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_CUDNN_PAD_FOR_CONVOLUTIONS_H_ 17 #define TENSORFLOW_COMPILER_XLA_SERVICE_GPU_CUDNN_PAD_FOR_CONVOLUTIONS_H_ 18 19 #include "tensorflow/compiler/xla/literal_util.h" 20 #include "tensorflow/compiler/xla/service/gpu/ir_emission_utils.h" 21 #include "tensorflow/compiler/xla/service/hlo_casting_utils.h" 22 #include "tensorflow/compiler/xla/service/hlo_pass_interface.h" 23 #include "tensorflow/compiler/xla/util.h" 24 #include "tensorflow/compiler/xla/window_util.h" 25 26 namespace xla { 27 namespace gpu { 28 29 // Two zero-paddings for CuDNN thunking are done in this transform: padding for 30 // tensor cores and padding for integer convolutions. This transform also 31 // add slice instruction to remove unnecessary output features. 32 class CudnnPadForConvolutions : public HloModulePass { 33 public: CudnnPadForConvolutions(se::CudaComputeCapability compute_capability)34 explicit CudnnPadForConvolutions(se::CudaComputeCapability compute_capability) 35 : compute_capability_(compute_capability) {} 36 name()37 absl::string_view name() const override { 38 return "cudnn_pad_for_convolutions"; 39 } 40 // Run PadForConvolutions on the given module and return if any change is made 41 using HloPassInterface::Run; 42 StatusOr<bool> Run( 43 HloModule* module, 44 const absl::flat_hash_set<absl::string_view>& execution_threads) override; 45 46 private: 47 const se::CudaComputeCapability compute_capability_; 48 }; 49 50 } // namespace gpu 51 } // namespace xla 52 53 #endif // TENSORFLOW_COMPILER_XLA_SERVICE_GPU_CUDNN_PAD_FOR_CONVOLUTIONS_H_ 54