| COMMIT |
0.15 |
PR #33269: Addin a knob to control the limitation of async-c |
|
Slightly formal phrasing, some ChatGPT-l |
2026-04-01 |
| COMMIT |
0.10 |
Automated Code Change |
|
Generic 'Automated Code Change'; could i |
2026-04-01 |
| PR |
0.10 |
Fix integer overflow in stablehlo_reduce_window output dimen |
|
Structured technical write-up, but no AI |
2026-04-01 |
| COMMIT |
0.00 |
[ReplicaGroupV3][Refactor][18/n] migrate more XLA functions |
|
Terse, technical, commit-style message w |
2026-04-01 |
| COMMIT |
0.00 |
[XLA] Relax memory ordering of `entry->next` access in `~Obj |
|
Technical writeup, natural phrasing, ref |
2026-04-01 |
| COMMIT |
0.00 |
Remove `allow_new_incarnation_to_reconnect` config from coor |
|
Brief removal message; no AI signals, do |
2026-04-01 |
| COMMIT |
0.00 |
Make an explicit ref type for PjRtDeviceEvent to allow easil |
|
Specific technical intent, colloquial wo |
2026-04-01 |
| COMMIT |
0.00 |
[ReplicaGroupV3][Refactor][17/n] migrate functions in spmd/ |
|
Refactor message, technical context, ter |
2026-04-01 |
| COMMIT |
0.00 |
Add PID metadata to XPlanes and TraceMeConsumers to enable p |
|
Describes technical change in one clear |
2026-04-01 |
| COMMIT |
0.00 |
[XLA] Make `ObjectPool::Entry::next` atomic |
|
Short, domain-specific commit; no AI-lik |
2026-04-01 |
| COMMIT |
0.00 |
Benchmark for traceme_encode. |
|
Minimal benchmark addition, no AI langua |
2026-04-01 |
| COMMIT |
0.00 |
[tfrt] Introduce DeviceInterconnectResource to bridge Global |
|
Contains some structure, but detailed te |
2026-04-01 |
| COMMIT |
0.00 |
Update Shardy to the latest commit. |
|
Routine update, minimal content, highly |
2026-04-01 |
| COMMIT |
0.00 |
[XLA] Remove obsolete shmap & sdy manual computation inliner |
|
Technical commit message with revert and |
2026-04-01 |
| COMMIT |
0.00 |
Grant users of XLA access to gpu_topology_util |
|
Brief, jargon-heavy message; no AI phras |
2026-04-01 |
| COMMIT |
0.00 |
Add a test for `jit(lambda x: x, out_shardings=unreduced)` a |
|
Concise, uses jargon and code refs, read |
2026-04-01 |
| COMMIT |
0.00 |
Internal visibility change. |
|
Extremely terse and generic; typical hum |
2026-04-01 |
| COMMIT |
0.00 |
Drop fully deduplicating on outliner and leave it to unflatt |
|
Uses domain-specific concepts; informal |
2026-04-01 |
| COMMIT |
0.00 |
Add SymbolicExpr::ReplaceDims(absl::Span |
|
Technical addition; notes requirement, n |
2026-04-01 |
| COMMIT |
0.00 |
Fix dependency issues. |
|
Minimal and typical engineering commit m |
2026-04-01 |
| COMMIT |
0.00 |
Drop func and call sharding conflict resolution on outliner. |
|
Concise, abrupt phrasing; uses specific |
2026-04-01 |
| COMMIT |
0.00 |
[XLA:CPU] Lower Cbrt to Libm call. |
|
Detailed technical description, no AI st |
2026-04-01 |
| COMMIT |
0.00 |
[XLA:GPU] Simplify and retune the splitk heuristic. |
|
Matter-of-fact, technical, typical commi |
2026-04-01 |
| COMMIT |
0.00 |
Handle V3 named sharding in windowed input resharding. |
|
