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 MY_RANK_H
12 : #define MY_RANK_H
13 :
14 : #include <cstdint>
15 : #include <memory>
16 : #include <mutex>
17 : #include <string>
18 : #include <unordered_map>
19 : #include <utility>
20 : #include <vector>
21 :
22 : #include "hccl/hccl_types.h"
23 : #include "hccl/base.h"
24 : #include "hccl/hccl_res.h"
25 : #include "hccl_mem_defs.h"
26 : #include "acl/acl_rt.h"
27 : #include "socket_manager.h"
28 : #include "hcomm_res_defs.h"
29 : #include "hcomm_channel.h"
30 : #include "mem_host_pub.h"
31 : #include "rank_pair_mgr.h"
32 : #include "endpoint_mgr.h"
33 : #include "comm_config_pub.h"
34 : #include "manager_common.h"
35 : #include "common.h"
36 : #include "comm_mems/comm_mems.h"
37 : #include "engine_ctxs.h"
38 : #include "../../../dfx/ns_recovery/ns_recovery.h"
39 : #include "hdc_pub.h"
40 : #include "rank_graph.h"
41 : #include "orion_adapter_hccp.h"
42 : #include "coll_comm_config_consistency.h"
43 : #include "exchange_info_mgr.h"
44 :
45 : #include "ccu_types.h"
46 : #include "ccu_drv_handle.h"
47 : #include "ccu_device_res.h"
48 : #include "dev_type.h"
49 :
50 : namespace hccl {
51 :
52 : constexpr uint32_t DEFAULT_MODE = 0;
53 : constexpr uint32_t AICPU_TS_MODE = 2;
54 : constexpr uint32_t CCU_MS_MODE = 5;
55 : constexpr uint32_t CCU_SCHED_MODE = 6;
56 : // opExpansionMode 到 CcuInstanceType 的映射，仅作数据类型转换，不做逻辑处理。
57 : // 供 MyRank 内部及外部适配层（如 coll_comm_ccu_c_adpt.cc）共用。
58 13 : inline CcuInstanceType OpExpansionModeToCcuInstanceType(uint32_t opExpansionMode)
59 : {
60 13 : if (opExpansionMode == CCU_SCHED_MODE) {
61 3 : return CcuInstanceType::CCU_SCHED;
62 : }
63 10 : if (opExpansionMode == CCU_MS_MODE) {
64 7 : return CcuInstanceType::CCU_MS;
65 : }
66 3 : return CcuInstanceType::CCU_UNUSED;
67 : }
68 :
69 : /**
70 : * @note 职责：管理当前通信域下本Rank的信息和通信资源
71 : */
72 : class MyRank {
73 : public:
74 : MyRank(
75 : aclrtBinHandle binHandle, uint32_t rankId, const CommConfig& config, const ManagerCallbacks& callbacks,
76 : RankGraph* rankGraph, const Hccl::RankIpPortMapPtr& rankIpPortMap);
77 : ~MyRank();
78 :
79 : HcclResult Init(HcclMem cclBuffer, const uint32_t opExpansionMode, uint32_t rankNum);
80 :
81 36 : CommMems* GetCommMems() const { return commMems_.get(); }
82 :
83 17 : EngineCtxs* GetEngineCtxs() const { return engineCtxs_.get(); }
84 :
85 : HcclResult UnregMemByTag(const std::string& tag);
86 177 : uint32_t GetOpExpansionMode() { return opExpansionMode_; }
87 17 : CcuInsHandle GetCcuInstance() const { return ccuInsHandle_; }
88 3 : void SetCcuInstance(CcuInsHandle ccuInsHandle) { ccuInsHandle_ = ccuInsHandle; }
89 22 : CcuInsHandle GetAssignedCcuInstance() const { return assignedCcuInsHandle_; }
90 4 : void SetAssignedCcuInstance(CcuInsHandle ccuInsHandle) { assignedCcuInsHandle_ = ccuInsHandle; }
91 :
92 : CollCommConfigConsistency& GetCollCommConfigConsistency();
93 :
94 0 : hcomm::EndpointMgr* GetEndpointMgr() const { return endpointMgr_.get(); }
95 :
96 : HcclResult CreateChannels(
97 : CommEngine engine, const std::string& commTag, const HcclChannelDesc* channelDescs, uint32_t channelNum,
98 : ChannelHandle* channels);
99 :
100 : HcclResult
101 : QueryChannels(CommEngine engine, const HcclChannelDesc* channelDescs, uint32_t channelNum, ChannelHandle* channels);
102 : HcclResult DestroyChannels(const ChannelHandle* channels, uint32_t channelNum);
103 :
104 : HcclResult ChannelGetHcclBuffer(ChannelHandle channel, void** buffer, uint64_t* size);
105 : HcclResult
106 : ChannelGetRemoteMems(ChannelHandle channel, uint32_t* memNum, CommMem** remoteMem, char*** memTags) const;
