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 : #ifndef HCCLV2_AICPU_WQEMGR_WQE_H
11 : #define HCCLV2_AICPU_WQEMGR_WQE_H
12 :
13 : #include <string>
14 : #include "string_util.h"
15 : #include "hccl/base.h"
16 : #include "enum_factory_legacy.h"
17 : #include "task_param.h"
18 : #include "ip_address.h"
19 :
20 : namespace Hccl {
21 :
22 : #define UDMA_SQE_RMT_EID_SIZE 4
23 :
24 312 : MAKE_ENUM(
25 : UdmaSqOpcode, UDMA_OPC_SEND, UDMA_OPC_SEND_WITH_IMM, UDMA_OPC_SEND_WITH_INVALID, UDMA_OPC_WRITE,
26 : UDMA_OPC_WRITE_WITH_IMM, UDMA_OPC_READ = 0x6, UDMA_OPC_CAS, UDMA_OPC_FAA = 0xb, UDMA_OPC_NOP = 0x11,
27 : UDMA_OPC_INVALID = 0x12)
28 :
29 : MAKE_ENUM(
30 : UdmaDataOp, REDUCE_OP_MAX = 0x8, REDUCE_OP_MIN = 0x9, REDUCE_OP_ADD = 0xA, REDUCE_OP_EQUAL = 0xB,
31 : REDUCE_OP_RESEVERD = 0xC)
32 :
33 : MAKE_ENUM(
34 : UdmaDataType, REDUCE_TYPE_INT8 = 0x0, REDUCE_TYPE_INT16 = 0X1, REDUCE_TYPE_INT32 = 0x2, REDUCE_TYPE_UINT32 = 0X5,
35 : REDUCE_TYPE_FP16_NORMAL = 0X6, REDUCE_TYPE_FP32 = 0X7, REDUCE_TYPE_FP16 = 0X8, REDUCE_TYPE_FP16_SAT = 0X9,
36 : REDUCE_TYPE_RESEVERD = 0XA)
37 :
38 : struct UdmaNormalSge {
39 : uint32_t length;
40 : uint32_t tokenId;
41 : uint32_t dataAddrLow;
42 : uint32_t dataAddrHigh;
43 :
44 : std::string Desc() const
45 : {
46 : constexpr uint64_t UINT32_BIT_WIDTH = 32;
47 : const uint64_t dataAddr
48 : = (static_cast<uint64_t>(dataAddrHigh) << UINT32_BIT_WIDTH) | static_cast<uint64_t>(dataAddrLow);
49 : return StringFormat(
50 : "length = %u dataAddrLow = %u dataAddrHigh = %u dataAddr = 0x%016llx", length, dataAddrLow, dataAddrHigh,
51 : static_cast<unsigned long long>(dataAddr));
52 : }
53 : };
54 :
55 : struct UdmaInlineData {
56 : u8 data[16];
57 : };
58 :
59 : struct UdfExtDate { // UDF扩展数据
60 : uint32_t udfType : 8;
61 : uint32_t reduceType : 4;
62 : uint32_t reduceOp : 4;
63 : uint32_t rsv : 16;
64 : };
65 :
66 : struct UdmaSqe {
67 : uint32_t sqeBbIdx : 16;
68 : uint32_t placeOdr : 2;
69 : uint32_t compOrder : 1;
70 : uint32_t fence : 1;
71 : uint32_t se : 1;
72 : uint32_t cqe : 1;
73 : uint32_t inlineEn : 1;
74 : uint32_t rsv : 5;
75 : uint32_t tokenEn : 1;
76 : uint32_t rmtJettyType : 2;
77 : uint32_t owner : 1;
78 : uint32_t targetHint : 8;
79 : uint32_t opcode : 8;
80 : uint32_t rsv1 : 6;
81 : uint32_t inlineMsgLen : 10;
82 : uint32_t tpn : 24;
83 : uint32_t sgeNum : 8;
84 : uint32_t rmtObjId : 20;
85 : uint32_t rsv2 : 12;
86 : uint32_t rmtEid[UDMA_SQE_RMT_EID_SIZE];
87 : uint32_t rmtTokenValue;
88 : union {
89 : uint32_t rsv3;
90 : UdfExtDate udfData;
91 : } inlinedata;
92 :
93 : uint32_t rmtAddrLow;
94 : uint32_t rmtAddrHigh;
95 : union {
96 : UdmaNormalSge sge;
97 : UdmaInlineData inlineData;
98 : } u;
99 : };
100 :
101 : union LocalValueU {
102 : UdmaNormalSge sge;
103 : UdmaInlineData inlineData;
104 : };
105 :
106 : struct UdmaSqeNotify {
107 : uint32_t notifyTokenId : 20;
108 : uint32_t rsv : 12;
109 : uint32_t notifyTokenValue;
110 : uint32_t notifyAddrLow;
111 : uint32_t notifyAddrHigh;
112 : uint32_t notifyDataLow;
113 : uint32_t notifyDataHigh;
114 : std::string Desc() const
115 : {
116 : return StringFormat(
117 : "notifyAddrLow = %u notifyAddrHigh %u notifyDataLow = %u "
118 : "notifyDataHigh %u",
119 : notifyAddrLow, notifyAddrHigh, notifyDataLow, notifyDataHigh);
120 : }
121 : };
122 :
123 : struct UdmaSqeCommon {
124 : uint32_t sqeBbIdx : 16;
125 : uint32_t placeOdr : 2;
126 : uint32_t compOrder : 1;
127 : uint32_t fence : 1;
128 : uint32_t se : 1;
129 : uint32_t cqe : 1;
130 : uint32_t inlineEn : 1;
131 : uint32_t udfFlag : 1;
132 : uint32_t rsv : 4;
133 : uint32_t tokenEn : 1;
134 : uint32_t rmtJettyType : 2;
135 : uint32_t owner : 1;
136 : uint32_t targetHint : 8;
137 : uint32_t opcode : 8;
138 : uint32_t rsv1 : 6;
139 : uint32_t inlineMsgLen : 10;
140 : uint32_t tpn : 24;
141 : uint32_t sgeNum : 8;
142 : uint32_t rmtObjId : 20;
143 : uint32_t rsv2 : 12;
144 : uint32_t rmtEid[UDMA_SQE_RMT_EID_SIZE];
145 : uint32_t rmtTokenValue;
146 : union {
147 : uint32_t rsv3;
148 : UdfExtDate udfData;
149 : } inlinedata;
150 :
151 : uint32_t rmtAddrLow;
152 : uint32_t rmtAddrHigh;
153 : };
154 :
155 : struct UdmaSqeWriteWithNotify {
156 : struct UdmaSqeCommon comm;
157 : struct UdmaSqeNotify notify;
158 : uint32_t rsv1;
159 : uint32_t rsv2;
160 : union LocalValueU localU;
161 : };
162 :
163 : struct UdmaSqeWrite {
164 : struct UdmaSqeCommon comm;
165 : union LocalValueU u;
166 : };
167 :
168 : struct UdmaSqeRead {
169 : struct UdmaSqeCommon comm;
170 : union LocalValueU u;
171 : };
172 :
173 : // 定义WqeTask用于aicpu task cache
174 : // 注意: 不需要额外维护wqeType指定struct, 因为所有struct开头都是UdmaSqeCommon, 可以利用opCode判断wqe类型
175 : union WqeTask {
176 : struct UdmaSqeRead wqeRead; // 64B (48 + 16); ub_conn_lite.cc中暂不使用UdmaSqeRead
177 : struct UdmaSqeWrite wqeWrite; // 64B (48 + 16)
178 : struct UdmaSqeWriteWithNotify wqeWriteWithNotify; // 96B (48 + 32 + 16)
179 :
180 18 : WqeTask() {}
181 48 : ~WqeTask() {}
182 :
183 : explicit WqeTask(const struct UdmaSqeRead& tmpWqeRead) : wqeRead(tmpWqeRead) {}
184 0 : explicit WqeTask(const struct UdmaSqeWrite& tmpWqeWrite) : wqeWrite(tmpWqeWrite) {}
185 0 : explicit WqeTask(const struct UdmaSqeWriteWithNotify& tmpWqeWriteWithNotify)
186 0 : : wqeWriteWithNotify(tmpWqeWriteWithNotify)
187 0 : {}
188 : };
189 :
190 : // 记录DbSqe TaskParam中, 与WQE相关的profiling信息, 用于aicpu task cache适配profiling
191 : struct DbSqeProfInfo {
192 : bool isValid = false;
193 : TaskParamType taskParamType;
194 : uint64_t srcAddr; // TASK_UB, TASK_UB_REDUCE_INLINE, TASK_WRITE_WITH_NOTIFY, TASK_WRITE_REDUCE_WITH_NOTIFY
195 : uint64_t dstAddr; // TASK_UB_INLINE_WRITE, TASK_UB, TASK_UB_REDUCE_INLINE, TASK_WRITE_WITH_NOTIFY,
196 : // TASK_WRITE_REDUCE_WITH_NOTIFY
197 : uint64_t size; // TASK_UB_INLINE_WRITE, TASK_UB, TASK_UB_REDUCE_INLINE, TASK_WRITE_WITH_NOTIFY,
198 : // TASK_WRITE_REDUCE_WITH_NOTIFY
199 : Eid locEid; // TASK_UB_INLINE_WRITE, TASK_UB, TASK_UB_REDUCE_INLINE, TASK_WRITE_WITH_NOTIFY,
200 : // TASK_WRITE_REDUCE_WITH_NOTIFY
201 : Eid rmtEid; // TASK_UB_INLINE_WRITE, TASK_UB, TASK_UB_REDUCE_INLINE, TASK_WRITE_WITH_NOTIFY,
202 : // TASK_WRITE_REDUCE_WITH_NOTIFY
203 : DmaOp dmaOp; // TASK_UB
204 : HcclReduceOp reduceOp; // TASK_UB_REDUCE_INLINE, TASK_WRITE_REDUCE_WITH_NOTIFY
205 : HcclDataType dataType; // TASK_UB_REDUCE_INLINE, TASK_WRITE_REDUCE_WITH_NOTIFY
206 : u64 notifyId; // TASK_UB_INLINE_WRITE, TASK_WRITE_WITH_NOTIFY, TASK_WRITE_REDUCE_WITH_NOTIFY
207 : uint64_t jettyHandle{0}; // TASK_UB_INLINE_WRITE
208 : uint32_t jettyId{0}; // TASK_UB_INLINE_WRITE, TASK_UB, TASK_UB_REDUCE_INLINE, TASK_WRITE_WITH_NOTIFY,
209 : // TASK_WRITE_REDUCE_WITH_NOTIFY
210 : };
211 :
212 : } // namespace Hccl
213 : #endif // HCCL_AICPU_RESOURCE_AI_CPU_RESOUCES_H_
|