Technical, concise, domain-specific and |
2026-04-01 |
| COMMIT |
0.00 |
Add original func names on shardy outliner. |
|
Concise, technical, contains no AI style |
2026-04-01 |
| COMMIT |
0.00 |
Fix use-after-free in `SymbolicMap::GetResults()` |
|
Domain jargon, detailed technical explan |
2026-04-01 |
| COMMIT |
0.00 |
[XLA:GPU] Fix Triton loads dropping the mask and causing OOB |
|
Technical jargon and direct explanation, |
2026-04-01 |
| COMMIT |
0.00 |
Add if_onednn_async to tf_copts. |
|
Short, domain-specific, no AI style dete |
2026-04-01 |
| COMMIT |
0.00 |
Run shardy/google/integrate_latest.sh to integrate. |
|
Minimal description, human style, typica |
2026-04-01 |
| COMMIT |
0.00 |
Add reshards for when func has no argument shardings but the |
|
Concise, domain-specific, human phrasing |
2026-04-01 |
| COMMIT |
0.00 |
Allow creating a GPU PjRtClient without devices |
|
Technical, explains code change, referen |
2026-04-01 |
| COMMIT |
0.00 |
Handle one size axis in `IsReplicated` checks |
|
Very brief, domain-specific term, human |
2026-04-01 |
| COMMIT |
0.00 |
Enforce consistent simplification at creation time in Symbol |
|
Technical, with spelling errors and info |
2026-04-01 |
| COMMIT |
0.00 |
[XLA:CPU] Add benchmarks for Cbrt |
|
Short, jargon-filled changelog typical o |
2026-04-01 |
| COMMIT |
0.00 |
Wrap XNNPack delegate check macros in `do { ... } while(0)`. |
|
Technical phrasing and concise explanati |
2026-04-01 |
| COMMIT |
0.00 |
[XLA:GPU] Mark NVSHMEM thunks as deprecated. |
|
Direct and technical; no AI hallmarks pr |
2026-04-01 |
| COMMIT |
0.00 |
PR #39725: [xla] Add global per-process hang watchdog |
|
Very technical, includes human-style sum |
2026-04-01 |
| COMMIT |
0.00 |
Use utility to walk on call graphs. |
|
Terse commit message, strong human style |
2026-04-01 |
| COMMIT |
0.00 |
[XLA:GPU] Remove NCCL from dependencies if no_nccl definitio |
|
Technical feature flag; clear human-auth |
2026-04-01 |
| COMMIT |
0.00 |
Run shardy/google/integrate_latest.sh to integrate. |
|
Build automation statement, typical of h |
2026-04-01 |
| COMMIT |
0.00 |
[XLA:GPU] Sort the tiled instructions taking RT variables in |
|
Long, technical, with domain context and |
2026-04-01 |
| COMMIT |
0.00 |
[XLA:GPU] Actually use `ConcurrentRegionsHloOrdering` for co |
|
Direct and based in technical details, h |
2026-04-01 |
| COMMIT |
0.00 |
[XLA:GPU]: Unify collective codegen checks in one place. |
|
Concise and technical PR description, no |
2026-04-01 |
| COMMIT |
0.00 |
PR #39854: [ROCm] Use hermetic clang for rocm |
|
Uses template sections but free text is |
2026-04-01 |
| COMMIT |
0.00 |
Move utility for in/outliner into shardy. part 2. |
|
Technical, minimal human-style commit me |
2026-04-01 |
| COMMIT |
0.00 |
Automated Code Change |
|
Standard autogenerated commit, not AI-ge |
2026-04-01 |
| COMMIT |
0.00 |
PR #39601: [xla:gpu] Add support for dynamically sized packe |
|
Technical explanation with domain jargon |
2026-04-01 |
| COMMIT |
0.00 |
Automated Code Change |
