| commit |
0.70 |
Automated Code Change |
|
Explicitly labeled as 'Automated Code Ch |
2026-03-26 |
| commit |
0.70 |
Automated Code Change |
|
Title is explicitly 'Automated Code Chan |
2026-03-26 |
| commit |
0.70 |
Automated Code Change |
|
Title is explicitly 'Automated Code Chan |
2026-03-26 |
| commit |
0.20 |
Automated Code Change |
|
Generic phrase 'Automated Code Change' s |
2026-03-27 |
| commit |
0.00 |
emit BitonicSort as CustomKernelThunk |
|
Technical commit message; terse and spec |
2026-03-27 |
| commit |
0.00 |
Fix typo in `tsl::monitoring::Buckets::Exponential` document |
|
Concise human-style commit, domain-speci |
2026-03-27 |
| commit |
0.00 |
Remove unused `TransactionToken` and associated methods from |
|
Removal refactoring message; no AI signa |
2026-03-27 |
| commit |
0.00 |
Add support for Named sharding format in ShardingFormatPicke |
|
Detailed technical PR, domain terms, no |
2026-03-27 |
| commit |
0.00 |
Migrate tuple_test to PjRt runtime. |
|
Direct, technical test migration with mi |
2026-03-27 |
| commit |
0.00 |
Fix `GetMeshAxesPartitionGroupsForReplication` for NamedShar |
|
Technical fix explanation, clear domain |
2026-03-27 |
| commit |
0.00 |
[IFRT IR] Remove ifrt_mpmd_py bindings are they've been repl |
|
Informal commit, typo but clear domain k |
2026-03-27 |
| commit |
0.00 |
Simplify PjRtDeviceEvent by moving event tracking to CommonP |
|
Terse commit; technical detail, not AI-l |
2026-03-27 |
| commit |
0.00 |
[PjRt-IFRT] Remove spurious warning messages during executab |
|
Technical explanation; informative, not |
2026-03-27 |
| commit |
0.00 |
Integrate LLVM at llvm/llvm-project@9a0b003dde83 |
|
Repository sync message, standard for LL |
2026-03-27 |
| commit |
0.00 |
Roll forward with fix |
|
Very terse, standard revert commit messa |
2026-03-27 |
| commit |
0.00 |
New recognize_reduce_window_test unnecessarily depends on de |
|
Concise, correct, and technical human st |
2026-03-27 |
| commit |
0.00 |
round_trip_packed_literal_test should not depend on client_l |
|
Short message with project-specific term |
2026-03-27 |
| commit |
0.00 |
Fix kRecv resource limit violation in top-down scheduling. |
|
Direct and technical; common human commi |
2026-03-27 |
| commit |
0.00 |
Internal clean up. |
|
Extremely brief, typical human commit me |
2026-03-27 |
| commit |
0.00 |
[XLA:MSA] Allow conditional outputs in alternate memory. |
|
Uses project and domain-specific abbrevi |
2026-03-27 |
| commit |
0.00 |
[XLA:GPU]: Fix layout when doing a gathered load. |
|
Technical detail and informal list; natu |
2026-03-27 |
| commit |
0.00 |
Rewrite matching python_repo requirement entries with local |
|
Detailed but context-specific and techni |
2026-03-27 |
| commit |
0.00 |
[XLA:GPU] Emit copy instruction when Mosaic operation is use |
|
Specialized explanation, natural human w |
2026-03-27 |
| commit |
0.00 |
Integrate LLVM at llvm/llvm-project@5a403150b468 |
|
Standard integration message; explicit r |
2026-03-27 |
| commit |
0.00 |
Add `Label` to `//third_party/tensorflow/compiler/xla/tsl/ut |
|
Succinct, domain-specific, no AI signals |
2026-03-27 |
