Well, the main problem with OpenCL is the fact that there is no way to 
know what amount of GPU memory is factually available at a time. In your 
case I suspect that there is some other application running which 
reserves quite some amount even if it may not use it continuously 
(geeqie has been identified in the past as one candidate).

So you should first try again with as few other applications as possible 
running in parallel.

Ulrich

Am 20.02.2015 um 22:39 schrieb Fearsome:
> Ulrich, thanks for your suggestions.
>
> I've tried various settings for the two parameters you mentioned, but to
> no avail. For the number of event handles I tried 15 and 5, and for the
> headroom I tried 500 and 30. My defaults were 25 and 300, respectively.
> The out of resources error is encountered for the base curve module, in
> every case.
>
> I also tried processing CR2 files from other cameras than the 40D. The
> export for a file from a 50D went splendid, without any hick-ups in the
> OpenCL processing. The 50D's resolution is 1 MP less than that of the
> X100T. The export from a 60D did not go so well, however. The 60D's
> resolution is larger than that of the X100T by ~4 MP. The error was
> different this time, giving no reference to a specific module but
> instead referring to the copy-back:
>
> [opencl_pixelpipe (c)] late opencl error detected while copying back to
> cpu buffer: -5
>
> What does this mean? If the image was possible to transfer to the GPU,
> it should be possible to transfer _from_ the GPU.
>
>  From inspection of the X100T opencl export log, it seems like the whole
> image is transferred in one piece to the GPU and thus tiling is not
> possible. Then I suspect the whole pipe is redone on the CPU if there is
> a error from one of the modules on the GPU, which really hurts. Is it
> possible in the future to apply tiling for the GPU as well to avoid
> these problems?
>
> Any further thoughts on how I can enable error-free offloading to the
> GPU would be highly appreciated. Or is this just the way it is?
>
> /Fred
>
>
> On Fri, Feb 20, 2015 at 6:49 AM, Ulrich Pegelow
> <ulrich.pege...@tongareva.de <mailto:ulrich.pege...@tongareva.de>> wrote:
>
>     Hi,
>
>     that's not an X-Trans related error. In fact the specific demosaic
>     algorithms for the X-Trans sensor are not (yet) ported to OpenCL and
>     always run on the CPU.
>
>     The error you see is code -5: "out of resources". Have a look into the
>     user manual for some ideas what to do. You most likely need to tune one
>     of the config parameters opencl_number_event_handles or
>     opencl_memory_headroom.
>
>     Best wishes
>
>     Ulrich
>
>     Am 19.02.2015 um 22:51 schrieb Fearsome:
>      > Hi,
>      >
>      > I'm trying to set up OpenCL and I thought I had succeeded. It works
>      > nicely for exporting Canon CR2 files but for RAF files from a X100T
>      > there are errors in the OpenCL profiling output from DT. The export
>      > seems to fall back to the CPU for some, to unknown, reason. A log
>     of the
>      > output is attached. I would appreciate any help in understanding
>     what's
>      > happening and whether I've set up OpenCL properly or not.
>      >
>      > I'm running DT 1.6.2 on Ubuntu 14.10 with a Geforce GTX 960
>     installed.
>      >
>      > Cheers,
>      >
>      > Fred
>      >
>

------------------------------------------------------------------------------
Download BIRT iHub F-Type - The Free Enterprise-Grade BIRT Server
from Actuate! Instantly Supercharge Your Business Reports and Dashboards
with Interactivity, Sharing, Native Excel Exports, App Integration & more
Get technology previously reserved for billion-dollar corporations, FREE
http://pubads.g.doubleclick.net/gampad/clk?id=190641631&iu=/4140/ostg.clktrk
_______________________________________________
Darktable-users mailing list
Darktable-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/darktable-users

Reply via email to