How to enable OpenCL under Ubuntu 24.04 on Framework 16 (AMD Radeon™ RX 7700S)

Software darktable
,
1

I’m running Ubuntu 24.04.1 LTS on a Framework Laptop 16 (AMD Ryzen™ 7040 Series with dGPU (AMD Radeon™ RX 7700S)).

I’m completely new to OpenCL and currently $ clinfo --list gives me an empty output. I’d like to get OpenCL running to use it with darktable (deb installation). I understood, that there are (at least) three ways to use OpenCL:

So what is the recommended way to enable OpenCL? Any good tutorial you can suggest as a starting point? Thx for any support!

2

Try the rocm first. Check online how to enable it in Ubuntu. On Fedora it is easy: sudo dnf install rocm-opencl

2 Likes
3

Thanks for the recommendation, @g-man! I followed the official installation manual, which worked fine.

If I understand it correct darktable-cltest recognized both, the iGPU (as device 1: gfx1103) and the dGPU (as device 0: gfx1102):

darktable-cltest output 
$ darktable-cltest 
darktable 4.8.1
Copyright (C) 2012-2024 Johannes Hanika and other contributors.

Compile options:
  Bit depth              -> 64 bit
  Debug                  -> DISABLED
  SSE2 optimizations     -> ENABLED
  OpenMP                 -> ENABLED
  OpenCL                 -> ENABLED
  Lua                    -> ENABLED  - API version 9.3.0
  Colord                 -> ENABLED
  gPhoto2                -> ENABLED
  GMIC                   -> ENABLED  - Compressed LUTs are supported
  GraphicsMagick         -> ENABLED
  ImageMagick            -> DISABLED
  libavif                -> DISABLED
  libheif                -> ENABLED
  libjxl                 -> ENABLED
  OpenJPEG               -> ENABLED
  OpenEXR                -> ENABLED
  WebP                   -> ENABLED

See https://www.darktable.org/resources/ for detailed documentation.
See https://github.com/darktable-org/darktable/issues/new/choose to report bugs.

     0,0226 [dt_get_sysresource_level] switched to 2 as `large'
     0,0226   total mem:       88298MB
     0,0226   mipmap cache:    11037MB
     0,0226   available mem:   60360MB
     0,0227   singlebuff:      1379MB
     0.0251 [dt_dlopencl_init] could not find default opencl runtime library 'libOpenCL'
     0.0252 [dt_dlopencl_init] could not find default opencl runtime library 'libOpenCL.so'
     0.0252 [opencl_init] opencl library 'libOpenCL.so.1' found on your system and loaded, preference 'default path'
     2.1670 [opencl_init] found 1 platform
[opencl_init] found 2 devices

[dt_opencl_device_init]
   DEVICE:                   0: 'gfx1102'
   CONF KEY:                 cldevice_v5_amdacceleratedparallelprocessinggfx1102
   PLATFORM, VENDOR & ID:    AMD Accelerated Parallel Processing, Advanced Micro Devices, Inc., ID=4098
   CANONICAL NAME:           amdacceleratedparallelprocessinggfx1102
   DRIVER VERSION:           3635.0 (HSA1.1,LC)
   DEVICE VERSION:           OpenCL 2.0 
   DEVICE_TYPE:              GPU, dedicated mem
   GLOBAL MEM SIZE:          8176 MB
   MAX MEM ALLOC:            6950 MB
   MAX IMAGE SIZE:           16384 x 16384
   MAX WORK GROUP SIZE:      256
   MAX WORK ITEM DIMENSIONS: 3
   MAX WORK ITEM SIZES:      [ 1024 1024 1024 ]
   ASYNC PIXELPIPE:          NO
   PINNED MEMORY TRANSFER:   NO
   USE HEADROOM:             600Mb
   AVOID ATOMICS:            NO
   MICRO NAP:                250
   ROUNDUP WIDTH & HEIGHT    16x16
   CHECK EVENT HANDLES:      128
   TILING ADVANTAGE:         0.000
   DEFAULT DEVICE:           NO
   KERNEL BUILD DIRECTORY:   /usr/share/darktable/kernels
   KERNEL DIRECTORY:         /home/simon/.cache/darktable/cached_v3_kernels_for_AMDAcceleratedParallelProcessinggfx1102_36350HSA11LC
   CL COMPILER OPTION:       -cl-fast-relaxed-math
   CL COMPILER COMMAND:      -w -cl-fast-relaxed-math  -DAMD=1 -I"/usr/share/darktable/kernels"
   KERNEL LOADING TIME:       0.0171 sec

[dt_opencl_device_init]
   DEVICE:                   1: 'gfx1103'
   CONF KEY:                 cldevice_v5_amdacceleratedparallelprocessinggfx1103
   PLATFORM, VENDOR & ID:    AMD Accelerated Parallel Processing, Advanced Micro Devices, Inc., ID=4098
   CANONICAL NAME:           amdacceleratedparallelprocessinggfx1103
   DRIVER VERSION:           3635.0 (HSA1.1,LC)
   DEVICE VERSION:           OpenCL 2.0 
   DEVICE_TYPE:              GPU, unified mem
   GLOBAL MEM SIZE:          44149 MB
   MAX MEM ALLOC:            37527 MB
   MAX IMAGE SIZE:           16384 x 16384
   MAX WORK GROUP SIZE:      256
   MAX WORK ITEM DIMENSIONS: 3
   MAX WORK ITEM SIZES:      [ 1024 1024 1024 ]
   ASYNC PIXELPIPE:          NO
   PINNED MEMORY TRANSFER:   NO
   USE HEADROOM:             600Mb
   AVOID ATOMICS:            NO
   MICRO NAP:                250
   ROUNDUP WIDTH & HEIGHT    16x16
   CHECK EVENT HANDLES:      128
   TILING ADVANTAGE:         0.000
   DEFAULT DEVICE:           NO
   KERNEL BUILD DIRECTORY:   /usr/share/darktable/kernels
   KERNEL DIRECTORY:         /home/simon/.cache/darktable/cached_v3_kernels_for_AMDAcceleratedParallelProcessinggfx1103_36350HSA11LC
   CL COMPILER OPTION:       -cl-fast-relaxed-math
   CL COMPILER COMMAND:      -w -cl-fast-relaxed-math  -DAMD=1 -I"/usr/share/darktable/kernels"
   KERNEL LOADING TIME:       0.0151 sec
[opencl_init] OpenCL successfully initialized. internal numbers and names of available devices:
[opencl_init]		0	'AMD Accelerated Parallel Processing gfx1102'
[opencl_init]		1	'AMD Accelerated Parallel Processing gfx1103'
     2.5902 [opencl_init] FINALLY: opencl PREFERENCE=ON is AVAILABLE and ENABLED.
[opencl_init] opencl_scheduling_profile: 'default'
[opencl_init] opencl_device_priority: '*/!0,*/*/*/!0,*'
[opencl_init] opencl_mandatory_timeout: 400
[dt_opencl_update_priorities] these are your device priorities:
[dt_opencl_update_priorities] 		image	preview	export	thumbs	preview2
[dt_opencl_update_priorities]		0	1	0	0	1
[dt_opencl_update_priorities]		1	-1	1	1	-1
[dt_opencl_update_priorities] show if opencl use is mandatory for a given pixelpipe:
[dt_opencl_update_priorities] 		image	preview	export	thumbs	preview2
[dt_opencl_update_priorities]		0	0	0	0	0
[opencl_synchronization_timeout] synchronization timeout set to 200
   UNIFIED MEM SIZE:         22075 MB reserved for 'amdacceleratedparallelprocessinggfx1103'
[dt_opencl_update_priorities] these are your device priorities:
[dt_opencl_update_priorities] 		image	preview	export	thumbs	preview2
[dt_opencl_update_priorities]		0	1	0	0	1
[dt_opencl_update_priorities]		1	-1	1	1	-1
[dt_opencl_update_priorities] show if opencl use is mandatory for a given pixelpipe:
[dt_opencl_update_priorities] 		image	preview	export	thumbs	preview2
[dt_opencl_update_priorities]		0	0	0	0	0
[opencl_synchronization_timeout] synchronization timeout set to 200

Do I have to tell darktable somehow which GPU to prefer (couldn’t find an option in the settings)?

