Skip to content

Some helper additions - #21868

Open
jenshannoschwalm wants to merge 3 commits into
darktable-org:masterfrom
jenshannoschwalm:some_opencl_additions
Open

Some helper additions#21868
jenshannoschwalm wants to merge 3 commits into
darktable-org:masterfrom
jenshannoschwalm:some_opencl_additions

Conversation

@jenshannoschwalm

Copy link
Copy Markdown
Collaborator

Provide --opencl-no-tiling

Added the cli option --opencl-no-tiling to disable fast internal opencl tiling in case this recent OpenCL pipe feature doesn't work correctly. (Can also be switched on via dt_conf_get_bool("no_opencl_fast_tiling")


Improved importance cache hint

The module-in-focus should only give a performance hint if enabled so the OpenCL code might avoid the write-back to host.


Improved analysis of NaN pipe data

  1. If analyzing data from OpenCL cl_mem we should not overwrite data in output cacheline as that might be used otherwise but to a temp buffer.
  2. The analysing happens in OMP code for performance
  3. At least in OpenCL code some drivers might behave in a strange way if there are NaNs or infinites in channel 3 so we analyse and report that too.
  4. Simplified log output

Added the cli option `--opencl-no-tiling` to disable fast internal opencl tiling
in case this recent OpenCL pipe feature doesn't work correctly.

(Can also be switched on via dt_conf_get_bool("no_opencl_fast_tiling")
The module-in-focus should only give a performance hint if enabled so the OpenCL
code might avoid the write-back to host.
@jenshannoschwalm jenshannoschwalm added this to the 5.8 milestone Aug 16, 2026
@jenshannoschwalm jenshannoschwalm added feature: new new features to add scope: performance doing everything the same but faster OpenCL Related to darktable OpenCL code scope: debugging labels Aug 16, 2026
1. If analyzing data from OpenCL cl_mem we should not overwrite data in output cacheline
   as that might be used otherwise but to a temp buffer.
2. The analysing happens in OMP code for performance
3. At least in OpenCL code some drivers might behave in a strange way if there are NaNs or
   infinites in channel 3 so we analyse and report that too.
4. Simplified log output
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

feature: new new features to add OpenCL Related to darktable OpenCL code scope: debugging scope: performance doing everything the same but faster

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant