已合并
docs: 为aclnnPrecisionCompare、aclnnProd、aclnnProdDim添加计算公式，新增aclnnBitwiseAndTensorOut、aclnnInplaceBitwiseAndTensorOut文档 #1834
zhangqijia1创建于 3月23日
docs: 为aclnnPrecisionCompare、aclnnProd、aclnnProdDim添加计算公式，新增aclnnBitwiseAndTensorOut、aclnnInplaceBitwiseAndTensorOut文档 #1834
已合并
共 5 个文件变更+549-3
| @@ -53,6 +53,7 @@ ${INSTALL_DIR}表示CANN安装后文件路径；\$\{ops\_project\}表示算子 | |||
| 53 | | [aclnnBitwiseNot](../../math/bitwise_not/docs/aclnnBitwiseNot.md) | 输入为BOOL型tensor时，进行逻辑非运算；输入为INT型时进行按位非运算。 | 默认确定性实现| | 53 | | [aclnnBitwiseNot](../../math/bitwise_not/docs/aclnnBitwiseNot.md) | 输入为BOOL型tensor时，进行逻辑非运算；输入为INT型时进行按位非运算。 | 默认确定性实现| |
| 54 | | [aclnnBitwiseAndScalar](../../math/bitwise_and/docs/aclnnBitwiseAndScalar.md) | 计算输入tensor中每个元素和输入标量的按位与结果。 | 默认确定性实现| | 54 | | [aclnnBitwiseAndScalar](../../math/bitwise_and/docs/aclnnBitwiseAndScalar.md) | 计算输入tensor中每个元素和输入标量的按位与结果。 | 默认确定性实现| |
| 55 | | [aclnnBitwiseAndTensor](../../math/bitwise_and/docs/aclnnBitwiseAndTensor.md) | 输入为BOOL型tensor时，进行逻辑与运算；输入为INT型时，进行位与运算。 | 默认确定性实现| | 55 | | [aclnnBitwiseAndTensor](../../math/bitwise_and/docs/aclnnBitwiseAndTensor.md) | 输入为BOOL型tensor时，进行逻辑与运算；输入为INT型时，进行位与运算。 | 默认确定性实现| |
| 56 | +| [aclnnBitwiseAndTensorOut&aclnnInplaceBitwiseAndTensorOut](../../math/bitwise_and/docs/aclnnBitwiseAndTensorOut&aclnnInplaceBitwiseAndTensorOut.md) | 输入为BOOL型tensor时，进行逻辑与运算；输入为INT型时，进行位与运算。 | 默认确定性实现| | ||
| 56 | | [aclnnBitwiseOrScalar&aclnnInplaceBitwiseOrScalar](../../math/bitwise_or/docs/aclnnBitwiseOrScalar&aclnnInplaceBitwiseOrScalar.md) | 计算输入张量self中每个元素和输入标量other的按位或。 | 默认确定性实现| | 57 | | [aclnnBitwiseOrScalar&aclnnInplaceBitwiseOrScalar](../../math/bitwise_or/docs/aclnnBitwiseOrScalar&aclnnInplaceBitwiseOrScalar.md) | 计算输入张量self中每个元素和输入标量other的按位或。 | 默认确定性实现| |
| 57 | | [aclnnBitwiseOrTensor&aclnnInplaceBitwiseOrTensor](../../math/bitwise_or/docs/aclnnBitwiseOrTensor&aclnnInplaceBitwiseOrTensor.md) | 计算张量self中每个元素与other张量中对应位置的元素的按位或。 | 默认确定性实现| | 58 | | [aclnnBitwiseOrTensor&aclnnInplaceBitwiseOrTensor](../../math/bitwise_or/docs/aclnnBitwiseOrTensor&aclnnInplaceBitwiseOrTensor.md) | 计算张量self中每个元素与other张量中对应位置的元素的按位或。 | 默认确定性实现| |
| 58 | | [aclnnBitwiseXorScalar&aclnnInplaceBitwiseXorScalar](../../math/bitwise_xor/docs/aclnnBitwiseXorScalar&aclnnInplaceBitwiseXorScalar.md) | 计算输入张量self中每个元素和输入标量other的按位异或，输入self和other必须是整数或布尔类型，对于布尔类型，计算逻辑异或。 | 默认确定性实现| | 59 | | [aclnnBitwiseXorScalar&aclnnInplaceBitwiseXorScalar](../../math/bitwise_xor/docs/aclnnBitwiseXorScalar&aclnnInplaceBitwiseXorScalar.md) | 计算输入张量self中每个元素和输入标量other的按位异或，输入self和other必须是整数或布尔类型，对于布尔类型，计算逻辑异或。 | 默认确定性实现| |
| @@ -0,0 +1,521 @@ | |||
| 1 | +# aclnnBitwiseAndTensorOut&aclnnInplaceBitwiseAndTensorOut | ||
S | |||
| 2 | + | ||
| 3 | +## 产品支持情况 | ||
| 4 | + | ||
| 5 | +|产品 | 是否支持 | | ||
| 6 | +|:-------------------------|:----------:| | ||
| 7 | +| <term>Ascend 950PR/Ascend 950DT</term> | √ | | ||
| 8 | +| <term>Atlas A3 训练系列产品/Atlas A3 推理系列产品</term> | √ | | ||
| 9 | +| <term>Atlas A2 训练系列产品/Atlas A2 推理系列产品</term> | √ | | ||
| 10 | +| <term>Atlas 200I/500 A2 推理产品</term> | × | | ||
