CANN/catlass稀疏矩阵乘法示例

SparseMatmulTla Example Readme

【免费下载链接】catlass本项目是CANN的算子模板库，提供NPU上高性能矩阵乘及其相关融合类算子模板样例。项目地址: https://gitcode.com/cann/catlass

Code Organization

├── 41_sparse_matmul_tla │ ├── CMakeLists.txt #CMake build file │ ├── README.md │ ├── sparse_gen_data.py │ └── sparse_matmul_tla.cpp # Main file

Example

• After obtaining the code, compile the operator executable file. For details, see Template Library Quick Start.

• Runsparse_gen_data.pyto generate a test sample. The test sample needs to be input from the command line. After the command is executed, theinputandoutputdirectories are generated in the specified path, including the input data of the operator and the golden data used for precision verification.

• Then, execute the operator. Note that the input shape of the operator must match the shape of the data generated in the first step. In addition, this sample supports only theint8_tdata type for the input of matrix A or B.

The following is a complete shell script example (run in the project directory):

m=160 n=320 k=64 device=0 function build() { bash scripts/build.sh 41_sparse_matmul_tla } function gen_data() { cd examples/41_sparse_matmul_tla python3 sparse_gen_data.py $m $n $k echo "Data gen finished" } function run_kernel { echo 'Case: m=' $m ' k=' $k ' n=' $n cd ../../output/bin/ cp -r ../../examples/41_sparse_matmul_tla/input . cp -r ../../examples/41_sparse_matmul_tla/output . ./41_sparse_matmul_tla $m $n $k $device } build gen_data run_kernel

If the following result is displayed, precision verification is successful.

Compare success.

【免费下载链接】catlass本项目是CANN的算子模板库，提供NPU上高性能矩阵乘及其相关融合类算子模板样例。项目地址: https://gitcode.com/cann/catlass