典型案例
上板调试Vector算子
概述
展示如何使用msDebug工具来上板调试一个Vector算子，该Vector算子可实现两个向量相加并输出结果的功能。
前期准备
- 单击链接获取样例工程，为进行算子调试做准备。
- 完成相关环境变量配置，请参见MindStudio Debugger工具用户指南。
操作步骤
-
基于样例工程编译算子，获取可执行文件add.fatbin。
-
修改sample/normal_sample/vec_only/Makefile中的COMPILER_FLAG编译选项，将
-O2修改为-O0 -g --cce-ignore-always-inline=true，使能编译器调试功能。# Makefile ... COMPILER := $(ASCEND_HOME_PATH)/compiler/ccec_compiler/bin/ccec COMPILER_FLAG := -xcce -O0 -g --cce-ignore-always-inline=true -std=c++17 # 使能编译器调试功能 -
执行以下命令完成算子编译。
Note
非首次场景，可以使用make clean && make命令替代make命令。
cd ./mstt/sample/normal_sample/vec_only/ make clean && make
-
-
设置断点。
-
启动msDebug工具拉起算子程序，进入调试界面。
msdebug add.fatbin (msdebug) target create "add.fatbin" Current executable set to '/home/mindstudio/projects/mstt/sample/build/add.fatbin' (aarch64). (msdebug) -
该sample中核函数的代码实现位于add_kernel.cpp中，在此文件中，为需要的代码行设置NPU断点。
(msdebug) b add_kernel.cpp:69 Breakpoint 1: where = device_debugdata`::add_custom(uint8_t *, uint8_t *, uint8_t *) + 18804 [inlined] KernelAdd::Compute(int) + 5144 at add_kernel.cpp:69:9, address = 0x0000000000004974 (msdebug)
-
-
运行算子程序。
程序会开始运行直到命中第一个断点（add_kernel.cpp:69）后停下，msDebug检测到NPU核函数add_custom开始运行，运行在Device 0。
(msdebug) run Process 730254 launched [Launch of Kernel add_custom on Device 0] Process 730254 stopped [Switching to focus on Kernel add_custom, CoreId 13, Type aiv] * thread #1, name = 'add.fatbin', stop reason = breakpoint 2.1 frame #0: 0x0000000000004974 device_debugdata`::add_custom(uint8_t *, uint8_t *, uint8_t *) [inlined] KernelAdd::Compute(this=0x000000000019a930, progress=0) at add_kernel.cpp:69:9 66 // call Add instr for computation 67 Add(zLocal, xLocal, yLocal, TILE_LENGTH); 68 // enque the output tensor to VECOUT queue -> 69 outQueueZ.EnQue<int16_t>(zLocal); # 断点位置 70 // free input tensors for reuse 71 inQueueX.FreeTensor(xLocal); 72 inQueueY.FreeTensor(yLocal); (msdebug) -
检视信息。
-
使用ascend info cores命令查询NPU核信息。
(msdebug) ascend info cores CoreId Type Device Stream Task Block PC Exception * 13 aiv 0 3 0 0 0x1240c0034974 f0000000 14 aiv 0 3 0 1 0x1240c0034974 f0000000 15 aiv 0 3 0 2 0x1240c0034974 f0000000 20 aiv 0 3 0 3 0x1240c0034974 f0000000 21 aiv 0 3 0 4 0x1240c0034974 f0000000 22 aiv 0 3 0 5 0x1240c0034974 f0000000 23 aiv 0 3 0 6 0x1240c0034974 f0000000 24 aiv 0 3 0 7 0x1240c0034974 f0000000 (msdebug) -
使用print命令直接打印变量信息。
(msdebug) print progress (int32_t) $0 = 0 -
使用print命令与memory read命令配合可打印出Tensor变量中存放的值。
-
打印位于UB内存上的LocalTensor中存放的数据。
Note
UB内存打印起始地址需参考LocalTensor变量展示的address_字段中的bufferAddr参数。此处以变量xLocal为例，其内存起始地址为0。
