[Bf-cycles] opencl with ATI @ OSX 10.7.4 (better but still buggy)

Dalai Felinto dfelinto at gmail.com
Mon May 14 03:13:16 CEST 2012


Hi,
just for reference I asked Daniel Salazar (ZanQdo) to test it his linux
machine (nvidia there).
In cuda the result is perfect (both viewport and render).

In opencl the result is good in render, but in viewport he gets:
http://www.pasteall.org/pic/show.php?id=31740

Thanks,
Dalai

* http://www.pasteall.org/31802/diff this is the actual patch (the one I
sent is missing a '-').
* the example file does have  ui, you need to use ctrl+right to see a
screen with it (sorry Daniel ;)

2012/5/13 Dalai Felinto <dfelinto at gmail.com>

> Hi there,
> It's time for another round of OSX ATI debugging :)
>
> This week Jens started re-testing the opecl compiler after OSX 10.7.4
> updates.
> There is an overall progress, but I'm still having some issues. They may
> even be different than his'.
>
> My test device: AMD Radeon HD 6750M
>
> I had to disable CL_NO_FLOAT3 and enable KERNEL_SHADING.
> If the NO_FLOAT3 is on the kernels don't build (*)
>
> (1) good render: http://www.pasteall.org/pic/31736
> (2) bad gpu viewport: http://www.pasteall.org/pic/31738
> (3) terrible gpu render: http://www.pasteall.org/pic/31737
>
> ((2) and (3) also tested with screen size matching render size and tile
> size on its maximum)
>
> I remember running into the problem in (2) with opencl in linux and even
> with cuda. the problem in (3) may be something unique though.
>
> - patch to test: http://www.pasteall.org/31792/diff
> - sample file: http://www.pasteall.org/blend/13883
>
> In my computer the biggest problem is the shadow. Any clues on what to try?
> @ Jens, you mentioned having troubles with the emitting kernel. Can you
> send me a sample file? In my test and computer it worked fine.
>
> Cheers,
> Dalai
>
> (*) #####
> Compiling OpenCL kernel ...
> OpenCL build failed: errors in console
> Error getting function data from server
> ######
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.blender.org/pipermail/bf-cycles/attachments/20120513/8343a5c9/attachment.htm 


More information about the Bf-cycles mailing list