PDA

View Full Version : *nix client optimisation


alpha
12-14-2002, 07:29 AM
Hey all,

I don't doubt anybody hasn't already figured this out, but by tweaking the Makefile a little you can improve performance of the *nix client.

On my K6-2 400 FreeBSD box, I managed to increase performance by 3k iterations/s simply by adding a '-march=k6 -mcpu=k6' to the end of the line starting 'OPT3' in src/Makefile. I used k6 and not k6-2 because I am using gcc 2.95.4, not the 3.x series.

I intend to play with gcc 3.x later today (and it's many optimisation flags), has anyone else tried this yet? I managed to get the ECCp-109 *nix client to perform significantly faster using gcc 3.x, so I am hopeful.
alpha
12-20-2002, 02:28 AM
Doesn't compile with gcc 3.3.
rus6
12-20-2002, 05:53 AM
I tried compiling the code (some time ago) and it complained about zero's on the end of the line, I had a quick play but didn't manage to get it to compile? Maybe the code's been cleaned up...

Then the RC5-72 linux client came out...

I was using red-hat 6.x cartman I think...
alpha
12-21-2002, 03:57 AM
Originally posted by rus6
I tried compiling the code (some time ago) and it complained about zero's on the end of the line, I had a quick play but didn't manage to get it to compile? Maybe the code's been cleaned up...


To get it to compile you have to remove all the ^M's from each file that the compiler looks at. It's simple enough, fire up your favourite editor and do some find and replace'ing. You should only need to do this for:

./Makefile
./include/ecc2-109.h
./src/mult109_mmx.c

where . is the top directory of the ECC2 source.

Because I am using FreeBSD I also had to change 'make' to 'gmake' in one of the Makefiles.

I'm actually taking a break from this project right now, but I will return shortly ;)
alpha
01-05-2003, 06:07 AM
Just reached a personal milestone, I was trying to compile an optimized core for my FreeBSD box that would exceed 155000 iterations/sec.

Current: 155115.5 its/sec (4.13 secs), 304426176 its since last DP.

Compiling out-of-the-box gives a core which peaks at 149100 iterations/sec. So thats a performance increase of 4.03%, right?

This was done with gcc optimization flags alone. Any C hackers out there going to optimize the source? ;)
alpha
01-06-2003, 03:45 PM
Originally posted by alpha
Any C hackers out there going to optimize the source? ;)

Apparently so! (http://www.ecc2.com/forum/viewtopic.php?t=140&highlight=)