Skip to content

[pull] master from darktable-org:master#329

Merged
pull[bot] merged 5 commits intoMu-L:masterfrom
darktable-org:master
May 14, 2021
Merged

[pull] master from darktable-org:master#329
pull[bot] merged 5 commits intoMu-L:masterfrom
darktable-org:master

Conversation

@pull
Copy link
Copy Markdown

@pull pull bot commented May 14, 2021

See Commits and Changes for more details.


Created by pull[bot]

Can you help keep this open source service alive? 💖 Please sponsor : )

victoryforce and others added 5 commits May 14, 2021 14:25
* Avoid div by zero in dt_iop_clip_and_zoom_mosaic_half_size_plain

Fixes #8951

Although the file given in the issue is crippled we can avoid the crash.
In `dt_iop_clip_and_zoom_mosaic_half_size` and the sse friend there is possibly a div/0
problem that should be checked.

* Fixing same dib by zero in dt_iop_clip_and_zoom_mosaic_half_size_f

* Remove sse code for dt_iop_clip_and_zoom_mosaic... after testing performance

checked performance non-sse vs sse specific code
- with added local timers
- using gcc 10.2
- testing -t 1/4/8/16
- intel (xeon like 9900) with fixed clock rate

in
- dt_iop_clip_and_zoom_mosaic_half_size
- dt_iop_clip_and_zoom_mosaic_half_size_f
- dt_iop_clip_and_zoom_demosaic_passthrough_monochrome_f
- dt_iop_clip_and_zoom_demosaic_half_size_f

with consitant results. For all functions the sse specific code was somewhat slower (~20%)
than the vectorized compiler code. Number of omp cores didn't matter, just made the results
more measurable because of low execution times.

So i removed all the sse specific code for less code burden and better performance.

* Fix sse header plus div/0

At least for bayer images we absolutely want to be sure there is no div by zero as there might
be buggy dng files.
This is the version on Opensuse Tunderbird rolling release and build
was reported to be ok with it.
Tested performance on 9900k like intel xeon with fixed clock rate at omp threads 1-16 using gcc 10.2
Removing the smp specific code does not change the performance in a measurable way (<2%) so remove the
code for small maintenance burden.
@pull pull bot added the ⤵️ pull label May 14, 2021
@pull pull bot merged commit 5d8b764 into Mu-L:master May 14, 2021
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants