| COMMIT |
0.40 |
PR #40189: [XLA:GPU] add cuDNN ragged dot fusion support |
|
Contains 'This PR aims to', summary sect |
2026-04-30 |
| PR |
0.30 |
fix: the ctstring_internal in ctstring_internal.h |
|
Detailed vulnerability format, mentions |
2026-04-30 |
| COMMIT |
0.20 |
Automated Code Change |
|
Likely bot message; 'Automated Code Chan |
2026-04-30 |
| COMMIT |
0.20 |
Automated Code Change |
|
Same as #2: standard bot message, but no |
2026-04-30 |
| COMMIT |
0.20 |
Automated Code Change |
|
Identical to #2; generic automation note |
2026-04-30 |
| COMMIT |
0.20 |
Automated Code Change |
|
Same as previous automated messages, gen |
2026-04-30 |
| PR |
0.20 |
Fix tf.math.sign returning 0 for small nonzero complex numbe |
|
Technical, detailed with markdown; no ov |
2026-04-26 |
| PR |
0.10 |
[ROCm] configure hipcc before local_config_rocm |
|
Technical, minimal, no AI signals; compa |
2026-04-30 |
| PR |
0.10 |
Fix eigh_tridiagonal crashing in tf.function when select='v' |
|
Template-like, technical, no evident AI |
2026-04-02 |
| PR |
0.10 |
Fix MapUnstageNoKey crash on out-of-range index (#112757) |
|
Markdown/PR template, detailed bug/tech |
2026-04-29 |
| COMMIT |
0.00 |
AbstractTrackedDeviceBuffer now owns all three: raw buffer, |
|
Concise technical summary with clear dom |
2026-04-30 |
| COMMIT |
0.00 |
Increase memory limit for ASan builds of py_binary targets. |
|
Brief and typical engineering commit mes |
2026-04-30 |
| COMMIT |
0.00 |
Fix off-by-one error in index_flat_map offset validation. |
|
Terse, technical, and lacks AI-like phra |
2026-04-30 |
| COMMIT |
0.00 |
[XLA:GPU] Rollback: Color multi-host mosaic custom calls wit |
|
Direct technical rollback and specific j |
2026-04-30 |
| COMMIT |
0.00 |
Mitigate out-of-bound read in GroupByReducerDataset via corr |
|
Short, specific technical fix; no AI-lik |
2026-04-30 |
| COMMIT |
0.00 |
This change adds test cases to HloVerifier to validate the ` |
|
Clear technical context; grammar is fitt |
2026-04-30 |
| COMMIT |
0.00 |
[tf.data Security] Validate checkpoint values. |
|
Short security note; domain-specific and |
2026-04-30 |
| COMMIT |
0.00 |
Integrate StableHLO at openxla/stablehlo@806a6844 |
|
Typical integration note with precise te |
2026-04-30 |
| COMMIT |
0.00 |
Add subprocess registration hooks to XLA profiler libraries. |
|
Direct technical addition; informal and |
2026-04-30 |
| COMMIT |
0.00 |
Add a nullptr check for `split_provider_` in `TensorSliceDat |
|
Technical fix with explicit function/cla |
2026-04-30 |
| COMMIT |
0.00 |
Change stride type to int64_t in DecodeProtoOp. |
|
Very terse technical message, no AI hall |
2026-04-30 |
| COMMIT |
0.00 |
This change adds integrity check in `SnappyReader::ReadTenso |
|
Succinct and domain-specific, no AI indi |
2026-04-30 |
| COMMIT |
0.00 |
Reverts 4f553ff1bbf72a96011373a47b85afd1621cea52 |
|
Standard Git revert message, human typic |
2026-04-30 |
| COMMIT |
0.00 |
[XLA] keep compilation failures as config results |
|
Technical content with informal tone, cl |
2026-04-30 |
| COMMIT |
0.00 |
[XLA:GPU] Split peer pointer and symmetric memory versions o |