| 11 | +| <term>Atlas 推理系列产品</term> | × | | ||
| 12 | +| <term>Atlas 训练系列产品</term> | √ | | ||
| 13 | + | ||
| 14 | +## 功能说明 | ||
| 15 | + | ||
| 16 | +- 接口功能：输入为BOOL型tensor时，进行逻辑与运算；输入为INT型时，进行位与运算。 | ||
| 17 | + | ||
| 18 | +- 计算公式： | ||
| 19 | + | ||
| 20 | + $$ | ||
| 21 | + out_i = self_i \& other_i | ||
| 22 | + $$ | ||
| 23 | + | ||
| 24 | +## 函数原型 | ||
| 25 | + | ||
| 26 | +- aclnnBitwiseAndTensorOut和aclnnInplaceBitwiseAndTensorOut实现相同的功能，使用区别如下，请根据自身实际场景选择合适的算子。 | ||
| 27 | + - aclnnBitwiseAndTensorOut：需新建一个输出张量对象存储计算结果。 | ||
| 28 | + - aclnnInplaceBitwiseAndTensorOut：无需新建输出张量对象，直接在输入张量的内存中存储计算结果。 | ||
| 29 | +- 每个算子分为[两段式接口](../../../docs/zh/context/两段式接口.md)，必须先调用"aclnnBitwiseAndTensorOutGetWorkspaceSize"接口获取计算所需workspace大小以及包含了算子计算流程的执行器，再调用"aclnnBitwiseAndTensorOut"接口执行计算。 | ||
S 问题: 「两段式接口」说明与「函数原型」仅覆盖了 建议: 参照同仓库   | |||
| 30 | + | ||
| 31 | +```Cpp | ||
| 32 | +aclnnStatus aclnnBitwiseAndTensorOutGetWorkspaceSize( | ||
| 33 | + const aclTensor* self, | ||
| 34 | + const aclTensor* other, | ||
| 35 | + aclTensor* out, | ||
| 36 | + uint64_t* workspaceSize, | ||
| 37 | + aclOpExecutor** executor) | ||
| 38 | +``` | ||
| 39 | + | ||
| 40 | +```Cpp | ||
| 41 | +aclnnStatus aclnnBitwiseAndTensorOut( | ||
| 42 | + void* workspace, | ||
| 43 | + uint64_t workspaceSize, | ||
| 44 | + aclOpExecutor* executor, | ||
| 45 | + const aclrtStream stream) | ||
| 46 | +``` | ||
| 47 | + | ||
| 48 | +## aclnnBitwiseAndTensorOutGetWorkspaceSize | ||
| 49 | + | ||
| 50 | +- **参数说明：** | ||
| 51 | + | ||
| 52 | + <table style="undefined;table-layout: fixed; width: 1523px"><colgroup> | ||
| 53 | + <col style="width: 146px"> | ||
| 54 | + <col style="width: 120px"> | ||
| 55 | + <col style="width: 301px"> | ||
| 56 | + <col style="width: 219px"> | ||
| 57 | + <col style="width: 328px"> | ||
| 58 | + <col style="width: 120px"> | ||
| 59 | + <col style="width: 143px"> | ||
| 60 | + <col style="width: 146px"> | ||
| 61 | + </colgroup> | ||
| 62 | + <thead> | ||
| 63 | + <tr> | ||
| 64 | + <th>参数名</th> | ||
| 65 | + <th>输入/输出</th> | ||
| 66 | + <th>描述</th> | ||
| 67 | + <th>使用说明</th> | ||
| 68 | + <th>数据类型</th> | ||
| 69 | + <th>数据格式</th> | ||
| 70 | + <th>维度(shape)</th> | ||
| 71 | + <th>非连续Tensor</th> | ||
| 72 | + </tr></thead> | ||
| 73 | + <tbody> | ||
| 74 | + <tr> | ||
| 75 | + <td>self（aclTensor*）</td> | ||
| 76 | + <td>输入</td> | ||
| 77 | + <td>输入tensor，公式中的self。</td> | ||
| 78 | + <td>数据类型需要与other的数据类型满足<a href="../../../docs/zh/context/互推导关系.md" target="_blank">数据类型推导规则</a>。<br>shape需要与other满足<a href="../../../docs/zh/context/broadcast关系.md" target="_blank">broadcast关系</a>。</td> | ||
| 79 | + <td>INT16、UINT16、INT32、INT64、INT8、UINT8、BOOL</td> | ||
| 80 | + <td>ND</td> | ||
| 81 | + <td>0-8</td> | ||
| 82 | + <td>√</td> | ||
| 83 | + </tr> | ||
| 84 | + <tr> | ||
| 85 | + <td>other（aclTensor*）</td> | ||
| 86 | + <td>输入</td> | ||
| 87 | + <td>输入tensor，公式中的other。</td> | ||
| 88 | + <td>数据类型需要与self的数据类型满足<a href="../../../docs/zh/context/互推导关系.md" target="_blank">数据类型推导规则</a>。<br>shape需要与self满足<a href="../../../docs/zh/context/broadcast关系.md" target="_blank">broadcast关系</a>。</td> | ||
