已合并
【PR】: 新增Runtime API 设备全局变量symbol相关rt接口 #2467
huzeyong_rts创建于 5月28日
【PR】: 新增Runtime API 设备全局变量symbol相关rt接口 #2467
已合并
共 30 个文件变更+1077-16
| @@ -0,0 +1,278 @@ | |||
| 1 | # 11-11 Device变量内存操作 | ||
| 2 | |||
| 3 | 本章节描述Device变量内存操作相关接口,用户可以访问和操作Device变量。 | ||
| 4 | |||
| 5 | - [`aclError aclrtGetSymbolAddress(const void *symbol, void **devPtr)`](#aclrtGetSymbolAddress):获取Device变量的地址。 | ||
| 6 | - [`aclError aclrtGetSymbolSize(const void *symbol, size_t *size)`](#aclrtGetSymbolSize):获取Device变量占用的内存大小。 | ||
| 7 | - [`aclError aclrtMemcpyFromSymbol(void *dst, size_t dstMax, const void *symbol, size_t count, size_t offset, aclrtMemcpyKind kind)`](#aclrtMemcpyFromSymbol):实现Device变量的数据到Host的同步内存复制。 | ||
| 8 | - [`aclError aclrtMemcpyFromSymbolAsync(void *dst, size_t dstMax, const void *symbol, size_t count, size_t offset, aclrtMemcpyKind kind, aclrtStream stream)`](#aclrtMemcpyFromSymbolAsync):实现Device变量的数据到Host的异步内存复制。 | ||
| 9 | - [`aclError aclrtMemcpyToSymbol(const void *symbol, const void *src, size_t count, size_t offset, aclrtMemcpyKind kind)`](#aclrtMemcpyToSymbol):实现Host数据到Device变量的同步内存复制。 | ||
| 10 | - [`aclError aclrtMemcpyToSymbolAsync(const void *symbol, const void *src, size_t count, size_t offset, aclrtMemcpyKind kind, aclrtStream stream)`](#aclrtMemcpyToSymbolAsync):实现Host数据到Device变量的异步内存复制。 | ||
| 11 | |||
| 12 | <a id="aclrtGetSymbolAddress"></a> | ||
| 13 | |||
| 14 | ## aclrtGetSymbolAddress | ||
| 15 | |||
| 16 | ```c | ||
| 17 | aclError aclrtGetSymbolAddress(const void *symbol, void **devPtr) | ||
| 18 | ``` | ||
| 19 | |||
| 20 | ### 产品支持情况 | ||
| 21 | |||
| 22 | | 产品 | 是否支持 | | ||
| 23 | | --- | :---: | | ||
| 24 | | Ascend 950PR/Ascend 950DT | √ | | ||
| 25 | | Atlas A3 训练系列产品/Atlas A3 推理系列产品 | √ | | ||
| 26 | | Atlas A2 训练系列产品/Atlas A2 推理系列产品 | √ | | ||
| 27 | |||
| 28 | ### 功能说明 | ||
| 29 | |||
| 30 | 获取Device变量的地址。 | ||
| 31 | |||
| 32 | ### 参数说明 | ||
| 33 | |||
| 34 | | 参数名 | 输入/输出 | 说明 | | ||
| 35 | | --- | :---: | --- | | ||
| 36 | | symbol | 输入 | Device变量名。此处传入`__gm__`声明的变量名。 | | ||
| 37 | | devPtr | 输出 | Device变量的内存地址指针。 | | ||
| 38 | |||
| 39 | ### 返回值说明 | ||
| 40 | |||
| 41 | 返回0表示成功,返回其他值表示失败,请参见[aclError](25_数据类型及其操作接口.md#aclError)。 | ||
| 42 | |||
| 43 | ### 约束说明 | ||
| 44 | |||
| 45 | - 本接口仅适用于Ascend C语言开发自定义算子并基于毕昇编译器进行Host和Device代码混合编译的场景。 | ||
| 46 | - Device变量地址仅在当前Device有效,切换Device后需重新获取地址。 | ||
| 47 | - 仅支持AI Core算子中的Device变量,具体约束如下: | ||
| 48 | - 变量定义位置:支持main函数所在文件中定义的Device变量(如 `__gm__ float convWeights`)。 | ||
| 49 | - extern变量支持:跨文件引用场景,即文件A定义的Device变量可在文件B中通过extern关键字声明引用(如 `extern __gm__ float convWeights`)。此场景需使用毕昇编译器 `-dc` 编译模式,将多文件编译为单一算子二进制文件。 | ||
| 50 | - 数据类型:支持基础数据类型、函数指针、结构体及数组,不支持函数指针、class类型。 | ||
| 51 | |||
| 52 | |||
| 53 | <br> | ||
| 54 | <br> | ||
| 55 | <br> | ||
| 56 | |||
| 57 | <a id="aclrtGetSymbolSize"></a> | ||
| 58 | |||
| 59 | ## aclrtGetSymbolSize | ||
| 60 | |||
| 61 | ```c | ||
| 62 | aclError aclrtGetSymbolSize(const void *symbol, size_t *size) | ||
| 63 | ``` | ||
| 64 | |||
| 65 | ### 产品支持情况 | ||
| 66 | |||
| 67 | | 产品 | 是否支持 | | ||
| 68 | | --- | :---: | | ||
| 69 | | Ascend 950PR/Ascend 950DT | √ | | ||
| 70 | | Atlas A3 训练系列产品/Atlas A3 推理系列产品 | √ | | ||
| 71 | | Atlas A2 训练系列产品/Atlas A2 推理系列产品 | √ | | ||
| 72 | |||
| 73 | ### 功能说明 | ||
| 74 | |||
| 75 | 获取Device变量占用的内存大小。 | ||
| 76 | |||
| 77 | ### 参数说明 | ||
| 78 | |||
| 79 | | 参数名 | 输入/输出 | 说明 | | ||
| 80 | | --- | :---: | --- | | ||
| 81 | | symbol | 输入 | Device变量名。此处传入`__gm__`声明的变量名。 | | ||
| 82 | | size | 输出 | Device变量的大小,单位Byte。 | | ||
| 83 | |||
| 84 | ### 返回值说明 | ||
| 85 | |||
| 86 | 返回0表示成功,返回其他值表示失败,请参见[aclError](25_数据类型及其操作接口.md#aclError)。 | ||
| 87 | |||
| 88 | ### 约束说明 | ||
| 89 | |||
| 90 | - 同[aclrtGetSymbolAddress](#aclrtGetSymbolAddress)接口约束说明。 | ||
| 91 | |||
| 92 | |||
| 93 | <br> | ||
| 94 | <br> | ||
| 95 | <br> | ||
| 96 | |||
| 97 | <a id="aclrtMemcpyFromSymbol"></a> | ||
| 98 | |||
| 99 | ## aclrtMemcpyFromSymbol | ||
| 100 | |||
| 101 | ```c | ||
| 102 | aclError aclrtMemcpyFromSymbol(void *dst, size_t dstMax, const void *symbol, | ||
| 103 | size_t count, size_t offset, aclrtMemcpyKind kind) | ||
| 104 | ``` | ||
| 105 | |||
| 106 | ### 产品支持情况 | ||
| 107 | |||
| 108 | | 产品 | 是否支持 | | ||
| 109 | | --- | :---: | | ||
| 110 | | Ascend 950PR/Ascend 950DT | √ | | ||
| 111 | | Atlas A3 训练系列产品/Atlas A3 推理系列产品 | √ | | ||
| 112 | | Atlas A2 训练系列产品/Atlas A2 推理系列产品 | √ | | ||
| 113 | |||
| 114 | ### 功能说明 | ||
| 115 | |||
| 116 | 实现Device变量的数据到Host的同步内存复制。用于读取Device变量的数据。 | ||
| 117 | |||
| 118 | ### 参数说明 | ||
| 119 | |||
| 120 | | 参数名 | 输入/输出 | 说明 | | ||
| 121 | | --- | :---: | --- | | ||
| 122 | | dst | 输入 | 目的内存地址指针。 | | ||
| 123 | | dstMax | 输入 | 目标内存最大长度,单位Byte。需满足 dstMax ≥ count。 | | ||
| 124 | | symbol | 输入 | Device变量名。此处传入`__gm__`声明的变量名。 | | ||
| 125 | | count | 输入 | 内存复制的长度,单位Byte。需满足 offset + count ≤ Device变量大小,Device变量大小可通过 [aclrtGetSymbolSize](#aclrtGetSymbolSize)接口查询获取。 | | ||
| 126 | | offset | 输入 | Device变量地址偏移,单位Byte。需满足 offset + count ≤ Device变量大小,Device变量大小可通过 [aclrtGetSymbolSize](#aclrtGetSymbolSize)接口查询获取。 | | ||
| 127 | | kind | 输入 | 拷贝类型,类型定义请参见[aclrtMemcpyKind](25_数据类型及其操作接口.md#aclrtMemcpyKind)。本接口仅支持ACL_MEMCPY_DEVICE_TO_HOST和ACL_MEMCPY_DEFAULT。 | | ||
| 128 | |||
| 129 | ### 返回值说明 | ||
| 130 | |||
| 131 | 返回0表示成功,返回其他值表示失败,请参见[aclError](25_数据类型及其操作接口.md#aclError)。 | ||
| 132 | |||
| 133 | ### 约束说明 | ||
| 134 | |||
| 135 | - 同[aclrtGetSymbolAddress](#aclrtGetSymbolAddress)接口约束说明。 | ||
| 136 | |||
| 137 | |||
| 138 | <br> | ||
| 139 | <br> | ||
| 140 | <br> | ||
| 141 | |||
| 142 | <a id="aclrtMemcpyFromSymbolAsync"></a> | ||
| 143 | |||
| 144 | ## aclrtMemcpyFromSymbolAsync | ||
| 145 | |||
| 146 | ```c | ||
| 147 | aclError aclrtMemcpyFromSymbolAsync(void *dst, size_t dstMax, const void *symbol, | ||
| 148 | size_t count, size_t offset, aclrtMemcpyKind kind, | ||
| 149 | aclrtStream stream) | ||
| 150 | ``` | ||
| 151 | |||
| 152 | ### 产品支持情况 | ||
| 153 | |||
| 154 | | 产品 | 是否支持 | | ||
| 155 | | --- | :---: | | ||
| 156 | | Ascend 950PR/Ascend 950DT | √ | | ||
| 157 | | Atlas A3 训练系列产品/Atlas A3 推理系列产品 | √ | | ||
| 158 | | Atlas A2 训练系列产品/Atlas A2 推理系列产品 | √ | | ||
| 159 | |||
| 160 | ### 功能说明 | ||
| 161 | |||
| 162 | 实现Device变量的数据到Host的异步内存复制。用于读取Device变量的数据。 | ||
| 163 | |||
| 164 | 本接口中的Host内存支持锁页内存(例如通过aclrtMallocHost接口申请的内存)、非锁页内存(通过malloc接口申请的内存)。当Host内存是锁页内存时,本接口是异步接口,调用接口成功仅表示任务下发成功,不表示任务执行成功,调用本接口后,需调用同步等待接口(例如,[aclrtSynchronizeStream](06_Stream管理.md#aclrtSynchronizeStream))确保内存复制的任务已执行完成;当Host内存是非锁页内存时,本接口在内存复制任务完成后才返回。 | ||
| 165 | |||
| 166 | ### 参数说明 | ||
| 167 | |||
| 168 | | 参数名 | 输入/输出 | 说明 | | ||
| 169 | | --- | :---: | --- | | ||
| 170 | | dst | 输入 | 目的内存地址指针。 | | ||
| 171 | | dstMax | 输入 | 目标内存最大长度,单位Byte。需满足 dstMax ≥ count。 | | ||
| 172 | | symbol | 输入 | Device变量名。此处传入`__gm__`声明的变量名。 | | ||
| 173 | | count | 输入 | 内存复制的长度,单位Byte。需满足 offset + count ≤ Device变量大小,Device变量大小可通过 [aclrtGetSymbolSize](#aclrtGetSymbolSize)接口查询获取。 | | ||
| 174 | | offset | 输入 | Device变量地址偏移,单位Byte。需满足 offset + count ≤ Device变量大小,Device变量大小可通过 [aclrtGetSymbolSize](#aclrtGetSymbolSize)接口查询获取。 | | ||
| 175 | | kind | 输入 | 拷贝类型,类型定义请参见[aclrtMemcpyKind](25_数据类型及其操作接口.md#aclrtMemcpyKind)。本接口仅支持ACL_MEMCPY_DEVICE_TO_HOST和ACL_MEMCPY_DEFAULT。 | | ||
| 176 | | stream | 输入 | 指定执行内存复制任务的Stream。类型定义请参见[aclrtStream](25_数据类型及其操作接口.md#aclrtStream)。 | | ||
| 177 | |||
| 178 | ### 返回值说明 | ||
| 179 | |||
| 180 | 返回0表示成功,返回其他值表示失败,请参见[aclError](25_数据类型及其操作接口.md#aclError)。 | ||
| 181 | |||
| 182 | ### 约束说明 | ||
| 183 | |||
| 184 | - 同[aclrtGetSymbolAddress](#aclrtGetSymbolAddress)接口约束说明。 | ||
| 185 | - 本接口为异步接口,调用后需同步等待拷贝完成。 | ||
| 186 | |||
| 187 | |||
| 188 | <br> | ||
| 189 | <br> | ||
| 190 | <br> | ||
| 191 | |||
| 192 | <a id="aclrtMemcpyToSymbol"></a> | ||
| 193 | |||
| 194 | ## aclrtMemcpyToSymbol | ||
| 195 | |||
| 196 | ```c | ||
| 197 | aclError aclrtMemcpyToSymbol(const void *symbol, const void *src, | ||
| 198 | size_t count, size_t offset, aclrtMemcpyKind kind) | ||
| 199 | ``` | ||
| 200 | |||
| 201 | ### 产品支持情况 | ||
| 202 | |||
| 203 | | 产品 | 是否支持 | | ||
| 204 | | --- | :---: | | ||
| 205 | | Ascend 950PR/Ascend 950DT | √ | | ||
| 206 | | Atlas A3 训练系列产品/Atlas A3 推理系列产品 | √ | | ||
| 207 | | Atlas A2 训练系列产品/Atlas A2 推理系列产品 | √ | | ||
| 208 | |||
| 209 | ### 功能说明 | ||
| 210 | |||
| 211 | 实现Host数据到Device变量的同步内存复制。用于向Device变量写入数据。 | ||
| 212 | |||
| 213 | ### 参数说明 | ||
| 214 | |||
| 215 | | 参数名 | 输入/输出 | 说明 | | ||
| 216 | | --- | :---: | --- | | ||
| 217 | | symbol | 输入 | Device变量名。此处传入`__gm__`声明的变量名。 | | ||
| 218 | | src | 输入 | 源内存地址指针。 | | ||
| 219 | | count | 输入 | 内存复制的长度,单位Byte。需满足 offset + count ≤ Device变量大小,Device变量大小可通过 [aclrtGetSymbolSize](#aclrtGetSymbolSize)接口查询获取。 | | ||
| 220 | | offset | 输入 | Device变量地址偏移,单位Byte。需满足 offset + count ≤ Device变量大小,Device变量大小可通过 [aclrtGetSymbolSize](#aclrtGetSymbolSize)接口查询获取。 | | ||
| 221 | | kind | 输入 | 拷贝类型,类型定义请参见[aclrtMemcpyKind](25_数据类型及其操作接口.md#aclrtMemcpyKind)。本接口仅支持ACL_MEMCPY_HOST_TO_DEVICE和ACL_MEMCPY_DEFAULT。 | | ||
| 222 | |||
| 223 | ### 返回值说明 | ||
| 224 | |||
| 225 | 返回0表示成功,返回其他值表示失败,请参见[aclError](25_数据类型及其操作接口.md#aclError)。 | ||