|
Contains future plans and technical cont |
2026-04-30 |
| COMMIT |
0.00 |
Add overflow check for NonMaxSuppressionV5. |
|
Concise technical message, standard prac |
2026-04-30 |
| COMMIT |
0.00 |
Add algebraic simplification for `A/broadcast(B)` => `A*broa |
|
Brief, technical, domain-specific, no AI |
2026-04-30 |
| COMMIT |
0.00 |
Add --allow_pickle flag to saved_model_cli. |
|
Short, human style CLI flag addition. |
2026-04-30 |
| COMMIT |
0.00 |
Integrate LLVM at llvm/llvm-project@4d676e56f0ee |
|
Standard LLVM integration note, human wr |
2026-04-30 |
| COMMIT |
0.00 |
Roll back sharding conversions in `DotHandler` for operands |
|
Detailed technical summary, domain terms |
2026-04-30 |
| COMMIT |
0.00 |
[XLA:CPU] Add 'manual' tag to failing oneDNN tests. |
|
Concise commit message with Jira-style n |
2026-04-30 |
| COMMIT |
0.00 |
[XLA:GPU] Add a test for RaggedAllToAllKernel using NcclSymm |
|
Brief technical summary, standard commit |
2026-04-30 |
| COMMIT |
0.00 |
Sort backend configs in executable HLO text (second attempt) |
|
Revert commit, concise, domain-specific; |
2026-04-30 |
| COMMIT |
0.00 |
Update dependencies for Bzlmod build |
|
Slightly more explanation, but still cas |
2026-04-30 |
| COMMIT |
0.00 |
Rolling back XLA PR 41412 |
|
Simple rollback message, clear context, |
2026-04-30 |
| COMMIT |
0.00 |
Internal build change. |
|
Extremely brief and generic, likely huma |
2026-04-30 |
| COMMIT |
0.00 |
[XLA] fix bitcast that is TRT in tile propagation |
|
Technical shorthand ('TRT'), tersely wri |
2026-04-30 |
| COMMIT |
0.00 |
Move gemm-conv autotuner pass after fusion passes behind a f |
|
Straightforward technical description, n |
2026-04-30 |
| COMMIT |
0.00 |
Correct reservation size for unscheduled use count map. |
|
Technical explanation, slight increased |
2026-04-30 |
| COMMIT |
0.00 |
PR #41837: [XLA:Build] Fetch entire history for main for dif |
|
Template+imported PR info, no stylistic |
2026-04-30 |
| COMMIT |
0.00 |
Delete more lingering auto sharding deps in gpu compiler as |
|
A terse, technical commit message typica |
2026-04-30 |
| COMMIT |
0.00 |
[XLA:GPU] add checks and logs for tile_propagation |
|
Informal tone with human-typical phrasin |
2026-04-30 |
| COMMIT |
0.00 |
Add sharding conversions in `DotHandler` for operands used f |
|
Concise, domain-specific language; no AI |
2026-04-30 |
| COMMIT |
0.00 |
Support 4D gemms in the GPU dot cost model |
|
Technical explanation with natural phras |
2026-04-30 |
| COMMIT |
0.00 |
PR #40529: [XLA:GPU] Dockerized Hermetic Tests for oneAPI |
|
Template and changelog; no signs of AI-g |
2026-04-30 |
| COMMIT |
0.00 |
Introduce `DevicelessEstimateCubSortScratchSize` HLO Pass |
|
Human-like rationale and informal langua |
2026-04-30 |
| COMMIT |
0.00 |
[XLA:GPU] Pass block_k from tiled hlo to dot cost model |
|
Minimal commit message, typical of human |
2026-04-30 |
| COMMIT |
0.00 |
PR #41604: [xla:se] Add CudaUnifiedAllocator |
|
Mostly template changelog; human-typical |
2026-04-30 |
| COMMIT |
0.00 |
PR #41832: add_conv_window_restore_and_amax_support |
|