|
Automated generation, not AI-written nat |
2026-04-01 |
| COMMIT |
0.00 |
Fix scan rewriter to handle broadcasted initial values and c |
|
Technical summary with explicit jargon, |
2026-04-01 |
| COMMIT |
0.00 |
Move gpu_device_info_test to target_config directory. |
|
Concise and technical, clear human tone |
2026-04-01 |
| COMMIT |
0.00 |
PR #40180: [XLA:GPU][oneAPI] Remove context activation and u |
|
Well-structured technical explanation, n |
2026-04-01 |
| COMMIT |
0.00 |
[XLA:GPU] Prevent collective memory from being destructed un |
|
Terse, technical commit; no AI signs det |
2026-04-01 |
| COMMIT |
0.00 |
Automated Code Change |
|
Explicitly marked as automated code chan |
2026-04-01 |
| COMMIT |
0.00 |
Always populate target config in GpuTopology created in PjRt |
|
Detailed and technical, typical human en |
2026-04-01 |
| COMMIT |
0.00 |
Automated Code Change |
|
Explicitly marked as automated code chan |
2026-04-01 |
| COMMIT |
0.00 |
Make GpuTargetConfig device-agnostic |
|
Technical, contains domain specifics and |
2026-04-01 |
| COMMIT |
0.00 |
Remove ptxas regression test. |
|
Direct, informal, includes human reasoni |
2026-04-01 |
| COMMIT |
0.00 |
Remove test timeout long from grouped_convolution_test. |
|
Terse, direct, no AI-indicative phrasing |
2026-04-01 |
| COMMIT |
0.00 |
Automated Code Change |
|
Explicitly marked as automated code chan |
2026-04-01 |
| COMMIT |
0.00 |
Automated Code Change |
|
Explicitly marked as automated code chan |
2026-04-01 |
| COMMIT |
0.00 |
Migrate xla::ifrt::GetDeviceMemoryInBytes to `device_memory_ |
|
Technical commit message; no AI linguist |
2026-04-01 |
| COMMIT |
0.00 |
Remove friend classing in coordination service. |
|
Concise commit message typical of humans |
2026-04-01 |
| COMMIT |
0.00 |
Test elements shape and dtype in `testLargeBatchSize`. |
|
Short, direct test description; no AI ha |
2026-03-31 |
| COMMIT |
0.00 |
Add support for reshape and reduce fusions. |
|
Standard concise commit phrasing, entire |
2026-03-31 |
| COMMIT |
0.00 |
Migrate quantize_test to PjRt runtime. |
|
Technical migration note, brief, human p |
2026-03-31 |
| COMMIT |
0.00 |
Add missing bf16 reduce benchmarks similarly to f32 |
|
Concise, domain-specific language; clear |
2026-03-31 |
| COMMIT |
0.00 |
Fix aligned_alloc usage for older Android APIs |
|
Technical, succinct explanation with cas |
2026-03-31 |
| COMMIT |
0.00 |
Fix use of deprecated `xnn_define_static_constant_pad` |
|
Concise technical commit message with no |
2026-03-31 |
| COMMIT |
0.00 |
Fix WinDefFileParse on Linux in pywrap_library by disabling |
|
Clear technical explanation, no boilerpl |
2026-03-31 |
| COMMIT |
0.00 |
Migrate qr_test to PjRt runtime. |
|
Direct, terse message typical of human c |
2026-03-31 |
| COMMIT |
0.00 |
Add a test for dataset batching using `dtypes.int64.max` bat |
|
Technical, precise commit description wi |
2026-03-31 |
| COMMIT |
0.00 |
Correct strided slice output shape calculation. |
|
Efficient technical language; no signs o |