| 89 | + <td>INT16、UINT16、INT32、INT64、INT8、UINT8、BOOL</td> | ||
| 90 | + <td>ND</td> | ||
| 91 | + <td>0-8</td> | ||
| 92 | + <td>√</td> | ||
| 93 | + </tr> | ||
| 94 | + <tr> | ||
| 95 | + <td>out（aclTensor*）</td> | ||
| 96 | + <td>输出</td> | ||
| 97 | + <td>输出tensor，公式中的out。</td> | ||
| 98 | + <td>数据类型需要是self与other推导之后可转换的数据类型。<br>shape需要是self与other满足<a href="../../../docs/zh/context/broadcast关系.md" target="_blank">broadcast关系</a>之后的shape。</td> | ||
| 99 | + <td>BOOL、INT8、INT16、INT32、INT64、UINT8、UINT16、UINT32、UINT64</td> | ||
| 100 | + <td>ND</td> | ||
| 101 | + <td>0-8</td> | ||
| 102 | + <td>√</td> | ||
| 103 | + </tr> | ||
| 104 | + <tr> | ||
| 105 | + <td>workspaceSize（uint64_t*）</td> | ||
| 106 | + <td>输出</td> | ||
| 107 | + <td>返回需要在Device侧申请的workspace大小。</td> | ||
| 108 | + <td>-</td> | ||
| 109 | + <td>-</td> | ||
| 110 | + <td>-</td> | ||
| 111 | + <td>-</td> | ||
| 112 | + <td>-</td> | ||
| 113 | + </tr> | ||
| 114 | + <tr> | ||
| 115 | + <td>executor（aclOpExecutor**）</td> | ||
| 116 | + <td>输出</td> | ||
| 117 | + <td>返回op执行器，包含了算子计算流程。</td> | ||
| 118 | + <td>-</td> | ||
| 119 | + <td>-</td> | ||
| 120 | + <td>-</td> | ||
| 121 | + <td>-</td> | ||
| 122 | + <td>-</td> | ||
| 123 | + </tr> | ||
| 124 | + </tbody></table> | ||
| 125 | + | ||
| 126 | + | ||
| 127 | +- **返回值：** | ||
| 128 | + | ||
| 129 | + aclnnStatus：返回状态码，具体参见[aclnn返回码](../../../docs/zh/context/aclnn返回码.md)。 | ||
| 130 | + | ||
| 131 | + 第一段接口完成入参校验，出现以下场景时报错： | ||
| 132 | + | ||
| 133 | + <table style="undefined;table-layout: fixed; width: 1150px"><colgroup> | ||
| 134 | + <col style="width: 300px"> | ||
| 135 | + <col style="width: 134px"> | ||
| 136 | + <col style="width: 716px"> | ||
| 137 | + </colgroup> | ||
| 138 | + <thead> | ||
| 139 | + <tr> | ||
| 140 | + <th>返回码</th> | ||
| 141 | + <th>错误码</th> | ||
| 142 | + <th>描述</th> | ||
| 143 | + </tr></thead> | ||
| 144 | + <tbody> | ||
| 145 | + <tr> | ||
| 146 | + <td>ACLNN_ERR_PARAM_NULLPTR</td> | ||
| 147 | + <td>161001</td> | ||
| 148 | + <td>传入的self、other或out是空指针。</td> | ||
| 149 | + </tr> | ||
| 150 | + <tr> | ||
| 151 | + <td rowspan="5">ACLNN_ERR_PARAM_INVALID</td> | ||
| 152 | + <td rowspan="5">161002</td> | ||
| 153 | + <td>self和other的数据类型不在支持的范围之内。</td> | ||
| 154 | + </tr> | ||
| 155 | + <tr> | ||
| 156 | + <td>self和other不满足数据类型推导规则。</td> | ||
| 157 | + </tr> | ||
| 158 | + <tr> | ||
| 159 | + <td>self和other推导出的数据类型无法转换为指定输出out的类型。</td> | ||
| 160 | + </tr> | ||
| 161 | + <tr> | ||
| 162 | + <td>self和other的shape无法做broadcast。</td> | ||
| 163 | + </tr> | ||
| 164 | + <tr> | ||
| 165 | + <td>self、other、out的维度超过8维。</td> | ||
| 166 | + </tr> | ||
| 167 | + </tbody></table> | ||
| 168 | + | ||
| 169 | +## aclnnBitwiseAndTensorOut | ||
| 170 | + | ||
| 171 | +- **参数说明：** | ||
| 172 | + | ||
| 173 | + <table style="undefined;table-layout: fixed; width: 953px"><colgroup> | ||
| 174 | + <col style="width: 173px"> | ||
| 175 | + <col style="width: 112px"> | ||
| 176 | + <col style="width: 668px"> | ||
| 177 | + </colgroup> | ||
| 178 | + <thead> | ||
| 179 | + <tr> | ||
