文档

Checking the Operators Called by a PyTorch API

Prerequisite

  • Click Link to obtain the sample project for operator check.
    • This sample project supports only Python 3.9. To run it on other Python versions, change the Python version in the run_op_plugin.sh file in the ${git_clone_path}/samples/operator/ascendc/0_introduction/1_add_frameworklaunch/PytorchInvocation directory.
    • When downloading the code sample, run the following command to specify the branch version:
      git clone https://gitee.com/ascend/samples.git -b v0.2-8.0.0.beta1
  • Install the PyTorch framework and torch_npu plugin by referring to Ascend Extension for PyTorch Configuration and Installation .

Procedure

  1. Run the following command to generate a custom operator project and implement the operator on the host and kernel:
    bash install.sh -v Ascendxxxyy    # xxxyy indicates the processor type.
  2. Compile and deploy the operator by referring to Compiling and Deploying Operators.

    Edit the CMakeLists.txt file in the ${git_clone_path}/samples/operator/ascendc/0_introduction/1_add_frameworklaunch/CustomOp/op_kernel directory of the sample project and add the compilation option -sanitizer.

    1
    add_ops_compile_options(ALL OPTIONS -sanitizer )
  3. Go to PyTorch access project, call the AddCustom operator project in PyTorch mode, and complete compilation as instructed.
    1
2
3
4
    PytorchInvocation
├── op_plugin_patch         
├── run_op_plugin.sh      // Required for sample execution.
└── test_ops_custom.py    // Required for tool startup in step 5.
  4. Execute the sample. During the sample execution, test data is automatically generated. Run the PyTorch sample, and verify the running result.
     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
    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/ascend-toolkit/latest
-- 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%] Linking CCE executable ../../../add_npu
[100%] Built target add_npu
${INSTALL_DIR}/AddKernelInvocation
INFO: compile op on ONBOARD succeed!
INFO: execute op on ONBOARD succeed!
test pass
  5. Start the msSanitizer tool to start the Python program for exception detection.
    mssanitizer python3 test_ops_custom.py
  6. Analyze abnormal behavior by referring to Memory Exception Report Parsing, Analyzing Contention Check Report, and Parsing of Uninitialization Exception Reports.
Parent topic: Typical Cases