(msdebug) print xLocal (AscendC::LocalTensor<short>) $0 = { address_ = (dataLen = 256, bufferAddr = 0, bufferHandle = "", logicPos = '\t') shapeInfo_ = { shapeDim = '\0' originalShapeDim = '\0' shape = ([0] = 0, [1] = 0, [2] = 0, [3] = 0, [4] = 0, [5] = 0, [6] = 0, [7] = 0) originalShape = ([0] = 0, [1] = 0, [2] = 0, [3] = 0, [4] = 0, [5] = 0, [6] = 0, [7] = 0) dataFormat = ND } } (msdebug) memory read -m UB -f int16_t[] 0 -s 256 -c 1 0x00000000: {0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127} (msdebug) -
打印位于GM内存上的GlobalTensor中存放的数据。
Note
GM内存打印的起始地址需参考GlobalTensor变量展示的address_字段。此处以变量xGm为例，其内存起始地址为0x00001240c0015000。
(msdebug) print xGm (AscendC::GlobalTensor<short>) $0 = { bufferSize_ = 2048 shapeInfo_ = { shapeDim = '\0' originalShapeDim = '\0' shape = ([0] = 0, [1] = 0, [2] = 0, [3] = 0, [4] = 0, [5] = 0, [6] = 0, [7] = 0) originalShape = ([0] = 0, [1] = 0, [2] = 0, [3] = 0, [4] = 0, [5] = 0, [6] = 0, [7] = 0) dataFormat = ND } address_ = 0x00001240c0015000 } (msdebug) memory read -m GM -f int16_t[] 0x00001240c0015000 -s 256 -c 1 0x1240c0015000: {0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127}
-
-
进行核切换，切换至另一个aiv核，并打印需要的信息。
(msdebug) ascend aiv 24 // ascend info cores中选择block 7对应的coreId,此处为24 [Switching to focus on Kernel add_custom, CoreId 24, Type aiv] * thread #1, name = 'add.fatbin', stop reason = breakpoint 2.1 frame #0: 0x0000000000004974 device_debugdata`::add_custom(uint8_t *, uint8_t *, uint8_t *) [inlined] KernelAdd::Compute(this=0x00000000001c6930, progress=0) at add_kernel.cpp:69:9 66 // call Add instr for computation 67 Add(zLocal, xLocal, yLocal, TILE_LENGTH); 68 // enque the output tensor to VECOUT queue -> 69 outQueueZ.EnQue<int16_t>(zLocal); ^ 70 // free input tensors for reuse 71 inQueueX.FreeTensor(xLocal); 72 inQueueY.FreeTensor(yLocal); (msdebug) p xLocal (AscendC::LocalTensor<short>) $0 = { address_ = (dataLen = 256, bufferAddr = 0, bufferHandle = "", logicPos = '\t') shapeInfo_ = { shapeDim = '\0' originalShapeDim = '\0' shape = ([0] = 0, [1] = 0, [2] = 0, [3] = 0, [4] = 0, [5] = 0, [6] = 0, [7] = 0) originalShape = ([0] = 0, [1] = 0, [2] = 0, [3] = 0, [4] = 0, [5] = 0, [6] = 0, [7] = 0) dataFormat = ND } } (msdebug) memory read -m UB -f int16_t[] 0 -s 256 -c 1 0x00000000: {14336 14337 14338 14339 14340 14341 14342 14343 14344 14345 14346 14347 14348 14349 14350 14351 14352 14353 14354 14355 14356 14357 14358 14359 14360 14361 14362 14363 14364 14365 14366 14367 14368 14369 14370 14371 14372 14373 14374 14375 14376 14377 14378 14379 14380 14381 14382 14383 14384 14385 14386 14387 14388 14389 14390 14391 14392 14393 14394 14395 14396 14397 14398 14399 14400 14401 14402 14403 14404 14405 14406 14407 14408 14409 14410 14411 14412 14413 14414 14415 14416 14417 14418 14419 14420 14421 14422 14423 14424 14425 14426 14427 14428 14429 14430 14431 14432 14433 14434 14435 14436 14437 14438 14439 14440 14441 14442 14443 14444 14445 14446 14447 14448 14449 14450 14451 14452 14453 14454 14455 14456 14457 14458 14459 14460 14461 14462 14463} (msdebug)
