Commits

Branch selector

 master

User selector

thomasjoerg

All time

Commit History

Commits on Jun 14, 2024

• [XLA:GPU] Disable CuDnnFusionLevel2Test.ClampExecutesCorrectly which is failing with CUDNN_BACKEND_OPERATION: cudnnFinalize Failed.

  

  thomasjoerg

  authored and

  tensorflower-gardener

  committedJun 14, 2024
  2d4295e

Commits on Jun 4, 2024

• [XLA:GPU] Pass CUDA and cuDNN versions (or ROCm counterparts) explicitly into CudnnFusedConvRewriter.

  

  thomasjoerg

  authored and

  tensorflower-gardener

  committedJun 4, 2024
  967175b

Commits on May 29, 2024

• [XLA:GPU] Pre-factoring: Add GetDnnVersionInfo helper that returns StatusOr.

  

  thomasjoerg

  authored and

  tensorflower-gardener

  committedMay 29, 2024
  c1ace89

Commits on May 24, 2024

• Remove GOOGLE_CUDA and TENSORFLOW_USE_ROCM defines from topk_specializer.cc

  

  thomasjoerg

  authored and

  tensorflower-gardener

  committedMay 24, 2024
  d526a0c

• [XLA:GPU] Let TritonFusion depend on CUDA or ROCm headers (transitively).

  

  thomasjoerg

  authored and

  tensorflower-gardener

  committedMay 24, 2024
  4549f45

Commits on May 17, 2024

• Exclude //xla/service/gpu/... targets for CPU CI runs.

  

  thomasjoerg

  authored and

  tensorflower-gardener

  committedMay 17, 2024
  e22b292

Commits on May 16, 2024

• [XLA:GPU] Remove GOOGLE_CUDA and TENSORFLOW_USE_ROCM defines from gemm_algorithm_picker.

  

  thomasjoerg

  authored and

  tensorflower-gardener

  committedMay 16, 2024
  15da74a

• [XLA:GPU] Remove GOOGLE_CUDA and TENSORFLOW_USE_ROCM guards from hlo_fusion_analysis.

  

  thomasjoerg

  authored and

  tensorflower-gardener

  committedMay 16, 2024
  62fe318

• [XLA:GPU] Remove preprocessor #if GOOGLE_CUDAs and similar from GemmRewriter pass.

  

  thomasjoerg

  authored and

  tensorflower-gardener

  committedMay 16, 2024
  31c77e9

Commits on May 15, 2024

• [XLA:GPU] Pass the CUDA / ROCm toolkit version explicitly for autotuning and GEMM rewriting.

  

  thomasjoerg

  authored and

  tensorflower-gardener

  committedMay 15, 2024
  85be21a

Commits on May 14, 2024

• [XLA:GPU] Introduce GpuCompiler::GetToolkit method for subclasses to provide the CUDA or ROCm versions.

  

  thomasjoerg

  authored and

  tensorflower-gardener

  committedMay 14, 2024
  e0e64ad

Commits on Apr 26, 2024

• [XLA:GPU] Remove unnecessary preprocessor directive.

  

  thomasjoerg

  authored and

  tensorflower-gardener

  committedApr 26, 2024
  aa8f32e

Commits on Apr 4, 2024

• Bump the operands+outputs threshold to allow larger fusions.

  

  thomasjoerg

  authored and

  tensorflower-gardener

  committedApr 4, 2024
  07a0373

Commits on Feb 2, 2024

• [XLA:GPU] Set metadata on reduce ops created by split-k rewriting.

  

  thomasjoerg

  authored and

  tensorflower-gardener

  committedFeb 2, 2024
  01745ce

Commits on Jan 4, 2024

• [XLA:GPU] Split reductions with non-consecutive dimensions more aggressively.

  

  thomasjoerg

  authored and

  tensorflower-gardener

  committedJan 4, 2024
  c293a78

• [XLA:GPU] Fix ReductionSplitter test.

  

  thomasjoerg

  authored and

  tensorflower-gardener

  committedJan 4, 2024
  5a35def

Commits on Oct 16, 2023

• [XLA:GPU] Log more details when the Triton autotuner check fails during compilation.

  

  thomasjoerg

  authored and

  tensorflower-gardener

  committedOct 16, 2023
  7793f57

Commits on Feb 21, 2023

• [XLA:GPU] Fix crash in reduction emitter. Use int64_t to avoid an overflows.