| 226 | |||
| 227 | ### 约束说明 | ||
| 228 | |||
| 229 | - 同[aclrtGetSymbolAddress](#aclrtGetSymbolAddress)接口约束说明。 | ||
| 230 | |||
| 231 | |||
| 232 | <br> | ||
| 233 | <br> | ||
| 234 | <br> | ||
| 235 | |||
| 236 | <a id="aclrtMemcpyToSymbolAsync"></a> | ||
| 237 | |||
| 238 | ## aclrtMemcpyToSymbolAsync | ||
| 239 | |||
| 240 | ```c | ||
| 241 | aclError aclrtMemcpyToSymbolAsync(const void *symbol, const void *src, | ||
| 242 | size_t count, size_t offset, aclrtMemcpyKind kind, | ||
| 243 | aclrtStream stream) | ||
| 244 | ``` | ||
| 245 | |||
| 246 | ### 产品支持情况 | ||
| 247 | |||
| 248 | | 产品 | 是否支持 | | ||
| 249 | | --- | :---: | | ||
| 250 | | Ascend 950PR/Ascend 950DT | √ | | ||
| 251 | | Atlas A3 训练系列产品/Atlas A3 推理系列产品 | √ | | ||
| 252 | | Atlas A2 训练系列产品/Atlas A2 推理系列产品 | √ | | ||
| 253 | |||
| 254 | ### 功能说明 | ||
| 255 | |||
| 256 | 实现Host数据到Device变量的异步内存复制。用于向Device变量写入数据。 | ||
| 257 | |||
| 258 | 本接口中的Host内存支持锁页内存(例如通过aclrtMallocHost接口申请的内存)、非锁页内存(通过malloc接口申请的内存)。当Host内存是锁页内存时,本接口是异步接口,调用接口成功仅表示任务下发成功,不表示任务执行成功,调用本接口后,需调用同步等待接口(例如,[aclrtSynchronizeStream](06_Stream管理.md#aclrtSynchronizeStream))确保内存复制的任务已执行完成;当Host内存是非锁页内存时,本接口在内存复制任务完成后才返回。 | ||
| 259 | |||
| 260 | ### 参数说明 | ||
| 261 | |||
| 262 | | 参数名 | 输入/输出 | 说明 | | ||
| 263 | | --- | :---: | --- | | ||
| 264 | | symbol | 输入 | Device变量名。此处传入`__gm__`声明的变量名。 | | ||
| 265 | | src | 输入 | 源内存地址指针。 | | ||
| 266 | | count | 输入 | 内存复制的长度,单位Byte。需满足 offset + count ≤ Device变量大小,Device变量大小可通过 [aclrtGetSymbolSize](#aclrtGetSymbolSize)接口查询获取。 | | ||
| 267 | | offset | 输入 | Device变量地址偏移,单位Byte。需满足 offset + count ≤ Device变量大小,Device变量大小可通过 [aclrtGetSymbolSize](#aclrtGetSymbolSize)接口查询获取。 | | ||
| 268 | | kind | 输入 | 拷贝类型,类型定义请参见[aclrtMemcpyKind](25_数据类型及其操作接口.md#aclrtMemcpyKind)。本接口仅支持ACL_MEMCPY_HOST_TO_DEVICE和ACL_MEMCPY_DEFAULT。 | | ||
| 269 | | stream | 输入 | 指定执行内存复制任务的Stream。类型定义请参见[aclrtStream](25_数据类型及其操作接口.md#aclrtStream)。 | | ||
| 270 | |||
| 271 | ### 返回值说明 | ||
| 272 | |||
| 273 | 返回0表示成功,返回其他值表示失败,请参见[aclError](25_数据类型及其操作接口.md#aclError)。 | ||
| 274 | |||
| 275 | ### 约束说明 | ||
| 276 | |||
| 277 | - 同[aclrtGetSymbolAddress](#aclrtGetSymbolAddress)接口约束说明。 | ||
| 278 | - 本接口为异步接口,调用后需同步等待拷贝完成。 | ||
| @@ -17,15 +17,16 @@ | |||
| 17 | 17 | ||
| 18 | 本分类包含以下子类: | 18 | 本分类包含以下子类: |
| 19 | 19 | ||
| 20 | - [11-01 设备内存分配与释放](11-01_设备内存分配与释放.md)(10 个接口) | 20 | - [11-01 设备内存分配与释放](11-01_设备内存分配与释放.md) |
| 21 | - [11-02 主机内存管理](11-02_主机内存管理.md)(9 个接口) | 21 | - [11-02 主机内存管理](11-02_主机内存管理.md) |
| 22 | - [11-03 内存拷贝与设置](11-03_内存拷贝与设置.md)(15 个接口) | 22 | - [11-03 内存拷贝与设置](11-03_内存拷贝与设置.md) |
| 23 | - [11-04 虚拟内存管理](11-04_虚拟内存管理.md)(19 个接口) | 23 | - [11-04 虚拟内存管理](11-04_虚拟内存管理.md) |
| 24 | - [11-05 统一寻址](11-05_统一寻址.md)(8 个接口) | 24 | - [11-05 统一寻址](11-05_统一寻址.md) |
| 25 | - [11-06 CMO 缓存操作](11-06_CMO缓存操作.md)(8 个接口) | 25 | - [11-06 CMO 缓存操作](11-06_CMO缓存操作.md) |
| 26 | - [11-07 IPC 进程间内存共享](11-07_IPC进程间内存共享.md)(6 个接口) | 26 | - [11-07 IPC 进程间内存共享](11-07_IPC进程间内存共享.md) |
| 27 | - [11-08 自定义内存分配器](11-08_自定义内存分配器.md)(3 个接口) | 27 | - [11-08 自定义内存分配器](11-08_自定义内存分配器.md) |
| 28 | - [11-09 流内存操作](11-09_流内存操作.md)(2 个接口) | 28 | - [11-09 流内存操作](11-09_流内存操作.md) |
| 29 | - [11-10 Stream有序内存分配](11-10_Stream有序内存分配.md)(7 个接口) | 29 | - [11-10 Stream有序内存分配](11-10_Stream有序内存分配.md) |
| 30 | - [11-11 Device变量内存操作](11-11_Device变量内存操作.md) | ||
| 30 | 31 | ||
| 31 | 32 | ||
| @@ -346,6 +346,7 @@ typedef int aclError; | |||
| 346 | | #define ACL_ERROR_RT_MODEL_UPDATE_FAILED 107040 | 模型未完全捕获、模型正在执行中,或参数更新失败导致模型更新失败。 | 确保模型已成功捕获、执行完成或参数更新成功。 | | 346 | | #define ACL_ERROR_RT_MODEL_UPDATE_FAILED 107040 | 模型未完全捕获、模型正在执行中,或参数更新失败导致模型更新失败。 | 确保模型已成功捕获、执行完成或参数更新成功。 | |
| 347 | | #define ACL_ERROR_RT_CAPTURE_MODE_BLOCK_ASYNC 107041 | 当前的捕获模式不支持异步转同步操作。 | 不同捕获模式支持的操作范围不同,请参见aclmdlRICaptureThreadExchangeMode接口中的说明,并切换到正确的捕获模式。 | | 347 | | #define ACL_ERROR_RT_CAPTURE_MODE_BLOCK_ASYNC 107041 | 当前的捕获模式不支持异步转同步操作。 | 不同捕获模式支持的操作范围不同,请参见aclmdlRICaptureThreadExchangeMode接口中的说明,并切换到正确的捕获模式。 | |
| 348 | | #define ACL_ERROR_RT_SYMBOL_NOT_FOUND 107042 | 符号未找到。 | 请检查全局变量名称是否与算子二进制中的全局变量名称一致。 | | 348 | | #define ACL_ERROR_RT_SYMBOL_NOT_FOUND 107042 | 符号未找到。 | 请检查全局变量名称是否与算子二进制中的全局变量名称一致。 | |
| 349 | | #define ACL_ERROR_RT_INVALID_SYMBOL 107043 | 无效的符号。 | 请检查变量名是否在Device代码中声明 | | ||
| 349 | | #define ACL_ERROR_RT_FEATURE_NOT_SUPPORT 207000 | 特性不支持。 | 您可以获取日志后单击[Link](https://www.hiascend.com/support)联系技术支持。 | | 350 | | #define ACL_ERROR_RT_FEATURE_NOT_SUPPORT 207000 | 特性不支持。 | 您可以获取日志后单击[Link](https://www.hiascend.com/support)联系技术支持。 | |
| 350 | | #define ACL_ERROR_RT_MEMORY_ALLOCATION 207001 | 内存申请失败。 | 请检查硬件环境上的存储剩余情况。 | | 351 | | #define ACL_ERROR_RT_MEMORY_ALLOCATION 207001 | 内存申请失败。 | 请检查硬件环境上的存储剩余情况。 | |
| 351 | | #define ACL_ERROR_RT_MEMORY_FREE 207002 | 内存释放失败。 | 您可以获取日志后单击[Link](https://www.hiascend.com/support)联系技术支持。 | | 352 | | #define ACL_ERROR_RT_MEMORY_FREE 207002 | 内存释放失败。 | 您可以获取日志后单击[Link](https://www.hiascend.com/support)联系技术支持。 | |
| @@ -1,4 +1,4 @@ | |||
| 1 | # 26. C++ API Routines | 1 | # 26. C++扩展接口 |
| 2 | 2 | ||
| 3 | 本章节描述 C++ 扩展接口,包括函数重载和模板封装,用于简化 C++ 场景下的 API 调用并提供类型安全的内存操作。 | 3 | 本章节描述 C++ 扩展接口,包括函数重载和模板封装,用于简化 C++ 场景下的 API 调用并提供类型安全的内存操作。 |
| 4 | 4 | ||
| @@ -34,6 +34,12 @@ | |||
| 34 | - [`template <typename T> aclError aclrtMemAllocManaged(T **devPtr, size_t size, uint32_t flags = ACL_RT_MEM_ATTACH_GLOBAL)`](#aclrtMemAllocManaged):类型安全的统一内存分配。 | 34 | - [`template <typename T> aclError aclrtMemAllocManaged(T **devPtr, size_t size, uint32_t flags = ACL_RT_MEM_ATTACH_GLOBAL)`](#aclrtMemAllocManaged):类型安全的统一内存分配。 |
| 35 | - [`template <typename T> aclError aclrtMemManagedPrefetchAsync(const T *ptr, size_t size, aclrtMemManagedLocation location, uint32_t flags, aclrtStream stream)`](#aclrtMemManagedPrefetchAsync):预取统一内存。 | 35 | - [`template <typename T> aclError aclrtMemManagedPrefetchAsync(const T *ptr, size_t size, aclrtMemManagedLocation location, uint32_t flags, aclrtStream stream)`](#aclrtMemManagedPrefetchAsync):预取统一内存。 |
| 36 | - [`template <typename T> aclError aclrtMemManagedPrefetchBatchAsync(...)`](#aclrtMemManagedPrefetchBatchAsync):批量预取统一内存。 | 36 | - [`template <typename T> aclError aclrtMemManagedPrefetchBatchAsync(...)`](#aclrtMemManagedPrefetchBatchAsync):批量预取统一内存。 |
| 37 | - [`template <typename T> aclError aclrtGetSymbolAddress(...)`](#aclrtGetSymbolAddress):获取Device变量的地址。 | ||
| 38 | - [`template <typename T> aclError aclrtGetSymbolSize(...)`](#aclrtGetSymbolSize):获取Device变量占用的内存大小。 | ||
| 39 | - [`template <typename T> aclError aclrtMemcpyFromSymbol(...)`](#aclrtMemcpyFromSymbol):实现Device变量的数据到Host的同步内存复制。 | ||
| 40 | - [`template <typename T> aclError aclrtMemcpyFromSymbolAsync(...)`](#aclrtMemcpyFromSymbolAsync):实现Device变量的数据到Host的异步内存复制。 | ||
| 41 | - [`template <typename T> aclError aclrtMemcpyToSymbol(...)`](#aclrtMemcpyToSymbol):实现Host数据到Device变量的同步内存复制。 | ||
| 42 | - [`template <typename T> aclError aclrtMemcpyToSymbolAsync(...)`](#aclrtMemcpyToSymbolAsync):实现Host数据到Device变量的异步内存复制。 | ||
| 37 | 43 | ||
| 38 | --- | 44 | --- |
| 39 | 45 | ||
| @@ -1016,4 +1022,287 @@ aclError aclrtMemcpy2dAsync(T *dst, size_t dpitch, const U *src, size_t spitch, | |||
| 1016 | ### 约束说明 | 1022 | ### 约束说明 |
| 1017 | 1023 | ||
| 1018 | - 本接口仅适用于 C++ 程序。 | 1024 | - 本接口仅适用于 C++ 程序。 |
| 1019 | - 本接口为异步接口,调用后需同步等待拷贝完成。 | 1025 | - 本接口为异步接口,调用后需同步等待拷贝完成。 |
| 1026 | |||
| 1027 | --- | ||
| 1028 | |||
| 1029 | <br> | ||
| 1030 | <br> | ||
| 1031 | <br> | ||
| 1032 | |||
| 1033 | <a id="aclrtGetSymbolAddress"></a> | ||
| 1034 | |||
| 1035 | ## aclrtGetSymbolAddress | ||
| 1036 | |||
| 1037 | ```cpp | ||
| 1038 | template <typename T> | ||
| 1039 | aclError aclrtGetSymbolAddress(const T &symbol, void **devPtr) | ||
| 1040 | ``` | ||
| 1041 | |||
| 1042 | ### 产品支持情况 | ||
| 1043 | |||
| 1044 | | 产品 | 是否支持 | | ||
| 1045 | | --- | :---: | | ||
| 1046 | | Ascend 950PR/Ascend 950DT | √ | | ||
| 1047 | | Atlas A3 训练系列产品/Atlas A3 推理系列产品 | √ | | ||
| 1048 | | Atlas A2 训练系列产品/Atlas A2 推理系列产品 | √ | | ||
| 1049 | |||
| 1050 | ### 功能说明 | ||
| 1051 | |||
| 1052 | 获取Device变量的地址。 | ||
| 1053 | |||
| 1054 | ### 参数说明 | ||
| 1055 | |||
| 1056 | | 参数名 | 输入/输出 | 说明 | | ||
| 1057 | | --- | :---: | --- | | ||
| 1058 | | symbol | 输入 | Device变量名。此处传入`__gm__`声明的变量名。 | | ||
| 1059 | | devPtr | 输出 | Device变量的内存地址指针。 | | ||
| 1060 | |||
| 1061 | ### 返回值说明 | ||
| 1062 | |||
| 1063 | 返回0表示成功,返回其他值表示失败,请参见[aclError](25_数据类型及其操作接口.md#aclError)。 | ||
| 1064 | |||
| 1065 | ### 约束说明 | ||
| 1066 | |||
| 1067 | - 本接口仅适用于 C++ 程序。 | ||
| 1068 | - 本接口仅适用于Ascend C语言开发自定义算子并基于毕昇编译器进行Host和Device代码混合编译的场景。 | ||
| 1069 | - Device变量地址仅在当前Device有效,切换Device后需重新获取地址。 | ||
| 1070 | - 仅支持AI Core算子中的Device变量,具体约束如下: | ||
| 1071 | - 变量定义位置:支持main函数所在文件中定义的Device变量(如 `__gm__ float convWeights`)。 | ||
| 1072 | - extern变量支持:跨文件引用场景,即文件A定义的Device变量可在文件B中通过extern关键字声明引用(如 `extern __gm__ float convWeights`)。此场景需使用毕昇编译器 `-dc` 编译模式,将多文件编译为单一算子二进制文件。 | ||
| 1073 | - 数据类型:支持基础数据类型、函数指针、结构体及数组,不支持函数指针、class类型。 | ||
| 1074 | |||
| 1075 | --- | ||
| 1076 | |||
| 1077 | <br> | ||
| 1078 | <br> | ||
| 1079 | <br> | ||
| 1080 | |||
| 1081 | <a id="aclrtGetSymbolSize"></a> | ||
| 1082 | |||