107 : HcclResult ChannelGetRemoteMems(
108 : ChannelHandle channel, uint32_t* memNum, CommMem** remoteMem, std::vector<std::string>& memTags) const;
109 :
110 : // Ns recovery
111 : void SetKfcControlTransfer(
112 : std::shared_ptr<HDCommunicate> kfcControlTransferH2D, std::shared_ptr<HDCommunicate> kfcStatusTransferD2H);
113 : std::vector<ChannelHandle> GetAllChannelList();
114 : HcclResult StopLaunch();
115 : HcclResult Clean();
116 : HcclResult Resume();
117 :
118 : /**
119 : * @brief 批量预建 socket（server 监听 + client 连接），与非共享路径 MyRank::CreateChannels 一致。
120 : * 共享 jetty 路径在 createFunc 中调用，确保两端 socket 体系一致、可对接。
121 : * @param[in] channelDescs HCCL 层 channel desc 数组
122 : * @param[in] channelNum 数量
123 : * @param[in] socketTag socket 标签（commTag + "_engine_" + engine）
124 : * @param[out] hcommDescs 输出的 HcommChannelDesc 数组（调用前需用 ChannelDescHccl2Hcomm 填充基础字段，
125 : * 本方法补上 socket/role/port 字段）
126 : */
127 : HcclResult BatchCreateSockets(
128 : const HcclChannelDesc* channelDescs, uint32_t channelNum, const std::string& socketTag,
129 : std::vector<HcommChannelDesc>& hcommDescs);
130 :
131 : /**
132 : * @brief 在已建好的 socket 上执行通信域一致性校验交换（CheckFrameV2 + 用户信息）。
133 : * 与非共享路径 MyRank::CreateChannels 内部调用 exchangeInfoMgr_ 的逻辑一致，
134 : * 供共享 jetty 路径在 createFunc 中调用，确保两端无论 shared/非shared 配置是否对称，
135 : * 都会在 socket 上完成定长 CheckFrameV2（120B）的对称收发，避免一端死等。
136 : * @param[in] channelDescs HCCL 层 channel desc 数组
137 : * @param[in] hcommDescs 已通过 BatchCreateSockets 填充 socket 字段的 HcommChannelDesc 数组
138 : * @param[in] channelNum 数量
139 : * @param[in] newChannels 新建 channel 的 (idx, reuseIdx) 列表；空表示全部按新建处理
140 : * @param[in] engine 通信引擎
141 : * @note 仅 DEV_TYPE_950 实际执行交换，其它设备类型直接返回 SUCCESS，与非共享路径保持一致。
142 : */
143 : HcclResult BatchExchangeAndCheckConsistency(
144 : const HcclChannelDesc* channelDescs, const std::vector<HcommChannelDesc>& hcommDescs, uint32_t channelNum,
145 : const std::vector<std::pair<u32, u32>>& newChannels, CommEngine engine);
146 : HcclResult PrepareMemHandles(
147 : EndpointHandle epHandle, void** memHandles, uint32_t memHandleNum, std::vector<MemHandle>& memHandleVec);
148 :
149 : private:
150 : using ReuseSocketIdxMap = std::unordered_map<RankPair*, std::unordered_map<hcomm::EndpointPair*, u32>>;
151 : HcclResult GetEndpointPairFromChannel(
152 : const HcclChannelDesc& channelDesc, uint32_t channelIndex, uint32_t channelNum, uint32_t& remoteRank,
153 : hcomm::EndpointPair*& endpointPair, RankPair*& rankPair);
154 : HcclResult BatchServerInitForChannels(
155 : const HcclChannelDesc* channelDescs, uint32_t channelNum, const std::string& socketTag,
156 : ReuseSocketIdxMap& reuseSocketIdxMap);
157 : HcclResult BatchGetSocketsForChannels(
158 : const HcclChannelDesc* channelDescs, uint32_t channelNum, const std::string& socketTag,
159 : std::vector<HcommChannelDesc>& hcommDescs, ReuseSocketIdxMap& reuseSocketIdxMap);
160 : HcclResult BatchCreateChannels(
161 : CommEngine engine, const HcclChannelDesc* channelDescs, uint32_t channelNum,
162 : std::vector<HcommChannelDesc>& hcommDescs, ChannelHandle* channelHandles,
163 : std::vector<std::vector<MemHandle>>& allHandles);
164 : HcclResult
165 : BatchConnectChannels(const HcclChannelDesc* channelDescs, ChannelHandle* channelHandles, uint32_t channelNum);
166 : void LogChannelCreationInfo(
167 : CommEngine engine, const std::string& commTag, const HcclChannelDesc* channelDescs, uint32_t channelNum,
168 : ChannelHandle* hostChannelHandleList);
169 : HcclResult FinalizeChannelsByEngine(
170 : CommEngine engine, const std::string& commTag, const HcclChannelDesc* channelDescs, uint32_t channelNum,
