Bug 99553 - Tracker bug for runnning OpenCL applications on Clover
Summary: Tracker bug for runnning OpenCL applications on Clover
Status: NEW
Alias: None
Product: Mesa
Classification: Unclassified
Component: Drivers/Gallium/radeonsi (show other bugs)
Version: git
Hardware: Other All
: medium normal
Assignee: Default DRI bug account
QA Contact: Default DRI bug account
URL:
Whiteboard:
Keywords:
Depends on: 64225 68857 69897 72785 74140 74973 74974 77907 78163 80876 81896 82717 84232 87071 87738 88886 91305 91315 93370 93466 93977 94273 96296 96881 96897 96934 96935 97250 99510 99540 99764 99765 100105 100199 100212 100218 101584 102009 102179 102233 102595 103586 64201 70199 73510 78581 88669 90121 91556 94040 94503 94525 99488 99539 99552 99856 99950 100067 101415 101594 101952 102030
Blocks:
  Show dependency treegraph
 
Reported: 2017-01-26 19:40 UTC by Vedran Miletić
Modified: 2017-11-08 15:23 UTC (History)
3 users (show)

See Also:
i915 platform:
i915 features:


Attachments

Note You need to log in before you can comment on or make changes to this bug.
Description Vedran Miletić 2017-01-26 19:40:30 UTC
Let's track all different applications having issues with Clover and radeonsi/r600 here.
Comment 1 darkbasic 2017-01-26 23:08:56 UTC
It would be easier to track the few apps which work, unfortunately :(
Comment 2 Christoph Haag 2017-01-27 14:57:41 UTC
Not to forget my mandelbulber bug 82717.

It was actually working with funfunctor's partial image support found at https://cgit.freedesktop.org/~funfunctor/mesa/log/?h=clover-image-support-enabled

Currently it fails on llvm-svn 293174 with

OpenCL Build log:       <unknown>:0:0: in function fractal3D void (%struct.sClPixel addrspace(1)*, %struct.sClInBuff addrspace(1)*, %struct.sClInConstants addrspace(2)*, %struct.sClReflect addrspace(1)*, %opencl.image2d_ro_t.0 addrspace(2)*, i32): unsupported call to function get_global_id

but since it's already been working with this non-mainlined code, it's one that can be crossed off the list maybe very soon already.
Comment 3 Vedran Miletić 2017-01-27 17:48:01 UTC
Thanks for contributing. Very glad to hear about Mandelbulber almost working. I have added it to [1] as well.

[1] https://en.wikipedia.org/wiki/List_of_OpenCL_applications
Comment 4 darkbasic 2017-01-27 19:14:40 UTC
Sorry for asking, but the reason behind this bug report is because someone is finally willing to put some love into clover or nothing changed in that regard?
Comment 5 Vedran Miletić 2017-01-27 19:23:41 UTC
(In reply to darkbasic from comment #4)
> Sorry for asking, but the reason behind this bug report is because someone
> is finally willing to put some love into clover or nothing changed in that
> regard?

I am planning to do that in the coming weeks/months/(decades/millennia), now that GROMACS finally works. I won't promise anything, but I would love to get at least CP2K and LAMMPS working (bug 99539 and bug 99540), which might help other applications as well.

Hopefully it will go a bit faster than it did with GROMACS now that I have grasped how different layers of the OpenCL stack work and how to fix bugs there.
Comment 6 darkbasic 2017-01-27 19:47:21 UTC
Thanks, I would love to have at least Darktable working. With the CPU it takes 30+ seconds to process a single photo, compared to 0.5 seconds when using OpenCL.
Comment 7 Vedran Miletić 2017-01-27 20:28:01 UTC
(In reply to darkbasic from comment #6)
> Thanks, I would love to have at least Darktable working. With the CPU it
> takes 30+ seconds to process a single photo, compared to 0.5 seconds when
> using OpenCL.

If you are up for experimentation, the patches mentioned in comment #2 might be worth trying.
Comment 8 nixscripter 2017-01-29 06:52:21 UTC
This is probably not the best place to mention it, but the GalliumCompute feature status matrix was last updated in June of 2015.

https://dri.freedesktop.org/wiki/GalliumCompute/

I imagine there have been improvements, including atomics fixes I've gotten merged to fix ImageMagick bugs I opened in the past year.

Could that perhaps be updated to reflect the current status? It might inspire other broken applications to try OpenCL again, and file bugs that could be more narrowly targeted.
Comment 9 Vedran Miletić 2017-03-22 16:49:28 UTC
I have beeing fishing through the Bugzilla backlog on Clover to see what's there. Most of the bugs are fixed by now, but a couple are still there. Sorry for bugspam.
Comment 10 M. Edward (Ed) Borasky 2017-05-09 06:28:31 UTC
clpeak - various symptoms over the past months but now a solid crash:
    https://bugzilla.redhat.com/show_bug.cgi?id=1433632
    https://github.com/krrishnarraj/clpeak/issues/32
    https://bugs.freedesktop.org/show_bug.cgi?id=96897

luxmark - crashes: http://www.luxrender.net/forum/viewtopic.php?f=16&t=13353

Any system where you have more than one OpenCL vendor using the same LLVM version: "CommandLine Error: Option 'enable-value-profiling' registered more than once!": https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=852746
Comment 11 Luke A. Guest 2017-06-25 18:36:31 UTC
Blender Cycles OpenCL support also doesn't work.
Comment 12 M. Edward (Ed) Borasky 2017-06-25 18:48:16 UTC
(In reply to Luke A. Guest from comment #11)
> Blender Cycles OpenCL support also doesn't work.

Is there a bug filed in a bug tracker somewhere for this one?
Comment 13 Luke A. Guest 2017-06-26 09:51:55 UTC
Doesn't seem to be.
Comment 14 Luke A. Guest 2017-06-26 10:05:59 UTC
Added https://bugs.freedesktop.org/show_bug.cgi?id=101594
Comment 15 Vedran Miletić 2017-08-23 13:58:51 UTC
Jan Vesely is maintaining a table of test results: http://paul.rutgers.edu/~jv356/piglit/gcn-latest-3/problems.html
Comment 16 Vedran Miletić 2017-08-23 14:22:03 UTC
And an equivalent one for r600, sorry for double comment: http://paul.rutgers.edu/~jv356/piglit/radeon-latest-5/problems.html
Comment 17 darkbasic 2017-09-22 14:10:57 UTC
Hi, any news on the OpenCL image support? I really miss darktable support.


Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct.