hi,
at the moment we are trying to optimize VASP for an IBM power6 machine. This parameter -DCACHE_SIZE what is it actually for? And more important how can you find/calculate what value you should use(or start from as a good estimate)
A power6 machine ahe 64kb L1 cache, 4Mb L2 per core and 32Mb L3 per core...
thanks
Danny
DCACHE_SIZE
Moderators: Global Moderator, Moderator
-
Danny
- Full Member
- Posts: 201
- Joined: Thu Nov 02, 2006 4:35 pm
- License Nr.: 5-532
- Location: Ghent, Belgium
- Contact:
DCACHE_SIZE
Last edited by Danny on Thu Aug 14, 2008 8:12 pm, edited 1 time in total.
-
job
- Jr. Member
- Posts: 55
- Joined: Tue Aug 16, 2005 7:44 am
DCACHE_SIZE
It's ostensibly the size of the L1D cache in units of DP FP variables. That should give you a starting point. Compile with different values and benchmark to find the best.
Though if you use FFTW instead of the built-in FFT library it's shouldn't matter that much. You should probably be using FFTW, in my recent benchmarks on a quad-core Opteron using FFTW improved performance by about 25 %.
Though if you use FFTW instead of the built-in FFT library it's shouldn't matter that much. You should probably be using FFTW, in my recent benchmarks on a quad-core Opteron using FFTW improved performance by about 25 %.
Last edited by job on Fri Aug 15, 2008 10:53 am, edited 1 time in total.