4

If you edit darktablerc, you can control which GPU to use.

But, with your system setup, I recommend that you disable the iGPU in the darktablerc and set your system to use Very fast gpu instead of default.

1 Like
5

image
image1512×1243 89.1 KB

1 Like
6

If I understand the manual correct, the darktablerc-line to disable the iGPU and use the dGPU for everything should read:
opencl_device_priority=0/0/0/0/0,0
Correct?

I’m wondering if the AMD Ryzen™ 9 7940HS is really so slow, that waiting for/only using the dGPU is faster instead than using CPU + dGPU in parallel.

7

Dt process most modules in sequence since the output of one feeds the next module, so there is really not much parallel within a single processing pipe. Telling it to use the dGPU normally yields faster processing because there are more processing units in a GPU.

The area that can help parallel is the preview vs the full pipe. One can execute in parallel. But with a iGPU, it is using the same memory and processing as the CPU. Something has to wait if you funnel too many things thru the CPU.

You can use -d perf to evaluate how your system behaves when it exports or when you do a change in the darkroom (ideally at the start of the pipe).

1 Like
8

I might be mistaken, but to my understanding with above configuration (very fast GPU & forbidding iGPU) the dGPU is the only device left over for darktable to use. So I was wondering if a configuration similar to
opencl_device_priority=0/*/0/*/*,* (to allow offloading to the iGPU for everything except center image and export pipeline) and/or not using very fast GPU config (to allow offloading to the CPU) wouldn’t make more sense.

Memory shouldn’t be an issue here, having 96 GB of it and if I recall correctly 8 GB assigned to the iGPU in BIOS. iGPU and CPU using the same processing sounds counter-intutive to me (what would the iGPU consist of then?), but I’m no expert on that field. :slight_smile:

9

Just test it and see what works for you.

10

So I tried 10 different combinations of opencl_device_priority and opencl_scheduling_profile (default, very fast GPU) and the results were very close to each other, differences normally <0.01 secs on process_image and ~0.1 secs on process_export.
Only two exceptions were when I forced to use the iGPU (priority: 1/1/1/1/1) & profile default (export time more than tripled, process image time more than doubled) and when I used */!0,*/*/*/!0,* with profile default (image process time more than doubled).

Note: Probably the numbers are not trustworthy anyway as I did not

  • repeat the runs
  • use the second preview window
  • pay attention to cache
  • process and export more than one image each run
11

Use parameter h. to disable the each GPU. You will get more accurate results this way.

https://docs.darktable.org/usermanual/4.6/en/special-topics/mem-performance/#device-specific-opencl-configuration

12

I tired some more, numbers stay very similar if I allow the iGPU for additional tasks or not and are still a little fluid. I now settled on

cldevice_v5_amdacceleratedparallelprocessinggfx1102=0 0 0 16 16 1024 1 0 0.000 0.000 0.250
cldevice_v5_amdacceleratedparallelprocessinggfx1102_building=-cl-fast-relaxed-math
cldevice_v5_amdacceleratedparallelprocessinggfx1102_id0=0
cldevice_v5_amdacceleratedparallelprocessinggfx1103=0 250 0 16 16 128 0 0 0.000 0.000 0.250
cldevice_v5_amdacceleratedparallelprocessinggfx1103_building=-cl-fast-relaxed-math
cldevice_v5_amdacceleratedparallelprocessinggfx1103_id0=600
cldevice_v5_amdacceleratedparallelprocessinggfx1103_id1=600
opencl_device_priority=+0/*/+0/*/+0
opencl_scheduling_profile=very fast GPU

and decided to spend time rather on image editing than on further performance tuning. :slight_smile: