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 : #include <ios>
12 : #include <iostream>
13 :
14 : #include "ccu_temp_reduce_scatter_v_mesh_1D_mem2mem.h"
15 : #include "ccu_instruction_reduce_scatter_v_mesh1d_mem2mem.h"
16 : #include "ccu_context_reduce_scatter_v_mesh1d_mem2mem.h"
17 :
18 : #include "log.h"
19 : #include "ccu_rank_group.h"
20 : #include "ccu_ctx_creator_registry.h"
21 : #include "alg_data_trans_wrapper.h"
22 :
23 : namespace Hccl {
24 : static CcuInstRegister<CcuContextReduceScatterVMeshMem2Mem1D>
25 : g_registrarReduceScatterVmem2mem(CcuInstType::CCU_REDUCE_SCATTER_V_MESH_1D_MEM2MEM_DIRECT);
26 :
27 0 : CcuTempReduceScatterVMeshMem2Mem1D::CcuTempReduceScatterVMeshMem2Mem1D(const RankId virtualRank, const u32 tempRankSize,
28 : const std::vector<std::vector<RankId>> &tempVTopo,
29 0 : const std::map<RankId, u32> &tempVirtRankMap)
30 0 : : CcuAlgTemplateBase(virtualRank, tempRankSize, tempVTopo, tempVirtRankMap)
31 : {
32 0 : }
33 :
34 0 : CcuTempReduceScatterVMeshMem2Mem1D::~CcuTempReduceScatterVMeshMem2Mem1D()
35 : {
36 0 : }
37 :
38 0 : void CcuTempReduceScatterVMeshMem2Mem1D::InitReduceInfo(const ReduceOp &reduceOp, const DataType &dataType)
39 : {
40 0 : reduceOp_ = reduceOp;
41 0 : dataType_ = dataType;
42 0 : }
43 :
44 0 : HcclResult CcuTempReduceScatterVMeshMem2Mem1D::CalcRes(AlgTempResReq &tempResReq)
45 : {
46 0 : tempResReq.queNum = 1;
47 0 : tempResReq.streamNum = tempResReq.queNum;
48 0 : HCCL_INFO("[CalcRes] tempResReq.queNum[%u]", tempResReq.queNum);
49 0 : CHK_RET(CalcResLinksMesh(myRank_, tempRankSize_, tempVTopo_, linkNumBtwPeers_, tempResReq));
50 0 : return HcclResult::HCCL_SUCCESS;
51 : }
52 :
53 0 : HcclResult CcuTempReduceScatterVMeshMem2Mem1D::GenExtIns(const TempFuncs &tempFuncs,
54 : const TemplateDataParams &templateDataParams,
55 : const ResLinks &tempLinks, std::vector<InsQuePtr> &tempInsQues)
56 : {
57 0 : CHK_PRT_RET(tempInsQues.empty(),
58 : HCCL_ERROR("[CcuTempReduceScatterVMeshMem2Mem1D] empty queue"), HcclResult::HCCL_E_INTERNAL);
59 0 : CHK_PTR_NULL(tempInsQues[0]);
60 0 : opMode_ = tempFuncs.opMode;
61 0 : buffInfo_ = templateDataParams.buffInfo;
62 0 : CcuInstructionReduceScatterVMeshMem2Mem1D ccuIns;
63 0 : std::vector<uint64_t> dimSize;
64 0 : dimSize.push_back(tempRankSize_);
65 :
66 : uint64_t inputAddr
67 0 : = BufferTypeToAddr(templateDataParams.buffInfo.inBuffType) + templateDataParams.buffInfo.inBuffBaseOff;
68 : uint64_t outputAddr
69 0 : = BufferTypeToAddr(templateDataParams.buffInfo.outBuffType) + templateDataParams.buffInfo.outBuffBaseOff;
70 : uint64_t scratchAddr
71 0 : = BufferTypeToAddr(templateDataParams.buffInfo.scratBuffType) + templateDataParams.buffInfo.scratchBuffBaseOff;
72 :
73 0 : uint64_t mySliceSize = templateDataParams.sliceSize;
74 0 : uint64_t mySliceInputOffset = templateDataParams.inputSliceStride;
75 :
76 : uint64_t token;
77 0 : CHK_RET(GetToken(op_, token));
78 :
79 0 : ccuIns.Init(static_cast<uint32_t>(myRank_), inputAddr, outputAddr, scratchAddr, mySliceSize, mySliceSize, mySliceInputOffset, token, op_, tempVTopo_);
80 0 : HCCL_INFO("[CcuTempReduceScatterVMeshMem2Mem1D] Run Init: myRank_[%d], dimSize[%llu], inputAddr[%llu],"
81 : "outputAddr[%llu], mySliceSize[%llu], mySliceInputOffset[%llu]",
82 : myRank_, dimSize[0], inputAddr, outputAddr, mySliceSize, mySliceInputOffset);
83 :
84 0 : std::vector<LinkData> links;
85 0 : for (auto &pair : tempLinks) {
86 0 : if (pair.second.empty()) {
87 0 : continue;
88 : }
89 0 : links.push_back(pair.second[0]);
90 : }
91 0 : HCCL_INFO("[CcuTempReduceScatterVMeshMem2Mem1D] links.size[%zu]", links.size());
92 0 : ccuIns.SetLinks(links);
93 :
94 0 : RankGroup rankGroup;
95 0 : for (auto &peer : tempVTopo_[0]) {
96 0 : rankGroup.AddRank(peer);
97 : }
98 0 : ccuIns.SetRankGroup(rankGroup);
99 :
100 0 : u32 cntCkeNum = 3;
101 0 : ccuIns.SetCntCkeNum(cntCkeNum);
102 0 : tempInsQues[0]->Append(std::move(std::make_unique<CcuInstructionReduceScatterVMeshMem2Mem1D>(ccuIns)));
103 :
104 0 : return HcclResult::HCCL_SUCCESS;
105 0 : }
106 :
107 0 : u32 CcuTempReduceScatterVMeshMem2Mem1D::CalcScratchMultiple(BufferType input, BufferType output)
108 : {
109 : (void)input;
110 : (void)output;
111 : // one shot 场景,scratch Buffer 需要是 usrIn的rankSize倍
112 0 : return tempRankSize_;
113 : }
114 :
115 : } // namespace Hccl
|