950基础矩阵乘法TLA示例

950 Basic Matmul TLA 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/catlassNote: The community package does not currently support 950 capabilities. Stay tuned for a future supported version.Code Organization├── 43_ascend950_basic_matmul │ ├── CMakeLists.txt # CMake build file │ ├── README.md │ └── basic_matmul_tla.cpp # Main fileUsage ExampleAfter obtaining the code, build the corresponding operator executable. See Template Library Quick Start. This case is a 950 operator, and-DCATLASS_ARCH3510must be added during build.Run the operator.# Build the specified case bash scripts/build.sh 43_ascend950_basic_matmul -DCATLASS_ARCH3510 cd output/bin # Executable file name | matrix m axis | n axis | k axis | Device ID # Device ID is optional and defaults to 0 ./43_ascend950_basic_matmul 256 512 1024 0The execution result is as follows, indicating that the precision comparison succeeds.Compare success.Usage NotesThe DispatchPolicy MmadPingpong used by BasicMatmul by default supports the following template parameters:Template ParameterDefault ValueParameter DescriptionArchTagNoneSpecifies the architecture modelenableUnitFlagfalseSpecifies whether to enable UnitFlag. It must be set to false when L0C multi-buffering is enableduseHF32falseSpecifies whether to enable HF32. Only the float type is supportedl0CStages1Specifies the number of L0C buffers. Set it to 2 to enable L0C double bufferingenableL1ResidentfalseSpecifies whether to enable L1 residencyl1AStages2Number of buffers for loading matrix A on L1l1BStages2Number of buffers for loading matrix B on L1l0AStages2Number of buffers for loading matrix A on L0l0BStages2Number of buffers for loading matrix B on L0Assume the matrix Shape isM N K, the tile size on L1 ism1 n1 k1, the number of tiles in the M direction ismTiles CeilDiv(M, m1), the number of tiles in the N direction isnTiles CeilDiv(N, n1), and the total number of tasks istaskBlocks mTiles * nTiles. enableL1Resident can be enabled in the following two cases:mTiles 1,nTiles CoreNum, andK 2 * k1. In this case,l0CStages2can also be set (enableUnitFlag must be disabled). If there is not enough space andl0CStages2cannot be set, setn1to half of the original value.nTiles 1,mTiles CoreNum, andK 2 * k1. In this case,l0CStages2can also be set (enableUnitFlag must be disabled). If there is not enough space andl0CStages2cannot be set, setm1to half of the original value.BasicMatmul also supports DispatchPolicy MmadPreloadAsyncWithCallback, which supports the following template parameters:Template ParameterDefault ValueParameter DescriptionArchTagNoneSpecifies the architecture modelpreloadStagesNoneSpecifies the number of preloadsl1AStages2Number of buffers for loading matrix A on L1l1BStages2Number of buffers for loading matrix B on L1l0AStages2Number of buffers for loading matrix A on L0l0BStages2Number of buffers for loading matrix B on L0l0CStages1Specifies the number of L0C buffers. Set it to 2 to enable L0C double bufferingenableUnitFlagfalseSpecifies whether to enable UnitFlag. It must be set to false when L0C multi-buffering is enabledenableShuffleKfalseSpecifies whether to enable K-direction staggered readinguseHF32falseSpecifies whether to enable HF32. Only the float type is supportedenableL1ResidentfalseSpecifies whether to enable L1 residencyCompared withMmadPingpong,MmadPreloadAsyncWithCallbackhas two more template parameters. One ispreloadStages. This parameter is usually set to 1 and specifies the number of preloads. When this parameter is set to 1, the first loop only loads data and does not perform matmul computation. The second loop first loads the data for the second loop, and then completes the Matmul computation of the previous loop, and so on. After the final loop ends, one additional Matmul computation is performed. The benefit is that the data required for the current Matmul computation has already been moved in the previous loop. Therefore, instruction issue is advanced, which reduces the performance loss caused by instruction issue latency.The second parameter isenableShuffleK. This parameter is mainly used to avoid bandwidth loss caused by same-address access conflicts. The main principle is to stagger the data read addresses of each core. This parameter does not need to be enabled on 950.Compared withMmadPingpong,MmadPreloadAsyncWithCallbackhas more optimization points, but its logic is also more complex and has higher Scalar overhead. Use it based on the scenario, especially for small Shape scenarios.【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创作声明：本文部分内容由AI辅助生成（AIGC），仅供参考