I am an undergraduate student working on my thesis about parallel programming.
I am using OpenMP model. Now i want to use gem5 for measure performance.
That's why i install gem5 Full System successfully by following link:
http://cearial01.kaist.ac.kr/index.php/2016/08/26/gem5-documentation/
Now i want to compile & run a c program with OpenMP (matmul.c) using gem5.
How can i compile & run this program?
I mean inside which folder i stored this program file (matmul.c) for compile?
How i create object file of this program?
How can i change no of processor, cache memory size, no of cpu during running & compilation?
I am new student of this section. That's why my asking list is too big!
Hope anybody don't mind.
Best Regards,
Litu
How can I compile and run this program? I mean inside which folder I stored this program file (matmul.c) for compile? How I create object file of this program?
How to cross compile for an image is not gem5 specific, so I'll be brief.
First you must obtain a cross compiler for the image.
The best way to do that, is to get a cross compiler from the same source as the image to ensure compatibility.
My preferred approach is to use minimal Buildroot images. Buildroot:
- builds both the cross compiler and the image for me, and thus ensure compatibility
- makes it easy to automate building new software with its package system
- can produce very simple images, which are more suitable for gem5 and architecture research
This is my setup on GitHub. It contains a minimal OpenMP hello world which I have successfully run inside of gem5.
Another good option is to use https://crosstool-ng.github.io/
Since you want OpenMP support, you must build the GCC cross compiler compiler with support.
I think this is done in GCC with:
./configure --enable-libgomp
or through the Buildroot option:
BR2_GCC_ENABLE_OPENMP=y
And then when compiling, you must pass the -fopenmp option to gcc.
How can I change the number of processors and the cache memory size?
The best way to answer that question yourself is to use something like:
./build/ARM/gem5.opt configs/example/fs.py -h
and search the options.
You come across:
-n NUM_CPUS, --num-cpus=NUM_CPUS.
For ARM you also need to pass a .dtb with the corresponding core count, e.g.: ./system/arm/dt/armv7_gem5_v1_2cpu.dtb for 2 cores.
caches: you will find the following options easily:
--caches --l1d_size=1024 --l1i_size=1024 --l2cache --l2_size=1024 --l3_size=1024
But keep in mind that:
- the Linux kernel does not see the cache sizes correctly as of fbe63074e3a8128bdbe1a5e8f6509c565a3abbd4
How to switch CPU models in gem5 after restoring a checkpoint and then observe the difference?
- caches only affect certain CPU types, usually the more detailed such as ARM
HPI and x86 DerivO3CPU, but not AtomicSimpleCPU.
