Line data Source code
1 : /**
2 : * Copyright (c) 2025 Huawei Technologies Co., Ltd.
3 : * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4 : * CANN Open Software License Agreement Version 2.0 (the "License").
5 : * Please refer to the License for details. You may not use this file except in compliance with the License.
6 : * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7 : * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8 : * See LICENSE in the root of the software repository for the full text of the License.
9 : */
10 :
11 : #ifndef PROFILER_BASE_PUB_H
12 : #define PROFILER_BASE_PUB_H
13 :
14 : #include <memory>
15 : #include <mutex>
16 : #include <string>
17 : #include <map>
18 : #include <thread>
19 :
20 : #include <hccl/hccl_types.h>
21 : #include "hccl_common.h"
22 : #include "common.h"
23 : #include "workflow_pub.h"
24 : #include "dispatcher_task_types.h"
25 : #include "alg_profiling.h"
26 : #include "profiler_base_pub_extend.h"
27 :
28 : namespace hccl {
29 : enum class StepType {
30 : STEP_STAGE = 0,
31 : STEP_STEP,
32 : STEP_MAX
33 : };
34 :
35 : enum class OpDict {
36 : SUM = 0,
37 : PROD,
38 : MAX,
39 : MIN
40 : };
41 :
42 : enum class DataType {
43 : DINT8 = 0,
44 : DINT16,
45 : DINT32,
46 : DFP16,
47 : DFP32,
48 : DINT64,
49 : DUINT64
50 : };
51 :
52 : struct GroupRankInfo {
53 : u32 rankSize{0};
54 : u32 rankId{0};
55 : u32 remoteRankId{INVALID_VALUE_RANKSIZE};
56 : };
57 :
58 : struct OpDataInfo {
59 : u64 count{0};
60 : const void *src{nullptr};
61 : const void *dst{nullptr};
62 : u32 index{0};
63 : u32 rootId{0};
64 : u32 deviceId{0};
65 : HcclDataType dataType{HcclDataType::HCCL_DATA_TYPE_RESERVED};
66 : HcclReduceOp reduceType{HcclReduceOp::HCCL_REDUCE_RESERVED};
67 : struct timeval tv{0};
68 : };
69 :
70 : struct StreamRecordInfo {
71 : s32 planeId;
72 : AlgType algType;
73 : std::string tag;
74 27 : StreamRecordInfo() = default;
75 27 : StreamRecordInfo(s32 plane, const AlgType &type, const std::string &strTag) : planeId(plane), algType(type), tag(strTag) {}
76 0 : StreamRecordInfo(const StreamRecordInfo &that) : planeId(that.planeId), algType(that.algType), tag(that.tag) {}
77 27 : StreamRecordInfo &operator=(const StreamRecordInfo &that)
78 : {
79 27 : if (&that != this) {
80 27 : planeId = that.planeId;
81 27 : algType = that.algType;
82 27 : tag = that.tag;
83 : }
84 27 : return *this;
85 : }
86 : };
87 :
88 : class ProfilerBase {
89 : public:
90 : /* * 输出文本时, 获取op, dataType的字符串以及单位数据长度的数组 */
91 : static const std::array<uint32_t, HCCL_REDUCE_RESERVED> opString;
92 : static const std::array<uint32_t, HCCL_DATA_TYPE_RESERVED> dataTypeString;
93 : static const std::array<s32, HCCL_DATA_TYPE_RESERVED> sizeOf;
94 :
95 : explicit ProfilerBase(u32 deviceLogicId);
96 : virtual ~ProfilerBase();
97 :
98 : virtual HcclResult Run(const StepData &stepData) = 0;
99 : virtual HcclResult Flush() = 0;
100 : static HcclResult AddStream(s32 streamID, const std::string &tag, s32 planeID, const AlgType &algType);
101 : static HcclResult DelStream(s32 streamID);
102 : static HcclResult AddTag(const std::string &tag, const std::string &group, const HcclWorkflowMode &workFlowMode,
103 : bool isSendRecv = false, bool isAiv = false);
104 : static HcclResult DelTag(const std::string &tag);
105 : static HcclResult AddOpData(const std::string &tag, u64 count, const void *src, const void *dst,
106 : HcclDataType dataType, u32 rootId, const std::string &group, HcclReduceOp reduceType = HCCL_REDUCE_RESERVED);