2026-03-31 |
| COMMIT |
0.00 |
Support the XLA GPU compilation flags in Orbax |
|
Brief technical message, typical human a |
2026-03-31 |
| COMMIT |
0.00 |
Migrate prng_test to use PjRt. |
|
Terse migration note, standard in code m |
2026-03-31 |
| COMMIT |
0.00 |
Rename TSL Monitoring's `MakeStatic` factory functions to `N |
|
Specific technical rename, no AI-style l |
2026-03-31 |
| COMMIT |
0.00 |
Add GetOrCreateRuntimeError to PJRT Megascale Extension |
|
Short technical addition; natural commit |
2026-03-31 |
| COMMIT |
0.00 |
Fix an integer overflow in symbolic_tile_analysis. |
|
Concise issue-fix entry, typical human-w |
2026-03-31 |
| COMMIT |
0.00 |
Add PRNG seed to PjRt ExecuteOptions and add runtime support |
|
Terse commit message; domain-specific te |
2026-03-31 |
| COMMIT |
0.00 |
[XLA:GPU] Emit broadcast via the new tiling. |
|
Brief with XLA:GPU jargon; style typical |
2026-03-31 |
| COMMIT |
0.00 |
PR #39567: [ROCm] Fix fp8 on ROCm when xla_gpu_enable_cublas |
|
Includes PR template sections and engine |
2026-03-31 |
| COMMIT |
0.00 |
port away from TF_ASSIGN_OR_RETURN |
|
Very terse, imperative style; no AI hall |
2026-03-31 |
| COMMIT |
0.00 |
Use absl::HashOf for key_hash in LocalRendezvous. |
|
Contains detailed technical motivation a |
2026-03-31 |
| COMMIT |
0.00 |
Fix RunHandler::ScheduleIntraOpClosure to call the correct i |
|
Direct, informal commit message; no AI t |
2026-03-31 |
| COMMIT |
0.00 |
Allow non-kPadUpPolicy for priority-aware batching when max_ |
|
Technical language, concise, matches hum |
2026-03-31 |
| COMMIT |
0.00 |
Remove `kLeaveBarriersOnRecoverableAgentRestart` from coordi |
|
Short, practical description; no sign of |
2026-03-31 |
| COMMIT |
0.00 |
Add `stablehlo.future` to C API. |
|
Short, direct summary with specific cont |
2026-03-31 |
| COMMIT |
0.00 |
Add a specialized kernel for scans of many short rows. |
|
Detailed implementation and benchmark da |
2026-03-31 |
| COMMIT |
0.00 |
PR #39929: [ROCm] Fix parse empty rocm distro links |
|
PR uses a template with concise domain-s |
2026-03-31 |
| COMMIT |
0.00 |
Handle mixed V2/V3 shardings in PartialReplicateReshardCompa |
|
Direct, technical commit message; no AI |
2026-03-31 |
| COMMIT |
0.00 |
Remove legacy autotuner compile util. |
|
Brief, technical statement; very human s |
2026-03-31 |
| COMMIT |
0.00 |
Run shardy/google/integrate_latest.sh to integrate. |
|
Standard terse integration commit, no AI |
2026-03-31 |
| COMMIT |
0.00 |
Check that cached buffers offset and size do not point out o |
|
Technical, circumspect, lacks AI-style p |
2026-03-31 |
| COMMIT |
0.00 |
Rename test case in SpmdPartitionerTest. |
|
Minimal technical summary, clearly human |
2026-03-31 |
| COMMIT |
0.00 |
Simplify partial replication check in SPMD partitioner. |
|
Succinct technical phrasing, no sign of |
2026-03-31 |
| COMMIT |
0.00 |
PR #39537: [xla:gpu] Add hang watchdog to PJRT GPU client |
|
Free-text is technical and specific, lit |
2026-03-31 |
| COMMIT |
0.00 |
Automated Code Change |