| 1083 | ## aclrtGetSymbolSize | ||
| 1084 | |||
| 1085 | ```cpp | ||
| 1086 | template <typename T> | ||
| 1087 | aclError aclrtGetSymbolSize(const T &symbol, size_t *size) | ||
| 1088 | ``` | ||
| 1089 | |||
| 1090 | ### 产品支持情况 | ||
| 1091 | |||
| 1092 | | 产品 | 是否支持 | | ||
| 1093 | | --- | :---: | | ||
| 1094 | | Ascend 950PR/Ascend 950DT | √ | | ||
| 1095 | | Atlas A3 训练系列产品/Atlas A3 推理系列产品 | √ | | ||
| 1096 | | Atlas A2 训练系列产品/Atlas A2 推理系列产品 | √ | | ||
| 1097 | |||
| 1098 | ### 功能说明 | ||
| 1099 | |||
| 1100 | 获取Device变量占用的内存大小。 | ||
| 1101 | |||
| 1102 | ### 参数说明 | ||
| 1103 | |||
| 1104 | | 参数名 | 输入/输出 | 说明 | | ||
| 1105 | | --- | :---: | --- | | ||
| 1106 | | symbol | 输入 | Device变量名。此处传入`__gm__`声明的变量名。 | | ||
| 1107 | | size | 输出 | Device变量的大小,单位Byte。 | | ||
| 1108 | |||
| 1109 | ### 返回值说明 | ||
| 1110 | |||
| 1111 | 返回0表示成功,返回其他值表示失败,请参见[aclError](25_数据类型及其操作接口.md#aclError)。 | ||
| 1112 | |||
| 1113 | ### 约束说明 | ||
| 1114 | |||
| 1115 | - 同[aclrtGetSymbolAddress](#aclrtGetSymbolAddress)接口约束说明。 | ||
| 1116 | |||
| 1117 | --- | ||
| 1118 | |||
| 1119 | <br> | ||
| 1120 | <br> | ||
| 1121 | <br> | ||
| 1122 | |||
| 1123 | <a id="aclrtMemcpyFromSymbol"></a> | ||
| 1124 | |||
| 1125 | ## aclrtMemcpyFromSymbol | ||
| 1126 | |||
| 1127 | ```cpp | ||
| 1128 | template <typename T> | ||
| 1129 | aclError aclrtMemcpyFromSymbol(void *dst, size_t dstMax, const T &symbol, size_t count, | ||
| 1130 | size_t offset, aclrtMemcpyKind kind) | ||
| 1131 | ``` | ||
| 1132 | |||
| 1133 | ### 产品支持情况 | ||
| 1134 | |||
| 1135 | | 产品 | 是否支持 | | ||
| 1136 | | --- | :---: | | ||
| 1137 | | Ascend 950PR/Ascend 950DT | √ | | ||
| 1138 | | Atlas A3 训练系列产品/Atlas A3 推理系列产品 | √ | | ||
| 1139 | | Atlas A2 训练系列产品/Atlas A2 推理系列产品 | √ | | ||
| 1140 | |||
| 1141 | ### 功能说明 | ||
| 1142 | |||
| 1143 | 实现Device变量的数据到Host的同步内存复制。用于读取Device变量的数据。 | ||
| 1144 | |||
| 1145 | ### 参数说明 | ||
| 1146 | |||
| 1147 | | 参数名 | 输入/输出 | 说明 | | ||
| 1148 | | --- | :---: | --- | | ||
| 1149 | | dst | 输入 | 目的内存地址指针。 | | ||
| 1150 | | dstMax | 输入 | 目标内存最大长度,单位Byte。需满足 dstMax ≥ count。 | | ||
| 1151 | | symbol | 输入 | Device变量名。此处传入`__gm__`声明的变量名。 | | ||
| 1152 | | count | 输入 | 内存复制的长度,单位Byte。需满足 offset + count ≤ Device变量大小,Device变量大小可通过 [aclrtGetSymbolSize](#aclrtGetSymbolSize)接口查询获取。 | | ||
| 1153 | | offset | 输入 | Device变量地址偏移,单位Byte。需满足 offset + count ≤ Device变量大小,Device变量大小可通过 [aclrtGetSymbolSize](#aclrtGetSymbolSize)接口查询获取。 | | ||
| 1154 | | kind | 输入 | 拷贝类型,类型定义请参见[aclrtMemcpyKind](25_数据类型及其操作接口.md#aclrtMemcpyKind)。本接口仅支持ACL_MEMCPY_DEVICE_TO_HOST和ACL_MEMCPY_DEFAULT。 | | ||
| 1155 | |||
| 1156 | ### 返回值说明 | ||
| 1157 | |||
| 1158 | 返回0表示成功,返回其他值表示失败,请参见[aclError](25_数据类型及其操作接口.md#aclError)。 | ||
| 1159 | |||
| 1160 | ### 约束说明 | ||
| 1161 | |||
| 1162 | - 同[aclrtGetSymbolAddress](#aclrtGetSymbolAddress)接口约束说明。 | ||
| 1163 | |||
| 1164 | --- | ||
| 1165 | |||
| 1166 | <br> | ||
| 1167 | <br> | ||
| 1168 | <br> | ||
| 1169 | |||
| 1170 | <a id="aclrtMemcpyFromSymbolAsync"></a> | ||
| 1171 | |||
| 1172 | ## aclrtMemcpyFromSymbolAsync | ||
| 1173 | |||
| 1174 | ```cpp | ||
| 1175 | template <typename T> | ||
| 1176 | aclError aclrtMemcpyFromSymbolAsync(void *dst, size_t dstMax, const T &symbol, size_t count, | ||
| 1177 | size_t offset, aclrtMemcpyKind kind, aclrtStream stream) | ||
| 1178 | ``` | ||
| 1179 | |||
| 1180 | ### 产品支持情况 | ||
| 1181 | |||
| 1182 | | 产品 | 是否支持 | | ||
| 1183 | | --- | :---: | | ||
| 1184 | | Ascend 950PR/Ascend 950DT | √ | | ||
| 1185 | | Atlas A3 训练系列产品/Atlas A3 推理系列产品 | √ | | ||
| 1186 | | Atlas A2 训练系列产品/Atlas A2 推理系列产品 | √ | | ||
| 1187 | |||
| 1188 | ### 功能说明 | ||
| 1189 | |||
| 1190 | 实现Device变量的数据到Host的异步内存复制。用于读取Device变量的数据。 | ||
| 1191 | |||
| 1192 | 本接口中的Host内存支持锁页内存(例如通过aclrtMallocHost接口申请的内存)、非锁页内存(通过malloc接口申请的内存)。当Host内存是锁页内存时,本接口是异步接口,调用接口成功仅表示任务下发成功,不表示任务执行成功,调用本接口后,需调用同步等待接口(例如,[aclrtSynchronizeStream](06_Stream管理.md#aclrtSynchronizeStream))确保内存复制的任务已执行完成;当Host内存是非锁页内存时,本接口在内存复制任务完成后才返回。 | ||
| 1193 | |||
| 1194 | ### 参数说明 | ||
| 1195 | |||
| 1196 | | 参数名 | 输入/输出 | 说明 | | ||
| 1197 | | --- | :---: | --- | | ||
| 1198 | | dst | 输入 | 目的内存地址指针。 | | ||
| 1199 | | dstMax | 输入 | 目标内存最大长度,单位Byte。需满足 dstMax ≥ count。 | | ||
| 1200 | | symbol | 输入 | Device变量名。此处传入`__gm__`声明的变量名。 | | ||
| 1201 | | count | 输入 | 内存复制的长度,单位Byte。需满足 offset + count ≤ Device变量大小,Device变量大小可通过 [aclrtGetSymbolSize](#aclrtGetSymbolSize)接口查询获取。 | | ||
| 1202 | | offset | 输入 | Device变量地址偏移,单位Byte。需满足 offset + count ≤ Device变量大小,Device变量大小可通过 [aclrtGetSymbolSize](#aclrtGetSymbolSize)接口查询获取。 | | ||
| 1203 | | kind | 输入 | 拷贝类型,类型定义请参见[aclrtMemcpyKind](25_数据类型及其操作接口.md#aclrtMemcpyKind)。本接口仅支持ACL_MEMCPY_DEVICE_TO_HOST和ACL_MEMCPY_DEFAULT。 | | ||
| 1204 | | stream | 输入 | 指定执行内存复制任务的Stream。类型定义请参见[aclrtStream](25_数据类型及其操作接口.md#aclrtStream)。 | | ||
| 1205 | |||
| 1206 | ### 返回值说明 | ||
| 1207 | |||
| 1208 | 返回0表示成功,返回其他值表示失败,请参见[aclError](25_数据类型及其操作接口.md#aclError)。 | ||
| 1209 | |||
| 1210 | ### 约束说明 | ||
| 1211 | |||
| 1212 | - 同[aclrtGetSymbolAddress](#aclrtGetSymbolAddress)接口约束说明。 | ||
| 1213 | - 本接口为异步接口,调用后需同步等待拷贝完成。 | ||
| 1214 | |||
| 1215 | --- | ||
| 1216 | |||
| 1217 | <br> | ||
| 1218 | <br> | ||
| 1219 | <br> | ||
| 1220 | |||
| 1221 | <a id="aclrtMemcpyToSymbol"></a> | ||
| 1222 | |||
| 1223 | ## aclrtMemcpyToSymbol | ||
| 1224 | |||
| 1225 | ```cpp | ||
| 1226 | template <typename T> | ||
| 1227 | aclError aclrtMemcpyToSymbol(const T &symbol, const void *src, size_t count, size_t offset, aclrtMemcpyKind kind) | ||
| 1228 | ``` | ||
| 1229 | |||
| 1230 | ### 产品支持情况 | ||
| 1231 | |||
| 1232 | | 产品 | 是否支持 | | ||
| 1233 | | --- | :---: | | ||
| 1234 | | Ascend 950PR/Ascend 950DT | √ | | ||
| 1235 | | Atlas A3 训练系列产品/Atlas A3 推理系列产品 | √ | | ||
| 1236 | | Atlas A2 训练系列产品/Atlas A2 推理系列产品 | √ | | ||
| 1237 | |||
| 1238 | ### 功能说明 | ||
| 1239 | |||
| 1240 | 实现Host数据到Device变量的同步内存复制。用于向Device变量写入数据。 | ||
| 1241 | |||
| 1242 | ### 参数说明 | ||
| 1243 | |||
| 1244 | | 参数名 | 输入/输出 | 说明 | | ||
| 1245 | | --- | :---: | --- | | ||
| 1246 | | symbol | 输入 | Device变量名。此处传入`__gm__`声明的变量名。 | | ||
| 1247 | | src | 输入 | 源内存地址指针。 | | ||
| 1248 | | count | 输入 | 内存复制的长度,单位Byte。需满足 offset + count ≤ Device变量大小,Device变量大小可通过 [aclrtGetSymbolSize](#aclrtGetSymbolSize)接口查询获取。 | | ||
| 1249 | | offset | 输入 | Device变量地址偏移,单位Byte。需满足 offset + count ≤ Device变量大小,Device变量大小可通过 [aclrtGetSymbolSize](#aclrtGetSymbolSize)接口查询获取。 | | ||
| 1250 | | kind | 输入 | 拷贝类型,类型定义请参见[aclrtMemcpyKind](25_数据类型及其操作接口.md#aclrtMemcpyKind)。本接口仅支持ACL_MEMCPY_HOST_TO_DEVICE和ACL_MEMCPY_DEFAULT。 | | ||
| 1251 | |||
| 1252 | ### 返回值说明 | ||
| 1253 | |||
| 1254 | 返回0表示成功,返回其他值表示失败,请参见[aclError](25_数据类型及其操作接口.md#aclError)。 | ||
| 1255 | |||
| 1256 | ### 约束说明 | ||
| 1257 | |||
| 1258 | - 同[aclrtGetSymbolAddress](#aclrtGetSymbolAddress)接口约束说明。 | ||
| 1259 | |||
| 1260 | --- | ||
| 1261 | |||
| 1262 | <br> | ||
| 1263 | <br> | ||
| 1264 | <br> | ||
| 1265 | |||
| 1266 | <a id="aclrtMemcpyToSymbolAsync"></a> | ||
| 1267 | |||
| 1268 | ## aclrtMemcpyToSymbolAsync | ||
| 1269 | |||
| 1270 | ```cpp | ||
| 1271 | template <typename T> | ||
| 1272 | aclError aclrtMemcpyToSymbolAsync(const T &symbol, const void *src, size_t count, size_t offset, | ||
| 1273 | aclrtMemcpyKind kind, aclrtStream stream) | ||
| 1274 | ``` | ||
| 1275 | |||
| 1276 | ### 产品支持情况 | ||
| 1277 | |||
| 1278 | | 产品 | 是否支持 | | ||
| 1279 | | --- | :---: | | ||
| 1280 | | Ascend 950PR/Ascend 950DT | √ | | ||
| 1281 | | Atlas A3 训练系列产品/Atlas A3 推理系列产品 | √ | | ||
| 1282 | | Atlas A2 训练系列产品/Atlas A2 推理系列产品 | √ | | ||
| 1283 | |||
| 1284 | ### 功能说明 | ||
| 1285 | |||
| 1286 | 实现Host数据到Device变量的异步内存复制。用于向Device变量写入数据。 | ||
| 1287 | |||
| 1288 | 本接口中的Host内存支持锁页内存(例如通过aclrtMallocHost接口申请的内存)、非锁页内存(通过malloc接口申请的内存)。当Host内存是锁页内存时,本接口是异步接口,调用接口成功仅表示任务下发成功,不表示任务执行成功,调用本接口后,需调用同步等待接口(例如,[aclrtSynchronizeStream](06_Stream管理.md#aclrtSynchronizeStream))确保内存复制的任务已执行完成;当Host内存是非锁页内存时,本接口在内存复制任务完成后才返回。 | ||
| 1289 | |||
| 1290 | ### 参数说明 | ||
| 1291 | |||
| 1292 | | 参数名 | 输入/输出 | 说明 | | ||
| 1293 | | --- | :---: | --- | | ||
| 1294 | | symbol | 输入 | Device变量名。此处传入`__gm__`声明的变量名。 | | ||
| 1295 | | src | 输入 | 源内存地址指针。 | | ||
| 1296 | | count | 输入 | 内存复制的长度,单位Byte。需满足 offset + count ≤ Device变量大小,Device变量大小可通过 [aclrtGetSymbolSize](#aclrtGetSymbolSize)接口查询获取。 | | ||
| 1297 | | offset | 输入 | Device变量地址偏移,单位Byte。需满足 offset + count ≤ Device变量大小,Device变量大小可通过 [aclrtGetSymbolSize](#aclrtGetSymbolSize)接口查询获取。 | | ||
| 1298 | | kind | 输入 | 拷贝类型,类型定义请参见[aclrtMemcpyKind](25_数据类型及其操作接口.md#aclrtMemcpyKind)。本接口仅支持ACL_MEMCPY_HOST_TO_DEVICE和ACL_MEMCPY_DEFAULT。 | | ||
| 1299 | | stream | 输入 | 指定执行内存复制任务的Stream。类型定义请参见[aclrtStream](25_数据类型及其操作接口.md#aclrtStream)。 | | ||
| 1300 | |||
| 1301 | ### 返回值说明 | ||
| 1302 | |||
| 1303 | 返回0表示成功,返回其他值表示失败,请参见[aclError](25_数据类型及其操作接口.md#aclError)。 | ||
| 1304 | |||
| 1305 | ### 约束说明 | ||
| 1306 | |||
| 1307 | - 同[aclrtGetSymbolAddress](#aclrtGetSymbolAddress)接口约束说明。 | ||
| 1308 | - 本接口为异步接口,调用后需同步等待拷贝完成。 | ||
| @@ -21,6 +21,7 @@ | |||
| 21 | - ### [11-08 自定义内存分配器](11-08_自定义内存分配器.md) | 21 | - ### [11-08 自定义内存分配器](11-08_自定义内存分配器.md) |
| 22 | - ### [11-09 流内存操作](11-09_流内存操作.md) | 22 | - ### [11-09 流内存操作](11-09_流内存操作.md) |