| commit |
0.00 |
Add PjRtArray overloads to support xla::ifrt::PjRtLayout. |
|
Brief technical language, no sign of AI- |
2026-03-27 |
| commit |
0.00 |
Unifiy TargetConfig logic in GpuCompiler |
|
Slightly more explanation but informal a |
2026-03-27 |
| commit |
0.00 |
Fix for when adding reshard for a call with null result shar |
|
Concise bugfix language typical of commi |
2026-03-27 |
| commit |
0.00 |
Integrate StableHLO at openxla/stablehlo@fef90093 |
|
Standard integration commit, terse and d |
2026-03-27 |
| commit |
0.00 |
Pass the cpu host target config from gpu compiler through th |
|
Technical, terse, domain-specific phrasi |
2026-03-27 |
| commit |
0.00 |
Reverts ba9ed044f3a801c2396f37fd4a212e8fbaa3ba02 |
|
Very terse, standard revert message. |
2026-03-27 |
| commit |
0.00 |
The collapsed_slice_dims/inserted_window_dims may not be in |
|
Concise, context-specific jargon, non-AI |
2026-03-27 |
| commit |
0.00 |
[XLA:GPU] Return detailed estimates from the dot cost model |
|
Detailed but plain, technical domain, hu |
2026-03-27 |
| commit |
0.00 |
Remove HoistFusedBitcasts from autotuner pipeline. |
|
Technical changelog with domain referenc |
2026-03-27 |
| commit |
0.00 |
[XLA:GPU] Support triton gemm fusions in the indexing perf m |
|
Domain-specific vocabulary, technical ex |
2026-03-27 |
| commit |
0.00 |
Reverts df55d43fbd71db4ea8cbd5d014f7baed1297f56f |
|
Standard revert, extremely minimal, huma |
2026-03-27 |
| commit |
0.00 |
Simplify reduce_window_test setup. |
|
Concise summary, technical content, info |
2026-03-27 |
| commit |
0.00 |
Prefer transpose bitcasts over reshape bitcasts - when we ar |
|
Technical reasoning, ShapeUtil jargon, h |
2026-03-27 |
| commit |
0.00 |
Add comparison of device description strings to GpuTopology |
|
Terse explanation, domain-specific, no A |
2026-03-27 |
| commit |
0.00 |
[XLA:GPU] Default dot precision to BF16_BF16_F32 for F32 arg |
|
Technical, informal tone and domain-spec |
2026-03-27 |
| commit |
0.00 |
[XLA:GPU] Remove collective multimem registry since now we c |
|
Terse, technical commit typical of human |
2026-03-27 |
| commit |
0.00 |
[XLA:GPU] Add missing include in multi_gpu_barrier_test |
|
Brief, direct phrasing, clear human auth |
2026-03-27 |
| commit |
0.00 |
Fix proto reconstruction for RaggedAllToAllThunk |
|
Specific technical details and casual st |
2026-03-27 |
| commit |
0.00 |
[XLA:BUILD]: Check windows build failure fix |
|
Short, non-formal message; human commit |
2026-03-27 |
| commit |
0.00 |
[IFRT IR] Bump IFRT IR to version 0.2.0 |
|
Standard version bump message, human aut |
2026-03-27 |
| commit |
0.00 |
Integrate LLVM at llvm/llvm-project@3c8e9499ac4c |
|
Factual and terse; typical of human LLVM |
2026-03-27 |
| commit |
0.00 |
Increase linker memory for `tfrt_session_python_test` under |
|
Direct, technical, no AI stylistic signa |
2026-03-27 |
| commit |
0.00 |
Adds ```output_shape()``` in hlo_module.h for entry computat |
|
Concise phrasing, technical detail, huma |
2026-03-27 |
| commit |
0.00 |
Allow mixing V2 and V3 shardings in utility functions. |
|
Technical commit message with specific c |
