1 /* Copyright 2018 The TensorFlow Authors. All Rights Reserved. 2 3 Licensed under the Apache License, Version 2.0 (the "License"); 4 you may not use this file except in compliance with the License. 5 You may obtain a copy of the License at 6 7 http://www.apache.org/licenses/LICENSE-2.0 8 9 Unless required by applicable law or agreed to in writing, software 10 distributed under the License is distributed on an "AS IS" BASIS, 11 WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 12 See the License for the specific language governing permissions and 13 limitations under the License. 14 ==============================================================================*/ 15 16 #ifndef TENSORFLOW_COMPILER_XLA_SERVICE_GPU_CUBLAS_PAD_FOR_GEMMS_H_ 17 #define TENSORFLOW_COMPILER_XLA_SERVICE_GPU_CUBLAS_PAD_FOR_GEMMS_H_ 18 19 #include "tensorflow/compiler/xla/service/hlo_pass_interface.h" 20 21 namespace xla { 22 namespace gpu { 23 24 // Adds padding to dot operations to make them run faster on GPUs. 25 // 26 // 27 // This can be used to pad f16 dots on tensor cores, or s8 dots to multiples of 28 // four. 29 // 30 // This pass depends on xla::DotDecomposer pass, 31 // so it should go strictly later. 32 class CublasPadForGemms : public HloModulePass { 33 public: CublasPadForGemms(PrimitiveType datatype,int32_t pad_to_multiple_of)34 CublasPadForGemms(PrimitiveType datatype, int32_t pad_to_multiple_of) 35 : datatype_(datatype), pad_to_multiple_of_(pad_to_multiple_of) {} 36 name()37 absl::string_view name() const override { return "cublas-pad-for-gemms"; } 38 39 using HloPassInterface::Run; 40 StatusOr<bool> Run( 41 HloModule* module, 42 const absl::flat_hash_set<absl::string_view>& execution_threads) override; 43 44 private: 45 PrimitiveType datatype_; 46 int32_t pad_to_multiple_of_; 47 }; 48 49 } // namespace gpu 50 } // namespace xla 51 52 #endif // TENSORFLOW_COMPILER_XLA_SERVICE_GPU_CUBLAS_PAD_FOR_GEMMS_H_ 53