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 : /*!
12 : * \file kernel_base_check.h
13 : * \brief
14 : */
15 :
16 : #ifndef ASCENDC_BASE_CHECK_H
17 : #define ASCENDC_BASE_CHECK_H
18 :
19 : #include <vector>
20 : #include <string>
21 : #include "kernel_check_util.h"
22 :
23 : namespace AscendC {
24 : namespace check {
25 : const uint32_t MASK_ARRAY_LEN = 2;
26 : const uint32_t NAME_MAX_LEN = 128;
27 :
28 : enum class ModeType : uint8_t { NONE_MODE = 0, NORM_MODE = 1, COUNTER_MODE = 2 };
29 :
30 : struct TensorOverflowParams {
31 : TensorOverflowParams() = default;
32 :
33 248 : TensorOverflowParams(
34 : const uint64_t bufferSizeIn, const uint64_t dtypeSizeIn, const uint64_t repeatTimesIn,
35 : const uint64_t blkStrideIn, const uint64_t repStrideIn, const bool isCounterIn)
36 248 : : bufferSize(bufferSizeIn),
37 248 : dtypeSize(dtypeSizeIn),
38 248 : repeatTimes(repeatTimesIn),
39 248 : blkStride(blkStrideIn),
40 248 : repStride(repStrideIn),
41 248 : isCounter(isCounterIn)
42 248 : {}
43 :
44 : uint64_t bufferSize = 0;
45 : uint64_t dtypeSize = 1;
46 : uint64_t repeatTimes = 1;
47 : uint64_t blkStride = DEFAULT_BLK_STRIDE;
48 : uint64_t repStride = DEFAULT_REPEAT_STRIDE;
49 : bool isCounter = false; // only used for gathermask
50 : };
51 :
52 : /*
53 : * @funcname: GetMaskLength
54 : * @brief: the maximum number of elements calculated this repeat.
55 : * For example mask is 0x30fff + 0xffffffff. The maximum num per repeat is 64 + 18 = 82
56 : * @return: uint64_t
57 : */
58 : uint64_t GetMaskLength(std::vector<uint64_t>& maskArray, const uint32_t dtypeSize);
59 :
60 : /*
61 : * @funcname: CheckTensorSizeOverflow
62 : * @brief: common method used to compare tensor size and minimum needed size, and report error if overflow
63 : * mode 0: none, mode 1: norm mode, mode 2: counter mode
64 : * @return: true/false
65 : */
66 : bool CheckTensorSizeOverflow(
67 : uint64_t expectedSize, uint64_t tensorSize, const std::string& tensorName, const std::string& apiName,
68 : const ModeType mode = ModeType::NONE_MODE);
69 :
70 : void CounterSplitMainTail(
71 : std::vector<uint64_t>& maskArray, const uint32_t dtypeBytes, uint64_t& mainRepeatTimes, uint64_t& tailRepeatTimes,
72 : std::vector<uint64_t>& mainMaskArray, std::vector<uint64_t>& tailMaskArray);
73 :
74 : uint64_t CalculateVectorMaxOffset(
75 : const uint64_t repeatTimes, const uint64_t blkStride, const uint64_t repStride, const uint64_t maskLen,
76 : const uint64_t blockLen);
77 :
78 : class TikcppBaseCheck {
79 : public:
80 608 : explicit TikcppBaseCheck(const std::string& name) : apiName(name) {}
81 608 : virtual ~TikcppBaseCheck() {}
82 :
83 : /*
84 : * @funcname: CheckTensorScope
85 : * @brief: check whether the tensor hardware position equal to the expect position.
86 : * @params: tensorPos, tensor physical position
87 : * expectedPos, expected tensor physical position
88 : * tensorInfo, src0 / src1/ dst for print information
89 : * posInfo, expected position info for print information
90 : * @return: true/false
91 : */
92 : bool CheckTensorScope(
93 : const uint8_t logicPos, const uint8_t expectedPos, const std::string& tensorInfo,
94 : const std::string& posInfo) const;
95 :
96 : /*
97 : * @funcname: CheckBufferSizeOverFlow
98 : * @brief: check whether the tensor allocate size equal to the buffer limited size.
99 : * @return: true/false
100 : */
101 : bool CheckBufferSizeOverFlow(const uint64_t localSize, const uint64_t bufferSize, const std::string& errMsg) const;
102 :
103 : /*
104 : * @funcname: CheckMaskArray
105 : * @brief: check the mask in bits mode
106 : * @return: true/false
107 : */
108 : bool CheckMaskArray(std::vector<uint64_t> maskArray) const;
109 :
110 : /*
111 : * @funcname: CheckMaskImm
112 : * @brief: check the mask in continuous mode
113 : * @return: true/false
114 : */
115 : bool CheckMaskImm(const uint64_t mask) const;
116 :
117 : /*
118 : * @funcname: CheckTensorOverflowLow
119 : * @brief: check whether tensor used size over the allocated size in low api level
120 : * @return: true/false
121 : */
122 : bool CheckTensorOverflowLow(
123 : std::vector<uint64_t>& maskArray, const TensorOverflowParams& params, const std::string& tensorName) const;
124 :
125 : /*
126 : * @funcname: CheckTensorOverflowLowGatherMask
127 : * @brief: check whether tensor used size over the allocated size in low api level for GatherMask
128 : * @return: true/false
129 : */
130 : bool CheckTensorOverflowLowGathermask(
131 : std::vector<uint64_t>& maskArray, const TensorOverflowParams& params, const std::string& tensorName) const;
132 :
133 : /*
134 : * @funcname: CheckTensorOverflowLowBrcb
135 : * @brief: check whether tensor used size over the allocated size in low api level for Brcb
136 : * Note that there is no counter mode for brcb + mask is not used
137 : * @return: true/false
138 : */
139 : bool CheckTensorOverflowLowBrcb(const TensorOverflowParams& params, const std::string& tensorName) const;
140 :
141 : /*
142 : * @funcname: CheckTensorOverflowHigh
143 : * @brief: check whether tensor used size over the allocated size in high api level
144 : * @return: true/false
145 : */
146 : bool CheckTensorOverflowHigh(
147 : const uint32_t dtypeSize, const uint64_t bufferSize, const uint32_t calCount,
148 : const std::string& tensorName) const;
149 :
150 : /*
151 : * @funcname: UpdateMaskArrayAndCheck
152 : * @brief: If isSetMask = false, replace maskArray with maskHigh and maskLow value in registers.
153 : * Check the latest maskArray value is valid with given dtype
154 : * @params: maskArray, mask value given by user. Can be len 1 or len 2
155 : * maxByteLen, among all dtypes given by function, the largest value of sizeof(dtype)
156 : * @return: true/false
157 : */
158 : bool UpdateMaskArrayAndCheck(std::vector<uint64_t>& maskArray, const uint32_t maxByteLen) const;
159 :
160 : /*
161 : * @funcname: CheckTensorAddrAlign
162 : * @brief: Check tensor start address is aligned with alignBytes
163 : * @params: tensorAddr, tensor address
164 : * phyPos, tensor physical position, used to calculate real offset of tensor
165 : * alignBytes, 32B aligned / 512B aligned etc
166 : * tensorInfo, src0 / src1/ dst for print information
167 : * @return: true/false
168 : */
169 : bool CheckTensorAddrAlign(
170 : const uint64_t tensorAddr, const uint8_t phyPos, const uint64_t alignBytes,
171 : const std::string& tensorInfo) const;
172 :
173 : protected:
174 : std::string apiName = "";
175 : };
176 :
177 : } // namespace check
178 : } // namespace AscendC
179 : #endif
|