  

  thomasjoerg

  authored and

  tensorflower-gardener

  committedFeb 21, 2023
  3bea547

Commits on Feb 6, 2023

• [XLA:GPU] Do not expand Scatter ops that are deterministic when running with --xla_gpu_deterministic_ops.

  

  thomasjoerg

  authored and

  tensorflower-gardener

  committedFeb 6, 2023
  81a0a85

Commits on Jul 28, 2022

• [XLA:GPU] Do not fuse transposing ops unless they can be emitted as shared memory transpose.

  

  thomasjoerg

  authored and

  tensorflower-gardener

  committedJul 28, 2022
  3caa0d6

Commits on Jul 26, 2022

• [XLA:GPU] Helper to determine if a fusion would be supported by the shmem transpose emitter.

  

  thomasjoerg

  authored and

  tensorflower-gardener

  committedJul 26, 2022
  7f269b3

Commits on Jul 13, 2022

• [XLA:GPU] NFC: Prefactor CheckAndEmitHloWithTile021 moving helpers to shape_util.h and ir_emission_utils.h.

  

  thomasjoerg

  authored and

  tensorflower-gardener

  committedJul 13, 2022
  04cc493

• [XLA:GPU] NFC: Rename LayoutsAreReduceInputFusionFriendly to IsPhysicallyTransposing, remove a dead parameter, update documentation and tests.

  

  thomasjoerg

  authored and

  tensorflower-gardener

  committedJul 13, 2022
  be4130e

Commits on May 31, 2022

• [XLA:GPU] Summarize cwise HLO ops when producing fusion stats.

  

  thomasjoerg

  authored and

  tensorflower-gardener

  committedMay 31, 2022
  043be71

Commits on May 30, 2022

• XLA:GPU Add HLO pass to log stats about HLO fusion ops.

  

  thomasjoerg

  authored and

  tensorflower-gardener

  committedMay 30, 2022
  7a3455b

Commits on Dec 15, 2020

• [XLA:GPU] Add --xla_gpu_deterministic_ops to force run-to-run determinism.

  

  thomasjoerg

  authored and

  tensorflower-gardener

  committedDec 15, 2020
  bb4a8a8

Commits on Dec 10, 2020

• [XLA:GPU] Document the GpuMultiOutputFusion in its header file.

  

  thomasjoerg

  authored and

  tensorflower-gardener

  committedDec 10, 2020
  9ca469e

Commits on Dec 9, 2020

• [XLA:GPU] Update and improve documentation of the FusionMerger HLO pass.

  

  thomasjoerg

  authored and

  tensorflower-gardener

  committedDec 9, 2020
  5163667

Commits on Nov 17, 2020

• [XLA:GPU] Re-enable horizontal input fusion. Use a minimum of 64 threads per block for multi-output fusions.

  

  thomasjoerg

  authored and

  tensorflower-gardener

  committedNov 17, 2020
  67adc58

Commits on Oct 29, 2020

• Integrate LLVM at llvm/llvm-project@6648414b2b7d

  

  thomasjoerg

  authored and

  tensorflower-gardener

  committedOct 29, 2020
  c23b1ad

• Integrate LLVM at llvm/llvm-project@4d11daa659a1

  

  thomasjoerg

  authored and

  tensorflower-gardener

  committedOct 29, 2020
  6889aba

Commits on Oct 27, 2020

• Integrate LLVM at llvm/llvm-project@0fc1aa22ee6a

  

  thomasjoerg

  authored and

  tensorflower-gardener

  committedOct 27, 2020
  30994be

Commits on Oct 22, 2020

• [XLA:GPU] Disable the GpuHorizontalInputFusion pass.

  

  thomasjoerg

  authored and

  tensorflower-gardener

  committedOct 22, 2020
  20cfbea

Commits on Oct 14, 2020

• [XLA:GPU] Do not create multi-output fusions that would cause exponential compile behavior (time and size).

  

  thomasjoerg

  authored and

  tensorflower-gardener

  committedOct 14, 2020
  379bcfa

Commits on Sep 25, 2020

• Revert: Implement horizontal input fusion.

  

  thomasjoerg

  authored and

  tensorflower-gardener

  committedSep 25, 2020
  e093e96

Pagination

PreviousNext