| 23 | - ### [11-10 Stream有序内存分配](11-10_Stream有序内存分配.md) | 23 | - ### [11-10 Stream有序内存分配](11-10_Stream有序内存分配.md) |
| 24 | - ### [11-11 Device变量内存操作](11-11_Device变量内存操作.md) | ||
| 24 | - ## [12. 执行控制](12_执行控制.md) | 25 | - ## [12. 执行控制](12_执行控制.md) |
| 25 | - ## [13. 异常处理](13_异常处理.md) | 26 | - ## [13. 异常处理](13_异常处理.md) |
| 26 | - ## [14. Kernel加载与执行](14_Kernel加载与执行.md) | 27 | - ## [14. Kernel加载与执行](14_Kernel加载与执行.md) |
| @@ -42,4 +43,4 @@ | |||
| 42 | - ## [23. 日志接口](23_日志接口.md) | 43 | - ## [23. 日志接口](23_日志接口.md) |
| 43 | - ## [24. 其他接口](24_其他接口.md) | 44 | - ## [24. 其他接口](24_其他接口.md) |
| 44 | - ## [25. 数据类型及其操作接口](25_数据类型及其操作接口.md) | 45 | - ## [25. 数据类型及其操作接口](25_数据类型及其操作接口.md) |
| 45 | - ## [26. C++ API Routines](26_C++ API Routines.md) | 46 | - ## [26. C++扩展接口](26_C++扩展接口.md) |
| @@ -61,6 +61,7 @@ extern "C" { | |||
| 61 | 61 | ||
| 62 | 62 | ||
| 63 | 63 | ||
| 64 | |||
| 64 | 65 | ||
| 65 | 66 | ||
| 66 | 67 | ||
| @@ -449,6 +449,19 @@ RTS_API rtError_t rtFusionLaunch(void * const fusionInfo, rtStream_t const stm, | |||
| 449 | */ | 449 | */ |
| 450 | RTS_API rtError_t rtNopTask(rtStream_t stm); | 450 | RTS_API rtError_t rtNopTask(rtStream_t stm); |
| 451 | 451 | ||
| 452 | /** | ||
| 453 | * @ingroup rt_kernel | ||
| 454 | * @brief register device variable symbol mapping relationship. | ||
| 455 | * @param [in] binHandle binary handle (Program pointer). | ||
| 456 | * @param [in] hostVar device symbol reference. | ||
| 457 | * @param [in] deviceVarName device symbol name. | ||
| 458 | * @param [in] size size of object associated with symbol. | ||
| 459 | * @param [in] flags symbol attribute flags (bit0: ext; bit1: global). | ||
| 460 | * @param [in] reserve reserved parameter. | ||
| 461 | */ | ||
| 462 | RTS_API void rtRegisterVariable(void *binHandle, const void *hostVar, const char *deviceVarName, | ||
| 463 | size_t size, uint32_t flags, void *reserve); | ||
| 464 | |||
| 452 | 465 | ||
| 453 | } | 466 | } |
| 454 | 467 | ||
| @@ -89,6 +89,18 @@ RTS_API rtError_t rtFunctionGetParamInfo(const void *func, size_t paramIndex, | |||
| 89 | */ | 89 | */ |
| 90 | RTS_API rtError_t rtFunctionGetAvailDynUbufPerBlock(void *func, uint32_t flags, size_t *dynamicUbufSize); | 90 | RTS_API rtError_t rtFunctionGetAvailDynUbufPerBlock(void *func, uint32_t flags, size_t *dynamicUbufSize); |
| 91 | 91 | ||
| 92 | /** | ||
| 93 | * @ingroup rt_kernel | ||
| 94 | * @brief find the device address and size associated with symbol. | ||
| 95 | * @param [in] hostVar device symbol reference. | ||
| 96 | * @param [out] devPtr device pointer associated with symbol. | ||
| 97 | * @param [out] size size of object associated with symbol. | ||
| 98 | * @retval ACL_RT_SUCCESS for ok. | ||
| 99 | * @retval RT_ERROR_INVALID_SYMBOL invalid device symbol. | ||
| 100 | * @retval ACL_ERROR_RT_PARAM_INVALID for error input. | ||
| 101 | */ | ||
| 102 | RTS_API rtError_t rtSymbolLookup(const void *hostVar, void **devPtr, size_t *size); | ||
| 103 | |||
| 92 | /** | 104 | /** |
| 93 | * @ingroup rts_kernel | 105 | * @ingroup rts_kernel |
| 94 | * @brief Get global symbol address and size from binary. | 106 | * @brief Get global symbol address and size from binary. |
| @@ -163,6 +163,10 @@ public: | |||
| 163 | virtual rtError_t DependencyRegister(Program * const mProgram, Program * const sProgram) = 0; | 163 | virtual rtError_t DependencyRegister(Program * const mProgram, Program * const sProgram) = 0; |
| 164 | virtual rtError_t FunctionRegister(Program * const prog, const void * const stubFunc, | 164 | virtual rtError_t FunctionRegister(Program * const prog, const void * const stubFunc, |
| 165 | const char_t * const stubName, const void * const kernelInfoExt, const uint32_t funcMode) = 0; | 165 | const char_t * const stubName, const void * const kernelInfoExt, const uint32_t funcMode) = 0; |
| 166 | virtual rtError_t RegisterVariable(void * const binHandle, const void * const hostVar, | ||
| 167 | const char_t * const deviceVarName, const size_t size, const uint32_t flags) = 0; | ||
| 168 | virtual rtError_t SymbolLookup(const void * const hostVar, void ** const devPtr, | ||
| 169 | size_t * const size) = 0; | ||
| 166 | virtual rtError_t GetFunctionByName(const char_t * const stubName, void ** const stubFunc) = 0; | 170 | virtual rtError_t GetFunctionByName(const char_t * const stubName, void ** const stubFunc) = 0; |
| 167 | virtual rtError_t GetAddrByFun(const void * const stubFunc, void ** const addr) = 0; | 171 | virtual rtError_t GetAddrByFun(const void * const stubFunc, void ** const addr) = 0; |
| 168 | virtual rtError_t GetAddrAndPrefCntWithHandle(void * const hdl, const void * const kernelInfoExt, | 172 | virtual rtError_t GetAddrAndPrefCntWithHandle(void * const hdl, const void * const kernelInfoExt, |
| @@ -1393,6 +1393,32 @@ rtError_t rtBinaryGetGlobal(const rtBinHandle binHandle, const char *name, void | |||
| 1393 | return ACL_RT_SUCCESS; | 1393 | return ACL_RT_SUCCESS; |
| 1394 | } | 1394 | } |
| 1395 | 1395 | ||
| 1396 | VISIBILITY_DEFAULT | ||
| 1397 | void rtRegisterVariable(void *binHandle, const void *hostVar, const char *deviceVarName, | ||
H | |||
| 1398 | size_t size, uint32_t flags, void *reserve) | ||
| 1399 | { | ||
| 1400 | UNUSED(reserve); | ||
| 1401 | Api * const apiInstance = Api::Instance(); | ||
| 1402 | if (unlikely(apiInstance == nullptr)) { | ||
| 1403 | RT_LOG(RT_LOG_ERROR, "Null apiInstance pointer"); | ||
| 1404 | return; | ||
| 1405 | } | ||
| 1406 | |||
| 1407 | (void)apiInstance->RegisterVariable(binHandle, hostVar, deviceVarName, size, flags); | ||
| 1408 | return; | ||
| 1409 | } | ||
| 1410 | |||
| 1411 | VISIBILITY_DEFAULT | ||
| 1412 | rtError_t rtSymbolLookup(const void *hostVar, void **devPtr, size_t *size) | ||
| 1413 | { | ||
| 1414 | Api * const apiInstance = Api::Instance(); | ||
| 1415 | NULL_RETURN_ERROR_WITH_EXT_ERRCODE(apiInstance); | ||
| 1416 | |||
| 1417 | const rtError_t error = apiInstance->SymbolLookup(hostVar, devPtr, size); | ||
| 1418 | ERROR_RETURN_WITH_EXT_ERRCODE(error); | ||
| 1419 | return ACL_RT_SUCCESS; | ||
| 1420 | } | ||
| 1421 | |||
| 1396 | VISIBILITY_DEFAULT | 1422 | VISIBILITY_DEFAULT |
| 1397 | rtError_t rtFunctionGetParamCount(const void *func, size_t *paramCount) | 1423 | rtError_t rtFunctionGetParamCount(const void *func, size_t *paramCount) |
| 1398 | { | 1424 | { |
| @@ -1254,6 +1254,27 @@ rtError_t rtMemManagedPrefetchBatchAsync(const void** ptrs, size_t* sizes, size_ | |||
| 1254 | return ACL_ERROR_RT_FEATURE_NOT_SUPPORT; | 1254 | return ACL_ERROR_RT_FEATURE_NOT_SUPPORT; |
| 1255 | } | 1255 | } |
| 1256 | 1256 | ||
| 1257 | VISIBILITY_DEFAULT | ||
| 1258 | void rtRegisterVariable(void *binHandle, const void *hostVar, const char *deviceVarName, | ||
| 1259 | size_t size, uint32_t flags, void *reserve) | ||
| 1260 | { | ||
| 1261 | (void)binHandle; | ||
| 1262 | (void)hostVar; | ||
| 1263 | (void)deviceVarName; | ||
| 1264 | (void)size; | ||
| 1265 | (void)flags; | ||
| 1266 | (void)reserve; | ||
| 1267 | } | ||
| 1268 | |||
| 1269 | VISIBILITY_DEFAULT | ||
| 1270 | rtError_t rtSymbolLookup(const void *hostVar, void **devPtr, size_t *size) | ||
| 1271 | { | ||
| 1272 | (void)hostVar; | ||
| 1273 | (void)devPtr; | ||
| 1274 | (void)size; | ||
| 1275 | return ACL_ERROR_RT_FEATURE_NOT_SUPPORT; | ||
| 1276 | } | ||
| 1277 | |||
| 1257 | 1278 | ||
| 1258 | VISIBILITY_DEFAULT | 1279 | VISIBILITY_DEFAULT |
| 1259 | rtError_t rtBinaryGetGlobal(const rtBinHandle binHandle, const char *name, void **dptr, size_t *size) | 1280 | rtError_t rtBinaryGetGlobal(const rtBinHandle binHandle, const char *name, void **dptr, size_t *size) |
| @@ -386,6 +386,7 @@ set(libruntime_cmodel_src_files | |||
| 386 | ${RUNTIME_CORE_DIR}/src/kernel/module.cc | 386 | ${RUNTIME_CORE_DIR}/src/kernel/module.cc |
| 387 | ${RUNTIME_CORE_DIR}/src/kernel/program.cc | 387 | ${RUNTIME_CORE_DIR}/src/kernel/program.cc |
| 388 | ${RUNTIME_CORE_DIR}/src/kernel/program_common.cc | 388 | ${RUNTIME_CORE_DIR}/src/kernel/program_common.cc |
| 389 | ${RUNTIME_CORE_DIR}/src/kernel/symbol_table.cc | ||
| 389 | ${RUNTIME_CORE_DIR}/src/kernel/kernel_utils.cc | 390 | ${RUNTIME_CORE_DIR}/src/kernel/kernel_utils.cc |
| 390 | ${RUNTIME_FEATURE_DIR}/soma/soma.cc | 391 | ${RUNTIME_FEATURE_DIR}/soma/soma.cc |
| 391 | ${RUNTIME_FEATURE_DIR}/soma/stream_mem_pool.cc | 392 | ${RUNTIME_FEATURE_DIR}/soma/stream_mem_pool.cc |
| @@ -509,6 +510,7 @@ set(libruntime_cmodel_v200_src_files | |||
| 509 | ${RUNTIME_CORE_DIR}/src/kernel/module.cc | 510 | ${RUNTIME_CORE_DIR}/src/kernel/module.cc |
| 510 | ${RUNTIME_CORE_DIR}/src/kernel/program.cc | 511 | ${RUNTIME_CORE_DIR}/src/kernel/program.cc |
| 511 | ${RUNTIME_CORE_DIR}/src/kernel/program_common.cc | 512 | ${RUNTIME_CORE_DIR}/src/kernel/program_common.cc |
| 513 | ${RUNTIME_CORE_DIR}/src/kernel/symbol_table.cc | ||
| 512 | ${RUNTIME_CORE_DIR}/src/kernel/kernel_utils.cc | 514 | ${RUNTIME_CORE_DIR}/src/kernel/kernel_utils.cc |
| 513 | ${RUNTIME_FEATURE_DIR}/soma/soma.cc | 515 | ${RUNTIME_FEATURE_DIR}/soma/soma.cc |
| 514 | ${RUNTIME_FEATURE_DIR}/soma/stream_mem_pool.cc | 516 | ${RUNTIME_FEATURE_DIR}/soma/stream_mem_pool.cc |
