[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Thu, 4 Jun 2015 02:05:55 +0200
From: Agnieszka Bielec <bielecagnieszka8@...il.com>
To: john-dev@...ts.openwall.com
Subject: Re: PHC: Parallel in OpenCL
2015-06-04 0:45 GMT+02:00 Lukas Odzioba <lukas.odzioba@...il.com>:
> She also implemented splitted kernel and it itself also degradated
> performance (from 28k to 27k c/s).
2015-06-03 21:59 GMT+02:00 Lukas Odzioba <lukas.odzioba@...il.com>:
> 2015-06-02 22:03 GMT+02:00 Agnieszka Bielec <bielecagnieszka8@...il.com>:
>> [a@...er run]$ ./john --test --format=parallel-opencl --dev=1 --skip-self-test
>> Speed for cost 1 (s) of 0, cost 2 (p) of 0
>> Many salts: 28493 c/s real, 3276K c/s virtual
>> Only one salt: 28493 c/s real, 3276K c/s virtual
>
> Ok, at least it is clear now that we have lost 1000 c/s by splitting kernel.
> Maybe we could limit code size by using gotos, but it would be a stunt
> and I don't recomend it.
test with value 28493 is for the version with split (all the time I've
been working on version with split kernels since I created it)
kernels.
value 28493 comes from http://www.openwall.com/lists/john-dev/2015/06/02/33
I don't know where the value 27k is from.
speed on --dev=5:
for the code with add 0 optimization without split kernels - 37/38k
IIRC, maybe 35k.
for the code with add 0 optimization with split kernels - 40k
speed on 960m:
for the code with add 0 optimization without split kernels - 32k
for the code with add 0 optimization with split kernels - 37k
on amd gcn speed decreased after making two the same functions SHA512
and SHA512_Z even without add 0 optimization (without split kernels)
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.