| 180 | + <th>参数名</th> | ||
| 181 | + <th>输入/输出</th> | ||
| 182 | + <th>描述</th> | ||
| 183 | + </tr></thead> | ||
| 184 | + <tbody> | ||
| 185 | + <tr> | ||
| 186 | + <td>workspace</td> | ||
| 187 | + <td>输入</td> | ||
| 188 | + <td>在Device侧申请的workspace内存地址。</td> | ||
| 189 | + </tr> | ||
| 190 | + <tr> | ||
| 191 | + <td>workspaceSize</td> | ||
| 192 | + <td>输入</td> | ||
| 193 | + <td>在Device侧申请的workspace大小，由第一段接口aclnnBitwiseAndTensorOutGetWorkspaceSize获取。</td> | ||
| 194 | + </tr> | ||
| 195 | + <tr> | ||
| 196 | + <td>executor</td> | ||
| 197 | + <td>输入</td> | ||
| 198 | + <td>op执行器，包含了算子计算流程。</td> | ||
| 199 | + </tr> | ||
| 200 | + <tr> | ||
| 201 | + <td>stream</td> | ||
| 202 | + <td>输入</td> | ||
| 203 | + <td>指定执行任务的Stream。</td> | ||
| 204 | + </tr> | ||
| 205 | + </tbody> | ||
| 206 | + </table> | ||
| 207 | + | ||
| 208 | + | ||
| 209 | +- **返回值：** | ||
| 210 | + | ||
| 211 | + aclnnStatus：返回状态码，具体参见[aclnn返回码](../../../docs/zh/context/aclnn返回码.md)。 | ||
| 212 | + | ||
| 213 | +## aclnnInplaceBitwiseAndTensorOutGetWorkspaceSize | ||
| 214 | + | ||
| 215 | +- **参数说明：** | ||
| 216 | + | ||
| 217 | + <table style="undefined;table-layout: fixed; width: 1523px"><colgroup> | ||
| 218 | + <col style="width: 146px"> | ||
| 219 | + <col style="width: 120px"> | ||
| 220 | + <col style="width: 301px"> | ||
| 221 | + <col style="width: 219px"> | ||
| 222 | + <col style="width: 328px"> | ||
| 223 | + <col style="width: 120px"> | ||
| 224 | + <col style="width: 143px"> | ||
| 225 | + <col style="width: 146px"> | ||
| 226 | + </colgroup> | ||
| 227 | + <thead> | ||
| 228 | + <tr> | ||
| 229 | + <th>参数名</th> | ||
| 230 | + <th>输入/输出</th> | ||
| 231 | + <th>描述</th> | ||
| 232 | + <th>使用说明</th> | ||
| 233 | + <th>数据类型</th> | ||
| 234 | + <th>数据格式</th> | ||
| 235 | + <th>维度(shape)</th> | ||
| 236 | + <th>非连续Tensor</th> | ||
| 237 | + </tr></thead> | ||
| 238 | + <tbody> | ||
| 239 | + <tr> | ||
| 240 | + <td>selfRef（aclTensor*）</td> | ||
| 241 | + <td>输入/输出</td> | ||
| 242 | + <td>输入和输出tensor，公式中的selfRef。</td> | ||
| 243 | + <td>数据类型需要与other的数据类型满足<a href="../../../docs/zh/context/互推导关系.md" target="_blank">数据类型推导规则</a>。<br>shape需要与other满足<a href="../../../docs/zh/context/broadcast关系.md" target="_blank">broadcast关系</a>。</td> | ||
| 244 | + <td>INT16、UINT16、INT32、INT64、INT8、UINT8、BOOL</td> | ||
| 245 | + <td>ND</td> | ||
| 246 | + <td>0-8</td> | ||
| 247 | + <td>√</td> | ||
| 248 | + </tr> | ||
| 249 | + <tr> | ||
| 250 | + <td>other（aclTensor*）</td> | ||
| 251 | + <td>输入</td> | ||
| 252 | + <td>输入tensor，公式中的other。</td> | ||
| 253 | + <td>数据类型需要与selfRef的数据类型满足<a href="../../../docs/zh/context/互推导关系.md" target="_blank">数据类型推导规则</a>。<br>shape需要与selfRef满足<a href="../../../docs/zh/context/broadcast关系.md" target="_blank">broadcast关系</a>。</td> | ||
| 254 | + <td>INT16、UINT16、INT32、INT64、INT8、UINT8、BOOL</td> | ||
| 255 | + <td>ND</td> | ||
| 256 | + <td>0-8</td> | ||
| 257 | + <td>√</td> | ||
| 258 | + </tr> | ||
| 259 | + <tr> | ||
| 260 | + <td>workspaceSize（uint64_t*）</td> | ||
| 261 | + <td>输出</td> | ||
| 262 | + <td>返回需要在Device侧申请的workspace大小。</td> | ||
