|
Message-ID: <CA+TsHUAYK0rzCcJDWf7xsvd1_QhR6S8+MBfqkK=b1sj0wDKZRQ@mail.gmail.com>
Date: Fri, 1 Jun 2012 15:29:55 +0000
From: SAYANTAN DATTA <std2048@...il.com>
To: john-dev@...ts.openwall.com
Subject: Re: clBuildProgram on a multi-GPU environment
On Fri, Jun 1, 2012 at 3:05 PM, Claudio André <claudioandre.br@...il.com>wrote:
> Hi, this might make sense to get into the new release. In common_opencl.c:
>
> - build_code = clBuildProgram(program[dev_id], 0, NULL,+ build_code = clBuildProgram(program[dev_id], 1, &devices[dev_id],
>
>
> Seems to me the patch is better on a multi-GPU environment. The situation
> on John now is NULL on device_list parameter of clBuildProgram. But "If
> device_list is NULL value, the program executable is built for all
> devices ..." and we are going to run the software on only 1.
>
>
> https://github.com/claudioandre/magnum-jumbo/commit/9bf37df59e503ccd60705200446b9a1e18c9ee92
>
> Claudio
>
> [1]
> http://www.khronos.org/registry/cl/sdk/1.0/docs/man/xhtml/clBuildProgram.html
>
It would build for only one device because the program object is created
for only one device. But still your patch seems more appropriate.
Content of type "text/html" skipped
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.