| @@ -876,4 +878,4 @@ foreach(product_type ${PRODUCT_TYPE_LIST}) | |||
| 876 | ) | 878 | ) |
| 877 | 879 | ||
| 878 | endforeach() | 880 | endforeach() |
| 879 | endif() | 881 | endif() |
| @@ -286,6 +286,7 @@ set(libruntime_v100_src_files | |||
| 286 | ${RUNTIME_CORE_DIR}/src/kernel/module.cc | 286 | ${RUNTIME_CORE_DIR}/src/kernel/module.cc |
| 287 | ${RUNTIME_CORE_DIR}/src/kernel/program.cc | 287 | ${RUNTIME_CORE_DIR}/src/kernel/program.cc |
| 288 | ${RUNTIME_CORE_DIR}/src/kernel/program_common.cc | 288 | ${RUNTIME_CORE_DIR}/src/kernel/program_common.cc |
| 289 | ${RUNTIME_CORE_DIR}/src/kernel/symbol_table.cc | ||
| 289 | ${RUNTIME_CORE_DIR}/src/kernel/kernel_utils.cc | 290 | ${RUNTIME_CORE_DIR}/src/kernel/kernel_utils.cc |
| 290 | ${RUNTIME_FEATURE_DIR}/soma/soma.cc | 291 | ${RUNTIME_FEATURE_DIR}/soma/soma.cc |
| 291 | ${RUNTIME_FEATURE_DIR}/soma/stream_mem_pool.cc | 292 | ${RUNTIME_FEATURE_DIR}/soma/stream_mem_pool.cc |
| @@ -173,6 +173,7 @@ set(libruntime_src_files_include_for_tiny | |||
| 173 | ${RUNTIME_DIR}/src/runtime/driver/npu_driver_tiny_stub.cc | 173 | ${RUNTIME_DIR}/src/runtime/driver/npu_driver_tiny_stub.cc |
| 174 | ${RUNTIME_CORE_DIR}/src/engine/engine_factory_tiny_stub.cc | 174 | ${RUNTIME_CORE_DIR}/src/engine/engine_factory_tiny_stub.cc |
| 175 | ${RUNTIME_CORE_DIR}/src/kernel/binary_loader_tiny_stub.cc | 175 | ${RUNTIME_CORE_DIR}/src/kernel/binary_loader_tiny_stub.cc |
| 176 | ${RUNTIME_CORE_DIR}/src/kernel/symbol_table_tiny_stub.cc | ||
| 176 | ${RUNTIME_CORE_DIR}/src/task/tiny/rdma_task_tiny_stub.cc | 177 | ${RUNTIME_CORE_DIR}/src/task/tiny/rdma_task_tiny_stub.cc |
| 177 | ${RUNTIME_FEATURE_DIR}/ffts/ffts_task_tiny_stub.cc | 178 | ${RUNTIME_FEATURE_DIR}/ffts/ffts_task_tiny_stub.cc |
| 178 | ${RUNTIME_CORE_DIR}/src/task/tiny/task_tiny_stub.cc | 179 | ${RUNTIME_CORE_DIR}/src/task/tiny/task_tiny_stub.cc |
| @@ -307,6 +307,7 @@ set(libruntime_v200_v201_common_src_files | |||
| 307 | ${RUNTIME_CORE_DIR}/src/kernel/module.cc | 307 | ${RUNTIME_CORE_DIR}/src/kernel/module.cc |
| 308 | ${RUNTIME_CORE_DIR}/src/kernel/program.cc | 308 | ${RUNTIME_CORE_DIR}/src/kernel/program.cc |
| 309 | ${RUNTIME_CORE_DIR}/src/kernel/program_common.cc | 309 | ${RUNTIME_CORE_DIR}/src/kernel/program_common.cc |
| 310 | ${RUNTIME_CORE_DIR}/src/kernel/symbol_table.cc | ||
| 310 | ${RUNTIME_CORE_DIR}/src/kernel/kernel_utils.cc | 311 | ${RUNTIME_CORE_DIR}/src/kernel/kernel_utils.cc |
| 311 | ${RUNTIME_FEATURE_DIR}/soma/soma.cc | 312 | ${RUNTIME_FEATURE_DIR}/soma/soma.cc |
| 312 | ${RUNTIME_FEATURE_DIR}/soma/stream_mem_pool.cc | 313 | ${RUNTIME_FEATURE_DIR}/soma/stream_mem_pool.cc |
| @@ -219,6 +219,7 @@ constexpr int32_t RT_ERROR_KERNEL_DUPLICATE = 0x07080007; | |||
| 219 | constexpr int32_t RT_ERROR_KERNEL_UNREGISTERING = 0x07080008; | 219 | constexpr int32_t RT_ERROR_KERNEL_UNREGISTERING = 0x07080008; |
| 220 | constexpr int32_t RT_ERROR_KERNEL_INVALID = 0x07080009; | 220 | constexpr int32_t RT_ERROR_KERNEL_INVALID = 0x07080009; |
| 221 | constexpr int32_t RT_ERROR_SYMBOL_NOT_FOUND = 0x0708000A; | 221 | constexpr int32_t RT_ERROR_SYMBOL_NOT_FOUND = 0x0708000A; |
| 222 | constexpr int32_t RT_ERROR_INVALID_SYMBOL = 0x0708000B; | ||
| 222 | 223 | ||
| 223 | constexpr int32_t RT_ERROR_PROGRAM_BASE = 0x07090000; | 224 | constexpr int32_t RT_ERROR_PROGRAM_BASE = 0x07090000; |
| 224 | constexpr int32_t RT_ERROR_PROGRAM_NULL = 0x07090001; | 225 | constexpr int32_t RT_ERROR_PROGRAM_NULL = 0x07090001; |
| @@ -0,0 +1,48 @@ | |||
| 1 | /** | ||
| 2 | * Copyright (c) 2026 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 | |||
| 13 | |||
| 14 | |||
| 15 | |||
| 16 | |||
| 17 | |||
| 18 | namespace cce { | ||
| 19 | namespace runtime { | ||
| 20 | |||
| 21 | struct SymbolEntry { | ||
| 22 | void *binHandle; | ||
| 23 | const void *hostVar; | ||
| 24 | const char *deviceVarName; | ||
| 25 | uint32_t deviceSymbolOffset; | ||
| 26 | uint32_t deviceSymbolSize; | ||
| 27 | uint32_t flags; | ||
| 28 | }; | ||
| 29 | |||
| 30 | class SymbolTable : public NoCopy { | ||
| 31 | public: | ||
| 32 | SymbolTable() = default; | ||
| 33 | |||
| 34 | rtError_t Register(void *binHandle, const void *hostVar, const char *deviceVarName, | ||
| 35 | size_t size, uint32_t flags); | ||
| 36 | rtError_t Lookup(const void *hostVar, SymbolEntry &entry); | ||
| 37 | rtError_t GetDeviceAddress(const void *hostVar, uint32_t deviceId, void **devPtr); | ||
| 38 | rtError_t GetSize(const void *hostVar, size_t *size); | ||
| 39 | |||
| 40 | private: | ||
| 41 | std::map<const void *, SymbolEntry> symbolMap_; | ||
| 42 | SpinLock symbolMapLock_; | ||
| 43 | }; | ||
| 44 | |||
| 45 | } // namespace runtime | ||
| 46 | } // namespace cce | ||
| 47 | |||
| 48 | |||
| @@ -14,6 +14,7 @@ | |||
| 14 | 14 | ||
| 15 | 15 | ||
| 16 | 16 | ||
| 17 | |||
| 17 | 18 | ||
| 18 | 19 | ||
| 19 | 20 | ||
| @@ -226,6 +227,10 @@ public: | |||
| 226 | return xpuCtxt_; | 227 | return xpuCtxt_; |
| 227 | } | 228 | } |
| 228 | 229 | ||
| 230 | SymbolTable &GetSymbolTable() { | ||
| 231 | return symbolTable_; | ||
| 232 | } | ||
| 233 | |||
| 229 | rtError_t MallocProgramAndReg(const rtDevBinary_t *const bin, Program **const newProg) const; | 234 | rtError_t MallocProgramAndReg(const rtDevBinary_t *const bin, Program **const newProg) const; |
| 230 | rtKernelAttrType Magic2KernelAttrType(const uint32_t magic) const; | 235 | rtKernelAttrType Magic2KernelAttrType(const uint32_t magic) const; |
| 231 | rtKernelAttrType GetDefaultKernelAttrType(void) const; | 236 | rtKernelAttrType GetDefaultKernelAttrType(void) const; |
| @@ -243,6 +248,7 @@ public: | |||
| 243 | const void * const kernelInfoExt, const uint32_t funcMode, const char_t *kernelName); | 248 | const void * const kernelInfoExt, const uint32_t funcMode, const char_t *kernelName); |
| 244 | const Kernel *KernelLookup(const void * const stub); | 249 | const Kernel *KernelLookup(const void * const stub); |
| 245 | const void *StubFuncLookup(const char_t * const name); | 250 | const void *StubFuncLookup(const char_t * const name); |
| 251 | |||
| 246 | rtError_t LookupAddrByFun(const void * const stubFunc, Context * const ctx, void ** const addr); | 252 | rtError_t LookupAddrByFun(const void * const stubFunc, Context * const ctx, void ** const addr); |
| 247 | rtError_t LookupAddrAndPrefCntWithHandle(const void * const handlePtr, const void * const kernelInfoExt, | 253 | rtError_t LookupAddrAndPrefCntWithHandle(const void * const handlePtr, const void * const kernelInfoExt, |
| 248 | Context * const ctx, void ** const addr, uint32_t * const prefetchCnt); | 254 | Context * const ctx, void ** const addr, uint32_t * const prefetchCnt); |
| @@ -691,6 +697,7 @@ public: | |||
| 691 | 697 | ||
| 692 | DriverFactory driverFactory_; | 698 | DriverFactory driverFactory_; |
| 693 | KernelTable kernelTable_; | 699 | KernelTable kernelTable_; |
| 700 | SymbolTable symbolTable_; | ||
| 694 | FacadeDriver facadeDriver_; | 701 | FacadeDriver facadeDriver_; |
| 695 | rtErrorCallback excptCallBack_; | 702 | rtErrorCallback excptCallBack_; |
| 696 | uint32_t deviceCnt = 0U; | 703 | uint32_t deviceCnt = 0U; |
| @@ -80,6 +80,18 @@ rtError_t ApiDecorator::FunctionRegister(Program * const prog, const void * cons | |||
| 80 | return impl_->FunctionRegister(prog, stubFunc, stubName, kernelInfoExt, funcMode); | 80 | return impl_->FunctionRegister(prog, stubFunc, stubName, kernelInfoExt, funcMode); |
| 81 | } | 81 | } |
| 82 | 82 | ||
| 83 | rtError_t ApiDecorator::RegisterVariable(void * const binHandle, const void * const hostVar, | ||
| 84 | const char_t * const deviceVarName, const size_t size, const uint32_t flags) | ||
| 85 | { | ||
| 86 | return impl_->RegisterVariable(binHandle, hostVar, deviceVarName, size, flags); | ||
| 87 | } | ||
| 88 | |||
| 89 | rtError_t ApiDecorator::SymbolLookup(const void * const hostVar, void ** const devPtr, | ||
| 90 | size_t * const size) | ||
| 91 | { | ||
| 92 | return impl_->SymbolLookup(hostVar, devPtr, size); | ||
| 93 | } | ||
| 94 | |||
| 83 | rtError_t ApiDecorator::GetFunctionByName(const char_t * const stubName, void ** const stubFunc) | 95 | rtError_t ApiDecorator::GetFunctionByName(const char_t * const stubName, void ** const stubFunc) |
| 84 | { | 96 | { |
| 85 | return impl_->GetFunctionByName(stubName, stubFunc); | 97 | return impl_->GetFunctionByName(stubName, stubFunc); |
| @@ -62,6 +62,10 @@ public: | |||
| 62 | rtError_t DependencyRegister(Program * const mProgram, Program * const sProgram) override; | 62 | rtError_t DependencyRegister(Program * const mProgram, Program * const sProgram) override; |
| 63 | rtError_t FunctionRegister(Program * const prog, const void * const stubFunc, const char_t * const stubName, | 63 | rtError_t FunctionRegister(Program * const prog, const void * const stubFunc, const char_t * const stubName, |
| 64 | const void * const kernelInfoExt, const uint32_t funcMode) override; | 64 | const void * const kernelInfoExt, const uint32_t funcMode) override; |
| 65 | rtError_t RegisterVariable(void * const binHandle, const void * const hostVar, | ||
