1 /* Copyright 2015 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_CORE_COMMON_RUNTIME_GPU_GPU_DEBUG_ALLOCATOR_H_ 17 #define TENSORFLOW_CORE_COMMON_RUNTIME_GPU_GPU_DEBUG_ALLOCATOR_H_ 18 19 #include <memory> 20 #include <string> 21 #include <unordered_map> 22 23 #include "tensorflow/core/common_runtime/gpu/gpu_id.h" 24 #include "tensorflow/core/framework/allocator.h" 25 #include "tensorflow/core/platform/macros.h" 26 #include "tensorflow/core/platform/stream_executor.h" 27 #include "tensorflow/core/platform/types.h" 28 29 namespace tensorflow { 30 31 // An allocator that wraps a GPU allocator and adds debugging 32 // functionality that verifies that users do not write outside their 33 // allocated memory. 34 class GPUDebugAllocator : public Allocator { 35 public: 36 explicit GPUDebugAllocator(Allocator* allocator, 37 PlatformGpuId platform_gpu_id); 38 ~GPUDebugAllocator() override; Name()39 string Name() override { return "gpu_debug"; } 40 void* AllocateRaw(size_t alignment, size_t num_bytes) override; 41 void DeallocateRaw(void* ptr) override; 42 bool TracksAllocationSizes() const override; 43 size_t RequestedSize(const void* ptr) const override; 44 size_t AllocatedSize(const void* ptr) const override; 45 int64 AllocationId(const void* ptr) const override; 46 absl::optional<AllocatorStats> GetStats() override; 47 void ClearStats() override; 48 49 // For testing. 50 bool CheckHeader(void* ptr); 51 bool CheckFooter(void* ptr); 52 53 private: 54 Allocator* base_allocator_ = nullptr; // owned 55 56 se::StreamExecutor* stream_exec_; // Not owned. 57 58 TF_DISALLOW_COPY_AND_ASSIGN(GPUDebugAllocator); 59 }; 60 61 // An allocator that wraps a GPU allocator and resets the memory on 62 // allocation and free to 'NaN', helping to identify cases where the 63 // user forgets to initialize the memory. 64 class GPUNanResetAllocator : public Allocator { 65 public: 66 explicit GPUNanResetAllocator(Allocator* allocator, 67 PlatformGpuId platform_gpu_id); 68 ~GPUNanResetAllocator() override; Name()69 string Name() override { return "gpu_nan_reset"; } 70 void* AllocateRaw(size_t alignment, size_t num_bytes) override; 71 void DeallocateRaw(void* ptr) override; 72 size_t RequestedSize(const void* ptr) const override; 73 size_t AllocatedSize(const void* ptr) const override; 74 absl::optional<AllocatorStats> GetStats() override; 75 void ClearStats() override; 76 77 private: 78 Allocator* base_allocator_ = nullptr; // owned 79 80 se::StreamExecutor* stream_exec_; // Not owned. 81 82 TF_DISALLOW_COPY_AND_ASSIGN(GPUNanResetAllocator); 83 }; 84 85 } // namespace tensorflow 86 87 #endif // TENSORFLOW_CORE_COMMON_RUNTIME_GPU_GPU_DEBUG_ALLOCATOR_H_ 88