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 #include "cuda/include/cuda.h" 18 #include "tensorflow/stream_executor/cuda/cuda_activation.h" 19 #endif // GOOGLE_CUDA 20 21 #include "tensorflow/core/common_runtime/gpu/gpu_cudamalloc_allocator.h" 22 23 #include "tensorflow/core/common_runtime/gpu/gpu_id.h" 24 #include "tensorflow/core/common_runtime/gpu/gpu_id_utils.h" 25 #include "tensorflow/core/common_runtime/gpu/gpu_init.h" 26 #include "tensorflow/core/platform/stream_executor.h" 27 28 namespace tensorflow { 29 30 GPUcudaMallocAllocator::GPUcudaMallocAllocator(VisitableAllocator* allocator, 31 CudaGpuId cuda_gpu_id) 32 : base_allocator_(allocator) { 33 stream_exec_ = GpuIdUtil::ExecutorForCudaGpuId(cuda_gpu_id).ValueOrDie(); 34 } 35 36 GPUcudaMallocAllocator::~GPUcudaMallocAllocator() { delete base_allocator_; } 37 38 void* GPUcudaMallocAllocator::AllocateRaw(size_t alignment, size_t num_bytes) { 39 #ifdef GOOGLE_CUDA 40 // allocate with cudaMalloc 41 gpu::cuda::ScopedActivateExecutorContext scoped_activation{stream_exec_}; 42 CUdeviceptr rv = 0; 43 CUresult res = cuMemAlloc(&rv, num_bytes); 44 if (res != CUDA_SUCCESS) { 45 LOG(ERROR) << "cuMemAlloc failed to allocate " << num_bytes; 46 return nullptr; 47 } 48 return reinterpret_cast<void*>(rv); 49 #else 50 return nullptr; 51 #endif // GOOGLE_CUDA 52 } 53 void GPUcudaMallocAllocator::DeallocateRaw(void* ptr) { 54 #ifdef GOOGLE_CUDA 55 // free with cudaFree 56 CUresult res = cuMemFree(reinterpret_cast<CUdeviceptr>(ptr)); 57 if (res != CUDA_SUCCESS) { 58 LOG(ERROR) << "cuMemFree failed to free " << ptr; 59 } 60 #endif // GOOGLE_CUDA 61 } 62 63 void GPUcudaMallocAllocator::AddAllocVisitor(Visitor visitor) { 64 return base_allocator_->AddAllocVisitor(visitor); 65 } 66 67 void GPUcudaMallocAllocator::AddFreeVisitor(Visitor visitor) { 68 return base_allocator_->AddFreeVisitor(visitor); 69 } 70 71 bool GPUcudaMallocAllocator::TracksAllocationSizes() { return false; } 72 73 } // namespace tensorflow 74