Performance Count

Collect hardware count and temperature when running benchmark

Very simple instruction of installation:

Prerequisites:

System：Ubuntu 20.04.5 LTS
cpu:i5-6500
third-party tools:
(1)cpupower, to control the running frequency
(2)lm-sensors, to read temperature sensor results
hardware: The CPU should have built-in temperature sensors

Installation:

change the CMakeLists.txt in the top directory, if necessary

by default, the two feature, reading hardware temperature sensors and estimating temperature based on performance counters, are both enabled. The user can disable and enable them by changing the line 3 and 4 of the CMakeLists.txt in the top directory. For example, to disable reading hardware sensors, change line 3 to: set(HARD_TEMPERATURE_SENSOR_ENABLE "OFF")

pre-complile PAPI, if you want to use the software temperature estimator

cd third_party_lib/papi-5.5.1/src

./configure

make

sudo make install-all

cd ../../../

make and go to the program build directory

mkdir build

cd build

cmake

cmake ..

compile the program

make

then move mcftp to the main folder

run the program:

first change the cpu frquency governor as userspace, if you want to control the cpu frequency

a. sudo cpupower -c all frequency-set -g userspace

set FREQUENCY that your device support

b. sudo cpupower -c all frequency-set -f FREQUENCY

then go to the program directory

c. cd ../main

then make

d. make

run the program and collecting data

e. sudo python test.py

notice

A file saving performance counts and related hardware temperatures will be generated in this directory one hour later

You can modify the running time of the program and the frequency of data collection in Examplebenchmark.xml,test.py and preset_native_test.c

​