Openwall GNU/*/Linux - a small security-enhanced Linux distro for servers
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Fri, 20 Apr 2012 11:54:18 +0200
From: magnum <john.magnum@...hmail.com>
To: john-dev@...ts.openwall.com
Subject: Re: [Off-topic] Experiments!

On 04/19/2012 08:38 PM, Claudio André wrote:
> Hi, i'm not sure what are you guys expecting about results in (my)
> OpenCL implementation.

I think all OpenCL formats should be labeled "experimental" right now,
including my own.

> Anyway, i can say i saw a lot of crazy things while playing with AMD
> compiler. I put Ivan's opinion below. Lukas said something some time
> ago. Somehow, i can even put the compiler in loop, something even i
> (myself) think is hard to believe.

I'm starting to get really frustrated too. Not only our formats are
immature, all OpenCL SDK's and tools are too. Bugs, weird "features",
undocumented behavior. An example: after days of trying to figure out
why I could not bump global work size in RAR, it turned out that
"__global const" is apparently treated as "__constant" by nvidia - thus
getting a serious size limitation. But the end result was not warnings
or build errors - just a crashing runtime. This is unacceptable, I'm
thinking maybe I should focus on CUDA until OpenCL is more mature.

> Not sure about your plans for the source code or if you want to keep it.

Sure. Some day I might try to fix it for nvidia. I would not be
surprised if it's a bug or quirk in nvidia's OpenCL rather than a
problem with your code.

magnum

Powered by blists - more mailing lists

Your e-mail address:

Powered by Openwall GNU/*/Linux - Powered by OpenVZ