|
Explicitly flagged as an automated (not |
2026-03-31 |
| COMMIT |
0.00 |
PR #39744: [ROCm] Porting CUB sort FFI handler consolidation |
|
Somewhat verbose but technical, not Chat |
2026-03-31 |
| COMMIT |
0.00 |
Remove gen_gpu_hlo_compile_tests bazel rule. |
|
Very terse commit, human-like, uses inte |
2026-03-31 |
| COMMIT |
0.00 |
[xla:cpu] Roll-back small_while_loop_hoisting_pass threshold |
|
Terse summary, includes internal revert |
2026-03-31 |
| COMMIT |
0.00 |
Fully serialize stream_executor::DeviceDescription to/from p |
|
Detailed, technical explanation, domain- |
2026-03-31 |
| COMMIT |
0.00 |
[XLA:GPU] Simplify xla_aot_compile_gpu_test by using HloPjRt |
|
Technical change note with informal tone |
2026-03-31 |
| PR |
0.00 |
Populate xla_input_shapes for unpacked inputs in IfrtServing |
|
— |
2026-03-26 |
| PR |
0.00 |
[IFRT] Add more general `Layout::ByteSize` static methods |
|
— |
2026-04-01 |
| PR |
0.00 |
[IFRT] Add `Value::ByteSize` method |
|
— |
2026-04-01 |
| PR |
0.00 |
[XLA] Put BufferAssignment dumps on a diet |
|
— |
2026-03-28 |
| PR |
0.00 |
[ReplicaGroupV3][Refactor][18/n] migrate more XLA functions |
|
— |
2026-04-01 |
| PR |
0.00 |
PR #35733: [XLA:GPU][oneAPI] SYCL executor and tests |
|
— |
2026-01-07 |
| PR |
0.00 |
Integrate StableHLO at openxla/stablehlo@05fdca09 |
|
— |
2026-03-30 |
| PR |
0.00 |
[GPU] Use CheckFabricHomogeneity in RaggedAllToAllThunk |
|
— |
2026-03-31 |
| PR |
0.00 |
[XLA] Relax memory ordering of `entry->next` access in `~Obj |
|
— |
2026-04-01 |
| PR |
0.00 |
Reverts f6b594dc036bed5a4d59ee60e69698e8219501b2 |
|
— |
2026-03-31 |
| PR |
0.00 |
Fix heap-use-after-free in ConstraintPropagator::PropagateCo |
|
— |
2026-04-01 |
| PR |
0.00 |
Remove `allow_new_incarnation_to_reconnect` config from coor |
|
— |
2026-04-01 |
| PR |
0.00 |
Make an explicit ref type for PjRtDeviceEvent to allow easil |
|
— |
2026-03-27 |
| PR |
0.00 |
Mark coordination service not copyable or movable. |
|
— |
2026-04-01 |
| PR |
0.00 |
[ReplicaGroupV3][Refactor][17/n] migrate functions in spmd/ |
|
— |
2026-04-01 |
| PR |
0.00 |
Add PID metadata to XPlanes and TraceMeConsumers to enable p |
|
— |
2025-12-09 |
| PR |
0.00 |
Consolidate sanitizer build checks for tfrt_session_python_t |
|
— |
2026-04-01 |
| PR |
0.00 |
Erase async collectives in XLA CPU. |
|
— |
2026-04-01 |
| PR |
0.00 |
Add option to allow resplitting for priority aware scheduler |
|
— |
2026-03-13 |
| PR |
0.00 |
[XLA] Make `ObjectPool::Entry::next` atomic |
|
— |
2026-04-01 |
| PR |
0.00 |
Update rules_ml_toolchain to the version with Hermetic Pytho |
|
— |
2026-04-01 |
| PR |
0.00 |
Benchmarks for traceme and connected_traceme. |
|
— |
2026-04-01 |
| PR |
0.00 |
Automated Code Change |
|
— |
2026-01-12 |
| PR |
0.00 |
Support For Roofline Analysis Of Pallas Kernels |
|
— |
2025-10-02 |
| PR |
0.00 |
Benchmark for traceme_encode. |
|
— |
2026-03-30 |
| PR |
0.00 |
Remove unused `TransactionToken` and associated methods from |
|
— |
2026-04-01 |
| PR |
0.00 |
