[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Fri, 6 Jul 2012 17:31:22 -0500
From: "jfoug" <jfoug@....net>
To: <john-dev@...ts.openwall.com>
Subject: RE: Between 1.7.9-jumbo-3 and 1.7.9-jumbo-4 performance of --format=bf dropped by more than 20%
>From: Frank Dittrich [mailto:frank_dittrich@...mail.com]
>
>Not that 32bit Linux is that important, but on my atom netbook (fedora
>16, gcc 4.6.3) i got:
>
>
>$ ./john-1.7.9-jumbo-3-sse2 --test=10 --format=bf
>Benchmarking: OpenBSD Blowfish (x32) [32/32]... DONE
>Raw: 216 c/s real, 217 c/s virtual
>
>$ ./john-1.7.9-jumbo-4-sse2 --test=10 --format=bf
>Benchmarking: OpenBSD Blowfish (x32) [32/32 X2]... DONE
>Raw: 167 c/s real, 167 c/s virtual
>
>and latest git:
>
>Unfortunately, on github there is no commit between jumbo-3 and jumbo-4,
>so I cannot bisect to find out what caused this drop in performance.
Look at the algorithm. Goes from [32/32] to [32/32 X2]. Different
algorithms. The X2 must have been added for some reason, but on your build,
or your CPU (Celeron??), or your memory, or 'something', it is slower than
the 1x.
The change is in x86-sse.h
// current for 32 bit sse builds:
#define BF_ASM 0
#define BF_X2 1
// to get back to [32/32] change them to this:
#define BF_ASM 1
#define BF_X2 0
Here are timings on my system:
Benchmarking: OpenBSD Blowfish (x32) [32/32 X2]... DONE
Raw: 601 c/s real, 608 c/s virtual
Benchmarking: OpenBSD Blowfish (x32) [32/32]... DONE
Raw: 417 c/s real, 419 c/s virtual
Both of these timing were from magnum-bleeding. The [32/32] build I simply
changed the x86-sse.h file.
Jim.
Powered by blists - more mailing lists
Confused about mailing lists and their use?
Read about mailing lists on Wikipedia
and check out these
guidelines on proper formatting of your messages.