-
-
查询并删除断点，恢复程序运行。
(msdebug) breakpoint list Current breakpoints: 1: name = 'main', locations = 1, resolved = 1, hit count = 1 1.1: where = add.fatbin`main + 36 at main.cpp:39:12, address = 0x0000aaaaaab0f568, resolved, hit count = 1 2: file = 'add_kernel.cpp', line = 69, exact_match = 0, locations = 1, resolved = 1, hit count = 1 2.1: where = device_debugdata`::add_custom(uint8_t *, uint8_t *, uint8_t *) + 18804 [inlined] KernelAdd::Compute(int) + 5144 at add_kernel.cpp:69:9, address = 0x0000000000004974, resolved, hit count = 1 (msdebug) breakpoint delete 2 1 breakpoints deleted; 0 breakpoint locations disabled. (msdebug) continue Process 730254 resuming 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 Process 730254 exited with status = 0 (0x00000000) -
调试完以后，执行q命令并输入Y或y结束调试。
(msdebug) q Quitting LLDB will kill one or more processes. Do you really want to proceed: [Y/n] y
调用Ascend CL单算子
前期准备
单击链接获取算子样例工程，为进行算子调试做准备。
Note
-
此样例工程不支持Atlas A3 训练系列产品。
-
下载代码样例时，需执行以下命令指定分支版本。
git clone https://gitee.com/ascend/samples.git -b v0.2-8.0.0.beta1
操作步骤
-
切换到msOpGen脚本install.sh所在目录。
cd ${git_clone_path}/samples/operator/ascendc/0_introduction/1_add_frameworklaunch -
执行以下命令，生成自定义算子工程，并进行Host侧和Kernel侧的算子实现。
bash install.sh -v Ascendxxxyy # xxxyy为用户实际使用的具体芯片类型 -
在${git_clone_path}/samples/operator/ascendc/0_introduction/1_add_frameworklaunch/CustomOp目录下修改CMakePresets.json文件的cacheVariables的配置项，将
"Release"修改为"Debug"。"cacheVariables": { "CMAKE_BUILD_TYPE": { "type": "STRING", "value": "Debug" }, ... } -
参考算子编译部署完成算子的编译部署。
-
切换到msOpGen脚本install.sh所在目录，并参考README编译单算子调用应用并得到可执行文件execute_add_op。
cd ${git_clone_path}/samples/operator/ascendc/0_introduction/1_add_frameworklaunch/AclNNInvocation -
导入算子动态加载路径。
将自定义算子工程编译后输出在build_out目录下Kernel侧的
.o文件路径导入环境变量。export LAUNCH_KERNEL_PATH=/{path_to_kernel}/kernel_name.o # {path_to_kernel}表示对算子Kernel侧实现编译后生成的算子二进制文件*.o所在路径，请根据实际情况进行替换Note
算子的多个dtype在Kernel侧可能会编译出多个
.o文件，请选择4示例中所调用的.o文件进行导入。 -
使用msDebug工具加载5中得到的单算子可执行文件execute_add_op。
export LD_LIBRARY_PATH=$ASCEND_HOME_PATH/opp/vendors/customize/op_api/lib:$LD_LIBRARY_PATH cd AclNNInvocation/output msdebug execute_add_op (msdebug) target create "execute_add_op" Current executable set to '/home/AclNNInvocation/output/execute_add_op' (aarch64). (msdebug) -
