[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Sat, 23 Jun 2012 21:32:25 +0200
From: Frank Dittrich <frank_dittrich@...mail.com>
To: john-dev@...ts.openwall.com
Subject: Re: Several enhancements for relbench
On 06/23/2012 09:07 PM, Solar Designer wrote:
> Frank -
>
> On Sat, Jun 23, 2012 at 07:17:50PM +0200, Frank Dittrich wrote:
>> If a particular benchmark appears several times in the same file,
>> print a message to STDERR,
>
> OK.
>
>> and pick the higher values for
>> comparision (higher real value is considered here, only if the
>> real values are the same, the higher virtual value is considered).
>
> I think it's better to include ALGORITHM_NAME into the names being
> compared (for both formats) when an otherwise-matching format name is
> detected. Only if that still leaves matching format names, print a
> second warning and do something else (like what you implemented).
I am not sure I understand what you mean.
This change was intended to solve the following problem:
File1 contains:
Benchmarking: Raw MD5 [32/32]... DONE
Raw: 1489K c/s real, 1489K c/s virtual
File2 contains
Benchmarking: Raw MD5 [SSE2i 12x]... DONE
Raw: 5702K c/s real, 5702K c/s virtual
Benchmarking: Raw MD5 [SSE2i 10x4x3]... DONE
Raw: 5122K c/s real, 5122K c/s virtual
The current relbench version would compare the benchmark of File1 with
the second benchmark in File2 where the format name matches the one in
File1.
I thought it would be more fair to use the fastest of multiple
benchmarks in File2.
So I think this change is an improvement over what we currently have.
What scenario do you have in mind?
Should I always include the algorithm name when trying to find a match
between File1 and File2, and only if that doesn't work, try matching
format names only, and in that case pick the fastest of multiple
benchmarks in each File?
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.
