[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CABob6io_TAZv=R+_5YmAnw5phSsCd1t97YxdXpOfU7qijFfJUQ@mail.gmail.com>
Date: Tue, 26 Nov 2013 06:41:19 +0100
From: Lukas Odzioba <lukas.odzioba@...il.com>
To: john-dev@...ts.openwall.com
Subject: Re: md5crypt-opencl
2013/11/25 magnum <john.magnum@...hmail.com>:
> With that patch and a similar hack for ctx_update() I get 1881K c/s on
> Tahiti and 2383K c/s on the Titan. Like I said it can probably be done
> better, it's just a QnD hack for showing how bad 8-bit copies are.
For the record, this change leads to nice result on 570: 1240K c/s.
I removed intermediate copy to alt_result inside inner loop:
Tahiti@...er: 2465K c/s
TITAN: 2621K c/s
GTX 570: 1184K c/s
Tahiti@...l: 2120K c/s
After removing PUTCHAR(0x80) from md5_digest():
Tahiti@...er: 2660K c/s
TITAN: 2621K c/s
GTX 570: 1198K c/s
Tahiti@...l: 2359K c/s
Seems that we've reached (or close to that) alu bound on nv, so it
will be hard to make it much faster. On amd obviously we have to
improve memory operations.
I'll try to get rid of 8-bit r/w now.
Lukas
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.