Bullet-list with code details; no ChatGP |
2026-04-30 |
| COMMIT |
0.00 |
Reverts 4404f546b880e66d018fd2a78339c244323dd841 |
|
Default revert message; mechanical and h |
2026-04-30 |
| COMMIT |
0.00 |
[XLA:GPU] Color multi-host mosaic custom calls with collecti |
|
Concise and technical commit message; no |
2026-04-30 |
| COMMIT |
0.00 |
Support 3D GEMMS in the dot cost model |
|
Informal tone, domain-specific language |
2026-04-30 |
| COMMIT |
0.00 |
Rollback of PR #41399 |
|
Terse human rollback message; explicit r |
2026-04-30 |
| COMMIT |
0.00 |
PR #41787: [xla:gpu] Use strongly typed RankId for peer_addr |
|
Structured PR import, but technical and |
2026-04-30 |
| COMMIT |
0.00 |
PR #41822: [XLA:Build]: Run clang-tidy actions on RBE |
|
Standard PR import message with minimal |
2026-04-30 |
| COMMIT |
0.00 |
[XLA:GPU] Add an ability to override dot tiling in TilingFro |
|
Succinct technical explanation; direct a |
2026-04-30 |
| COMMIT |
0.00 |
configure hipcc before local_config_rocm |
|
Very terse, imperative commit message st |
2026-04-30 |
| COMMIT |
0.00 |
PR #41652: [xla:se] Make CudaVmmAllocator compatible with NC |
|
Highly technical and detailed; no AI-sug |
2026-04-30 |
| COMMIT |
0.00 |
Add loop overhead penalty to L2 time in GPU dot fusion cost |
|
Technical language and clear domain cont |
2026-04-30 |
| COMMIT |
0.00 |
[XLA:GPU] Disable warp-specialization for GEMMs that have a |
|
Technical commit message with domain-spe |
2026-04-30 |
| COMMIT |
0.00 |
Integrate Triton up to [4da2e268dc](https://github.com/trito |
|
Human style with specific references and |
2026-04-30 |
| COMMIT |
0.00 |
Update layout comparison in RemapPlan validation. |
|
Human-written, technical explanation and |
2026-04-30 |
| COMMIT |
0.00 |
[NCCL] Upgrade XLA NCCL version to 2.29.7 |
|
Succinct commit, conventional technical |
2026-04-30 |
| COMMIT |
0.00 |
PR #41758: [XLA:Build]: Add clang-tidy workflow. |
|
Standard PR merge description with techn |
2026-04-30 |
| COMMIT |
0.00 |
PR #41782: [ROCm] Skip F64 dot operation paths in Triton bac |
|
Technical PR, includes jargon, human ton |
2026-04-30 |
| COMMIT |
0.00 |
Automated Code Change |
|
Generic 'Automated Code Change' with no |
2026-04-30 |
| COMMIT |
0.00 |
[XLA:Runtime]: Hide error contexts behind a mutex. |
|
Informal tone, domain-specific reference |
2026-04-30 |
| COMMIT |
0.00 |
PR #41497: [xla] Add a separate pass to propagate metadata a |
|
Technical jargon, informal tone, no sign |
2026-04-30 |
| COMMIT |
0.00 |
Automated Code Change |
|
Generic 'Automated Code Change' with no |
2026-04-30 |
| COMMIT |
0.00 |
Automated Code Change |
|
Generic 'Automated Code Change' with no |
2026-04-30 |
| COMMIT |
0.00 |
PR #41652: [xla:se] Make CudaVmmAllocator compatible with NC |
|
Contains technical details and jargon, n |
2026-04-30 |
| COMMIT |
0.00 |
PR #41399: Gpu profiler deadlock ending up in DelayKernel ti |
|
Informal, technical, uses emoji, clearly |
2026-04-30 |
| COMMIT |
0.00 |
Automated Code Change |
|
Generic 'Automated Code Change' with no |
2026-04-30 |
| COMMIT |
0.00 |
Automated Code Change |
|
Generic 'Automated Code Change' with no |
2026-04-30 |
| COMMIT |
0.00 |
Automated Code Change |
|
Generic 'Automated Code Change' with no |
2026-04-30 |
| COMMIT |
0.00 |
[XLA:TPU] Allocate host buffer directly on host for one dus |
|
Technical commit summary, no AI indicato |
2026-04-30 |
| COMMIT |
0.00 |
[XLA] Add method that allows to split replica groups based o |
|
Technical and domain-specific, no AI phr |
2026-04-30 |
| COMMIT |
0.00 |
[XLA:GPU] Fix the test with old devices. |
|
Brief, specific issue fix with no AI sig |
2026-04-30 |
| COMMIT |
0.00 |
[ReplicaGroupV3] Prefer iota device order in GetMeshFromShar |
|
Technical commit message with domain-spe |
2026-04-30 |
| COMMIT |
0.00 |
Automated Code Change |
|
Clearly marked as automated, not necessa |
2026-04-30 |
| COMMIT |
0.00 |
[XLA:Original Value] Propagate original value for HLOs with |
|
Technical explanation with domain jargon |
2026-04-30 |
| COMMIT |
0.00 |
Automated Code Change |
|
Marked as automated, not necessarily AI- |
2026-04-30 |
| COMMIT |
0.00 |
Automated Code Change |
|
Labeled as an automated change, not AI t |
2026-04-30 |
| COMMIT |
0.00 |
[XLA] Handle partitioning constants a bit more elegantly. |
|
Casual, domain-specific message, no AI s |
2026-04-30 |
| COMMIT |
0.00 |
Pass `xla_cpu_enable_fast_math` state to `AlgebraicSimplifie |
|
Brief technical statement, no AI-generat |
2026-04-30 |
| COMMIT |
0.00 |
* Removes a dead comment in `buffer_assignment.h`. |
|
Succinct changelog, domain-specific, no |
2026-04-30 |
| COMMIT |
0.00 |
Prepare for merging targets by adding the header file |
|
Terse, technical commit message; entirel |
2026-04-30 |
| COMMIT |
0.00 |
[tf.data Security] Check nullptr before calling StartCancel( |
|
Short, direct summary with issue context |
2026-04-30 |
| COMMIT |
0.00 |
Hoist `num_devices()` call out of a loop over all devices. |
|
Casual tone, code reference, no AI style |
2026-04-30 |
| COMMIT |
0.00 |
Enforce start index is strictly less than end index for Iter |
|
Direct procedural wording, no AI phrasin |
2026-04-30 |
| COMMIT |
0.00 |
Fix: Raise OutOfRange error for negative indices in CacheDat |
|
Standard fix summary; short and technica |
2026-04-30 |
| COMMIT |
0.00 |
Update XLA to support async collectives, partially. |
|
Slightly more explanatory, but informal |
2026-04-30 |
| COMMIT |
0.00 |
[XLA:CPU] Fix crash in HLO profiling for N-way conditionals |
|
Technical, specific, informal — human st |
2026-04-29 |
| COMMIT |
0.00 |
Add bounds check for shard_num in GetNextFromShard. |
|
Brevity and domain language, not AI-gene |
2026-04-29 |
| COMMIT |
0.00 |
[tf.data Security] Set query_connection_initialized_ after c |
|
Terse patch description, contextually ap |
2026-04-29 |
| COMMIT |
0.00 |
[tf.data Security] Check null pointer reference in TensorSli |
|
Terse message, clear domain context, no |
2026-04-29 |
| COMMIT |
0.00 |
Add type and shape checks to `GetScalar` in serialization ut |
|
Technical, concise; no signs of AI-gener |
2026-04-29 |
| COMMIT |
0.00 |
[tf.data Security] Fix unchecked vector indexing in concaten |
|
Direct fix description, domain-specific, |
2026-04-29 |