| 263 | + <td>-</td> | ||
| 264 | + <td>-</td> | ||
| 265 | + <td>-</td> | ||
| 266 | + <td>-</td> | ||
| 267 | + <td>-</td> | ||
| 268 | + </tr> | ||
| 269 | + <tr> | ||
| 270 | + <td>executor（aclOpExecutor**）</td> | ||
| 271 | + <td>输出</td> | ||
| 272 | + <td>返回op执行器，包含了算子计算流程。</td> | ||
| 273 | + <td>-</td> | ||
| 274 | + <td>-</td> | ||
| 275 | + <td>-</td> | ||
| 276 | + <td>-</td> | ||
| 277 | + <td>-</td> | ||
| 278 | + </tr> | ||
| 279 | + </tbody></table> | ||
| 280 | + | ||
| 281 | + | ||
| 282 | +- **返回值：** | ||
| 283 | + | ||
| 284 | + aclnnStatus：返回状态码，具体参见[aclnn返回码](../../../docs/zh/context/aclnn返回码.md)。 | ||
| 285 | + | ||
| 286 | + 第一段接口完成入参校验，出现以下场景时报错： | ||
| 287 | + | ||
| 288 | + <table style="undefined;table-layout: fixed; width: 1150px"><colgroup> | ||
| 289 | + <col style="width: 300px"> | ||
| 290 | + <col style="width: 134px"> | ||
| 291 | + <col style="width: 716px"> | ||
| 292 | + </colgroup> | ||
| 293 | + <thead> | ||
| 294 | + <tr> | ||
| 295 | + <th>返回码</th> | ||
| 296 | + <th>错误码</th> | ||
| 297 | + <th>描述</th> | ||
| 298 | + </tr></thead> | ||
| 299 | + <tbody> | ||
| 300 | + <tr> | ||
| 301 | + <td>ACLNN_ERR_PARAM_NULLPTR</td> | ||
| 302 | + <td>161001</td> | ||
| 303 | + <td>传入的selfRef或other是空指针。</td> | ||
| 304 | + </tr> | ||
| 305 | + <tr> | ||
| 306 | + <td rowspan="5">ACLNN_ERR_PARAM_INVALID</td> | ||
| 307 | + <td rowspan="5">161002</td> | ||
| 308 | + <td>selfRef和other的数据类型不在支持的范围之内。</td> | ||
| 309 | + </tr> | ||
| 310 | + <tr> | ||
| 311 | + <td>selfRef和other不满足数据类型推导规则。</td> | ||
| 312 | + </tr> | ||
| 313 | + <tr> | ||
| 314 | + <td>selfRef和other推导出的数据类型无法转换为指定输出selfRef的类型。</td> | ||
| 315 | + </tr> | ||
| 316 | + <tr> | ||
| 317 | + <td>selfRef和other的shape无法做broadcast。</td> | ||
| 318 | + </tr> | ||
| 319 | + <tr> | ||
| 320 | + <td>selfRef、other的维度超过8维。</td> | ||
| 321 | + </tr> | ||
| 322 | + </tbody></table> | ||
| 323 | + | ||
| 324 | +## aclnnInplaceBitwiseAndTensorOut | ||
| 325 | + | ||
| 326 | +- **参数说明：** | ||
| 327 | + | ||
| 328 | + <table style="undefined;table-layout: fixed; width: 1151px"><colgroup> | ||
| 329 | + <col style="width: 184px"> | ||
| 330 | + <col style="width: 134px"> | ||
| 331 | + <col style="width: 833px"> | ||
| 332 | + </colgroup> | ||
| 333 | + <thead> | ||
| 334 | + <tr> | ||
| 335 | + <th>参数名</th> | ||
| 336 | + <th>输入/输出</th> | ||
| 337 | + <th>描述</th> | ||
| 338 | + </tr></thead> | ||
| 339 | + <tbody> | ||
| 340 | + <tr> | ||
| 341 | + <td>workspace</td> | ||
| 342 | + <td>输入</td> | ||
| 343 | + <td>在Device侧申请的workspace内存地址。</td> | ||
| 344 | + </tr> | ||
| 345 | + <tr> | ||
| 346 | + <td>workspaceSize</td> | ||
| 347 | + <td>输入</td> | ||
| 348 | + <td>在Device侧申请的workspace大小，由第一段接口aclnnInplaceBitwiseAndTensorOutGetWorkspaceSize获取。</td> | ||
| 349 | + </tr> | ||
| 350 | + <tr> | ||
| 351 | + <td>executor</td> | ||
| 352 | + <td>输入</td> | ||
| 353 | + <td>op执行器，包含了算子计算流程。</td> | ||
| 354 | + </tr> | ||
| 355 | + <tr> | ||
| 356 | + <td>stream</td> | ||
| 357 | + <td>输入</td> | ||
| 358 | + <td>指定执行任务的Stream。</td> | ||
| 359 | + </tr> | ||
| 360 | + </tbody> | ||
