Migration Procedure
- Install KAIL based on "Installing KAIL" in Kunpeng AI Library Developer Guide.
- Obtain the adaptation code of KAIL_DNN v2.0.0.
git clone -b v2.0.0 https://gitee.com/openeuler/kail_dnn_adapter
- Access kail_dnn_adapter and compile oneDNN.
- New Kunpeng 920 processor model
cd kail_dnn_adapter sh build.sh --kml_dir=/path/to/HPCKit/latest/kml/gcc
Path to libdnnl.so: out/oneDNN-3.4/build/src/
Paths to the dependency libraries:- Path to .so files related to ACL: out/ComputeLibrary-23.11/build/
- Path to .so files related to KAIL: /usr/local/kail/lib/libkdnn.so
- Paths to .so files related to KML:
- /path/to/HPCKit/latest/gcc/kml/lib/sve/kblas/multi/libkblas_armv9p_v25.0.0.so
- /path/to/HPCKit/latest/gcc/kml/lib/noarch/kvml/multi/libkvml.so.25.0.0
- /path/to/HPCKit/latest/gcc/kml/lib/noarch/libkm.so.25.0.0
You can use all interfaces of oneDNN v3.4.0 by linking libdnnl.so.
The --kml_dir option in the compilation command specifies the installation path of KML.
In the .so paths related to KML, /path/to/ indicates the HPCKit installation path.
- New Kunpeng 920 processor model
Parent topic: Migrating oneDNN to KAIL_DNN