| COMMIT |
0.00 |
[tf.data Security] Null Pointer Dereference in Prefetch Asyn |
|
Technical and concise; lacks AI-style ph |
2026-04-29 |
| COMMIT |
0.00 |
Scan bodies do not need instruction fusion, as scan bodies d |
|
Informal style, explicit reasoning, no f |
2026-04-29 |
| COMMIT |
0.00 |
[tf.data Security] Null Pointer Dereference in Interleave Ca |
|
Direct, brief, technical; not AI-generat |
2026-04-29 |
| COMMIT |
0.00 |
Fix undefined behavior in activity_watcher by adding a fallb |
|
Specific, concise bug fix; no AI charact |
2026-04-29 |
| COMMIT |
0.00 |
Run shardy/google/integrate_latest.sh to integrate. |
|
Standard short integration note, no AI i |
2026-04-29 |
| COMMIT |
0.00 |
Move shardy unflattener into mlir::sdy pipeline. |
|
Technical reasoning with jargon, informa |
2026-04-29 |
| COMMIT |
0.00 |
[XLA:GPU][NFC] drop handling of nested fusions in TilingFrom |
|
Casual language and abbreviations, no AI |
2026-04-29 |
| PR |
0.00 |
PR #41561: [tsl:concurrency] Add automatic Future::Map flatt |
|
— |
2026-04-27 |
| PR |
0.00 |
Reverts 2bebbd3ce2e0b0cfb8508a404d178d89d938c436 |
|
— |
2026-04-30 |
| PR |
0.00 |
[IFRT] Do not register dialects on IFRT IR compilation |
|
— |
2026-04-30 |
| PR |
0.00 |
[IFRT IR] Remove SpmdExpandableInterface |
|
— |
2026-04-30 |
| PR |
0.00 |
The new unified tracked buffer code triggers is_error codepa |
|
— |
2026-04-30 |
| PR |
0.00 |
This CL adds explicit state and bounds validation within `Re |
|
— |
2026-04-30 |
| PR |
0.00 |
Rolling back build breakage. |
|
— |
2026-04-30 |
| PR |
0.00 |
This CL adds robust bounds checking for `provider_index` in |
|
— |
2026-04-30 |
| PR |
0.00 |
Add zizmor security analysis and fix identified issues in Gi |
|
— |
2026-04-28 |
| PR |
0.00 |
Add CreateUsageEventSet, WaitOnStream and DeleteAfter to Com |
|
— |
2026-04-30 |
| PR |
0.00 |
Enable SchedulingState to be passed into ScheduleWithPrefere |
|
— |
2025-08-11 |
| PR |
0.00 |
[PjRt-IFRT] Fill in parameter information in SerializedXlaEx |
|
— |
2026-03-27 |
| PR |
0.00 |
Add check for empty input vector in TensorSliceDataset. |
|
— |
2026-04-29 |
| PR |
0.00 |
Validate dataset IDs in tf.data service dispatcher. |
|
— |
2026-04-30 |
| PR |
0.00 |
build fixes linux arm64 |
|
— |
2026-04-30 |
| PR |
0.00 |
Add a check for file size being smaller than header + footer |
|
— |
2026-04-30 |
| PR |
0.00 |
* Enable variadic operand parsing and verification for kAsyn |
|
— |
2026-04-29 |
| PR |
0.00 |
Support batch padding policy in PriorityTaskQueue |
|
— |
2026-04-30 |
| PR |
0.00 |
introducing Cuda13 build target |
|
— |
2026-04-24 |
| PR |
0.00 |
Support 64-bit work element counts in gather kernels. |
|
— |
2026-04-24 |
| PR |
0.00 |
Add out-of-bounds checks in VariantTensorDataReader. |
|
— |
2026-04-30 |
| PR |
0.00 |
Add a check for tensor count mismatch in snapshot metadata. |
|
— |
2026-04-30 |
| PR |
0.00 |
Use safe integer operations in PoolAllocator::AllocateRaw. |
|
— |
2026-04-30 |
| PR |
0.00 |
Add bounds check before accessing split string in ParseInput |
|
— |
2026-04-30 |
| PR |
0.00 |
Make shape inference for TopKWithUnique more robust to unkno |