断点设置。
b add_custom.cpp:55 -
运行算子程序，等待直到命中断点。
(msdebug) r Process 1385976 launched: '$HOME/shelltest/test/samples/operator/ascendc/0_introduction/1_add_frameworklaunch/AclNNInvocationNaive/build/execute_add_op' (aarch64) [Launch of Kernel anonymous on Device 0] Process 1385976 stopped [Switching to focus on Kernel anonymous, CoreId 24, Type aiv] * thread #1, name = 'execute_add_op', stop reason = breakpoint 1.1 frame #0: 0x0000000000001564 AddCustom_1e04ee05ab491cc5ae9c3d5c9ee8950b.o`KernelAdd::Compute(this=0x000000000028f8a8, progress=0) (.vector) at add_custom.cpp:55:19 52 LocalTensor<DTYPE_Y> yLocal = inQueueY.DeQue<DTYPE_Y>(); 53 LocalTensor<DTYPE_Z> zLocal = outQueueZ.AllocTensor<DTYPE_Z>(); 54 Add(zLocal, xLocal, yLocal, this->tileLength); -> 55 outQueueZ.EnQue<DTYPE_Z>(zLocal); 56 inQueueX.FreeTensor(xLocal); 57 inQueueY.FreeTensor(yLocal); 58 } (msdebug)Note
后续调试过程可参考导入调试信息、内存与变量打印功能介绍及核切换功能介绍等，与其操作一致。
调试PyTorch接口调用的算子
概述
展示如何使用msDebug工具来上板调试一个PyTorch接口调用的add算子，该add算子可实现两个向量相加并输出结果的功能。
前期准备
-
单击链接获取样例工程，为进行算子调试做准备。
Note
- 此样例工程仅支持Python3.9，若要在其他Python版本上运行，需要修改${git_clone_path}/samples/operator/ascendc/0_introduction/1_add_frameworklaunch/PytorchInvocation目录下run_op_plugin.sh文件中的Python版本。
- 此样例工程不支持Atlas A3 训练系列产品。
- 下载代码样例时，需执行以下命令指定分支版本。
git clone https://gitee.com/ascend/samples.git -b v0.2-8.0.0.beta1
- 已参考《Ascend Extension for PyTorch 软件安装指南》，完成PyTorch框架和torch_npu插件的安装。
- 完成相关环境变量配置，请参见MindStudio Debugger工具用户指南。
操作步骤
-
执行以下命令，可生成自定义算子工程，并进行Host侧和Kernel侧的算子实现。
bash install.sh -v Ascendxxxyy # xxxyy为用户实际使用的具体芯片类型 -
在${git_clone_path}/samples/operator/ascendc/0_introduction/1_add_frameworklaunch/CustomOp目录下修改CMakePresets.json文件的cacheVariables的配置项，将
"Release"修改为"Debug"。"cacheVariables": { "CMAKE_BUILD_TYPE": { "type": "STRING", "value": "Debug" }, ... } -
参考算子编译部署，完成算子的编译部署。
-
进入到样例目录，以命令行方式下载样例代码。参考README使用PyTorch调用方式调用AddCustom算子工程，并按照指导完成编译。
cd ${git_clone_path}/samples/operator/ascendc/0_introduction/1_add_frameworklaunch/PytorchInvocationNote
PyTorch接入工程的样例工程目录如下：
PytorchInvocation ├── op_plugin_patch ├── README.md //使用PyTorch调用方式调用AddCustom算子工程的注册样例 ├── run_op_plugin.sh // 执行样例时，需要使用 └── test_ops_custom.py // 启动工具时,需要使用 └── test_ops_custom_register_in_graph.py // 执行torch.compile模式下用例脚本
-
执行样例，样例执行过程中会自动生成测试数据，然后运行PyTorch样例，最后检验运行结果。