2026-03-27 |
| commit |
0.00 |
Implicitly replicate axes in V2ToV3. |
|
Very brief technical commit message abou |
2026-03-27 |
| commit |
0.00 |
Bump rules_ml_toolchain to version with changes for ROCM |
|
Concise technical change statement with |
2026-03-27 |
| commit |
0.00 |
Improve readability of TSL Monitoring test-failure logs |
|
Direct technical explanation with clear |
2026-03-26 |
| commit |
0.00 |
Make involuntary full rematerialization an fatal error when |
|
Terse, domain-specific error handling de |
2026-03-26 |
| commit |
0.00 |
Add optimization pass to recognize reductions and form highe |
|
Specific optimization pass addition with |
2026-03-26 |
| commit |
0.00 |
Reduce unique ways of calling PjRt's Execute in HloRunnerPjR |
|
Concise refactoring goal using jargon an |
2026-03-26 |
| commit |
0.00 |
Update PjRtDeviceEvent subclasses to not need to track metho |
|
Direct, brief API/subclass refactoring s |
2026-03-26 |
| commit |
0.00 |
Fully switch to ExecuteReplicated in ClientLibraryTestRunner |
|
Standardizing effort explained with acro |
2026-03-26 |
| commit |
0.00 |
Replace `.value()` with `*` for accessing `std::optional` va |
|
Specific code style change referencing i |
2026-03-26 |
| commit |
0.00 |
[XLA] Re-enable scalar LICM and make collective pipeliner lo |
|
XLA-specific optimization re-enablement |
2026-03-26 |
| commit |
0.00 |
Add `stablehlo.async_start` and `stablehlo.async_done`. |
|
Detailed technical addition with PR refe |
2026-03-26 |
| commit |
0.00 |
Fix HloDCE pruning root multi-output fusions with side-effec |
|
Detailed technical analysis of a bug fix |
2026-03-26 |
| commit |
0.00 |
Add scatter partitioning without conflict resolution. |
|
Brief, direct description with standard |
2026-03-26 |
| commit |
0.00 |
[XLA:CPU] Add Softmax regression test for AArch64 (Pass for |
|
Simple test addition with TODO marker - |
2026-03-26 |
| commit |
0.00 |
Add ```NamedSharding``` support to `PartialReplicateReshardC |
|
Technical feature addition with specific |
2026-03-26 |
| commit |
0.00 |
[TSL] Move MaybeOwning to TSL. |
|
Minimal code movement commit - typical h |
2026-03-26 |
| commit |
0.00 |
Integrate LLVM at llvm/llvm-project@f2f94597671f |
|
Standard LLVM integration commit format |
2026-03-26 |
| commit |
0.00 |
[Convolution] Derives cuDNN fusion conv kind from HLO instea |
|
Technical refactoring description with b |
2026-03-26 |
| commit |
0.00 |
[IFRT IR] Modify -ifrt-dump-atom-programs to dump valid IFRT |
|
Brief tool modification description - ty |
2026-03-26 |
| commit |
0.00 |
Rollforward with fixes as described here: cl/888943300 |
|
Rollforward commit referencing internal |
2026-03-26 |
| commit |
0.00 |
Add method to get implicitly replicated axes in ``NamedShard |
|
Technical method addition description wi |
2026-03-26 |
| commit |
0.00 |
Add more unit tests to unflatten for empty and no shardings |
|
Brief, technical commit message typical |
2026-03-26 |
| commit |
0.00 |
[XLA:GPU] Move GPU-only tests from xla/tests to xla/backends |
|
Concise XLA/GPU directory move statement |
2026-03-26 |
| commit |
0.00 |
