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 DUMP_ARGS_H
11 : #define DUMP_ARGS_H
12 :
13 : #include <string>
14 : #include <vector>
15 : #include <sstream>
16 : #include "adump_pub.h"
17 : #include "adump_api.h"
18 : #include "dump_operator.h"
19 : #include "runtime/rt.h"
20 : #include "kernel_info_collector.h"
21 : #include "log/adx_log.h"
22 : #include "adx_exception_callback.h"
23 :
24 : namespace Adx {
25 : #define IDE_CHECK_RET(ret, action) \
26 : do { \
27 : if ((ret) != ADUMP_SUCCESS) { \
28 : action; \
29 : } \
30 : } while (0)
31 :
32 : constexpr uint16_t TYPE_L0_EXCEPTION_DFX = 4U;
33 : constexpr uint16_t TYPE_L0_EXCEPTION_DFX_ARGS_INFO = 5U;
34 : constexpr uint16_t TYPE_L0_EXCEPTION_DFX_IS_TIK = 6U;
35 : constexpr uint64_t TENSOR_TYPE_MASK = 0x0FFFF; // 0~15 bit
36 : constexpr uint64_t POINTER_TYPE_MASK = 0x0FFFF0000; // 16~31 bit
37 : constexpr uint64_t TENSOR_COUNT_MASK = 0xFFFFFFFF00000000; // high 32 bits
38 : constexpr uint64_t TENSOR_DIMENSION_MASK = 0x0FFFFFFFF; // low 32 bits
39 : constexpr uint32_t TENSOR_COUNT_SHIFT_BITS = 32;
40 : constexpr uint32_t POINTER_TYPE_SHIFT_BITS = 16;
41 : constexpr int32_t ELF_DATA2MSB = 2;
42 : constexpr uint32_t BITS_PER_BYTE = 8U;
43 :
44 : enum class DfxPointerType : uint16_t {
45 : INVALID_POINTER = 0,
46 : LEVEL_1_POINTER = 1,
47 : LEVEL_2_POINTER,
48 : LEVEL_2_POINTER_WITH_SHAPE,
49 : SHAPE_TENSOR_PLACEHOLD
50 : };
51 :
52 : struct InputBuffer {
53 125 : InputBuffer(const void *argAddr, uint64_t len, uint32_t index) : addr(argAddr), length(len), argIndex(index) {}
54 : const void *addr;
55 : uint64_t length;
56 : uint32_t argIndex;
57 : };
58 :
59 : struct TensorBuffer {
60 50 : TensorBuffer(const void *argAddr, uint32_t index, DfxTensorType dfxTensorType, DfxPointerType dfxPointerType)
61 50 : : addr(argAddr),
62 50 : argIndex(index),
63 50 : tensorType(dfxTensorType),
64 50 : pointerType(dfxPointerType)
65 : {
66 50 : }
67 : const void *addr{nullptr};
68 : uint64_t size{0U};
69 : uint64_t dataTypeSize{1U}; // 无data type size情况下，默认值为1，size为实际内存大小
70 : bool isDataTypeSizeByte{true}; // 标记dataTypeSize以bit还是byte为单位
71 : uint64_t dimension{0U};
72 : std::vector<uint64_t> shape;
73 : uint32_t argIndex;
74 : DfxTensorType tensorType;
75 : DfxPointerType pointerType;
76 42 : uint64_t GetTotalByteSize() const
77 : {
78 42 : if (isDataTypeSizeByte) {
79 34 : return dataTypeSize * size;
80 : }
81 8 : uint64_t totalBits = dataTypeSize * size;
82 8 : uint32_t remainder = totalBits % BITS_PER_BYTE == 0 ? 0 : 1; // 向上取整
83 8 : return totalBits / BITS_PER_BYTE + remainder;
84 : }
85 : };
86 :
87 : extern uint64_t g_chunk[RING_CHUNK_SIZE + MAX_TENSOR_NUM];
88 :
89 : class DumpArgs {
90 : public:
91 67 : DumpArgs()
92 134 : : sizeInfo_(nullptr),
93 67 : sizeBeginIndex_(0),
94 67 : skipNum_(0),
95 67 : inputNum_(0),
96 67 : argAddr_(nullptr),
97 67 : argSize_(0),
98 67 : dumpWithDfxFlag_(false),
99 67 : isTik_(false),
100 67 : exceptionDfxPtr_(nullptr),
101 67 : exceptionDfxSize_(0),
102 67 : kernelCollector_(std::make_shared<KernelInfoCollector>()){};
103 67 : ~DumpArgs() = default;
104 : int32_t LoadArgsExceptionInfo(const rtExceptionInfo &exception);
105 : int32_t DumpArgsExceptionInfo(const uint32_t deviceId, const std::string &dumpPath);
106 :
107 : bool DumpArgsDumpWithDfxFlag() const;
108 : const std::vector<InputBuffer> &DumpArgsGetInputBuffer() const;
109 : const std::vector<TensorBuffer> &DumpArgsGetTensorBuffer() const;
110 : const std::vector<DumpWorkspace> &DumpArgsGetWorkSpace() const;
111 :
112 : private:
113 : int32_t DumpArgsExceptionFile(const uint32_t deviceId, const std::string &dumpPath);
114 : std::string GetDumpFilePath(const std::string &dumpPath) const;
115 : int32_t CheckParam(const rtExceptionArgsInfo_t &exceptionArgsInfo,
116 : const rtExceptionExpandInfo_t &exceptionExpandInfo) const;
117 : int32_t InitAttributes(const rtExceptionArgsInfo_t &exceptionArgsInfo,
118 : const rtExceptionExpandInfo_t &exceptionExpandInfo);
119 : int32_t InitTensorModeInfo(const uint8_t *exceptionDfxPtr);
120 : int32_t InitTensorModeInfoInner(const uint8_t *exceptionDfxPtr, uint64_t &currDfxSize, uint32_t currArgsIndex);
121 : int32_t LoadArgsInfoWithDfx(const rtExceptionArgsInfo_t &exceptionArgsInfo);
122 : int32_t LoadArgsInfoWithSizeInfo(const rtExceptionArgsInfo_t &exceptionArgsInfo,
123 : const rtExceptionExpandInfo_t &exceptionExpandInfo);
124 : void LogArgsInfo(const void **argOnHost, uint32_t maxArgNum);
125 : int32_t LoadInputBuffer(const void **argOnHost, const uint32_t argIndex, uint64_t &sizeInfoIdx);
126 : int32_t LoadPointerTensor(const void **argOnHost, const uint32_t argIndex, uint64_t &sizeInfoIdx);
127 : void LoadTilingData();
128 :
129 : template <typename T>
130 : int32_t GetPointerValueByBigEndian(const uint8_t **ptr, T &value, uint64_t ¤tDfxSize,
131 : uint16_t totalDfxSize) const;
132 : template <typename T>
133 : int32_t GetPointerValueByLittleEndian(const uint8_t **ptr, T &value, uint64_t ¤tDfxSize,
134 : uint16_t totalDfxSize) const;
135 : int32_t FindExceptionDfx(const rtExceptionArgsInfo_t &exceptionArgsInfo);
136 : int32_t CheckAddressOverArgs(const uint64_t *address, const void **argOnHost, uint64_t maxArgNum) const;
137 : int32_t GetAddressBias(uint64_t &addrBias, const void *argAddr, void *baseAddr, uint64_t argsSize) const;
138 : void RecordCurrentLog();
139 : std::string taskId_;
140 : std::string streamId_;
141 : std::vector<InputBuffer> inputBuffer_;
142 : std::vector<InputBuffer> tilingData_;
143 : std::vector<TensorBuffer> tensorBuffer_;
144 : std::vector<DumpWorkspace> workspace_;
145 : std::vector<DumpWorkspace> mc2Space_;
146 : uint64_t *sizeInfo_;
147 : uint32_t sizeBeginIndex_;
148 : uint32_t skipNum_;
149 : uint32_t inputNum_;
150 : void *argAddr_;
151 : uint64_t argSize_;
152 : bool dumpWithDfxFlag_;
153 : bool isTik_;
154 : const uint8_t *exceptionDfxPtr_;
155 : uint16_t exceptionDfxSize_;
156 : std::shared_ptr<KernelInfoCollector> kernelCollector_;
157 : std::vector<std::string> logRecord_;
158 : std::ostringstream oss_;
159 : };
160 : } // namespace Adx
161 : #endif
|