bash run_op_plugin.sh -- CMAKE_CCE_COMPILER: ${INSTALL_DIR}/toolkit/tools/ccec_compiler/bin/ccec -- CMAKE_CURRENT_LIST_DIR: ${INSTALL_DIR}/AddKernelInvocation/cmake/Modules -- ASCEND_PRODUCT_TYPE: Ascendxxxyy -- ASCEND_CORE_TYPE: VectorCore -- ASCEND_INSTALL_PATH: /usr/local/Ascend/cann -- The CXX compiler identification is GNU 10.3.1 -- Detecting CXX compiler ABI info -- Detecting CXX compiler ABI info - done -- Check for working CXX compiler: /usr/bin/c++ - skipped -- Detecting CXX compile features -- Detecting CXX compile features - done -- Configuring done -- Generating done -- Build files have been written to: ${INSTALL_DIR}/AddKernelInvocation/build Scanning dependencies of target add_npu ... [100%] Built target add_npu INFO: Ascend C Add Custom SUCCESS ... INFO: Ascend C Add Custom in torch.compile graph SUCCESS -
手动导入算子调试信息，示例如下。
Note
- ${INSTALL_DIR}请替换为CANN软件安装后文件存储路径。以root安装举例，则安装后文件存储路径为：/usr/local/Ascend/cann。
- 非Atlas A3 训练系列产品/Atlas A3 推理系列产品：在安装昇腾AI处理器的服务器执行
npu-smi info命令进行查询，获取Chip Name信息。实际配置值为AscendChip Name，例如Chip Name取值为xxxyy，实际配置值为Ascend_xxxyy_。当Ascendxxxyy为代码样例的路径时，需要配置为ascendxxxyy。 - Atlas A3 训练系列产品/Atlas A3 推理系列产品：在安装昇腾AI处理器的服务器执行
npu-smi info -t board -i id -c chip_id命令进行查询，获取Chip Name和NPU Name信息，实际配置值为Chip Name_NPU Name。例如Chip Name取值为Ascendxxx，NPU Name取值为1234，实际配置值为Ascendxxx_1234。当Ascendxxx_1234为代码样例的路径时，需要配置为ascendxxx_1234。 其中：- id：设备id，通过
npu-smi info -l命令查出的NPU ID即为设备id。 - chip_id：芯片id，通过
npu-smi info -m命令查出的Chip ID即为芯片id。
- id：设备id，通过
export LAUNCH_KERNEL_PATH=${INSTALL_DIR}/opp/vendors/customize/op_impl/ai_core/tbe/kernel/SOC_VERSION/add_custom/AddCustom_1e04ee05ab491cc5ae9c3d5c9ee8950b.o -
启动msDebug工具拉起Python程序，进入调试界面。
msdebug python3 test_ops_custom.py (msdebug) target create "python3" Current executable set to '/home/mindstudio/miniconda3/envs/py39/bin/python3' (aarch64). (msdebug) settings set -- target.run-args "test_ops_custom.py" (msdebug) -
设置断点。
根据指定源码文件与对应行号，在核函数中设置NPU断点。
(msdebug) b add_custom.cpp:60 Breakpoint 1: where = AddCustom_1e04ee05ab491cc5ae9c3d5c9ee8950b.o`::AddCustom_1e04ee05ab491cc5ae9c3d5c9ee8950b_1(uint8_t *, uint8_t *, uint8_t *, uint8_t *, uint8_t *) + 9912 [inlined] KernelAdd::Compute(int) + 3400 at add_custom.cpp:60:9, address = 0x00000000000026b8 -
运行程序，等待直到命中断点。
(msdebug) r Process 197189 launched: '/home/miniconda3/envs/py39/bin/python3' (aarch64) Process 197189 stopped and restarted: thread 1 received signal: SIGCHLD ... [Launch of Kernel anonymous on Device 0] Process 197189 stopped [Switching to focus on Kernel anonymous, CoreId 8, Type aiv] * thread #1, name = 'python3', stop reason = breakpoint 2.1 frame #0: 0x00000000000026b8 AddCustom_1e04ee05ab491cc5ae9c3d5c9ee8950b.o`::AddCustom_1e04ee05ab491cc5ae9c3d5c9ee8950b_1(uint8_t *, uint8_t *, uint8_t *, uint8_t *, uint8_t *) [inlined] KernelAdd::Compute(this=0x000000000020efb8, progress=1) at add_custom.cpp:60:9 57 LocalTensor<DTYPE_Y> yLocal = inQueueY.DeQue<DTYPE_Y>(); 58 LocalTensor<DTYPE_Z> zLocal = outQueueZ.AllocTensor<DTYPE_Z>(); 59 Add(zLocal, xLocal, yLocal, this->tileLength); -> 60 outQueueZ.EnQue<DTYPE_Z>(zLocal); 61 inQueueX.FreeTensor(xLocal); 62 inQueueY.FreeTensor(yLocal); 63 } (msdebug)Note
