Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Mon, 11 May 2015 17:36:31 +0200
From: Frank Dittrich <frank.dittrich@...lbox.org>
To: john-dev@...ts.openwall.com
Subject: Re: Adding OpenMP support to SunMD5

On 05/11/2015 05:17 PM, Solar Designer wrote:
> On Sat, May 09, 2015 at 10:39:34PM +0800, Lei Zhang wrote:
>> [lei@...er src]$ GOMP_CPU_AFFINITY=0-31 ../run/john --test --format=sunmd5
>> Will run 32 OpenMP threads
>> Benchmarking: SunMD5 [MD5 128/128 AVX 4x3]... (32xOMP) DONE
>> Speed for cost 1 (iteration count) of 5000
>> Raw:	8302 c/s real, 259 c/s virtual
>>
>> The newer version is still slower.
> 
> Both are still slower than our target speed, which I measured at around
> 8800 c/s based on cumulative performance for --fork=32 with a version
> from just prior to your work.

I tried on super with latest bleeding-jumbo (now that magnum pulled the
changes):


$ scl enable devtoolset-2 bash
$ export GOMP_CPU_AFFINITY=0-31
$ export GOMP_SPINCOUNT=10000

$ for i in 31 32 64 128 256 512; do OMP_NUM_THREADS=$i ../run/john
--test=10 --format=sunmd5; done
Will run 31 OpenMP threads
Benchmarking: SunMD5 [MD5 128/128 AVX 4x3]... (31xOMP) DONE
Speed for cost 1 (iteration count) of 5000
Raw:	7939 c/s real, 290 c/s virtual

Will run 32 OpenMP threads
Benchmarking: SunMD5 [MD5 128/128 AVX 4x3]... (32xOMP) DONE
Speed for cost 1 (iteration count) of 5000
Raw:	7581 c/s real, 285 c/s virtual

Will run 64 OpenMP threads
Benchmarking: SunMD5 [MD5 128/128 AVX 4x3]... (64xOMP) DONE
Speed for cost 1 (iteration count) of 5000
Raw:	8597 c/s real, 295 c/s virtual

Will run 128 OpenMP threads
Benchmarking: SunMD5 [MD5 128/128 AVX 4x3]... (128xOMP) DONE
Speed for cost 1 (iteration count) of 5000
Raw:	8969 c/s real, 298 c/s virtual

Will run 256 OpenMP threads
Benchmarking: SunMD5 [MD5 128/128 AVX 4x3]... (256xOMP) DONE
Speed for cost 1 (iteration count) of 5000
Raw:	9282 c/s real, 301 c/s virtual

Will run 512 OpenMP threads
Benchmarking: SunMD5 [MD5 128/128 AVX 4x3]... (512xOMP) DONE
Speed for cost 1 (iteration count) of 5000
Raw:	9309 c/s real, 300 c/s virtual

This (higher c/s rate for OMP_NUM_THREADS >> number of cores) matches my
experience for sunmd5 on my hardware.

Frank

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.