Function
When developing template library operators, you can use the msKPP APIs to quickly generate, compile, and run kernel delivery code in Python scripts.
When tuning the operator performance, adjust the template parameters (for example, the L0shape size) in the kernel for multiple times and compare the performance results. To improve the tuning efficiency, msKPP provides autotune APIs, which allows developers to efficiently replace, compile, run, and compare the performance of the code.
The auto tuning function is supported only by the
Restrictions
- For a single device, only one msKPP tool can be used for auto tuning. You are not advised to run other operator programs at the same time.
- Ensure that import mskpp is performed before import acl. Otherwise, set environment variables before running the tool.
export LD_PRELOAD=${INSTALL_DIR}/lib64/libmspti.so
Parent topic: Auto Tuning