| 66 | const char_t * const deviceVarName, const size_t size, const uint32_t flags) override; | ||
| 67 | rtError_t SymbolLookup(const void * const hostVar, void ** const devPtr, | ||
| 68 | size_t * const size) override; | ||
| 65 | rtError_t GetFunctionByName(const char_t * const stubName, void ** const stubFunc) override; | 69 | rtError_t GetFunctionByName(const char_t * const stubName, void ** const stubFunc) override; |
| 66 | rtError_t GetAddrByFun(const void * const stubFunc, void ** const addr) override; | 70 | rtError_t GetAddrByFun(const void * const stubFunc, void ** const addr) override; |
| 67 | rtError_t GetAddrAndPrefCntWithHandle(void * const hdl, const void * const kernelInfoExt, | 71 | rtError_t GetAddrAndPrefCntWithHandle(void * const hdl, const void * const kernelInfoExt, |
| @@ -119,6 +119,32 @@ rtError_t ApiErrorDecorator::FunctionRegister(Program * const prog, const void * | |||
| 119 | return error; | 119 | return error; |
| 120 | } | 120 | } |
| 121 | 121 | ||
| 122 | rtError_t ApiErrorDecorator::RegisterVariable(void * const binHandle, const void * const hostVar, | ||
| 123 | const char_t * const deviceVarName, const size_t size, const uint32_t flags) | ||
| 124 | { | ||
| 125 | NULL_PTR_RETURN_MSG_OUTER(binHandle, RT_ERROR_INVALID_VALUE); | ||
| 126 | NULL_PTR_RETURN_MSG_OUTER(hostVar, RT_ERROR_INVALID_VALUE); | ||
| 127 | NULL_PTR_RETURN_MSG_OUTER(deviceVarName, RT_ERROR_INVALID_VALUE); | ||
| 128 | |||
| 129 | const auto len = strnlen(deviceVarName, static_cast<size_t>(NAME_MAX_LENGTH)); | ||
| 130 | COND_RETURN_AND_MSG_OUTER_WITH_PARAM(len >= NAME_MAX_LENGTH, RT_ERROR_INVALID_VALUE, len, | ||
| 131 | "less than " + std::to_string(NAME_MAX_LENGTH)); | ||
| 132 | |||
| 133 | const rtError_t error = impl_->RegisterVariable(binHandle, hostVar, deviceVarName, size, flags); | ||
| 134 | ERROR_RETURN(error, "Register variable failed, hostVar=%p, deviceVarName=%s.", hostVar, deviceVarName); | ||
| 135 | return error; | ||
| 136 | } | ||
| 137 | |||
| 138 | rtError_t ApiErrorDecorator::SymbolLookup(const void * const hostVar, void ** const devPtr, | ||
| 139 | size_t * const size) | ||
| 140 | { | ||
| 141 | NULL_PTR_RETURN_MSG_OUTER(devPtr, RT_ERROR_INVALID_VALUE); | ||
| 142 | NULL_PTR_RETURN_MSG_OUTER(size, RT_ERROR_INVALID_VALUE); | ||
| 143 | const rtError_t error = impl_->SymbolLookup(hostVar, devPtr, size); | ||
| 144 | ERROR_RETURN(error, "Symbol lookup failed, hostVar=%p.", hostVar); | ||
| 145 | return error; | ||
| 146 | } | ||
| 147 | |||
| 122 | rtError_t ApiErrorDecorator::GetFunctionByName(const char_t * const stubName, void ** const stubFunc) | 148 | rtError_t ApiErrorDecorator::GetFunctionByName(const char_t * const stubName, void ** const stubFunc) |
| 123 | { | 149 | { |
| 124 | NULL_PTR_RETURN_MSG_OUTER(stubFunc, RT_ERROR_INVALID_VALUE); | 150 | NULL_PTR_RETURN_MSG_OUTER(stubFunc, RT_ERROR_INVALID_VALUE); |
| @@ -5277,7 +5303,7 @@ rtError_t ApiErrorDecorator::ImportFromShareableHandleV2(const void *shareableHa | |||
| 5277 | (handleType != RT_MEM_SHARE_HANDLE_TYPE_DEFAULT && handleType != RT_MEM_SHARE_HANDLE_TYPE_FABRIC), | 5303 | (handleType != RT_MEM_SHARE_HANDLE_TYPE_DEFAULT && handleType != RT_MEM_SHARE_HANDLE_TYPE_FABRIC), |
| 5278 | RT_ERROR_INVALID_VALUE, handleType, std::to_string(RT_MEM_SHARE_HANDLE_TYPE_DEFAULT) | 5304 | RT_ERROR_INVALID_VALUE, handleType, std::to_string(RT_MEM_SHARE_HANDLE_TYPE_DEFAULT) |
| 5279 | + " or " + std::to_string(RT_MEM_SHARE_HANDLE_TYPE_FABRIC)); | 5305 | + " or " + std::to_string(RT_MEM_SHARE_HANDLE_TYPE_FABRIC)); |
| 5280 | COND_RETURN_AND_MSG_OUTER_WITH_PARAM((flags != 0U), RT_ERROR_INVALID_VALUE, flags, "0"); | 5306 | COND_RETURN_AND_MSG_OUTER_WITH_PARAM((flags != 0U), RT_ERROR_INVALID_VALUE, flags, "0"); |
| 5281 | int32_t realDeviceId = 0; | 5307 | int32_t realDeviceId = 0; |
| 5282 | rtError_t error = Runtime::Instance()->ChgUserDevIdToDeviceId( | 5308 | rtError_t error = Runtime::Instance()->ChgUserDevIdToDeviceId( |
| 5283 | static_cast<uint32_t>(devId), RtPtrToPtr<uint32_t *>(&realDeviceId)); | 5309 | static_cast<uint32_t>(devId), RtPtrToPtr<uint32_t *>(&realDeviceId)); |
| @@ -6245,7 +6271,7 @@ rtError_t ApiErrorDecorator::FunctionGetAvailDynUbufPerBlock(Kernel *funcHandle, | |||
| 6245 | { | 6271 | { |
| 6246 | NULL_PTR_RETURN_MSG_OUTER(funcHandle, RT_ERROR_INVALID_VALUE); | 6272 | NULL_PTR_RETURN_MSG_OUTER(funcHandle, RT_ERROR_INVALID_VALUE); |
| 6247 | NULL_PTR_RETURN_MSG_OUTER(dynamicUbufSize, RT_ERROR_INVALID_VALUE); | 6273 | NULL_PTR_RETURN_MSG_OUTER(dynamicUbufSize, RT_ERROR_INVALID_VALUE); |
| 6248 | COND_RETURN_AND_MSG_OUTER_WITH_PARAM((flags != 0U), RT_ERROR_INVALID_VALUE, flags, "0"); | 6274 | COND_RETURN_AND_MSG_OUTER_WITH_PARAM((flags != 0U), RT_ERROR_INVALID_VALUE, flags, "0"); |
| 6249 | 6275 | ||
| 6250 | Program * const prog = funcHandle->Program_(); | 6276 | Program * const prog = funcHandle->Program_(); |
| 6251 | NULL_PTR_RETURN_MSG(prog, RT_ERROR_PROGRAM_NULL); | 6277 | NULL_PTR_RETURN_MSG(prog, RT_ERROR_PROGRAM_NULL); |
| @@ -60,6 +60,10 @@ public: | |||
| 60 | rtError_t DependencyRegister(Program * const mProgram, Program * const sProgram) override; | 60 | rtError_t DependencyRegister(Program * const mProgram, Program * const sProgram) override; |
| 61 | rtError_t FunctionRegister(Program * const prog, const void * const stubFunc, const char_t * const stubName, | 61 | rtError_t FunctionRegister(Program * const prog, const void * const stubFunc, const char_t * const stubName, |
| 62 | const void * const kernelInfoExt, const uint32_t funcMode) override; | 62 | const void * const kernelInfoExt, const uint32_t funcMode) override; |
| 63 | rtError_t RegisterVariable(void * const binHandle, const void * const hostVar, | ||
| 64 | const char_t * const deviceVarName, const size_t size, const uint32_t flags) override; | ||
| 65 | rtError_t SymbolLookup(const void * const hostVar, void ** const devPtr, | ||
| 66 | size_t * const size) override; | ||
| 63 | rtError_t GetFunctionByName(const char_t * const stubName, void ** const stubFunc) override; | 67 | rtError_t GetFunctionByName(const char_t * const stubName, void ** const stubFunc) override; |
| 64 | rtError_t GetAddrByFun(const void * const stubFunc, void ** const addr) override; | 68 | rtError_t GetAddrByFun(const void * const stubFunc, void ** const addr) override; |
| 65 | rtError_t GetAddrAndPrefCntWithHandle(void * const hdl, const void * const kernelInfoExt, | 69 | rtError_t GetAddrAndPrefCntWithHandle(void * const hdl, const void * const kernelInfoExt, |
| @@ -85,6 +85,7 @@ | |||
| 85 | 85 | ||
| 86 | 86 | ||
| 87 | 87 | ||
| 88 | |||
| 88 | 89 | ||
| 89 | 90 | ||
| 90 | 91 | ||
| @@ -412,6 +413,29 @@ rtError_t ApiImpl::FunctionRegister(Program * const prog, const void * const stu | |||
| 412 | return Runtime::Instance()->KernelRegister(prog, stubFunc, stubName, kernelInfoExt, funcMode); | 413 | return Runtime::Instance()->KernelRegister(prog, stubFunc, stubName, kernelInfoExt, funcMode); |
| 413 | } | 414 | } |
| 414 | 415 | ||
| 416 | rtError_t ApiImpl::RegisterVariable(void * const binHandle, const void * const hostVar, | ||
| 417 | const char_t * const deviceVarName, const size_t size, const uint32_t flags) | ||
| 418 | { | ||
| 419 | RT_LOG(RT_LOG_INFO, "register variable, hostVar=%p, deviceVarName=%s, size=%zu, flags=%u.", | ||
| 420 | hostVar, deviceVarName, size, flags); | ||
| 421 | return Runtime::Instance()->GetSymbolTable().Register(binHandle, hostVar, deviceVarName, size, flags); | ||
| 422 | } | ||
| 423 | |||
| 424 | rtError_t ApiImpl::SymbolLookup(const void * const hostVar, void ** const devPtr, | ||
| 425 | size_t * const size) | ||
| 426 | { | ||
| 427 | Context * const curCtx = CurrentContext(); | ||
| 428 | CHECK_CONTEXT_VALID_WITH_RETURN(curCtx, RT_ERROR_CONTEXT_NULL); | ||
| 429 | NULL_PTR_RETURN_MSG(curCtx->Device_(), RT_ERROR_DEVICE_NULL); | ||
| 430 | const uint32_t deviceId = static_cast<uint32_t>(curCtx->Device_()->Id_()); | ||
| 431 | |||
| 432 | RT_LOG(RT_LOG_DEBUG, "lookup symbol, hostVar=%p, device_id=%u.", hostVar, deviceId); | ||
| 433 | rtError_t error = Runtime::Instance()->GetSymbolTable().GetDeviceAddress(hostVar, deviceId, devPtr); | ||
| 434 | COND_RETURN_WITH_NOLOG(error != RT_ERROR_NONE, error); | ||
| 435 | error = Runtime::Instance()->GetSymbolTable().GetSize(hostVar, size); | ||
| 436 | return error; | ||
| 437 | } | ||
| 438 | |||
| 415 | rtError_t ApiImpl::GetFunctionByName(const char_t * const stubName, void ** const stubFunc) | 439 | rtError_t ApiImpl::GetFunctionByName(const char_t * const stubName, void ** const stubFunc) |
| 416 | { | 440 | { |
| 417 | RT_LOG(RT_LOG_DEBUG, "name=%s.", (stubName != nullptr) ? stubName : "(none)"); | 441 | RT_LOG(RT_LOG_DEBUG, "name=%s.", (stubName != nullptr) ? stubName : "(none)"); |
| @@ -70,6 +70,10 @@ public: | |||
| 70 | rtError_t DependencyRegister(Program * const mProgram, Program * const sProgram) override; | 70 | rtError_t DependencyRegister(Program * const mProgram, Program * const sProgram) override; |
| 71 | rtError_t FunctionRegister(Program * const prog, const void * const stubFunc, const char_t * const stubName, | 71 | rtError_t FunctionRegister(Program * const prog, const void * const stubFunc, const char_t * const stubName, |
| 72 | const void * const kernelInfoExt, const uint32_t funcMode) override; | 72 | const void * const kernelInfoExt, const uint32_t funcMode) override; |
| 73 | rtError_t RegisterVariable(void * const binHandle, const void * const hostVar, | ||
| 74 | const char_t * const deviceVarName, const size_t size, const uint32_t flags) override; | ||
| 75 | rtError_t SymbolLookup(const void * const hostVar, void ** const devPtr, | ||
| 76 | size_t * const size) override; | ||
| 73 | rtError_t GetFunctionByName(const char_t * const stubName, void ** const stubFunc) override; | 77 | rtError_t GetFunctionByName(const char_t * const stubName, void ** const stubFunc) override; |
| 74 | rtError_t GetAddrByFun(const void * const stubFunc, void ** const addr) override; | 78 | rtError_t GetAddrByFun(const void * const stubFunc, void ** const addr) override; |
| 75 | rtError_t GetAddrAndPrefCntWithHandle(void * const hdl, const void * const kernelInfoExt, | 79 | rtError_t GetAddrAndPrefCntWithHandle(void * const hdl, const void * const kernelInfoExt, |
| @@ -222,6 +222,7 @@ void ErrorcodeManage::InitRtErrCodeMap() | |||
| 222 | rtErrMap_[RT_ERROR_KERNEL_UNREGISTERING] = {ACL_ERROR_RT_KERNEL_UNREGISTERING, "kernel unregistering"}; | 222 | rtErrMap_[RT_ERROR_KERNEL_UNREGISTERING] = {ACL_ERROR_RT_KERNEL_UNREGISTERING, "kernel unregistering"}; |
| 223 | rtErrMap_[RT_ERROR_KERNEL_INVALID] = {ACL_ERROR_RT_INVALID_HANDLE, "invalid funcHandle"}; | 223 | rtErrMap_[RT_ERROR_KERNEL_INVALID] = {ACL_ERROR_RT_INVALID_HANDLE, "invalid funcHandle"}; |
| 224 | rtErrMap_[RT_ERROR_SYMBOL_NOT_FOUND] = {ACL_ERROR_RT_SYMBOL_NOT_FOUND, "global symbol not found"}; | 224 | rtErrMap_[RT_ERROR_SYMBOL_NOT_FOUND] = {ACL_ERROR_RT_SYMBOL_NOT_FOUND, "global symbol not found"}; |
| 225 | rtErrMap_[RT_ERROR_INVALID_SYMBOL] = {ACL_ERROR_RT_INVALID_SYMBOL, "invalid device symbol"}; | ||
| 225 | 226 | ||
| 226 | // program error | 227 | // program error |
| 227 | rtErrMap_[RT_ERROR_PROGRAM_BASE] = {ACL_ERROR_RT_INTERNAL_ERROR, "program base error"}; | 228 | rtErrMap_[RT_ERROR_PROGRAM_BASE] = {ACL_ERROR_RT_INTERNAL_ERROR, "program base error"}; |
| @@ -0,0 +1,104 @@ | |||
| 1 | /** | ||
| 2 | * Copyright (c) 2026 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 | |||
| 13 | |||
| 14 | |||
| 15 | |||
| 16 | |||
| 17 | namespace cce { | ||
| 18 | namespace runtime { | ||
| 19 | |||
| 20 | rtError_t SymbolTable::Register(void *binHandle, const void *hostVar, const char *deviceVarName, | ||
| 21 | size_t size, uint32_t flags) | ||
| 22 | { | ||
| 23 | Program *prog = static_cast<Program *>(binHandle); | ||
| 24 | ElfProgram * const elfProg = dynamic_cast<ElfProgram *>(prog); | ||
| 25 | COND_RETURN_ERROR_MSG_INNER(elfProg == nullptr, RT_ERROR_INVALID_VALUE, "can't dynamic_cast program."); | ||
| 26 | |||
| 27 | uint64_t symbolOffset = 0ULL; | ||
| 28 | uint64_t symbolSizeFromBin = 0ULL; | ||
| 29 | rtError_t ret = elfProg->GetGlobalSymbol(deviceVarName, &symbolOffset, &symbolSizeFromBin); | ||
| 30 | ERROR_RETURN(ret, "Get global symbol=%s failed, retCode=%#x.", deviceVarName, ret); | ||
| 31 | |||
| 32 | SymbolEntry entry = {}; | ||
| 33 | entry.binHandle = binHandle; | ||
| 34 | entry.hostVar = hostVar; | ||
| 35 | entry.deviceVarName = deviceVarName; | ||
| 36 | entry.deviceSymbolOffset = static_cast<uint32_t>(symbolOffset); | ||
| 37 | entry.deviceSymbolSize = static_cast<uint32_t>(size); | ||
| 38 | entry.flags = flags; | ||
| 39 | |||
| 40 | symbolMapLock_.Lock(); | ||
| 41 | auto it = symbolMap_.find(hostVar); | ||
| 42 | if (it != symbolMap_.end()) { | ||
| 43 | symbolMapLock_.Unlock(); | ||
| 44 | RT_LOG(RT_LOG_WARNING, "Symbol hostVar=%p, binHandle=%p already registered, skip duplicate registration.", hostVar, binHandle); | ||
| 45 | return RT_ERROR_NONE; | ||
| 46 | } | ||
| 47 | symbolMap_.emplace(hostVar, entry); | ||
| 48 | symbolMapLock_.Unlock(); | ||
| 49 | |||
| 50 | RT_LOG(RT_LOG_DEBUG, "Register symbol success, hostVar=%p, deviceVarName=%s, offset=%llu, size=%llu", | ||
| 51 | hostVar, deviceVarName, entry.deviceSymbolOffset, entry.deviceSymbolSize); | ||
| 52 | return RT_ERROR_NONE; | ||
| 53 | } | ||
| 54 | |||
| 55 | rtError_t SymbolTable::Lookup(const void *hostVar, SymbolEntry &entry) | ||
| 56 | { | ||
| 57 | symbolMapLock_.Lock(); | ||
| 58 | auto it = symbolMap_.find(hostVar); | ||
| 59 | if (it == symbolMap_.end()) { | ||
| 60 | symbolMapLock_.Unlock(); | ||
| 61 | RT_LOG_OUTER_MSG_IMPL(ErrorCode::EE1017, __func__, "symbol", | ||
| 62 | "The corresponding device variable cannot be found through the symbol"); | ||
| 63 | return RT_ERROR_INVALID_SYMBOL; | ||
| 64 | } | ||
| 65 | |||
| 66 | entry = it->second; | ||
| 67 | symbolMapLock_.Unlock(); | ||
| 68 | return RT_ERROR_NONE; | ||
| 69 | } | ||
| 70 | |||
| 71 | rtError_t SymbolTable::GetDeviceAddress(const void *hostVar, uint32_t deviceId, void **devPtr) | ||
| 72 | { | ||
| 73 | SymbolEntry entry = {}; | ||
| 74 | rtError_t ret = Lookup(hostVar, entry); | ||
| 75 | COND_RETURN_WITH_NOLOG(ret != RT_ERROR_NONE, ret); | ||
| 76 | |||
| 77 | ElfProgram * const elfProg = dynamic_cast<ElfProgram *>(static_cast<Program *>(entry.binHandle)); | ||
| 78 | COND_RETURN_ERROR_MSG_INNER(elfProg == nullptr, RT_ERROR_INVALID_VALUE, "can't dynamic_cast program."); | ||
| 79 | |||
| 80 | ret = elfProg->Load2Device(); | ||
| 81 | ERROR_RETURN(ret, "load program to device_id=%d failed, retCode=%#x", deviceId, ret); | ||
| 82 | |||
| 83 | const void *baseAddr = elfProg->GetBinAlignBaseAddr(deviceId); | ||
| 84 | COND_RETURN_ERROR_MSG_INNER(baseAddr == nullptr, RT_ERROR_PROGRAM_DATA, | ||
| 85 | "binary not loaded to device, deviceId=%u", deviceId); | ||
| 86 | |||
| 87 | *devPtr = RtValueToPtr<void *>(RtPtrToValue(baseAddr) + entry.deviceSymbolOffset); | ||
| 88 | RT_LOG(RT_LOG_DEBUG, "Get symbol device address, hostVar=%p, deviceId=%u, devPtr=%p", | ||
| 89 | hostVar, deviceId, *devPtr); | ||
| 90 | return RT_ERROR_NONE; | ||
| 91 | } | ||
| 92 | |||
| 93 | rtError_t SymbolTable::GetSize(const void *hostVar, size_t *size) | ||
| 94 | { | ||
| 95 | SymbolEntry entry = {}; | ||
| 96 | rtError_t ret = Lookup(hostVar, entry); | ||
| 97 | COND_RETURN_WITH_NOLOG(ret != RT_ERROR_NONE, ret); | ||
| 98 | |||
| 99 | *size = static_cast<size_t>(entry.deviceSymbolSize); | ||
| 100 | return RT_ERROR_NONE; | ||
| 101 | } | ||
| 102 | |||
| 103 | } // namespace runtime | ||
| 104 | } // namespace cce | ||
| @@ -0,0 +1,49 @@ | |||
| 1 | /** | ||
| 2 | * Copyright (c) 2026 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 | |||
| 13 | namespace cce { | ||
| 14 | namespace runtime { | ||
| 15 | rtError_t SymbolTable::Register(void *binHandle, const void *hostVar, const char *deviceVarName, | ||
| 16 | size_t size, uint32_t flags) | ||
| 17 | { | ||
| 18 | UNUSED(binHandle); | ||
| 19 | UNUSED(hostVar); | ||
| 20 | UNUSED(deviceVarName); | ||
| 21 | UNUSED(size); | ||
| 22 | UNUSED(flags); | ||
| 23 | return RT_ERROR_FEATURE_NOT_SUPPORT; | ||
| 24 | } | ||
| 25 | |||
| 26 | rtError_t SymbolTable::Lookup(const void *hostVar, SymbolEntry &entry) | ||
| 27 | { | ||
| 28 | UNUSED(hostVar); | ||
| 29 | UNUSED(entry); | ||
| 30 | return RT_ERROR_FEATURE_NOT_SUPPORT; | ||
| 31 | } | ||
| 32 | |||
| 33 | rtError_t SymbolTable::GetDeviceAddress(const void *hostVar, uint32_t deviceId, void **devPtr) | ||
| 34 | { | ||
| 35 | UNUSED(hostVar); | ||
| 36 | UNUSED(deviceId); | ||
| 37 | UNUSED(devPtr); | ||
| 38 | return RT_ERROR_FEATURE_NOT_SUPPORT; | ||
| 39 | } | ||
| 40 | |||
| 41 | rtError_t SymbolTable::GetSize(const void *hostVar, size_t *size) | ||
| 42 | { | ||
| 43 | UNUSED(hostVar); | ||
| 44 | UNUSED(size); | ||
| 45 | return RT_ERROR_FEATURE_NOT_SUPPORT; | ||
| 46 | } | ||
| 47 | |||
| 48 | } // namespace runtime | ||
| 49 | } // namespace cce | ||
| @@ -408,6 +408,7 @@ set(runtimeut_src_tiny_exclude_list | |||
| 408 | ${TOP_DIR}/src/runtime/core/src/task/task_info/event/notify_task_v200.cc | 408 | ${TOP_DIR}/src/runtime/core/src/task/task_info/event/notify_task_v200.cc |
| 409 | ${TOP_DIR}/src/runtime/core/src/memory/common_memset_d32.cpp | 409 | ${TOP_DIR}/src/runtime/core/src/memory/common_memset_d32.cpp |
| 410 | ${TOP_DIR}/src/runtime/core/src/memory/simd_memsetd32.cpp | 410 | ${TOP_DIR}/src/runtime/core/src/memory/simd_memsetd32.cpp |
| 411 | ${TOP_DIR}/src/runtime/core/src/kernel/symbol_table.cc | ||
| 411 | ) | 412 | ) |
| 412 | 413 | ||
| 413 | set(runtimeut_src_tiny_include_list | 414 | set(runtimeut_src_tiny_include_list |
| @@ -425,6 +426,7 @@ set(runtimeut_src_tiny_include_list | |||
| 425 | ${TOP_DIR}/src/runtime/core/src/engine/engine_factory_tiny_stub.cc | 426 | ${TOP_DIR}/src/runtime/core/src/engine/engine_factory_tiny_stub.cc |
| 426 | ${TOP_DIR}/src/runtime/core/src/kernel/binary_loader_tiny_stub.cc | 427 | ${TOP_DIR}/src/runtime/core/src/kernel/binary_loader_tiny_stub.cc |
| 427 | ${TOP_DIR}/src/runtime/core/src/kernel/program_common.cc | 428 | ${TOP_DIR}/src/runtime/core/src/kernel/program_common.cc |