PR #39763: [ROCm] Extend select_threshold macro to be more s |
|
Uses PR template with structured heading |
2026-03-26 |
| commit |
0.00 |
Integrate Triton up to [fabd2eea04](https://github.com/trito |
|
Technical integration description with c |
2026-03-26 |
| commit |
0.00 |
[XLA:GPU] Allow to VLOG command buffer profile names and the |
|
Technical XLA/GPU feature description wi |
2026-03-26 |
| commit |
0.00 |
[XLA:GPU] NFC: Move TilingFromAnnotatedFusion to a new libra |
|
Informal explanation with 'NFC' and casu |
2026-03-26 |
| commit |
0.00 |
PR #39168: [GPU] Clean up setting of kernel dynamic shared m |
|
PR template used but free-text is terse |
2026-03-26 |
| commit |
0.00 |
[XLA:GPU] use triton-licm pass instead of mlir |
|
Direct technical statement about making |
2026-03-26 |
| commit |
0.00 |
Resolve conflicts by adding reshards also when outsharding o |
|
Detailed technical explanation with refe |
2026-03-26 |
| commit |
0.00 |
PR #39692: Skip python_hlo_runner_test when transformer_engi |
|
Clear technical purpose with PR template |
2026-03-26 |
| commit |
0.00 |
[Convolution] Add support for Convolution layout normalizati |
|
Brief, technical changelog style; common |
2026-03-26 |
| commit |
0.00 |
[CublasLt] Bump cache key. |
|
Very terse commit message; typical human |
2026-03-26 |
| commit |
0.00 |
PR #39730: [ROCm] Skip denorm fp16 dots on MI200 architcture |
|
Follows PR template; free text contains |
2026-03-26 |
| commit |
0.00 |
Improve symbolic pretty printing of Symbolic Expressions in |
|
Human technical explanation with direct |
2026-03-26 |
| commit |
0.00 |
Add manual axes to call op only based on the named computati |
|
Human-written; domain-specific terms, ca |
2026-03-26 |
| commit |
0.00 |
PR #39535: [XLA:GPU] Add end-to-end test for VMM memory allo |
|
Detailed technical changelog; domain-spe |
2026-03-26 |
| commit |
0.00 |
Add manual axes to created copy ops on func/call sharding co |
|
Brief technical note; direct and lacks A |
2026-03-26 |
| commit |
0.00 |
Add and use manual axes on func ops for dedupping funcs. |
|
Terse, domain-specific; human-like commi |
2026-03-26 |
| commit |
0.00 |
PR #39097: [GPU] Enable PDL by default. |
|
Minimal summary with domain content; not |
2026-03-26 |
| commit |
0.00 |
[Autotuner] Capture all output in profiler to improve correc |
|
Single technical statement; brief and to |
2026-03-26 |
| commit |
0.00 |
PR #39797: [xla:gpu] Track while loop state in Thunk progers |
|
Casual, explanatory tone; lacks AI-annou |
2026-03-26 |
| commit |
0.00 |
PR #38756: [ROCm] Modify fusion_emitter_large_test to work o |
|
Some template headings, but free-text is |
2026-03-26 |
| commit |
0.00 |
PR #39742: [xla:gpu] Update GPU memory colorer to handle cus |
|
Technical, concise; domain-specific with |
2026-03-26 |
| commit |
0.00 |
PR #39673: [ROCm] Fix ShardedAutotuningWorks test for ROCm p |
|
Mostly technical, has template sections; |
2026-03-26 |
| commit |
0.00 |
Fix 6 IncludeCleaner findings |
|
Very short, direct technical fix; no AI |
2026-03-26 |
| commit |
0.00 |
Remove unused flag xla_ignore_channel_ids. |
|
Brief, technical commit message with dom |