171 : std::vector<HcommChannelDesc>& hcommDescs, ChannelHandle* hostChannelHandleList, ChannelHandle* channelHandles);
172 : HcclResult CheckChannelParam(CommEngine engine, const HcclChannelDesc* channelDesc, uint32_t channelNum) const;
173 : HcclResult QueryListenPort(
174 : uint32_t localRank, uint32_t remoteRank, const EndpointDesc& localEndpointDesc,
175 : const EndpointDesc& remoteEndpointDesc, uint32_t& listenPort, HcommChannelDesc& hcommDesc);
176 : HcclResult GetLocalTlsStatus(Hccl::TlsStatus& tlsStatus) const;
177 : HcclResult RegisterCommMemsToEndpoint(EndpointHandle epHandle);
178 : HcclResult TryInitCcuInstance();
179 : HcclResult ReserveCcuMsCommOrFallback();
180 : HcclResult TryInitCcuInstanceOnDemand();
181 : void ReconcileCcuMsCommReservation(HcclResult initRet);
182 : void ReleaseCcuMsCommReservation();
183 : HcclResult ConfigSqDepthByExpansionMode(CommEngine engine, HcommChannelDesc& hcommDesc) const;
184 : HcclResult DestroyNewChannels(
185 : CommEngine engine, const HcclChannelDesc* channelDescs, const std::vector<std::pair<u32, u32>>& newChannels);
186 : HcclResult
187 : QueryOneChannel(CommEngine engine, const HcclChannelDesc& channelDesc, u32 reuseIdx, ChannelHandle& handle);
188 : HcclResult
189 : DestroyOneChannel(ChannelHandle userHandle, u32 index, HcclResult& firstErr, u32& invalidHandleCnt, u32& failedCnt);
190 : aclrtBinHandle binHandle_{nullptr};
191 : uint32_t rankId_{};
192 : int32_t devLogicId_{};
193 : CommConfig config_{};
194 :
195 : // 当前通信域初始化没有处理CommConfig，暂时只使用展开模式
196 : uint32_t opExpansionMode_{0};
197 :
198 : std::unique_ptr<RankPairMgr> rankPairMgr_{nullptr};
199 : std::unique_ptr<hcomm::EndpointMgr> endpointMgr_{nullptr};
200 : std::unique_ptr<CommMems> commMems_{nullptr};
201 : std::unique_ptr<EngineCtxs> engineCtxs_{nullptr};
202 :
203 : CcuInsHandle ccuInsHandle_{0}; // 按固定量CCU资源方式创建的 ccuInsHandle（HcclCommQueryCcuIns）（兼容旧版本）
204 : CcuInsHandle assignedCcuInsHandle_{0}; // 按需申请CCU资源方式绑定的 ccuInsHandle（HcommCcuInsCreate + Assign）
205 :
206 : ManagerCallbacks callbacks_;
207 :
208 : // RankGraph (临时放在myRank里面，后面会随着createchannel整体迁移到RankPairMgr上)
209 : RankGraph* rankGraph_{nullptr};
210 :
211 : // 记录每次调用BatchCreateChannels时新增的channelIndex, reuseIdx
212 : std::vector<std::pair<u32, u32>> newChannels_{};
213 :
214 : // channelHandle(host) -> EndpointPair 裸指针反查索引：EndpointPair 由 EndpointPairMgr 以 unique_ptr 持有，
215 : // 其生命周期与 RankPairMgr 一致（MyRank 析构时先 clear 反查表再释放 rankPairMgr_）
216 : std::unordered_map<ChannelHandle, hcomm::EndpointPair*> handleToEpPair_{};
217 :
218 : // 保护 newChannels_ / handleToEpPair_ 的并发访问
219 : std::mutex channelIndexMtx_{};
220 :
221 : // Ns recovery
222 : std::unique_ptr<NsRecoveryProcessor> nsRecoveryProcessor_{nullptr};
223 : // 内部获取 port 的方法，根据 mode_ 区分 v1/v2
224 : HcclResult GetDevicePortInternal(uint32_t rank, uint32_t* devPort, EndpointLocType locType);
225 :
226 : Hccl::RankIpPortMapPtr rankIpPortMap_;
227 :
228 : CollCommConfigConsistency collCommConfigConsistency_;
229 : ExchangeInfoMgr exchangeInfoMgr_;
230 : std::shared_ptr<hcomm::CcuDrvHandle> ccuDrvHandle_{};
231 : bool ccuMsCommReserved_{false};
232 : Hccl::HcclMainboardId mainBoardType_{Hccl::HcclMainboardId::MAINBOARD_OTHERS};
233 : };
234 :
235 : } // namespace hccl
236 :
237 : namespace MyRankUtils {
238 :
239 : HcommChannelDesc ChannelDescHccl2Hcomm(const HcclChannelDesc& hcclDesc, const hccl::CommConfig& commConfig);
240 :
241 : HcclResult
242 : FillRoceSrcPortList(const HcclChannelDesc& hcclDesc, HcommChannelDesc& hcommDesc, std::vector<uint16_t>& srcPortBuf);
243 :
244 : } // namespace MyRankUtils
245 :
246 : #endif // MY_RANK_H
|