| 361 | + </table> | ||
| 362 | + | ||
| 363 | + | ||
| 364 | +- **返回值：** | ||
| 365 | + | ||
| 366 | + aclnnStatus：返回状态码，具体参见[aclnn返回码](../../../docs/zh/context/aclnn返回码.md)。 | ||
| 367 | + | ||
| 368 | +## 约束说明 | ||
| 369 | + | ||
| 370 | +- 确定性计算： | ||
| 371 | + - aclnnBitwiseAndTensorOut默认确定性实现。 | ||
| 372 | + - aclnnInplaceBitwiseAndTensorOut默认确定性实现。 | ||
| 373 | + | ||
| 374 | +## 调用示例 | ||
| 375 | + | ||
| 376 | +示例代码如下，仅供参考，具体编译和执行过程请参考[编译与运行样例](../../../docs/zh/context/编译与运行样例.md)。 | ||
| 377 | + | ||
| 378 | +```Cpp | ||
| 379 | +#include <iostream> | ||
| 380 | +#include <vector> | ||
| 381 | +#include "acl/acl.h" | ||
| 382 | +#include "aclnnop/aclnn_bitwise_and_tensor.h" | ||
S 问题: 调用示例中的头文件与 建议: | |||
| 383 | + | ||
| 384 | +#define CHECK_RET(cond, return_expr) \ | ||
| 385 | + do { \ | ||
| 386 | + if (!(cond)) { \ | ||
| 387 | + return_expr; \ | ||
| 388 | + } \ | ||
| 389 | + } while (0) | ||
| 390 | + | ||
| 391 | +#define LOG_PRINT(message, ...) \ | ||
| 392 | + do { \ | ||
| 393 | + printf(message, ##__VA_ARGS__); \ | ||
| 394 | + } while (0) | ||
| 395 | + | ||
| 396 | +int64_t GetShapeSize(const std::vector<int64_t>& shape) { | ||
| 397 | + int64_t shape_size = 1; | ||
| 398 | + for (auto i : shape) { | ||
| 399 | + shape_size *= i; | ||
| 400 | + } | ||
| 401 | + return shape_size; | ||
| 402 | +} | ||
| 403 | + | ||
| 404 | +int Init(int32_t deviceId, aclrtStream* stream) { | ||
| 405 | + // 固定写法，资源初始化 | ||
| 406 | + auto ret = aclInit(nullptr); | ||
| 407 | + CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclInit failed. ERROR: %d\n", ret); return ret); | ||
| 408 | + ret = aclrtSetDevice(deviceId); | ||
| 409 | + CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtSetDevice failed. ERROR: %d\n", ret); return ret); | ||
| 410 | + ret = aclrtCreateStream(stream); | ||
| 411 | + CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtCreateStream failed. ERROR: %d\n", ret); return ret); | ||
| 412 | + | ||
| 413 | + return 0; | ||
| 414 | +} | ||
| 415 | + | ||
| 416 | +template <typename T> | ||
| 417 | +int CreateAclTensor(const std::vector<T>& hostData, const std::vector<int64_t>& shape, void** deviceAddr, | ||
| 418 | + aclDataType dataType, aclTensor** tensor) { | ||
| 419 | + auto size = GetShapeSize(shape) * sizeof(T); | ||
| 420 | + // 调用aclrtMalloc申请device侧内存 | ||
| 421 | + auto ret = aclrtMalloc(deviceAddr, size, ACL_MEM_MALLOC_HUGE_FIRST); | ||
| 422 | + CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtMalloc failed. ERROR: %d\n", ret); return ret); | ||
| 423 | + | ||
| 424 | + // 调用aclrtMemcpy将host侧数据拷贝到device侧内存上 | ||
| 425 | + ret = aclrtMemcpy(*deviceAddr, size, hostData.data(), size, ACL_MEMCPY_HOST_TO_DEVICE); | ||
| 426 | + CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtMemcpy failed. ERROR: %d\n", ret); return ret); | ||
| 427 | + | ||
| 428 | + // 计算连续tensor的strides | ||
| 429 | + std::vector<int64_t> strides(shape.size(), 1); | ||
| 430 | + for (int64_t i = shape.size() - 2; i >= 0; i--) { | ||
| 431 | + strides[i] = shape[i + 1] * strides[i + 1]; | ||