2026-03-26 |
| commit |
0.00 |
[XLA:GPU] update test names |
|
Informal tone; succinct technical domain |
2026-03-26 |
| commit |
0.00 |
Move tests from service/gpu/tests to backends/gpu/tests |
|
Direct, specific technical action with b |
2026-03-26 |
| commit |
0.00 |
allow CustomKernelThunk conversion for command buffer |
|
Short, technical; no AI phrasing or form |
2026-03-26 |
| commit |
0.00 |
Fix the algebraic simplifier conv->multiply pass to not cons |
|
Direct technical explanation; domain-spe |
2026-03-26 |
| pr |
0.00 |
[XLA:GPU] Get information about NVLink domain size from devi |
|
— |
2026-03-27 |
| pr |
0.00 |
[XLA:MSA] Allow conditional inputs with uses after the condi |
|
— |
2026-03-01 |
| pr |
0.00 |
PR #40017: [xla:gpu] Correctly track thunk progress in prese |
|
— |
2026-03-27 |
| pr |
0.00 |
Populate xla_input_shapes for unpacked inputs in IfrtServing |
|
— |
2026-03-26 |
| pr |
0.00 |
Internal change, adds a dependency. |
|
— |
2026-01-09 |
| pr |
0.00 |
emit BitonicSort as CustomKernelThunk |
|
— |
2026-03-27 |
| pr |
0.00 |
Enhance IfrtRestoreVariableOp to support non-variable tensor |
|
— |
2026-03-27 |
| pr |
0.00 |
Fix typo in `tsl::monitoring::Buckets::Exponential` document |
|
— |
2026-03-26 |
| pr |
0.00 |
Automated Code Change |
|
— |
2026-03-27 |
| pr |
0.00 |
[SC Serving] Support variable valencies and skip-scatter in |
|
— |
2026-03-27 |
| pr |
0.00 |
Remove unused `TransactionToken` and associated methods from |
|
— |
2026-03-13 |
| pr |
0.00 |
Make CPU devices report as cpu:0 instead of TFRT_CPU_0. |
|
— |
2026-03-27 |
| pr |
0.00 |
Add support for Named sharding format in ShardingFormatPicke |
|
— |
2026-03-25 |
| pr |
0.00 |
[SC Serving] Clamps embedding IDs strictly to `[0, vocab_siz |
|
— |
2026-03-27 |
| pr |
0.00 |
[SC Serving] Bypass `sample_ids` scatter for univalent featu |
|
— |
2026-03-27 |
| pr |
0.00 |
[XLA] Remove obsolete shmap & sdy manual computation inliner |
|
— |
2026-03-26 |
| pr |
0.00 |
Add a `device_memory_limit_bytes` attribute to PjRt GPU devi |
|
— |
2026-03-27 |
| pr |
0.00 |
Migrate tuple_test to PjRt runtime. |
|
— |
2026-03-27 |
| pr |
0.00 |
Fix `GetMeshAxesPartitionGroupsForReplication` for NamedShar |
|
— |
2026-03-25 |
| pr |
0.00 |
[IFRT IR] Add ifrt_ir_compile_options to compile_mpmd() |
|
— |
2026-03-27 |
| pr |
0.00 |
Support passing a creation mode for TSL `RecursivelyCreateDi |
|
— |
2026-03-26 |
| pr |
0.00 |
[IFRT IR] Remove ifrt_mpmd_py bindings are they've been repl |
|
— |
2026-03-27 |
| pr |
0.00 |
Make an explicit ref type for PjRtDeviceEvent to allow easil |
|
— |
2026-03-27 |
| pr |
0.00 |
Simplify PjRtDeviceEvent by moving event tracking to CommonP |
|
— |
2026-03-27 |
| pr |
0.00 |
[PjRt-IFRT] Remove spurious warning messages during executab |
|
— |
2026-03-27 |
| pr |
0.00 |
an internal change |
|
— |
2026-01-09 |
| pr |
0.00 |
Integrate LLVM at llvm/llvm-project@9a0b003dde83 |
|
— |
2026-03-27 |
| pr |
0.00 |
Fix aligned_alloc usage for older Android APIs |
|
— |
2026-03-27 |
| pr |
0.00 |
[XLA:GPU] form regions in symbolic analysis |
|
— |
2026-01-09 |
| pr |
0.00 |
Roll forward with fix |
|
— |
