[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <10de44f1c84cbdebe36d3afaa8dfd673@smtp.hushmail.com>
Date: Thu, 26 Apr 2012 00:14:20 +0200
From: magnum <john.magnum@...hmail.com>
To: john-dev@...ts.openwall.com
Subject: Re: OpenCL optimisations for different cards
On 04/25/2012 11:25 PM, Samuele Giovanni Tonon wrote:
> On 04/25/12 22:21, Lukas Odzioba wrote:
>> 2012/4/25 magnum <john.magnum@...hmail.com>:
>>> On 04/25/2012 09:21 PM, SAYANTAN DATTA wrote:
>>>> Hi magnum,
>>>>
>>>> Is it okay with everyone if we enable profiling while creating the coomand
>>>> queue in common_opencl.c?
>>>
>>> And leave it on? Why? I suspect it might degrade performance. We'll see
>>> what the others say.
>>
>> I agree with magnum. In fact there is nothing about performance in
>> OpenCL specification, but it does not confess us from using it.
>> What's the purpose of this change?
>
> maybe that could help to integrate find_best_workgroup and find_best_kpc
> in common_opencl .
Would it make sense to move the queue creation from dev_init() to [the
format's] create_clobj()? Then we could add an argument to the latter so
it's create_clobj(int kpc, int prof) where prof is used as a boolean
for profiling or not.
I think this makes sense as we create and destory clobj during the
find_best functions.
magnum
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.
