[CPMD-list] NCACHE size

[Kozin, I (Igor)]

Hello,
I'd appreciate to hear any comments on varing NCACHE size 
(as given in mltfft.F) on various platforms.
Particularly Xeon, Itanium2, Opteron, Power4,5, NEC sx6.

In the file above NCACHE is given as 1024*N where N varies.
So I'd guess it has to do with L1 cache.
For any i386 NCACHE=1024*10.
Xeon's L1 cache is 8 KB. So should N be 8?

Itanium2: NCACHE=1024*8 but L1 cache size is 16 KB.
This is the only machine I've experimented with so far.
Taking N = 8, 16, 32 I found that N = 16 is marginally quicker
(wat32 benchmark, run 1: 1372 s, 1363 s, 1388 s).
So basically no need to bother.

More interesting cases:
Itanium2 + HPUX: NCACHE=1024*64 
(obviously the L1 cache is the same as above)

Operon: NCACHE=1024*2 (default) but L1 cache is 64 KB.

BTW, are there any efforts in attempting to use ACML or MKL
on AMD and Intel respectively for FFT instead of default FFT?

Thanks,

Igor Kozin
Computational Science & Engineering Dept.
CCLRC Daresbury Laboratory
Keckwick Lane
Warrington
WA4 4AD
UK

i. kozin at dl.ac.uk
+44 (0) 1925 603308
http://www.cse.clrc.ac.uk/disco