler for priority-aware batching in all c |
|
— |
2026-03-24 |
| PR |
0.00 |
Log configs in coordination service `client_server_test.cc` |
|
— |
2026-03-25 |
| PR |
0.00 |
Remove unused device info related code from coordination ser |
|
— |
2026-03-24 |
| PR |
0.00 |
[mGPU] Pass CPU target machine options to the Mosaic GPU ins |
|
— |
2026-03-19 |
| PR |
0.00 |
[PjRt-IFRT] Remove AdjustExecutableDevicesForPmap now that j |
|
— |
2026-03-24 |
| PR |
0.00 |
Set call sharding also when func results have no shardings b |
|
— |
2026-03-25 |
| PR |
0.00 |
Fix/resource sparse adagrad dtype check |
|
Short technical statement, typo present, |
2026-03-25 |
| PR |
0.00 |
Support the XLA GPU compilation flags in Orbax |
|
— |
2026-03-23 |
| PR |
0.00 |
Add option to allow resplitting for priority aware scheduler |
|
— |
2026-03-13 |
| PR |
0.00 |
Add method to get implicitly replicated axes in ``NamedShard |
|
— |
2026-03-20 |
| PR |
0.00 |
Hosting builds for pull requests check on GitHub - Windows |
|
— |
2026-03-21 |
| PR |
0.00 |
Update comments to reflect the new name of the CSE pass. |
|
— |
2026-03-24 |
| PR |
0.00 |
Reverts 94e0685f03ee7e7d13e2d6a04025cbe8b00b21f1 |
|
— |
2026-03-25 |
| PR |
0.00 |
Remove unused `TransactionToken` and associated methods from |
|
— |
2026-03-13 |
| PR |
0.00 |
PR #38756: [ROCm] Modify fusion_emitter_large_test to work o |
|
— |
2026-03-25 |
| PR |
0.00 |
PR #38756: [ROCm] Modify fusion_emitter_large_test to work o |
|
— |
2026-03-25 |
| PR |
0.00 |
Inline addCommonPreImportPasses. |
|
— |
2026-03-24 |
| PR |
0.00 |
Use utils to resolve conflicts on func and call results duri |
|
— |
2026-03-25 |
| PR |
0.00 |
Pass the cpu host target config from gpu compiler through th |
|
— |
2026-03-25 |
| PR |
0.00 |
Resolve func and call sharding conflicts only during unflatt |
|
— |
2026-03-25 |
| PR |
0.00 |
[CublasLt] Bump cache key. |
|
— |
2026-03-25 |
| PR |
0.00 |
Hoist HoistFusedBitcasts above GemmFusionSwapOperands |
|
— |
2026-03-12 |
| PR |
0.00 |
Fix unsorted_segment_max NaN propagation on GPU (Issue #1066 |
|
Technical summary with test details, cas |
2026-03-25 |
| PR |
0.00 |
PR #39622: [ROCm] Use BFCAllocator for scratch allocations n |
|
— |
2026-03-25 |
| PR |
0.00 |
[CollectiveCombiner] Deduplicate redundant buffers sent |
|
— |
2026-03-21 |
| PR |
0.00 |
Sample text to avoid presubmit check error |
|
— |
2026-03-24 |
| PR |
0.00 |
Migrate dynamic_reshape_test to PjRt runtime. |
|
— |
2026-03-24 |
| PR |
0.00 |
[XLA:GPU] Allow to VLOG command buffer profile names and the |
|
— |
2026-03-25 |
| PR |
0.00 |
493236662: Support to CUDA 13 |
|
Brief, domain-specific commit title; cle |
2026-03-23 |
| PR |
0.00 |
PR #39797: [xla:gpu] Track while loop state in Thunk progers |
|
— |
2026-03-24 |
| PR |
0.00 |
[XLA:GPU] Support triton gemm fusions in the indexing perf m |
|
— |
2026-03-25 |
| PR |
0.00 |
[Convolution] Derives cuDNN fusion conv kind from HLO instea |
|
— |
2026-02-17 |
| PR |
0.00 |
Simplify utils for inserting reshards to resolve conflicts o |
|
— |
2026-03-24 |
| PR |
0.00 |
Resolve conflicts by adding reshards also when outsharding o |
|
— |
2026-03-25 |