其他调试操作可参考导入调试信息、内存与变量打印功能介绍、调试信息展示功能介绍及核切换功能介绍等，与其操作一致。
-
删除断点，具体操作请参见断点设置功能介绍。
-
调试完以后，执行q命令并输入Y或y结束调试。
(msdebug) q Quitting LLDB will kill one or more processes. Do you really want to proceed: [Y/n] y
上板调试模板库的算子
概述
展示如何使用msDebug工具来上板调试一个模板库算子（matmul），该算子可实现两个矩阵相乘并输出结果的功能。
- 单击链接获取样例工程，为进行算子调试做准备。
- 完成相关环境变量配置，请参见MindStudio Debugger工具用户指南。
操作步骤
-
基于前期准备中的样例工程编译算子，获取可执行文件00_basic_matmul。
执行以下命令完成算子编译，编译完成后，在build/bin目录下生成可执行文件00_basic_matmul。
bash ./scripts/build.sh 00_basic_matmul --debug --msdebug -
启动msDebug工具拉起算子程序，进入调试界面。
msdebug ./build/bin/00_basic_matmul 256 512 1024 0 (msdebug) target create "./build/bin/00_basic_matmul" Current executable set to '/home/mindstudio/projects/ascendc-templates/build/bin/00_basic_matmul' (aarch64). (msdebug) -
设置断点。
该用例中核函数的代码实现位于basic_matmul.hpp中，在此文件中，为需要的代码行设置NPU断点。
(msdebug) b basic_matmul.hpp:121 Breakpoint 1: 2 locations. (msdebug) -
运行算子程序，等待直到命中断点。
程序会开始运行直到命中第一个断点（basic_matmul.hpp:127）后停下，msDebug检测到NPU核函数开始运行，运行在Device 0。
(msdebug) run Process 3344307 launched: '/home/mindstudio/projects/ascendc-templates/build/bin/00_basic_matmul' (aarch64) [Launch of Kernel _ZN7Catlass13KernelAdapterINS_4Gemm6Kernel11BasicMatmulINS1_5Blo on Device 0] Process 3344307 stopped [Switching to focus on Kernel _ZN7Catlass13KernelAdapterINS_4Gemm6Kernel11BasicMatmulINS1_5Blo, CoreId 21, Type aic] * thread #1, name = '00_basic_matmul', stop reason = breakpoint 1.1 frame #0: 0x0000000000001c38 device_debugdata`_ZN7Catlass13KernelAdapterINS_4Gemm6Kernel11BasicMatmulINS1_5Block9BlockMmadINS1_19MmadAtlasA2PingpongILb1EEENS_9GemmShapeILj128ELj256ELj256EEENS8_ILj128ELj256ELj64EEENS1_8GemmTypeIDhNS_6layout8RowMajorELN7AscendC9TPositionE0EEESG_SG_vNS1_4Tile8TileCopyINS_4Arch7AtlasA2ESG_SG_SG_vEENSH_8TileMmadISK_SG_SG_vEEEEvNS4_24GemmIdentityBlockSwizzleILj3ELj0EEEEEEEvNT_6ParamsE_mix_aic at basic_matmul.hpp:121:71 118 119 for (uint32_t loopIdx = AscendC::GetBlockIdx(); loopIdx < coreLoops; loopIdx += AscendC::GetBlockNum()) { 120 // Compute block location -> 121 GemmCoord blockCoord = matmulBlockScheduler.GetBlockCoord(loopIdx); 122 GemmCoord actualBlockShape = matmulBlockScheduler.GetActualBlockShape(blockCoord); 123 124 // Compute initial location in logical coordinates (msdebug)Note
_ZN7Catlass13KernelAdapterINS_4Gemm6Kernel11BasicMatmulINS1_5Blo为模板库的kernel名字，示例仅显示前面64位。