| 432 | + } | ||
| 433 | + | ||
| 434 | + // 调用aclCreateTensor接口创建aclTensor | ||
| 435 | + *tensor = aclCreateTensor(shape.data(), shape.size(), dataType, strides.data(), 0, aclFormat::ACL_FORMAT_ND, | ||
| 436 | + shape.data(), shape.size(), *deviceAddr); | ||
| 437 | + return 0; | ||
| 438 | +} | ||
| 439 | + | ||
| 440 | +int main() { | ||
| 441 | + // 1. （固定写法）device/stream初始化, 参考acl对外接口列表 | ||
| 442 | + // 根据自己的实际device填写deviceId | ||
| 443 | + int32_t deviceId = 0; | ||
| 444 | + aclrtStream stream; | ||
| 445 | + auto ret = Init(deviceId, &stream); | ||
| 446 | + // check根据自己的需要处理 | ||
| 447 | + CHECK_RET(ret == 0, LOG_PRINT("Init acl failed. ERROR: %d\n", ret); return ret); | ||
| 448 | + | ||
| 449 | + // 2. 构造输入与输出，需要根据API的接口自定义构造 | ||
| 450 | + std::vector<int64_t> selfShape = {4, 2}; | ||
| 451 | + std::vector<int64_t> otherShape = {4, 2}; | ||
| 452 | + std::vector<int64_t> outShape = {4, 2}; | ||
| 453 | + void* selfDeviceAddr = nullptr; | ||
| 454 | + void* otherDeviceAddr = nullptr; | ||
| 455 | + void* outDeviceAddr = nullptr; | ||
| 456 | + aclTensor* self = nullptr; | ||
| 457 | + aclTensor* other = nullptr; | ||
| 458 | + aclTensor* out = nullptr; | ||
| 459 | + std::vector<int64_t> selfHostData = {0, 1, 2, 3, 4, 5, 6, 7}; | ||
| 460 | + std::vector<int64_t> otherHostData = {1, 1, 2, 3, 3, 3, 4, 4}; | ||
| 461 | + std::vector<int64_t> outHostData = {0, 0, 0, 0, 0, 0, 0, 0}; | ||
| 462 | + | ||
| 463 | + // 创建self aclTensor | ||
| 464 | + ret = CreateAclTensor(selfHostData, selfShape, &selfDeviceAddr, aclDataType::ACL_INT64, &self); | ||
| 465 | + CHECK_RET(ret == ACL_SUCCESS, return ret); | ||
| 466 | + // 创建other aclTensor | ||
| 467 | + ret = CreateAclTensor(otherHostData, otherShape, &otherDeviceAddr, aclDataType::ACL_INT64, &other); | ||
| 468 | + CHECK_RET(ret == ACL_SUCCESS, return ret); | ||
| 469 | + // 创建out aclTensor | ||
| 470 | + ret = CreateAclTensor(outHostData, outShape, &outDeviceAddr, aclDataType::ACL_INT64, &out); | ||
| 471 | + CHECK_RET(ret == ACL_SUCCESS, return ret); | ||
| 472 | + | ||
| 473 | + // 3. 调用CANN算子库API，需要修改为具体的API | ||
| 474 | + uint64_t workspaceSize = 0; | ||
| 475 | + aclOpExecutor* executor; | ||
| 476 | + // 调用aclnnBitwiseAndTensorOut第一段接口 | ||
| 477 | + ret = aclnnBitwiseAndTensorOutGetWorkspaceSize(self, other, out, &workspaceSize, &executor); | ||
| 478 | + CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnBitwiseAndTensorOutGetWorkspaceSize failed. ERROR: %d\n", ret); return ret); | ||
| 479 | + // 根据第一段接口计算出的workspaceSize申请device内存 | ||
| 480 | + void* workspaceAddr = nullptr; | ||
| 481 | + if (workspaceSize > 0) { | ||
| 482 | + ret = aclrtMalloc(&workspaceAddr, workspaceSize, ACL_MEM_MALLOC_HUGE_FIRST); | ||
| 483 | + CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("allocate workspace failed. ERROR: %d\n", ret); return ret;); | ||
| 484 | + } | ||
| 485 | + // 调用aclnnBitwiseAndTensorOut第二段接口 | ||
| 486 | + ret = aclnnBitwiseAndTensorOut(workspaceAddr, workspaceSize, executor, stream); | ||
| 487 | + CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnBitwiseAndTensorOut failed. ERROR: %d\n", ret); return ret); | ||
