[EMAIL PROTECTED] wrote:
> Simon Burge writes (Hi, Simon!):
Howdy!
> The 2.6a source has been moved to the /pub/archived directory, for those
> of you who want to see what changes I made to the source to squeeze more
> speed out of the code.
I'm curious - there seems to be definite patterns in the sections of
code (especially when comparing the different length radix subroutines).
Is any of this code machine generated, or is it all hand generated? In
the past I attempted to get a C version of your code working but didn't
quite get there, and then Gord Palameta did a version last year. I was
thinking of taking a stab at trying 2.6b again...
> I look forward to the revised (and hopefully improved) ev6 timings!
Here's some more times with a few explainitory notes below:
Platform/per-iteration time (sec)
500MHz 21264 500MHz 21264 500MHz 21264
64kB I-cache 64kB I-cache 64kB I-cache
64kB D-cache 64kB D-cache 64kB D-cache
4MB L2 4MB L2 4MB L2
Mlucas2.6a[1] Mlucas2.6b[2] Mlucas2.6b[3] [4]
FFT length: ------------- ------------- ------------- -------
128K 0.043 0.040 0.041 0.001
160K 0.051 0.049 0.053 0.003
192K 0.062 0.062 0.066 0.004
224K 0.081 0.081 0.081 0.001
256K 0.10 0.089 0.093 0.003
320K 0.12 0.112 0.118 0.005
384K 0.16 0.14 0.15 0.009
448K 0.19 0.18 0.18 0.003
512K 0.21 0.19 0.20 0.004
640K 0.28 0.25 0.26 0.015
768K 0.38 0.34 0.36 0.011
896K 0.40 0.38 0.38 0.006
1024K 0.46 0.42
[1] - This is the average for the first N iters INCLUDING SETUP TIME.
N ranges from 8000 for a 128K FFT to 500 for a 1024K FFT.
[2] - This is the average for the time difference between a 100 iter
run and a 200 iter run as described in the README file and
completely ignored by me the first time around :-) This using
my binary compiled with:
f90 -arch ev6 -tune ev6 -fast -O5 -assume accuracy_sensitive
[3] - As for [2], but using Ernst's Mlucas_2.6b.exe.ev6 binary. In
both [2] and [3], the residues matched.
[4] - The speed increase in secs/iter between my binary and Ernst's
binary. Ernst - what compiler options did you use?
Simon.
_________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
