shaders/colorspace: improve peak detection performance

Substantially improves performance on systems that are constrained by global atomic pressure, like intel iGPUs.

A value of 12 was experimentally determined to maximize throughput on an UHD770. More powerful discrete GPU systems don't benefit from this either way.

changed milestone to %v6.283+

added 1 commit

• ae107984 - shaders/colorspace: parallelize peak detection atomics

Compare with previous version

Rewrote this commit to parallelize all atomics, not just histogram

On my system it seems any value of 4 or higher provides maximum benefit, testing even up to e.g. 64 I see zero change in GPU timings whatsoever.

added 1 commit

• 12b11c1a - shaders/colorspace: parallelize peak detection atomics

Compare with previous version

added 3 commits

• 516541c9 - renderer: remove unnecessary need_fbo force
• 9bad17e3 - renderer: hook PL_HOOK_SCALED even on free sampling
• 33807934 - renderer: update HDR peak at downscaled resolution

Compare with previous version

added 1 commit

• 1da0eefa - renderer: update HDR peak at downscaled resolution

Compare with previous version

changed title from shaders/colorspace: parallelize histogram bins to shaders/colorspace: improve peak detection performance

mentioned in merge request !492 (merged)

approved this merge request

added 5 commits

• e158e307 - 1 commit from branch videolan:master
• c9d40038 - shaders/colorspace: parallelize peak detection atomics
• a0ba2382 - renderer: remove unnecessary need_fbo force
• ad052218 - renderer: hook PL_HOOK_SCALED even on free sampling
• fcf81635 - renderer: update HDR peak at downscaled resolution

Compare with previous version

merged