-
检视信息。
-
使用ascend info cores命令查询NPU核信息。
(msdebug) ascend info cores CoreId Type Device Stream Task Block PC stop reason * 21 aic 0 48 0 0 0x12c0c00d6c38 breakpoint 1.1 22 aic 0 48 0 1 0x12c0c00d6c38 breakpoint 1.1 23 aic 0 48 0 2 0x12c0c00d6c38 breakpoint 1.1 24 aic 0 48 0 3 0x12c0c00d6c38 breakpoint 1.1 (msdebug) -
使用print命令直接打印gmA变量信息。
(msdebug) print gmA (AscendC::GlobalTensor<__fp16>) $0 = { AscendC::BaseGlobalTensor<__fp16> = { address_ = 0x000012c0c0013000 oriAddress_ = 0x000012c0c0013000 } bufferSize_ = 0 cacheMode_ = CACHE_MODE_NORMAL } -
继续使用memory read命令可打印出gmA变量中存放的值。
-
打印位于GM内存上的gmA中存放的数据。
(msdebug) memory read -m GM 0x12c0c0013000 -f float16[] -s 256 -c 1 0x12c0c0013000: {3.40234 -1.05664 2.83008 2.98438 4.11719 -3.02539 -1.64746 2.68164 -2.22266 0.539551 -0.226074 1.28906 -1.35254 0.134033 4.52344 4.16016 1.35742 2.17383 -3.58398 1.06934 -4.83594 -2.57031 -3.62695 3.04102 -3.43359 -0.990723 -3.70117 -3.91211 4.98828 -2.81836 0.129272 3.39062 1.12598 -2.03906 1.37598 0.24292 -0.0641479 4.72656 -2.07422 2.71289 0.267334 2.69922 -0.997559 3.91602 -2.16602 -1.47559 3.07812 4.19141 -4.30078 4.49219 0.26001 -4.14062 -3.07812 1.63184 3.90234 -1.51074 -4.35938 -4.80078 -0.423096 -4.36719 -2.61719 4.70703 4.02344 3.50977 -2.33398 0.397705 -1.24805 2.60156 0.125366 1.67676 0.316162 -4.60547 -0.623535 4.31641 4.30859 2.20898 -2.15625 2.38477 1.39941 -1.45996 1.87891 -3.33984 -0.599121 3.80078 3.29297 -1.69629 -2.71094 3.93359 -1.49609 1.86621 4.56641 0.88623 1.57324 3.58594 -0.604492 4.23828 -1.01562 3.14844 1.8418 4.10938 -0.175049 -2.8418 4.50391 4.20312 -3.52344 3.81055 1.41113 -0.680664 1.19629 -2.18945 2.85938 -1.92578 -0.529785 -2.73828 -3.125 -2.23828 0.564453 -0.834961 -3.30469 4.06641 -3.96875 -3.73828 -0.0455627 2.60547 4.84766 4.35156 1.84473 -1.16797} (msdebug)
-
-
进行核切换，切换至另一个aic核，并打印需要的信息。
(msdebug) ascend aic 24 // ascend info cores中选择block 3对应的coreId,此处为24 [Switching to focus on Kernel _ZN7Catlass13KernelAdapterINS_4Gemm6Kernel11BasicMatmulINS1_5Blo, CoreId 24, Type aic] * thread #1, name = '00_basic_matmul', stop reason = breakpoint 1.1 frame #0: 0x0000000000001c38 device_debugdata`_ZN7Catlass13KernelAdapterINS_4Gemm6Kernel11BasicMatmulINS1_5Block9BlockMmadINS1_19MmadAtlasA2PingpongILb1EEENS_9GemmShapeILj128ELj256ELj256EEENS8_ILj128ELj256ELj64EEENS1_8GemmTypeIDhNS_6layout8RowMajorELN7AscendC9TPositionE0EEESG_SG_vNS1_4Tile8TileCopyINS_4Arch7AtlasA2ESG_SG_SG_vEENSH_8TileMmadISK_SG_SG_vEEEEvNS4_24GemmIdentityBlockSwizzleILj3ELj0EEEEEEEvNT_6ParamsE_mix_aic at basic_matmul.hpp:121:71 118 119 for (uint32_t loopIdx = AscendC::GetBlockIdx(); loopIdx < coreLoops; loopIdx += AscendC::GetBlockNum()) { 120 // Compute block location -> 121 GemmCoord blockCoord = matmulBlockScheduler.GetBlockCoord(loopIdx); 122 GemmCoord actualBlockShape = matmulBlockScheduler.GetActualBlockShape(blockCoord); 123 124 // Compute initial location in logical coordinates (msdebug) p loopIdx (uint32_t) $1 = 0
