1 /* Copyright 2017 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 #ifdef GOOGLE_CUDA 17 #define EIGEN_USE_GPU 18 #include "tensorflow/core/common_runtime/gpu/gpu_managed_allocator.h" 19 #endif 20 21 #include "tensorflow/core/kernels/ops_testutil.h" 22 23 namespace tensorflow { 24 25 void OpsTestBase::SetDevice(const DeviceType& device_type, 26 std::unique_ptr<Device> device) { 27 CHECK(device_.get()) << "No device provided"; 28 device_type_ = device_type; 29 device_ = std::move(device); 30 #ifdef GOOGLE_CUDA 31 if (device_type == DEVICE_GPU) { 32 managed_allocator_.reset(new GpuManagedAllocator()); 33 allocator_ = managed_allocator_.get(); 34 } else { 35 managed_allocator_.reset(); 36 allocator_ = device_->GetAllocator(AllocatorAttributes()); 37 } 38 #else 39 CHECK_NE(device_type, DEVICE_GPU) 40 << "Requesting GPU on binary compiled without GOOGLE_CUDA."; 41 #endif 42 } 43 44 Tensor* OpsTestBase::GetOutput(int output_index) { 45 CHECK_LT(output_index, context_->num_outputs()); 46 Tensor* output = context_->mutable_output(output_index); 47 #ifdef GOOGLE_CUDA 48 if (device_type_ == DEVICE_GPU) { 49 managed_outputs_.resize(context_->num_outputs()); 50 // Copy the output tensor to managed memory if we haven't done so. 51 if (!managed_outputs_[output_index]) { 52 Tensor* managed_output = 53 new Tensor(allocator(), output->dtype(), output->shape()); 54 auto src = output->tensor_data(); 55 auto dst = managed_output->tensor_data(); 56 context_->eigen_gpu_device().memcpy(const_cast<char*>(dst.data()), 57 src.data(), src.size()); 58 context_->eigen_gpu_device().synchronize(); 59 managed_outputs_[output_index] = managed_output; 60 } 61 output = managed_outputs_[output_index]; 62 } 63 #endif 64 return output; 65 } 66 67 } // namespace tensorflow 68