> Althought he FFT code itself probably has large influence on memory access
> patterns.

Yes! Good point! My FFT's in the optical code are all rather large'ish 2-D
FFT's (numbering 256x256 on average). The vendor supplied code is a 1-D FFT
wrapped by a hand coded C orchestrator to split these large 2-D arrays into
multiple threads of concurrent 1-D FFT's. But the sheer size of these arrays
will most likely flush the cache of anything needed by the higher level
OCaml analysis routines.

This machine is a consumer grade P4 and so it still incorporates a PCI bus,
and the cache is probably minimally sized. Such is the cost of doing science
on a shoestring budget.... But I ought not complain... I have a machine on
my desk with nearly 10^6 as much memory, and a thousand times faster, than I
once had, and I don't have to wait around all day for the printed program
output from my tray of cards representing a Fortran program...

- DM


-------------------
To unsubscribe, mail caml-list-request@inria.fr Archives: http://caml.inria.fr
Bug reports: http://caml.inria.fr/bin/caml-bugs FAQ: http://caml.inria.fr/FAQ/
Beginner's list: http://groups.yahoo.com/group/ocaml_beginners