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 // The temporary-memory-manager is a helper class for a Stream to keep track of 17 // temporary allocations. These allocations defer their deallocation to the next 18 // Stream::BlockHostUntilDone call for efficiency purposes (as deallocation 19 // itself generally forces synchronization to occur). 20 21 #ifndef TENSORFLOW_STREAM_EXECUTOR_TEMPORARY_MEMORY_MANAGER_H_ 22 #define TENSORFLOW_STREAM_EXECUTOR_TEMPORARY_MEMORY_MANAGER_H_ 23 24 #include <map> 25 #include <memory> 26 27 #include "tensorflow/stream_executor/device_memory.h" 28 #include "tensorflow/stream_executor/lib/status.h" 29 #include "tensorflow/stream_executor/lib/statusor.h" 30 #include "tensorflow/stream_executor/platform/mutex.h" 31 #include "tensorflow/stream_executor/platform/thread_annotations.h" 32 #include "tensorflow/stream_executor/temporary_device_memory.h" 33 34 namespace stream_executor { 35 namespace internal { 36 37 // Record used inside the TemporaryMemoryManager as metadata for a given device 38 // memory region. 39 struct TemporaryMemoryRecord { 40 // What "generation" this record was allocated in. 41 // 42 // Currently the generation counter is bumped for every allocation, but this 43 // could be made coarser if necessary. 44 uint64 allocation_generation; 45 46 // Notes whether the temporary memory has been marked as finalized, such that 47 // we can release the DeviceMemory associated with this record at 48 // synchronization time. 49 bool finalized; 50 }; 51 52 // Manages temporary memories associated with a stream -- keeps records of 53 // outstanding temporaries and their state, and can deallocate them 54 // appropriately at points in the Stream lifecycle (e.g. BlockHostUntilDone, 55 // destruction). 56 class TemporaryMemoryManager { 57 public: 58 explicit TemporaryMemoryManager(Stream* stream) : stream_(stream) {} 59 60 // Allocates a temporary array that is then managed by this object. 61 template <typename T> 62 port::StatusOr<std::unique_ptr<TemporaryDeviceMemory<T>>> AllocateArray( 63 uint64 element_count); 64 65 // Forces deallocation of all managed temporary memory regions. 66 // 67 // Called, for example, when the Stream owning this temporary memory manager 68 // is destroyed. 69 // 70 // Note: These calls to Deallocate will likely force synchronization. 71 void ForceDeallocateAll(); 72 73 // Marks the given memory region as finalized. 74 // 75 // If must_exist is set, this will check-fail if the temporary memory record 76 // is not found. 77 void MarkFinalized(const DeviceMemoryBase& device_memory, uint64 generation, 78 bool must_exist); 79 80 // Deallocates temporary memories that have been finalized. 81 // 82 // Note: These calls to Deallocate will likely force synchronization, so it is 83 // meant to be called before a "BlockHostUntilDone" is about to be performed. 84 void DeallocateFinalizedTemporaries(); 85 86 // Returns whether the provided device_memory is finalized. 87 // 88 // In the vacuous case where the device memory doesn't appear in the temporary 89 // memory records, it is either not a temporary at all, or has already been 90 // deallocated, and thus returns true. 91 bool IsFinalized(const DeviceMemoryBase& device_memory, 92 uint64 allocation_generation) const; 93 94 // Returns whether the manager has a live allocation record for the given 95 // device memory pointer with the given generation counter. 96 // 97 // Note: this is a polling call -- there is no guarantee that the region is 98 // still allocated once the call has completed. 99 bool HasAllocated(const DeviceMemoryBase& device_memory, 100 uint64 generation) const; 101 102 private: 103 // Allocates an array without type parameterization, so that the 104 // implementation can live in the source file. Without this base allocation 105 // method, we incur a circular dependency between the StreamExecutor 106 // definition and this class' definition. 107 port::StatusOr<std::unique_ptr<TemporaryDeviceMemoryBase>> AllocateArrayBase( 108 uint64 element_count, uint64 element_size); 109 110 // Mutex to guard temporary record state. 111 mutable mutex mutex_; 112 113 // Mapping from device memory to the current (live) temporary memory record. 114 // 115 // If a device memory is not in this mapping, it is not a temporary currently 116 // allocated and owned by this temporary memory manager. 117 std::map<DeviceMemoryBase, TemporaryMemoryRecord> records_ GUARDED_BY(mutex_); 118 119 // Allocation generation -- we bump this counter to distinguish temporary 120 // memory handles that have been deallocated and later reallocated at the same 121 // device memory address. 122 uint64 generation_ GUARDED_BY(mutex_); 123 124 // The stream (parent object) for this temporary memory manager -- allocations 125 // are performed through this stream handle. 126 Stream* stream_; 127 128 SE_DISALLOW_COPY_AND_ASSIGN(TemporaryMemoryManager); 129 }; 130 131 //////////// 132 // Inlines 133 134 template <typename T> 135 port::StatusOr<std::unique_ptr<TemporaryDeviceMemory<T>>> 136 TemporaryMemoryManager::AllocateArray(uint64 element_count) { 137 port::StatusOr<std::unique_ptr<TemporaryDeviceMemoryBase>> temporary_memory = 138 AllocateArrayBase(element_count, sizeof(T)); 139 if (!temporary_memory.ok()) { 140 return temporary_memory.status(); 141 } 142 143 return std::unique_ptr<TemporaryDeviceMemory<T>>( 144 reinterpret_cast<TemporaryDeviceMemory<T>*>( 145 temporary_memory.ConsumeValueOrDie().release())); 146 } 147 148 } // namespace internal 149 } // namespace stream_executor 150 151 #endif // TENSORFLOW_STREAM_EXECUTOR_TEMPORARY_MEMORY_MANAGER_H_ 152
