[gmx-developers] Gromacs FFT

[Knox, Kent]

Hi all,

I'm interested in gromacs to find out how it uses FFT's.  I've
downloaded the latest tarballs from the website

Gromacs-4.0.2.tar.gz
Gmxtest-3.3.3.tgz
Gmxbench-3.0.tar.gz

built everything single-threaded single-precision, grabbed a profile of
the d.lzm benchmark, which I understand is the only included bench that
uses FFT's.

After Oprofiling the resulting executables, I am only finding about 2%
of the time is spent in fft's.

CPU: AMD64 family10, speed 2300 MHz (estimated)
Counted CPU_CLK_UNHALTED events (Cycles outside of halt state) with a
unit mask of 0x00 (No unit mask) count 100000
CPU_CLK_UNHALT...|
  samples|      %|
------------------
 11992213 41.3899 mdrun
  6312771 21.7879 libmd.so.5.0.0
  5121677 17.6770 libgmx.so.5.0.0
  3659315 12.6298 no-vmlinux
  1201312  4.1462 libm-2.4.so
   620405  2.1413 libfftw3f.so.3.2.2
    43580  0.1504 oprofiled
    20128  0.0695 libc-2.4.so

Does this sound typical?  Are FFT's not a bottleneck?  Maybe there are
better workloads available that stress FFT computation?

For disclosure, I do not know much about Molecular Dynamics, but am
interested in interesting and useful workloads that stress math
computation.

Thanks in advance for any of your time.

Kent Knox
Member of Technical Staff; AMD ACML