This program aims to compute machin-like approximation of pi and measure performance between host only implementation and host with epiphany implementation.
This example uses host and epiphany. To print csv data, provide -D STAT in the
compilation stage.
to run
./build.sh && ./main.elf
It is possible to passes as arguments the main number of iteration and the sub number of iterations :
./main.elf [main number] [sub number]
The sub number of iterations is the iterations for the eCores while the main number of iterations is the number of time the host will invoke the eCores.
Expected output :
PI = 3.141584 time spent 7.880000s
With stat option (gcc with -D STAT) :
10000, 1000, 7.880000, 3.1415
A bash script is provided and aims to outputs performance stats as csv files. Be sure first to enable csv output in the build script (with -D STAT).
The first script (stats.sh) takes no parameter and produces data in rows :
./stats.sh
And will produce output stats.csv.
The second script (stats_chunck.sh) produces data in a progressiv way along rows and cols of a square. It takes as parameter the start and the end of the portion :
./stats_chunck.sh 31 50
And will produce a file output stats_chunck_31_50.csv.
./speedup.sh
Will run the program with different number of cores and output a csv file.
Tips to run on different screen, type screen to start a new session. Then
leave it with CTRL-A CTRL-D, come back with screen -r.
That step is useful to launch stats from ssh without ending it on leaving ssh
session.
MIT