[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <32d902e565430c834d4e288de2ee45a2@smtp.hushmail.com>
Date: Wed, 23 May 2012 22:38:53 +0200
From: magnum <john.magnum@...hmail.com>
To: john-dev@...ts.openwall.com
Subject: Re: Nvidia compiler bug
On 05/23/2012 03:58 PM, Claudio André wrote:
> Em 22-05-2012 16:39, magnum escreveu:
>> So this made the compiler actually use local memory? Weird. I'll try
>> that in RAR.
>
> The compiler was using local memory, but it wasn't showing the memory
> usage on its verbosity (It can't on compile time using 1st syntax). And
> my fault, i glanced at it and start thinking the wrong way and i tried
> to fight against the compiler. I failed miserably.
Ah, yes of course. BTW when you query for local memory usage with
CL_KERNEL_LOCAL_MEM_SIZE, some implementations will include dynamically
allocated local memory (even without your workaround) while others will
just show static allocations. I think nvidia did not show dynamic while
AMD did. Obviously for this to work I had to set the kernel arguments
before doing this query (or it would simply not have a chance to know
how much was dynamically allocated).
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.