| 429 | ${TOP_DIR}/src/runtime/core/src/kernel/symbol_table_tiny_stub.cc | ||
| 428 | ${TOP_DIR}/src/runtime/core/src/kernel/kernel_utils.cc | 430 | ${TOP_DIR}/src/runtime/core/src/kernel/kernel_utils.cc |
| 429 | ${TOP_DIR}/src/runtime/core/src/profiler/api_profile_log_decoratoc_tiny_stub.cc | 431 | ${TOP_DIR}/src/runtime/core/src/profiler/api_profile_log_decoratoc_tiny_stub.cc |
| 430 | ${TOP_DIR}/src/runtime/core/src/profiler/api_profile_decorator_tiny_stub.cc | 432 | ${TOP_DIR}/src/runtime/core/src/profiler/api_profile_decorator_tiny_stub.cc |
| @@ -504,6 +506,7 @@ set(runtimeut_src_list | |||
| 504 | ${TOP_DIR}/src/runtime/core/src/kernel/program_common.cc | 506 | ${TOP_DIR}/src/runtime/core/src/kernel/program_common.cc |
| 505 | ${TOP_DIR}/src/runtime/core/src/kernel/v100/program_plat.cc | 507 | ${TOP_DIR}/src/runtime/core/src/kernel/v100/program_plat.cc |
| 506 | ${TOP_DIR}/src/runtime/core/src/kernel/program.cc | 508 | ${TOP_DIR}/src/runtime/core/src/kernel/program.cc |
| 509 | ${TOP_DIR}/src/runtime/core/src/kernel/symbol_table.cc | ||
| 507 | ${TOP_DIR}/src/runtime/core/src/kernel/kernel_utils.cc | 510 | ${TOP_DIR}/src/runtime/core/src/kernel/kernel_utils.cc |
| 508 | ${TOP_DIR}/src/runtime/feature/soma/soma.cc | 511 | ${TOP_DIR}/src/runtime/feature/soma/soma.cc |
| 509 | ${TOP_DIR}/src/runtime/feature/soma/stream_mem_pool.cc | 512 | ${TOP_DIR}/src/runtime/feature/soma/stream_mem_pool.cc |
| @@ -669,6 +672,7 @@ set(runtimeut_david_src_list | |||
| 669 | ${TOP_DIR}/src/runtime/core/src/kernel/kernel.cc | 672 | ${TOP_DIR}/src/runtime/core/src/kernel/kernel.cc |
| 670 | ${TOP_DIR}/src/runtime/core/src/kernel/module.cc | 673 | ${TOP_DIR}/src/runtime/core/src/kernel/module.cc |
| 671 | ${TOP_DIR}/src/runtime/core/src/kernel/program.cc | 674 | ${TOP_DIR}/src/runtime/core/src/kernel/program.cc |
| 675 | ${TOP_DIR}/src/runtime/core/src/kernel/symbol_table.cc | ||
| 672 | ${TOP_DIR}/src/runtime/feature/soma/soma.cc | 676 | ${TOP_DIR}/src/runtime/feature/soma/soma.cc |
| 673 | ${TOP_DIR}/src/runtime/feature/soma/stream_mem_pool.cc | 677 | ${TOP_DIR}/src/runtime/feature/soma/stream_mem_pool.cc |
| 674 | ${TOP_DIR}/src/runtime/core/src/launch/label.cc | 678 | ${TOP_DIR}/src/runtime/core/src/launch/label.cc |
| @@ -727,6 +731,7 @@ set(runtimeut_david_src_list | |||
| 727 | ${TOP_DIR}/src/runtime/core/src/runtime_v200/runtime_adapt.cc | 731 | ${TOP_DIR}/src/runtime/core/src/runtime_v200/runtime_adapt.cc |
| 728 | ${TOP_DIR}/src/runtime/feature/xpu/program_plat.cc | 732 | ${TOP_DIR}/src/runtime/feature/xpu/program_plat.cc |
| 729 | ${TOP_DIR}/src/runtime/core/src/kernel/program.cc | 733 | ${TOP_DIR}/src/runtime/core/src/kernel/program.cc |
| 734 | ${TOP_DIR}/src/runtime/core/src/kernel/symbol_table.cc | ||
| 730 | ${TOP_DIR}/src/runtime/core/src/kernel/program_common.cc | 735 | ${TOP_DIR}/src/runtime/core/src/kernel/program_common.cc |
| 731 | ${TOP_DIR}/src/runtime/core/src/kernel/kernel_utils.cc | 736 | ${TOP_DIR}/src/runtime/core/src/kernel/kernel_utils.cc |
| 732 | ${TOP_DIR}/src/runtime/core/src/kernel/json_parse.cc | 737 | ${TOP_DIR}/src/runtime/core/src/kernel/json_parse.cc |
| @@ -843,6 +848,7 @@ add_executable(runtime_utest | |||
| 843 | test/rt_utest_module.cc | 848 | test/rt_utest_module.cc |
| 844 | test/rt_utest_elf.cc | 849 | test/rt_utest_elf.cc |
| 845 | test/rt_utest_program.cc | 850 | test/rt_utest_program.cc |
| 851 | test/rt_utest_symbol.cc | ||
| 846 | test/rt_utest_runtime.cc | 852 | test/rt_utest_runtime.cc |
| 847 | test/rt_utest_kernel.cc | 853 | test/rt_utest_kernel.cc |
| 848 | test/rt_utest_config.cc | 854 | test/rt_utest_config.cc |
| @@ -1505,6 +1511,7 @@ set(runtimeut_v201_src_list | |||
| 1505 | ${TOP_DIR}/src/runtime/core/src/kernel/module.cc | 1511 | ${TOP_DIR}/src/runtime/core/src/kernel/module.cc |
| 1506 | ${TOP_DIR}/src/runtime/core/src/kernel/kernel.cc | 1512 | ${TOP_DIR}/src/runtime/core/src/kernel/kernel.cc |
| 1507 | ${TOP_DIR}/src/runtime/core/src/kernel/program.cc | 1513 | ${TOP_DIR}/src/runtime/core/src/kernel/program.cc |
| 1514 | ${TOP_DIR}/src/runtime/core/src/kernel/symbol_table.cc | ||
| 1508 | ${TOP_DIR}/src/runtime/core/src/runtime_v200/runtime_adapt.cc | 1515 | ${TOP_DIR}/src/runtime/core/src/runtime_v200/runtime_adapt.cc |
| 1509 | ${TOP_DIR}/src/runtime/feature/xpu/program_plat.cc | 1516 | ${TOP_DIR}/src/runtime/feature/xpu/program_plat.cc |
| 1510 | ${TOP_DIR}/src/runtime/feature/soma/soma.cc | 1517 | ${TOP_DIR}/src/runtime/feature/soma/soma.cc |
| @@ -279,6 +279,7 @@ set(runtimeut_910b_src_list | |||
| 279 | ${TOP_DIR}/src/runtime/core/src/kernel/kernel.cc | 279 | ${TOP_DIR}/src/runtime/core/src/kernel/kernel.cc |
| 280 | ${TOP_DIR}/src/runtime/core/src/kernel/module.cc | 280 | ${TOP_DIR}/src/runtime/core/src/kernel/module.cc |
| 281 | ${TOP_DIR}/src/runtime/core/src/kernel/program.cc | 281 | ${TOP_DIR}/src/runtime/core/src/kernel/program.cc |
| 282 | ${TOP_DIR}/src/runtime/core/src/kernel/symbol_table.cc | ||
| 282 | ${TOP_DIR}/src/runtime/feature/soma/soma.cc | 283 | ${TOP_DIR}/src/runtime/feature/soma/soma.cc |
| 283 | ${TOP_DIR}/src/runtime/feature/soma/stream_mem_pool.cc | 284 | ${TOP_DIR}/src/runtime/feature/soma/stream_mem_pool.cc |
| 284 | ${TOP_DIR}/src/runtime/core/src/launch/cmo_barrier_common.cc | 285 | ${TOP_DIR}/src/runtime/core/src/launch/cmo_barrier_common.cc |
| @@ -0,0 +1,117 @@ | |||
| 1 | /** | ||
| 2 | * Copyright (c) 2026 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 | |||
| 13 | |||
| 14 | |||
| 15 | |||
| 16 | |||
| 17 | |||
| 18 | |||
| 19 | |||
| 20 | using namespace testing; | ||
| 21 | using namespace cce::runtime; | ||
| 22 | |||
| 23 | class SymbolTest : public testing::Test { | ||
| 24 | protected: | ||
| 25 | static void SetUpTestCase() | ||
| 26 | { | ||
| 27 | std::cout << "symbol test start" << std::endl; | ||
| 28 | } | ||
| 29 | |||
| 30 | static void TearDownTestCase() | ||
| 31 | { | ||
| 32 | std::cout << "symbol test end" << std::endl; | ||
| 33 | } | ||
| 34 | |||
| 35 | virtual void SetUp() | ||
| 36 | { | ||
| 37 | rtSetDevice(0); | ||
| 38 | prog_ = CreateMockElfProgram(); | ||
| 39 | Runtime::Instance()->GetSymbolTable().Register(prog_, &hostVar_, "test_symbol", sizeof(hostVar_), 0U); | ||
| 40 | } | ||
| 41 | |||
| 42 | virtual void TearDown() | ||
| 43 | { | ||
| 44 | delete prog_; | ||
| 45 | prog_ = nullptr; | ||
| 46 | rtDeviceReset(0); | ||
| 47 | GlobalMockObject::verify(); | ||
| 48 | } | ||
| 49 | |||
| 50 | ElfProgram *CreateMockElfProgram() | ||
| 51 | { | ||
| 52 | ElfProgram *prog = new ElfProgram(); | ||
| 53 | prog->elfData_ = new rtElfData(); | ||
| 54 | prog->elfData_->globalSymbolMap["test_symbol"] = {0x100, sizeof(int)}; | ||
| 55 | prog->SetBinBaseAddr((void *)0x8000, 0); | ||
| 56 | prog->SetBinAlignBaseAddr((void *)0x8000, 0); | ||
| 57 | return prog; | ||
| 58 | } | ||
| 59 | |||
| 60 | ElfProgram *prog_ = nullptr; | ||
| 61 | int hostVar_ = 0; | ||
| 62 | }; | ||
| 63 | |||
| 64 | TEST_F(SymbolTest, rtSymbolLookup_SymbolNotFound) | ||
| 65 | { | ||
| 66 | int hostVarNotFound = 0; | ||
| 67 | void *devPtr = nullptr; | ||
| 68 | size_t size = 0; | ||
| 69 | |||
| 70 | rtError_t error = rtSymbolLookup(&hostVarNotFound, &devPtr, &size); | ||
| 71 | EXPECT_EQ(error, ACL_ERROR_RT_INVALID_SYMBOL); | ||
| 72 | } | ||
| 73 | |||
| 74 | TEST_F(SymbolTest, rtSymbolLookup_Success) | ||
| 75 | { | ||
| 76 | void *devPtr = nullptr; | ||
| 77 | size_t size = 0; | ||
| 78 | |||
| 79 | rtError_t error = rtSymbolLookup(&hostVar_, &devPtr, &size); | ||
| 80 | EXPECT_EQ(error, ACL_RT_SUCCESS); | ||
| 81 | EXPECT_EQ(devPtr, (void *)0x8100); | ||
| 82 | EXPECT_EQ(size, sizeof(hostVar_)); | ||
| 83 | } | ||
| 84 | |||
| 85 | TEST_F(SymbolTest, SymbolTable_Register_Duplicate) | ||
| 86 | { | ||
| 87 | rtError_t error = Runtime::Instance()->GetSymbolTable().Register(prog_, &hostVar_, "test_symbol", sizeof(hostVar_), 0U); | ||
| 88 | EXPECT_EQ(error, RT_ERROR_NONE); | ||
| 89 | } | ||
| 90 | |||
| 91 | TEST_F(SymbolTest, ApiRegisterVariable_DeviceVarNameTooLong) | ||
| 92 | { | ||
| 93 | Api *api = Api::Instance(); | ||
| 94 | int newHostVar = 0; | ||
| 95 | char longName[4097]; // NAME_MAX_LENGTH = 4096 | ||
| 96 | (void)memset_s(longName, sizeof(longName), 'a', 4096); | ||
| 97 | longName[4096] = '\0'; | ||
| 98 | |||
| 99 | rtError_t error = api->RegisterVariable(prog_, &newHostVar, longName, sizeof(newHostVar), 0U); | ||
| 100 | EXPECT_EQ(error, RT_ERROR_INVALID_VALUE); | ||
| 101 | } | ||
| 102 | |||
| 103 | TEST_F(SymbolTest, ApiRegisterVariable_ImplSuccess) | ||
| 104 | { | ||
| 105 | Api *api = Api::Instance(); | ||
| 106 | int newHostVar = 0; | ||
| 107 | prog_->elfData_->globalSymbolMap["new_symbol"] = {0x200, sizeof(int)}; | ||
| 108 | |||
| 109 | rtError_t error = api->RegisterVariable(prog_, &newHostVar, "new_symbol", sizeof(newHostVar), 0U); | ||
| 110 | EXPECT_EQ(error, RT_ERROR_NONE); | ||
| 111 | |||
| 112 | void *devPtr = nullptr; | ||
| 113 | size_t size = 0; | ||
| 114 | error = rtSymbolLookup(&newHostVar, &devPtr, &size); | ||
| 115 | EXPECT_EQ(error, ACL_RT_SUCCESS); | ||
| 116 | } | ||
| 117 | |||


rt接口没有返回值吗