Asynchronous GeometryTile deletion

[mwilsnd]

General performance augmentation. Defers destruction of a GeometryTile's mailbox to a worker thread. These tiles hold OpenGL objects which must be deleted on the correct thread, so those objects are deleted on the render thread.

A number of special concerns around synchronization had to be addressed.

The primary performance problem centers around the mailbox needing to take both mutex locks before it can be properly closed. As this was done on the render thread, it would stall the whole renderer when the map was in motion.

This started in #1928 and has since been moved to this development branch.

[github-actions]

Bloaty Results (iOS) 🐋

Compared to main

    FILE SIZE        VM SIZE    
 --------------  -------------- 
  +0.5% +79.0Ki  +0.5% +80.0Ki    TOTAL

Full report: https://maplibre-native.s3.eu-central-1.amazonaws.com/bloaty-results-ios/pr-2051-compared-to-main.txt

[codecov]

Codecov Report

Attention: Patch coverage is 56.54952% with 136 lines in your changes are missing coverage. Please review.

Project coverage is 58.38%. Comparing base (003c062) to head (937e261).

Files	Patch %	Lines
src/mbgl/util/thread_pool.cpp	52.72%	2 Missing and 24 partials ⚠️
src/mbgl/text/glyph_manager.cpp	47.61%	0 Missing and 22 partials ⚠️
src/mbgl/tile/tile_loader_impl.hpp	48.48%	2 Missing and 15 partials ⚠️
src/mbgl/renderer/image_manager.cpp	33.33%	0 Missing and 14 partials ⚠️
src/mbgl/tile/tile_cache.cpp	50.00%	0 Missing and 14 partials ⚠️
src/mbgl/tile/geometry_tile.cpp	50.00%	1 Missing and 6 partials ⚠️
src/mbgl/util/thread_pool.hpp	70.83%	0 Missing and 7 partials ⚠️
src/mbgl/renderer/render_source.cpp	53.84%	0 Missing and 6 partials ⚠️
src/mbgl/renderer/tile_pyramid.cpp	58.33%	0 Missing and 5 partials ⚠️
src/mbgl/actor/mailbox.cpp	84.00%	3 Missing and 1 partial ⚠️
... and 10 more

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2051      +/-   ##
==========================================
- Coverage   58.38%   58.38%   -0.01%     
==========================================
  Files         572      574       +2     
  Lines       28136    28340     +204     
  Branches    11270    11376     +106     
==========================================
+ Hits        16428    16545     +117     
+ Misses       4158     4157       -1     
- Partials     7550     7638      +88     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[github-actions]

Bloaty Results 🐋

Compared to main

    FILE SIZE        VM SIZE    
 --------------  -------------- 
  +0.5%  +741Ki  +0.3%  +104Ki    TOTAL

Full report: https://maplibre-native.s3.eu-central-1.amazonaws.com/bloaty-results/pr-2051-compared-to-main.txt

Compared to d387090 (legacy)

    FILE SIZE        VM SIZE    
 --------------  -------------- 
   +19% +22.2Mi  +401% +23.9Mi    TOTAL

Full report: https://maplibre-native.s3.eu-central-1.amazonaws.com/bloaty-results/pr-2051-compared-to-legacy.txt

[alexcristici]

Looks great! The android benchmark runs very smooth now.

[mwilsnd]

!benchmark android

[TimSylvester]

The additional locks in the managers are still a potential performance issue, I think, but I don't feel like we should mess around with it more unless they are observed to be relevant.

I'd prefer to further reduce direct use of the Scheduler static variables to help make it easier to reason about what can and must be using the same thread pool, but not enough to keep this open.

[TimSylvester]

ok, I think this one is finally ready