| 488 | + | ||
| 489 | + // 4. （固定写法）同步等待任务执行结束 | ||
| 490 | + ret = aclrtSynchronizeStream(stream); | ||
| 491 | + CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtSynchronizeStream failed. ERROR: %d\n", ret); return ret); | ||
| 492 | + | ||
| 493 | + // 5. 获取输出的值，将device侧内存上的结果拷贝至host侧，需要根据具体API的接口定义修改 | ||
| 494 | + auto size = GetShapeSize(outShape); | ||
| 495 | + std::vector<int64_t> resultData(size, 0); | ||
| 496 | + ret = aclrtMemcpy(resultData.data(), resultData.size() * sizeof(resultData[0]), outDeviceAddr, size * sizeof(int64_t), | ||
| 497 | + ACL_MEMCPY_DEVICE_TO_HOST); | ||
| 498 | + CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("copy result from device to host failed. ERROR: %d\n", ret); return ret); | ||
| 499 | + for (int64_t i = 0; i < size; i++) { | ||
| 500 | + LOG_PRINT("result[%ld] is: %ld\n", i, resultData[i]); | ||
| 501 | + } | ||
| 502 | + | ||
| 503 | + // 6. 释放aclTensor | ||
| 504 | + aclDestroyTensor(self); | ||
| 505 | + aclDestroyTensor(other); | ||
| 506 | + aclDestroyTensor(out); | ||
| 507 | + | ||
| 508 | + // 7. 释放device资源 | ||
| 509 | + aclrtFree(selfDeviceAddr); | ||
| 510 | + aclrtFree(otherDeviceAddr); | ||
| 511 | + aclrtFree(outDeviceAddr); | ||
| 512 | + if (workspaceSize > 0) { | ||
| 513 | + aclrtFree(workspaceAddr); | ||
| 514 | + } | ||
| 515 | + aclrtDestroyStream(stream); | ||
| 516 | + aclrtResetDevice(deviceId); | ||
| 517 | + aclFinalize(); | ||
| 518 | + | ||
| 519 | + return 0; | ||
| 520 | +} | ||
| 521 | +``` | ||
| @@ -15,7 +15,15 @@ | |||
| 15 | 15 | ||
| 16 | ## 功能说明 | 16 | ## 功能说明 |
| 17 | 17 | ||
| 18 | -计算两个Tensor元素是否相同，进行精度比对，返回算子执行的状态码，不会触发管理面故障上报。 | 18 | +- 接口功能：计算两个Tensor元素是否相同，进行精度比对，返回算子执行的状态码，不会触发管理面故障上报。 |
| 19 | + | ||
| 20 | +- 计算公式： | ||
| 21 | + | ||
| 22 | + $$ | ||
| 23 | + golden_i = realdata_i, \quad i = 0, 1, \ldots, n-1 | ||
| 24 | + $$ | ||
| 25 | + | ||
| 26 | + 逐元素比对golden与realdata是否相同，out返回比对结果状态码（0表示通过，非零表示异常）。 | ||
| 19 | 27 | ||
| 20 | ## 函数原型 | 28 | ## 函数原型 |
| 21 | 29 | ||
| @@ -16,7 +16,15 @@ | |||
| 16 | 16 | ||
| 17 | ## 功能说明 | 17 | ## 功能说明 |
| 18 | 18 | ||
| 19 | -返回输入tensor中所有元素的乘积。 | 19 | +- 接口功能：返回输入tensor中所有元素的乘积。 |
| 20 | + | ||
| 21 | +- 计算公式： | ||
| 22 | + | ||
| 23 | + $$ | ||
| 24 | + out = \prod_{i=0}^{n-1} self_i | ||
| 25 | + $$ | ||
| 26 | + | ||
| 27 | + 其中$n$为self中元素的总数，$self_i$为self中第$i$个元素。 | ||
| 20 | 28 | ||
| 21 | ## 函数原型 | 29 | ## 函数原型 |
| 22 | 30 | ||
| @@ -16,7 +16,15 @@ | |||
| 16 | 16 | ||
| 17 | ## 功能说明 | 17 | ## 功能说明 |
| 18 | 18 | ||
| 19 | -返回输入tensor给定维度上每行的乘积。 | 19 | +- 接口功能：返回输入tensor给定维度上每行的乘积。 |
| 20 | + | ||
| 21 | +- 计算公式： | ||
| 22 | + | ||
| 23 | + $$ | ||
| 24 | + out = \prod_{i=0}^{S_{dim}-1} self_{[..., \, i, \, ...]} | ||
| 25 | + $$ | ||
| 26 | + | ||
| 27 | + 其中$S_{dim}$为self在维度$dim$上的大小，$dim$为指定的缩减维度，$keepDim$控制输出是否保留该维度。 | ||
| 20 | 28 | ||
| 21 | ## 函数原型 | 29 | ## 函数原型 |
| 22 | 30 | ||
问题: 与同目录其他 aclnn 文档相比，标题下缺少「查看源码」链接，版式不统一。
建议: 在标题行下增加一行：
[📄 查看源码](https://gitcode.com/cann/ops-math/tree/master/math/bitwise_and)（与同系列文档一致）。