|
— |
2026-04-30 |
| PR |
0.00 |
Add a check for empty participants in NcclManager::AllToAll. |
|
— |
2026-04-30 |
| PR |
0.00 |
Replace DCHECK with CHECK and remove unused dependencies in |
|
— |
2026-04-29 |
| PR |
0.00 |
Delete experimental tensorflow/compiler/xla/hlo/experimental |
|
— |
2026-04-30 |
| PR |
0.00 |
[SDY] Update PjRt MLIR/HLO conversion API to accept an optio |
|
— |
2026-04-23 |
| PR |
0.00 |
AbstractTrackedDeviceBuffer now owns all three: raw buffer, |
|
— |
2026-04-27 |
| PR |
0.00 |
Upgrade the kotlin compiler to v2.3.20 and add GitHub workfl |
|
— |
2026-04-23 |
| PR |
0.00 |
Increase memory limit for ASan builds of py_binary targets. |
|
— |
2026-04-29 |
| PR |
0.00 |
Fix MockNcclClientWithGpuTopologyTest failures by explicitly |
|
— |
2026-04-29 |
| PR |
0.00 |
Fix off-by-one error in index_flat_map offset validation. |
|
— |
2026-04-30 |
| PR |
0.00 |
[XLA:GPU] Rollback: Color multi-host mosaic custom calls wit |
|
— |
2026-04-30 |
| PR |
0.00 |
Mitigate out-of-bound read in GroupByReducerDataset via corr |
|
— |
2026-04-30 |
| PR |
0.00 |
This change adds test cases to HloVerifier to validate the ` |
|
— |
2026-04-30 |
| PR |
0.00 |
PR #117311: [ROCm] configure hipcc before local_config_rocm |
|
— |
2026-04-30 |
| PR |
0.00 |
[XLA:GPU] Emit scaled dot in two stages |
|
— |
2026-04-30 |
| PR |
0.00 |
[tf.data Security] Validate checkpoint values. |
|
— |
2026-04-30 |
| PR |
0.00 |
Integrate StableHLO at openxla/stablehlo@806a6844 |
|
— |
2026-04-30 |
| PR |
0.00 |
[XLA:CPU] Split out subcomputation HLO for independent compi |
|
— |
2026-04-30 |
| PR |
0.00 |
Add subprocess registration hooks to XLA profiler libraries. |
|
— |
2026-04-29 |
| PR |
0.00 |
Add a nullptr check for `split_provider_` in `TensorSliceDat |
|
— |
2026-04-30 |
| PR |
0.00 |
Change stride type to int64_t in DecodeProtoOp. |
|
— |
2026-04-30 |
| PR |
0.00 |
This change adds integrity check in `SnappyReader::ReadTenso |
|
— |
2026-04-30 |
| PR |
0.00 |
Reverts 4f553ff1bbf72a96011373a47b85afd1621cea52 |
|
— |
2026-04-30 |
| PR |
0.00 |
[XLA] keep compilation failures as config results |
|
— |
2026-04-23 |
| PR |
0.00 |
Fix XLA CPU memory kind mismatch when pinned_host is request |
|
— |
2026-04-20 |
| PR |
0.00 |
Stop CreateAllocateBufferForDynamicUpdateSlice() when hittin |
|
— |
2026-04-30 |
| PR |
0.00 |
[XLA] Restrict instruction/multi-output fusion inside any kE |
|
— |
2026-04-30 |
| PR |
0.00 |
[XLA:GPU] Split peer pointer and symmetric memory versions o |
|
— |
2026-04-29 |
| PR |
0.00 |
Add overflow check for NonMaxSuppressionV5. |
|
— |
2026-04-30 |
| PR |
0.00 |
Add scalar shape checks to several TensorFlow kernel operati |
|
— |
2026-04-30 |
| PR |
0.00 |
Add algebraic simplification for `A/broadcast(B)` => `A*broa |
|
— |
2026-04-21 |
| PR |
0.00 |
Add --allow_pickle flag to saved_model_cli. |
|
— |
2026-04-30 |
| PR |
0.00 |
Fix async collective replacer. |
|
— |
2026-04-14 |
| PR |
0.00 |
Integrate LLVM at llvm/llvm-project@4d676e56f0ee |
|
— |
2026-04-28 |
| PR |
0.00 |
Roll back sharding conversions in `DotHandler` for operands |
|