Note
其他调试操作可参考内存与变量打印功能介绍、调试信息展示功能介绍及核切换功能介绍等，与其操作一致。
-
-
查询并删除断点，恢复程序运行。
(msdebug) breakpoint list Current breakpoints: 1: file = 'basic_matmul.hpp', line = 121, exact_match = 0, locations = 2, resolved = 2, hit count = 1 1.1: where = device_debugdata`_ZN7Catlass13KernelAdapterINS_4Gemm6Kernel11BasicMatmulINS1_5Block9BlockMmadINS1_19MmadAtlasA2PingpongILb1EEENS_9GemmShapeILj128ELj256ELj256EEENS8_ILj128ELj256ELj64EEENS1_8GemmTypeIDhNS_6layout8RowMajorELN7AscendC9TPositionE0EEESG_SG_vNS1_4Tile8TileCopyINS_4Arch7AtlasA2ESG_SG_SG_vEENSH_8TileMmadISK_SG_SG_vEEEEvNS4_24GemmIdentityBlockSwizzleILj3ELj0EEEEEEEvNT_6ParamsE_mix_aic + 4748 [inlined] _ZN7Catlass4Gemm6Kernel11BasicMatmulINS0_5Block9BlockMmadINS0_19MmadAtlasA2PingpongILb1EEENS_9GemmShapeILj128ELj256ELj256EEENS7_ILj128ELj256ELj64EEENS0_8GemmTypeIDhNS_6layout8RowMajorELN7AscendC9TPositionE0EEESF_SF_vNS0_4Tile8TileCopyINS_4Arch7AtlasA2ESF_SF_SF_vEENSG_8TileMmadISJ_SF_SF_vEEEEvNS3_24GemmIdentityBlockSwizzleILj3ELj0EEEEclILi1EEEvRKNSQ_6ParamsE_mix_aic + 4632 at basic_matmul.hpp:121:71, address = 0x0000000000001c38, resolved, hit count = 1 1.2: where = device_debugdata`_ZN7Catlass13KernelAdapterINS_4Gemm6Kernel11BasicMatmulINS1_5Block9BlockMmadINS1_19MmadAtlasA2PingpongILb1EEENS_9GemmShapeILj128ELj256ELj256EEENS8_ILj128ELj256ELj64EEENS1_8GemmTypeIDhNS_6layout8RowMajorELN7AscendC9TPositionE0EEESG_SG_vNS1_4Tile8TileCopyINS_4Arch7AtlasA2ESG_SG_SG_vEENSH_8TileMmadISK_SG_SG_vEEEEvNS4_24GemmIdentityBlockSwizzleILj3ELj0EEEEEEEvNT_6ParamsEm_mix_aic + 4772 [inlined] _ZN7Catlass4Gemm6Kernel11BasicMatmulINS0_5Block9BlockMmadINS0_19MmadAtlasA2PingpongILb1EEENS_9GemmShapeILj128ELj256ELj256EEENS7_ILj128ELj256ELj64EEENS0_8GemmTypeIDhNS_6layout8RowMajorELN7AscendC9TPositionE0EEESF_SF_vNS0_4Tile8TileCopyINS_4Arch7AtlasA2ESF_SF_SF_vEENSG_8TileMmadISJ_SF_SF_vEEEEvNS3_24GemmIdentityBlockSwizzleILj3ELj0EEEEclILi1EEEvRKNSQ_6ParamsE_mix_aic + 4632 at basic_matmul.hpp:121:71, address = 0x000000000000dd54, resolved, hit count = 0 (msdebug) breakpoint delete 1 1 breakpoints deleted; 0 breakpoint locations disabled. (msdebug) continue Process 3344307 resuming Compare success. Process 3344307 exited with status = 0 (0x00000000) -
调试完以后，执行q命令并输入Y或y结束调试。
(msdebug) q