背景

CANN上算子基础库编译遇到一个问题，如下：

-- The C compiler identification is GNU 10.3.1
-- The CXX compiler identification is GNU 10.3.1
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Check for working C compiler: /usr/bin/cc - skipped
-- Detecting C compile features
-- Detecting C compile features - done
-- 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
-- Found Python3: /usr/bin/python3.9 (found version "3.9.9") found components: Interpreter 
-- ASCEND_CANN_PACKAGE_PATH=/usr/local/Ascend/ascend-toolkit/latest
-- THIRD_PARTY_LIB_PATH=/root/zjun/cann-ops/third_party
-- ENABLE_CCACHE=ON, CUSTOM_CCACHE=
-- CCACHE_PROGRAM=CCACHE_PROGRAM-NOTFOUND
-- ASCEND_COMPUTE_UNIT=ascend310p;ascend310b;ascend910;ascend910b
-- ASCEND_OP_NAME=Addcmul
-- TILING_KEY=
-- TESTS_UT_OPS_TEST=
-- TESTS_EXAMPLE_OPS_TEST=
CMake Error at CMakeLists.txt:120 (add_library):
  No SOURCES given to target: opsproto


CMake Generate step failed.  Build files cannot be regenerated correctly.
make: *** No rule to make target 'prepare_build'.  Stop.
CMake Error at cmake/config.cmake:239 (message):
  Error: ops prepare build failed.
Call Stack (most recent call first):
  CMakeLists.txt:20 (include)

-- Configuring incomplete, errors occurred!

通过如下命令执行的：

[root@localhost cann-ops]# bash build.sh -n Addcmul

-n是执行单独的算子编译

[root@localhost cann-ops]# bash build.sh -h
Usage: build.sh [options]
Options:

-h|--help            Displays help message.

-n|--op-name         Specifies the compiled operator. If there are multiple values, separate them with semicolons and use quotation marks. The default is all.
                     For example: -n "flash_attention_score" or -n "flash_attention_score;flash_attention_score_grad"
...

该算子工程仓库地址：

https://gitee.com/ascend/cann-ops.git

原因分析

上述过程都是按照https://gitee.com/ascend/cann-ops/blob/master/QuickStart.md中的指导进行操作的，反复检查了下，并没有发现问题。
还是通过报错进行分析，打开CMakeLists.txt:120，查看，这里开始定义opsproto ，并在ASCEND_AUTOGEN_DIR下找proto.cpp文件。

119 # op proto
120 message( " -----1------------ASCEND_AUTOGEN_DIR: ${ASCEND_AUTOGEN_DIR}")
121 add_library(opsproto SHARED)
122 message( " -----2------------ASCEND_AUTOGEN_DIR: ${ASCEND_AUTOGEN_DIR}")
123 file(GLOB PROTO_AUTO_FILES
124        ${ASCEND_AUTOGEN_DIR}/*_proto.cpp
125     )
126 target_sources(opsproto PRIVATE
127        ${PROTO_AUTO_FILES}
128 )

结合报错No SOURCES given to target: opsproto，猜测是没有找到任何proto.cpp文件，通过在CMakeLists.txt中添加打印，查看ASCEND_AUTOGEN_DIR下的文件信息，确实如猜测所述：

[root@localhost cann-ops]# ls /root/zjun/cann-ops/build/autogen
custom_compile_options.ini  custom_opc_options.ini  custom_tiling_keys.ini

到这里，猜测是算子名字写错了吗？
对比了下，Addcmul这个名字单词并没拼错，只是名字大小写不一致，莫非是因为这个导致的？
修改编译命令：

[root@localhost cann-ops]# bash build.sh -n addcmul

果然，顺利的执行下去了

 -----1------------ASCEND_AUTOGEN_DIR: /root/zjun/cann-ops/build/autogen
 -----2------------ASCEND_AUTOGEN_DIR: /root/zjun/cann-ops/build/autogen
-- Configuring done (6.2s)
-- Generating done (0.0s)
CMake Warning:
  Manually-specified variables were not used by the project:

    ASCEND_PYTHON_EXECUTABLE
    CMAKE_CROSS_PLATFORM_COMPILER
    ENABLE_BINARY_PACKAGE
    ENABLE_CROSS_COMPILE
    ENABLE_SOURCE_PACKAGE
    ENABLE_TEST


-- Build files have been written to: /root/zjun/cann-ops/build
[  4%] Generating op_host_aclnn_inner_stub.cpp
[  4%] Generating op_host_aclnn_exc_stub.cpp
[  8%] Building CXX object CMakeFiles/optiling.dir/src/common/src/fallback_comm.cpp.o
[  8%] Building CXX object CMakeFiles/op_host_aclnn.dir/src/math/addcmul/op_host/addcmul.cpp.o
[ 12%] Building CXX object CMakeFiles/optiling.dir/src/math/addcmul/op_host/addcmul.cpp.o

查看ASCEND_AUTOGEN_DIR下的文件信息，确实也多了addcmul_proto.cpp等文件。

[root@localhost cann-ops]# ls /root/zjun/cann-ops/build/autogen
aclnn_addcmul.cpp  aclnn_addcmul.h  addcmul_proto.cpp  addcmul_proto.h  aic-ascend310b-ops-info.ini  aic-ascend910b-ops-info.ini  custom_compile_options.ini  custom_opc_options.ini  custom_tiling_keys.ini

解决办法

汗，最终原因就是编译算子的大小写拼音问题。
通过与CANN算子开发同学沟通，发现-n的使用规则如下：

-n后面跟算子的文件夹名，文件夹是什么样的（大小写、下划线等等），-n后面就写什么样的

记录一下吧，以防踩坑。

PS：研发的同学是不是应该在-n的help解释上多写上几句话的说明。