— |
2026-04-30 |
| PR |
0.00 |
[XLA:Python] Move bindings for XLA HLO types back to XLA. |
|
— |
2026-04-30 |
| PR |
0.00 |
Add a graph rewrite to fallback to fp32 when fp16 isn't supp |
|
— |
2026-03-13 |
| PR |
0.00 |
Print the fingerprint id string instead of it's value in err |
|
— |
2026-02-06 |
| PR |
0.00 |
[XLA] Vectorize loop in LiteralBase::Piece::EqualElements us |
|
— |
2026-04-30 |
| PR |
0.00 |
Add output tensor size checks for AllGather and AllToAll in |
|
— |
2026-04-30 |
| PR |
0.00 |
Add a check for IFRT call result size. |
|
— |
2026-04-30 |
| PR |
0.00 |
[XLA:CPU] Add 'manual' tag to failing oneDNN tests. |
|
— |
2026-04-30 |
| PR |
0.00 |
Add sharding conversions in `DotHandler` for operands used f |
|
— |
2026-04-30 |
| PR |
0.00 |
[XLA:GPU] Add a test for RaggedAllToAllKernel using NcclSymm |
|
— |
2026-04-30 |
| PR |
0.00 |
Sort backend configs in executable HLO text (second attempt) |
|
— |
2026-04-30 |
| PR |
0.00 |
Update dependencies for Bzlmod build |
|
— |
2026-04-27 |
| PR |
0.00 |
Avoid materializing Iota instructions in HloEvaluator when f |
|
— |
2026-04-29 |
| PR |
0.00 |
[benchmarks] Refactor HLO runner AOT compilation to use mult |
|
— |
2026-04-30 |
| PR |
0.00 |
Rolling back XLA PR 41412 |
|
— |
2026-04-30 |
| PR |
0.00 |
Internal build change. |
|
— |
2026-04-28 |
| PR |
0.00 |
[XLA] fix bitcast that is TRT in tile propagation |
|
— |
2026-04-29 |
| PR |
0.00 |
Move gemm-conv autotuner pass after fusion passes behind a f |
|
— |
2026-04-10 |
| PR |
0.00 |
Correct reservation size for unscheduled use count map. |
|
— |
2026-04-29 |
| PR |
0.00 |
PR #41837: [XLA:Build] Fetch entire history for main for dif |
|
— |
2026-04-30 |
| PR |
0.00 |
[XLA:GPU] Disable LLVM op diagnostics in CompileTritonToLLVM |
|
— |
2026-04-30 |
| PR |
0.00 |
[XLA:GPU] Introduce cost model config selection for triton g |
|
— |
2026-04-30 |
| PR |
0.00 |
Delete more lingering auto sharding deps in gpu compiler as |
|
— |
2026-04-29 |
| PR |
0.00 |
Introduce `DevicelessEstimateCubSortScratchSize` HLO Pass |
|
— |
2026-04-30 |
| PR |
0.00 |
[XLA:GPU] add checks and logs for tile_propagation |
|
— |
2026-04-29 |
| PR |
0.00 |
Automated Code Change |
|
— |
2026-04-30 |
| PR |
0.00 |
Add sharding conversions in `DotHandler` for operands used f |
|
— |
2026-04-29 |
| PR |
0.00 |
Add support for HLO Scan in Triton. |
|
— |
2026-04-22 |
| PR |
0.00 |
Remove usages of xla_gpu_experimental_aot_compiled_thunks |
|
— |
2026-04-30 |
| PR |
0.00 |
Automated Code Change |
|
— |
2026-04-30 |
| PR |
0.00 |
Register CuTe DSL XLA FFI handler |
|
— |
2026-04-30 |
| PR |
0.00 |
Support 4D gemms in the GPU dot cost model |
|
— |
2026-04-30 |
| PR |
0.00 |
Automated Code Change |
|
— |
2026-04-30 |
| PR |
0.00 |
Automated Code Change |
|
— |
2026-04-30 |
| PR |
0.00 |
Add compile-only mode to multihost_hlo_runner. |
|
— |
2026-04-29 |
| PR |
0.00 |
Fix int32 overflow in BiasAdd/BiasAddGrad GPU kernels |
|
All bullet points, domain details, no AI |
2026-02-04 |
| PR |
0.00 |
Fix tf.cond to validate both branches in eager mode |
|
Informal explanation, clearly human phra |
2026-02-03 |