| PR |
0.00 |
PR #39309: [ROCm] Add scope_range_id support to ROCm profile |
|
— |
2026-03-25 |
| PR |
0.00 |
PR #39730: [ROCm] Skip denorm fp16 dots on MI200 architcture |
|
— |
2026-03-25 |
| PR |
0.00 |
[SPMD] Add support for a multi pad operation |
|
— |
2026-03-25 |
| PR |
0.00 |
Add host target machine options to GpuTopology. |
|
— |
2026-03-17 |
| PR |
0.00 |
PR #34957: [ROCm] Enable matmul perf table implementation fo |
|
— |
2026-03-25 |
| PR |
0.00 |
Add overload of IsCompatibleWithTargetTopology that accepts |
|
— |
2026-03-25 |
| PR |
0.00 |
[XLA:GPU] Plumb through fast interconnect slice size overrid |
|
— |
2026-03-24 |
| PR |
0.00 |
[XLA:GPU] NFC: Move TilingFromAnnotatedFusion to a new libra |
|
— |
2026-03-25 |
| PR |
0.00 |
PR #39497: [XLA:GPU][oneAPI] Register base oneCCL collective |
|
— |
2026-03-25 |
| PR |
0.00 |
Add manual axes to call op only based on the named computati |
|
— |
2026-03-25 |
| PR |
0.00 |
Add a new collective permute cse pass |
|
— |
2026-03-25 |
| PR |
0.00 |
[XLA:GPU] Return detailed estimates from the dot cost model |
|
— |
2026-03-25 |
| PR |
0.00 |
Simplify reduce_window_test setup. |
|
— |
2026-03-25 |
| PR |
0.00 |
[XLA:GPU] Remove ragged dot fusion wrapper. |
|
— |
2026-03-25 |
| PR |
0.00 |
Add manual axes to created copy ops on func/call sharding co |
|
— |
2026-03-25 |
| PR |
0.00 |
Fix SIGSEGV from integer overflow in TFLite tensor dimension |
|
Technical description, domain detail, no |
2026-03-18 |
| PR |
0.00 |
Fix integer overflow in Detection PostProcess (heap buffer o |
|
Technical, pointed security fix; domain |
2026-03-18 |
| PR |
0.00 |
Fix integer overflow in Tile kernel MultiplyShapeDims |
|
Technical, concise, domain-specific lang |
2026-03-18 |
| PR |
0.00 |
[XLA:GPU] Default dot precision to BF16_BF16_F32 for F32 arg |
|
— |
2026-03-23 |
| PR |
0.00 |
Automated Code Change |
|
— |
2026-03-25 |
| PR |
0.00 |
[XLA:GPU]: Use error callbacks in the runtime to notify of a |
|
— |
2026-03-19 |
| PR |
0.00 |
[XLA:GPU] Exchange pointer is RaggedAllToAllThunk thunk with |
|
— |
2026-03-24 |
| PR |
0.00 |
Remove unused flag xla_ignore_channel_ids. |
|
— |
2026-03-25 |
| PR |
0.00 |
Automated Code Change |
|
— |
2026-03-25 |
| PR |
0.00 |
PR #39763: [ROCm] Extend select_threshold macro to be more s |
|
— |
2026-03-25 |
| PR |
0.00 |
[XLA:GPU] Rename multimem parameter property |
|
— |
2026-03-25 |
| PR |
0.00 |
[XLA:GPU]: Use the host callback registry for cuda stream. |
|
— |
2026-03-19 |
| PR |
0.00 |
Automated Code Change |
|
— |
2026-03-25 |
| PR |
0.00 |
Automated Code Change |
|
— |
2026-03-25 |
| PR |
0.00 |
Automated Code Change |
|
— |
2026-03-25 |
| PR |
0.00 |
Automated Code Change |
|
— |
2026-03-25 |
| PR |
0.00 |
Drop xla.sdy.import_mhlo_shardings frontend that is used onl |
|
— |
2026-03-24 |
| PR |
0.00 |
Automated Code Change |
|
— |
2026-03-25 |
| PR |
0.00 |
Integrate Triton up to [fabd2eea04](https://github.com/trito |
|
— |
2026-03-25 |
| PR |
0.00 |
[XLA:GPU]: Unify the compute capability b/w codegen and runt |
|
— |
2026-03-24 |
| PR |
0.00 |
Migrate Tiling Expressions from mlir::AffineExpr to xla::Sym |
|
— |
2026-03-24 |
