Build and Configuration

This demo uses the matmul binary as an example. Instrumentation code has been added to the demo based on the Instrumentation Guide.

Download and install the KML.

unzip -o BoostKit-kml_2.2.0.zip 
rpm -ivh boostkit-kml-2.2.0-1.aarch64.rpm 
export KML_INCLUDE=/usr/local/kml/include/ 
export KML_LIB=/usr/local/kml/lib/kblas/omp/

Download the demo and compile matmul.

git clone xxx 
cd xxx 
export RFEVENTS_INCLUDE=/usr/local/devkit/tuner/include/ 
export RFEVENTS_LIB=/usr/local/devkit/tuner/lib/ 
make matmul

/usr/local/devkit/tuner is an example path generated after installing the RPM package of the Kunpeng DevKit command line tool. Replace it with the actual path.

matmul has the following Makefile content:

gcc -g -march=armv8.2-a+fp16fml+simd -O3 -fopenmp $(SOURCES) -o matmul -DROOFLINE_EVENTS -I$(RFEVENTS_INCLUDE) -L$(RFEVENTS_LIB) -lrfevents -DENABLE_KML -I$(KML_INCLUDE) -L$(KML_LIB) -lkblas

Set the LD_LIBRARY_PATH environment variable.

export LD_LIBRARY_PATH=$KML_INCLUDE:$KML_LIB:/usr/local/devkit/tuner/lib:$LD_LIBRARY_PATH

Explicitly specify OMP_NUM_THREADS based on the number of physical cores.
1
export OMP_NUM_THREADS=128

Parent topic: Sample 1: Using Roofline Analysis to Tune Applications