2026-03-27 |
| pr |
0.00 |
New recognize_reduce_window_test unnecessarily depends on de |
|
— |
2026-03-27 |
| pr |
0.00 |
PR #39953: [xla:gpu] Delete dead code |
|
— |
2026-03-27 |
| pr |
0.00 |
round_trip_packed_literal_test should not depend on client_l |
|
— |
2026-03-27 |
| pr |
0.00 |
Fix kRecv resource limit violation in top-down scheduling. |
|
— |
2026-03-25 |
| pr |
0.00 |
[ReplicaGroupV3][JAX+stablehlo][3/n] add stablehlo bindings |
|
— |
2026-03-27 |
| pr |
0.00 |
[ReplicaGroupV3][JAX+stablehlo] Shardy updates for RGV3 |
|
— |
2026-03-27 |
| pr |
0.00 |
[ReplicaGroupV3][JAX+stablehlo][2/n] StableHLO changes to su |
|
— |
2026-03-27 |
| pr |
0.00 |
Support the XLA GPU compilation flags in Orbax |
|
— |
2026-03-23 |
| pr |
0.00 |
[XLA:CPU] Take advantage of xla_allow_excess_precision when |
|
— |
2026-03-27 |
| pr |
0.00 |
[XLA:GPU] remove nested gemm fusion code |
|
— |
2026-03-23 |
| pr |
0.00 |
Reduction detection, make more aggressive for overlapping sl |
|
— |
2026-03-27 |
| pr |
0.00 |
Make CPU devices report as cpu:0 instead of TFRT_CPU_0. |
|
— |
2026-03-27 |
| pr |
0.00 |
Push shardy outliner up to the beginning of Stablehlo export |
|
— |
2026-03-27 |
| pr |
0.00 |
Push shardy outliner up past ExportOps pass. |
|
— |
2026-03-27 |
| pr |
0.00 |
Internal clean up. |
|
— |
2026-03-27 |
| pr |
0.00 |
[XLA:GPU]: Reapply flag flip for one shot all-reduce after f |
|
— |
2026-03-26 |
| pr |
0.00 |
[XLA:MSA] Allow conditional outputs in alternate memory. |
|
— |
2026-03-01 |
| pr |
0.00 |
[XLA:GPU]: Fix layout when doing a gathered load. |
|
— |
2026-03-26 |
| pr |
0.00 |
Add PRNG seed to PjRt ExecuteOptions and add runtime support |
|
— |
2026-03-26 |
| pr |
0.00 |
test: do not keep hlo sharding constraints. |
|
— |
2026-03-27 |
| pr |
0.00 |
Introduce DeviceInterconnectResource to bridge Global Topolo |
|
— |
2026-03-26 |
| pr |
0.00 |
Rewrite matching python_repo requirement entries with local |
|
— |
2026-03-25 |
| pr |
0.00 |
[PjRt-IFRT] Fill in parameter information in SerializedXlaEx |
|
— |
2026-03-27 |
| pr |
0.00 |
[XLA:GPU] Rename multimem parameter property |
|
— |
2026-03-25 |
| pr |
0.00 |
Rename TSL Monitoring's `MakeStatic` factory functions to `N |
|
— |
2026-03-26 |
| pr |
0.00 |
[XLA:GPU] Emit copy instruction when Mosaic operation is use |
|
— |
2026-03-24 |
| pr |
0.00 |
Rename `XNNPACK_CACHE_NO_MMAP_FOR_TEST` to `XNNPACK_CACHE_NO |
|
— |
2026-03-27 |
| pr |
0.00 |
Add GetOrCreateRuntimeError to PJRT Megascale Extension |
|
— |
2026-03-26 |
| pr |
0.00 |
Integrate LLVM at llvm/llvm-project@5a403150b468 |
|
— |
2026-03-27 |
| pr |
0.00 |
[PjRt] Implement `PjRt(Loaded)Executable::GetParameterMemory |
|
— |
2026-03-27 |
| pr |
0.00 |
Add `Label` to `//third_party/tensorflow/compiler/xla/tsl/ut |
|
— |
2026-03-27 |
| pr |
0.00 |
Allow dots with multiple batch dimensions through GemmFusion |
|
— |
2026-03-27 |
| pr |
0.00 |
[XLA:GPU] Fix ReorderFilterAndBiasHloTest.TestCudnnReorderFi |
|
— |
2026-03-27 |
| pr |
0.00 |
[mGPU] Pass CPU target machine options to the Mosaic GPU ins |
|
— |
2026-03-19 |
| pr |
0.00 |
Remove TensorBoard dependency from TensorFlow |
|
— |