[tfrt] Introduce DeviceInterconnectResource to bridge Global |
|
— |
2026-03-31 |
| PR |
0.00 |
Use salted hashing for key_hash so that we have fewer collis |
|
— |
2026-04-01 |
| PR |
0.00 |
Add CUDA 13 presubmit job and update Python dependencies. |
|
— |
2026-03-30 |
| PR |
0.00 |
Add a cross-compilation test for Mosaic GPU using `xla_compi |
|
— |
2026-04-01 |
| PR |
0.00 |
Have autotuner control the register spilling strategy rather |
|
— |
2026-01-12 |
| PR |
0.00 |
Update Shardy to the latest commit. |
|
— |
2026-04-01 |
| PR |
0.00 |
[mpmd] Use reserved_hbm_bytes attribute to reserve per-fragm |
|
— |
2026-04-01 |
| PR |
0.00 |
[mGPU] TargetMachineOptions are always passed from XLA side |
|
— |
2026-03-19 |
| PR |
0.00 |
[IFRT] Add GetName() method to HloProgram |
|
— |
2026-04-01 |
| PR |
0.00 |
Integrate LLVM at llvm/llvm-project@bdf28a6d4811 |
|
— |
2026-04-01 |
| PR |
0.00 |
[ReplicaGroupV3][JAX+stablehlo][2/n] stablehlo and vhlo supp |
|
— |
2026-03-30 |
| PR |
0.00 |
Integrate LLVM at llvm/llvm-project@293623ce99d4 |
|
— |
2026-04-01 |
| PR |
0.00 |
Integrate LLVM at llvm/llvm-project@db80420930c3 |
|
— |
2026-03-31 |
| PR |
0.00 |
[XLA] Remove obsolete shmap & sdy manual computation inliner |
|
— |
2026-03-30 |
| PR |
0.00 |
[XLA:CPU] Add Opt Preset FAST_COMPILE for interactive and te |
|
— |
2026-03-31 |
| PR |
0.00 |
Grant users of XLA access to gpu_topology_util |
|
— |
2026-03-27 |
| PR |
0.00 |
Remove condition hints that are based on future's readiness |
|
— |
2026-04-01 |
| PR |
0.00 |
only change internal visibility |
|
— |
2026-04-01 |
| PR |
0.00 |
Roll-forward Migrate XLA GPU emitters and HWIR to use Symbol |
|
— |
2026-03-31 |
| PR |
0.00 |
Add a test for `jit(lambda x: x, out_shardings=unreduced)` a |
|
— |
2026-04-01 |
| PR |
0.00 |
Always pass TargetMachineOptions to FFI calls from CustomCal |
|
— |
2026-03-31 |
| PR |
0.00 |
Integrate LLVM at llvm/llvm-project@7ccd92e5e6e5 |
|
— |
2026-03-31 |
| PR |
0.00 |
Internal visibility change. |
|
— |
2026-03-31 |
| PR |
0.00 |
Avoid crashing when the sub mesh does not have the same devi |
|
— |
2026-04-01 |
| PR |
0.00 |
Fix missing dependencies and remove strict_deps workarounds |
|
— |
2026-03-25 |
| PR |
0.00 |
Always enable `strict_shape_checking` for HloRunnerPjRt. |
|
— |
2026-04-01 |
| PR |
0.00 |
Improve parentheses in SymbolicExpr serialization |
|
— |
2026-04-01 |
| PR |
0.00 |
PR #40117: Update XLA oneAPI presubmit build targets |
|
— |
2026-04-01 |
| PR |
0.00 |
[NFC] Remove obsolete `endian.patch`. |
|
— |
2026-01-12 |
| PR |
0.00 |
PR #31284: [ROCm] Add PJRT_Triton_Extension support |
|
— |
2026-01-12 |
| PR |
0.00 |
Drop fully deduplicating on outliner and leave it to unflatt |
|
— |
2026-03-28 |
| PR |
0.00 |
Original change is causing presubmit tests failure. Rolling |
|
— |
2026-04-01 |
| PR |
0.00 |
PR #38198: [oneDNN] Upgrade to oneDNN v3.11 |
|
— |
2026-04-01 |
| PR |
0.00 |
Bump Compute Library version from v24.12 to v52.8.0. |
|
Brief update summary; reviewer approval |
2026-03-26 |