107 : static HcclResult DelOpData(const std::string &tag);
108 : static HcclResult AddGroupRankInfo(const std::string &group, u32 rankSize, u32 rankId, bool isSendRecv = false,
109 : u32 remoteRankId = INVALID_VALUE_RANKSIZE);
110 : static HcclResult DelGroupRankInfo(const std::string &tag);
111 : static HcclResult GetTagByStream(u32 &streamID, std::string &tag);
112 : static HcclResult GetAlgTypeByStream(u32 &streamID, AlgType &algType);
113 : static HcclResult GetGroupNameByTag(const std::string &tag, std::string &group);
114 : static HcclResult GetRankInfoByGroup(const std::string &group, GroupRankInfo &groupRankInfo);
115 : static HcclResult GetOpDataInfoByTag(const std::string &tag, OpDataInfo &opDataInfo);
116 : static HcclResult AddGroupUdi(const std::string &group, const std::string &udi);
117 : static HcclResult DelGroupUdi(const std::string &group);
118 : static HcclResult GetUdiByGroup(const std::string &group, std::string &udi);
119 : static void GetSubmittedOpCnt(u32 &index);
120 : virtual HcclResult Save(u32 &streamID, u32 &taskID, TaskType &taskType, const TaskParaDMA ¶) = 0;
121 : virtual HcclResult Save(u32 &streamID, u32 &taskID, TaskType &taskType, const TaskParaReduce ¶) = 0;
122 : virtual HcclResult Save(u32 &streamID, u32 &taskID, TaskType &taskType, const TaskParaNotify ¶) = 0;
123 : virtual HcclResult Save(u32 streamID, u32 taskID, const TaskParaAiv ¶) = 0;
124 : virtual HcclResult Save(u32 &streamID, u32 &taskID, const void *descBuf = nullptr, size_t descBufLen = 0) = 0;
125 : virtual HcclResult Save(u32 captureStreamID, u32 streamID, u32 taskID, TaskType &taskType, const TaskParaDMA ¶) = 0;
126 : virtual HcclResult Save(u32 captureStreamID, u32 streamID, u32 taskID, TaskType &taskType, const TaskParaReduce ¶) = 0;
127 : virtual HcclResult Save(u32 captureStreamID, u32 streamID, u32 taskID, TaskType &taskType, const TaskParaNotify ¶) = 0;
128 : virtual HcclResult Save(u32 captureStreamID, u32 streamID, u32 taskID, const void *descBuf = nullptr, size_t descBufLen = 0) = 0;
129 : virtual HcclResult Save(u32 captureStreamID, u32 streamID, u32 taskID, const TaskParaAiv ¶Aiv) = 0;
130 : virtual HcclResult SaveToLog(const TaskParaHost ¶Host) = 0;
131 :
132 : protected:
133 : static std::array<std::map<s32, StreamRecordInfo>, MAX_MODULE_DEVICE_NUM> streamRecordInfoMap_;
134 : static std::array<std::map<const std::string, const std::string>, MAX_MODULE_DEVICE_NUM> tagGroupMap_;
135 : static std::array<std::map<const std::string, const HcclWorkflowMode>, MAX_MODULE_DEVICE_NUM> tagModeMap_;
136 : static std::array<std::mutex, MAX_MODULE_DEVICE_NUM> streamMutex_;
137 : static std::array<std::map<const std::string, GroupRankInfo>, MAX_MODULE_DEVICE_NUM> groupRankMap_;
138 : static std::array<std::map<const std::string, OpDataInfo>, MAX_MODULE_DEVICE_NUM> tagOpDataMap_;
139 : static std::array<std::map<const std::string, u32>, MAX_MODULE_DEVICE_NUM> groupIndexMap_;
140 : static std::array<std::map<const std::string, u32>, MAX_MODULE_DEVICE_NUM> aivGroupIndexMap_;
141 : static std::array<std::map<const std::string, u32>, MAX_MODULE_DEVICE_NUM> sendRecvGroupIndexMap_;
142 : static std::array<std::map<const std::string, std::string>, MAX_MODULE_DEVICE_NUM> groupUdiMap_;
143 : const u32 deviceLogicId_;
144 : static bool isSendRecv_[MAX_MODULE_DEVICE_NUM];
145 : static u32 index_[MAX_MODULE_DEVICE_NUM];
146 :
147 : private:
148 : };
149 : } // namespace hccl
150 :
151 : #endif /* PROFILER_BASE_PUB_H */
|
