[2359] in cryptography@c2.net mail archive
Re: comparing CPUs running blowfish [Re: fastest blowfish.asm?]
daemon@ATHENA.MIT.EDU (Adam Back)
Wed Mar 25 10:06:33 1998
Date: Wed, 25 Mar 1998 11:46:30 GMT
From: Adam Back <aba@dcs.ex.ac.uk>
To: eay@cryptsoft.com
CC: peterg@kcbbs.gen.nz, coderpunks@toad.com, cryptography@c2.net
In-reply-to: <Pine.GSO.3.96.980325150544.2087V-100000@pandora.cryptsoft.com>
(message from Eric Young on Wed, 25 Mar 1998 15:41:27 +1000 (EST))
Eric Young <eay@cryptsoft.com> writes:
> On Wed, 25 Mar 1998, Adam Back wrote:
> > This could be read as a story about the vagaries of attempting to tune
> > assembler language on the various pentiums and clones...
>
> :-), you think that is bad, try playing with the bf_opts.c program.
> It tweaks the different build options available for the inner loop in the C
> code. For my current version (0.8.2?), on a pentium 133, NT,
>
> options BF ecb/s
> <nothing> 243567.89 100.0%
> ptr 220833.43 90.7%
> ptr2 120301.05 49.4%
(was that a non MMX p133?)
!! I used ptr2 because it didn't seem to make much difference with the
AMD K6, and because this comment in the Makefile of 0.8.2b is wrong!:
#OPTS= -DBF_PTR2 # use for pentium
as is the force to auto-select BF_PTR2 in the bf_locl.h file. (Yes I
had noticed that -- it splurges redefinition warnings when compiling
with the script I posted, and I was trying to turn BF_PTR2 on not
off).
> I strongly suspect I have the default C options for gcc-x86 stuffed up in
> the 0.8.2 release.
Yes you do, you have missed off "-m486 -DCPU=pentium", though this
seemed to make little to no difference with -DBF_PTR2 / AMD k6. This
discovery was the reason for my explicit compile script, to force the
options to be the same for both distributions, I thought initially the
slower 082b code was the difference in compile options. It is not, as
I am compiling with -DBF_PTR2 -m486 -DCPU=pentium -fomit-frame-pointer
-O3 for all. So my statement on apparent slowness of 082b C code
holds but applys to the BF_PTR2 compiled C code on 082b being a lot
slower than the 072m C code compiled with the same options, which
is uninteresting though because BF_PTR2 is the worst option.
> > Follows is timings of a AMD K6 MMX, AMD K5 and Intel MMX all clocked
> > at 166 Mhz.
>
> I can do pentium 100/133 and ppro 200. What program are us using
> for timing? For the fast ciphers, the byte/pack/unpack overhead can
> cost alot, so the specific function being used matters quite a bit.
> I assume the speed program?
Yes bfspeed. I will retry without BF_PTR2.
Adam