| PR |
0.00 |
Support mixed serialization in ifrt under new ser-des versio |
|
— |
2026-04-01 |
| PR |
0.00 |
Remove obsolete `btree.patch`. |
|
— |
2026-01-12 |
| PR |
0.00 |
Add SymbolicExpr::ReplaceDims(absl::Span |
|
— |
2026-04-01 |
| PR |
0.00 |
Move shardy outliner into mlir sdy. Part 2. |
|
— |
2026-04-01 |
| PR |
0.00 |
Fix dependency issues. |
|
— |
2026-03-31 |
| PR |
0.00 |
Automated Code Change |
|
— |
2026-01-12 |
| PR |
0.00 |
Automated Code Change |
|
— |
2026-01-12 |
| PR |
0.00 |
[XLA:CPU] Lower Cbrt to Libm call. |
|
— |
2026-04-01 |
| PR |
0.00 |
Move shardy outliner into mlir sdy. Part 1. |
|
— |
2026-04-01 |
| PR |
0.00 |
Push shardy outliner up past ManualReductionCollectives pass |
|
— |
2026-03-27 |
| PR |
0.00 |
Move shardy inliner into mlir sdy. Part 2. |
|
— |
2026-04-01 |
| PR |
0.00 |
Migrate Tiling Expressions from mlir::AffineExpr to xla::Sym |
|
— |
2026-03-24 |
| PR |
0.00 |
Automated Code Change |
|
— |
2026-01-12 |
| PR |
0.00 |
[XLA:GPU] Simplify and retune the splitk heuristic. |
|
— |
2026-04-01 |
| PR |
0.00 |
Handle V3 named sharding in windowed input resharding. |
|
— |
2026-03-31 |
| PR |
0.00 |
Automated Code Change |
|
— |
2026-04-01 |
| PR |
0.00 |
Fix use-after-free in `SymbolicMap::GetResults()` |
|
— |
2026-03-31 |
| PR |
0.00 |
Fix AArch64 CPUIDInfo init |
|
Human reviewer interaction, direct style |
2025-10-14 |
| PR |
0.00 |
[XLA:GPU] Added concat to the allowlist of block-level emitt |
|
— |
2026-03-30 |
| PR |
0.00 |
[XLA:GPU] Fix Triton loads dropping the mask and causing OOB |
|
— |
2026-04-01 |
| PR |
0.00 |
Automated Code Change |
|
— |
2026-04-01 |
| PR |
0.00 |
Add if_onednn_async to tf_copts. |
|
— |
2026-04-01 |
| PR |
0.00 |
PR #36704: [XLA:GPU][oneAPI] Blas plugin for SYCL platform |
|
— |
2026-04-01 |
| PR |
0.00 |
Run shardy/google/integrate_latest.sh to integrate. |
|
— |
2026-04-01 |
| PR |
0.00 |
Automated Code Change |
|
— |
2026-01-12 |
| PR |
0.00 |
Automated Code Change |
|
— |
2026-01-12 |
| PR |
0.00 |
Add reshards for when func has no argument shardings but the |
|
— |
2026-03-27 |
| PR |
0.00 |
Automated Code Change |
|
— |
2026-01-12 |
| PR |
0.00 |
Allow creating a GPU PjRtClient without devices |
|
— |
2026-04-01 |
| PR |
0.00 |
Migrate indexing analysis from MLIR Affine to SymbolicMap |
|
— |
2026-03-12 |
| PR |
0.00 |
Automated Code Change |
|
— |
2026-04-01 |
| PR |
0.00 |
Migrate XLA codegen attributes and operations to the new Sym |
|
— |
2026-02-11 |
| PR |
0.00 |
Automated Code Change |
|
— |
2026-01-12 |
| PR |
0.00 |
Handle one size axis in `IsReplicated` checks |
|
— |
2026-04-01 |
| PR |
0.00 |
[XLA:GPU] Enable heuristic-based splitk pass, and disable au |
|
— |
2026-04-01 |
| PR |
0.00 |
Automated Code Change |
|
— |
2026-04-01 |
| PR |
0.00 |
Enforce consistent simplification at creation time in Symbol |
|
— |
2026-03-26 |
| PR |
0.00 |
[XLA:CPU] Add benchmarks for Cbrt |
|
— |
2026-04-01 |
| PR |
0.00 |
PR #40232: Add missing kAllGatherStart/kAllGatherDone verifi |
|
— |
2026-04-01 |