2026-03-26 |
| pr |
0.00 |
Add PjRtArray overloads to support xla::ifrt::PjRtLayout. |
|
— |
2026-03-18 |
| pr |
0.00 |
Unifiy TargetConfig logic in GpuCompiler |
|
— |
2026-03-27 |
| pr |
0.00 |
[Convolution] Handle layout assignment of convolutions. |
|
— |
2026-03-27 |
| pr |
0.00 |
Fix for when adding reshard for a call with null result shar |
|
— |
2026-03-26 |
| pr |
0.00 |
PR #39373: [ROCm] Fix issue with unsupported types combinati |
|
— |
2026-03-27 |
| pr |
0.00 |
Integrate StableHLO at openxla/stablehlo@fef90093 |
|
— |
2026-03-27 |
| pr |
0.00 |
PR #39843: Bump jwalton/gh-find-current-pr from 1.3.3 to 1.3 |
|
— |
2026-03-26 |
| pr |
0.00 |
Always populate target config in GpuTopology created in PjRt |
|
— |
2026-03-26 |
| pr |
0.00 |
Pass the cpu host target config from gpu compiler through th |
|
— |
2026-03-25 |
| pr |
0.00 |
Automated Code Change |
|
— |
2026-03-27 |
| pr |
0.00 |
Reverts ba9ed044f3a801c2396f37fd4a212e8fbaa3ba02 |
|
— |
2026-03-27 |
| pr |
0.00 |
[XLA:GPU] Reroute dots to a detailed cost model in the index |
|
— |
2026-03-26 |
| pr |
0.00 |
Remove ptxas regression test. |
|
— |
2026-03-27 |
| pr |
0.00 |
The collapsed_slice_dims/inserted_window_dims may not be in |
|
— |
2026-03-27 |
| pr |
0.00 |
[XLA:GPU] Return detailed estimates from the dot cost model |
|
— |
2026-03-25 |
| pr |
0.00 |
PR #39393: [xla:gpu] Add VA remapping for command buffer thu |
|
— |
2026-03-26 |
| pr |
0.00 |
[XLA:GPU] Add command buffer scheduling mode `CONCURRENT_REG |
|
— |
2026-03-27 |
| pr |
0.00 |
Remove HoistFusedBitcasts from autotuner pipeline. |
|
— |
2026-03-27 |
| pr |
0.00 |
[XLA:GPU] Support triton gemm fusions in the indexing perf m |
|
— |
2026-03-25 |
| pr |
0.00 |
Reverts df55d43fbd71db4ea8cbd5d014f7baed1297f56f |
|
— |
2026-03-27 |
| pr |
0.00 |
Simplify reduce_window_test setup. |
|
— |
2026-03-25 |
| pr |
0.00 |
PR #39854: [ROCm] Use hermetic clang for rocm |
|
— |
2026-03-27 |
| pr |
0.00 |
Prefer transpose bitcasts over reshape bitcasts - when we ar |
|
— |
2026-03-26 |
| pr |
0.00 |
Add comparison of device description strings to GpuTopology |
|
— |
2026-03-27 |
| pr |
0.00 |
Allow compatible sample GPUs for cross compilation in Pathwa |
|
— |
2026-03-27 |
| pr |
0.00 |
PR #39601: [xla:gpu] Add support for dynamically sized packe |
|
— |
2026-03-27 |
| pr |
0.00 |
[XLA:GPU] Default dot precision to BF16_BF16_F32 for F32 arg |
|
— |
2026-03-26 |
| pr |
0.00 |
[XLA:GPU] Remove collective multimem registry since now we c |
|
— |
2026-03-27 |
| pr |
0.00 |
Add reshards for when func has no argument shardings but the |
|
— |
2026-03-27 |
| pr |
0.00 |
[XLA:GPU] Add missing include in multi_gpu_barrier_test |
|
— |
2026-03-26 |
| pr |
0.00 |
Fix proto reconstruction for RaggedAllToAllThunk |
|
— |
2026-03-26 |
| pr |
0.00 |
[XLA:GPU] remove nested fusions support in symbolic analysis |
|
— |
2026-03-27 |
| pr |
0.00 |
PR #39871: [ROCm] Fix bf16 upcast handling for libdevice cal |
|
— |
2026-03-27 |
| pr |
0.00 |
[XLA:BUILD]: Check windows build failure fix |
|
— |
2026-03-27 |
| pr |
0.00 |
Hoist HoistFusedBitcasts above GemmFusionSwapOperands - this |
|
— |
2026-03-12 |
