1 /** 2 * Copyright 2019 Huawei Technologies Co., Ltd 3 * 4 * Licensed under the Apache License, Version 2.0 (the "License"); 5 * you may not use this file except in compliance with the License. 6 * You may obtain a copy of the License at 7 * 8 * http://www.apache.org/licenses/LICENSE-2.0 9 * 10 * Unless required by applicable law or agreed to in writing, software 11 * distributed under the License is distributed on an "AS IS" BASIS, 12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 13 * See the License for the specific language governing permissions and 14 * limitations under the License. 15 */ 16 17 #ifndef MINDSPORE_CCSRC_RUNTIME_DEVICE_GPU_GPU_DEVICE_MANAGER_H_ 18 #define MINDSPORE_CCSRC_RUNTIME_DEVICE_GPU_GPU_DEVICE_MANAGER_H_ 19 20 #include <cudnn.h> 21 #include <cublas_v2.h> 22 #include <cusolverDn.h> 23 #include <cusparse.h> 24 #include <vector> 25 #include <set> 26 #include <memory> 27 #include "plugin/device/gpu/hal/device/cuda_driver.h" 28 #include "plugin/device/gpu/hal/device/gpu_memory_allocator.h" 29 30 namespace mindspore { 31 namespace device { 32 namespace gpu { 33 class GPUDeviceManager { 34 public: 35 void InitDevice(); 36 void ReleaseDevice(); 37 38 int device_count() const; 39 bool set_cur_device_id(uint32_t device_id); 40 uint32_t cur_device_id() const; 41 bool is_device_id_init() const; 42 43 bool CreateStream(CudaDeviceStream *stream); 44 bool CreateStream(size_t *stream_id); 45 bool CreateStreamWithPriority(size_t *stream_id, int32_t priority); 46 bool DestroyStream(size_t stream_id); 47 CudaDeviceStream GetStream(size_t stream_id) const; 48 size_t QueryStreamSize() const; 49 std::vector<uint32_t> GetStreamIds() const; 50 void set_current_stream(size_t stream_id); 51 size_t current_stream() const; 52 bool QueryStream(size_t stream_id); 53 bool SyncStream(size_t stream_id) const; 54 bool SyncStream(const CudaDeviceStream &stream) const; 55 bool SyncAllStreams() const; 56 bool SyncNotDefaultStreams() const; 57 // Sync all streams except the streams in except_streams. 58 bool SyncExceptStreamsInList(const std::set<CudaDeviceStream> &except_streams) const; 59 const CudaDeviceStream &default_stream() const; 60 size_t default_stream_id() const; 61 62 const cudnnHandle_t &GetCudnnHandle() const; 63 const cublasHandle_t &GetCublasHandle() const; 64 const cusolverDnHandle_t &GetCusolverDnHandle() const; 65 const cusparseHandle_t &GetCuSparseHandle() const; 66 67 bool CopyDeviceMemToHost(const HostMemPtr &dst, const DeviceMemPtr &src, size_t size) const; 68 bool CopyHostMemToDevice(const DeviceMemPtr &dst, const void *src, size_t size) const; 69 70 bool CopyDeviceMemToHostAsync(const HostMemPtr &dst, const void *src, size_t size, CudaDeviceStream stream) const; 71 bool CopyHostMemToDeviceAsync(const DeviceMemPtr &dst, const void *src, size_t size, CudaDeviceStream stream) const; 72 bool CopyDeviceMemToDeviceAsync(const DeviceMemPtr &dst, const void *src, size_t size, CudaDeviceStream stream) const; 73 bool CopyHostMemToHost(const HostMemPtr &dst, const void *src, size_t size) const; 74 75 static GPUDeviceManager &GetInstance(); single_op_multi_stream_enable()76 bool single_op_multi_stream_enable() const { return single_op_multi_stream_enable_; } set_single_op_multi_stream_enable(bool single_op_multi_stream_enable)77 void set_single_op_multi_stream_enable(bool single_op_multi_stream_enable) { 78 single_op_multi_stream_enable_ = single_op_multi_stream_enable; 79 } 80 81 private: GPUDeviceManager()82 GPUDeviceManager() : dev_id_init_(false), cur_dev_id_(0), dev_alive_(false) {} 83 ~GPUDeviceManager() = default; 84 GPUDeviceManager(const GPUDeviceManager &) = delete; 85 GPUDeviceManager &operator=(const GPUDeviceManager &) = delete; 86 87 // Ensure the thread safety for creating and destroying stream. 88 std::mutex stream_mutex_; 89 90 // default CUDA stream used for all the kernels. 91 CudaDeviceStream default_stream_{nullptr}; 92 93 size_t default_stream_id_{0}; 94 95 size_t current_stream_id_{0}; 96 97 // all gpu CUDA streams including default_stream_. 98 std::vector<CudaDeviceStream> gpu_streams_; 99 100 // handle used for cuDNN kernels. 101 cudnnHandle_t cudnn_handle_{nullptr}; 102 103 // handle used for cuBLAS kernels. 104 cublasHandle_t cublas_handle_{nullptr}; 105 106 // handle used for cusolver dn kernels; 107 cusolverDnHandle_t cusolver_dn_handle_{nullptr}; 108 109 // handle used for cusparse kernels; 110 cusparseHandle_t cusparse_handle_{nullptr}; 111 bool dev_id_init_; 112 uint32_t cur_dev_id_; 113 bool dev_alive_; 114 bool single_op_multi_stream_enable_{false}; 115 }; 116 } // namespace gpu 117 } // namespace device 118 } // namespace mindspore 119 120 #endif // MINDSPORE_CCSRC_RUNTIME_DEVICE_GPU_GPU_DEVICE_MANAGER_H_ 121