[core] TilePyramid has sorted render tiles #13739
Conversation
src/mbgl/renderer/tile_pyramid.cpp
Outdated
| std::sort(sortedTiles.begin(), sortedTiles.end(), [](const RenderTile& a, const RenderTile& b) { | ||
| return std::tie(a.id.canonical.z, a.id.canonical.y, a.id.wrap, a.id.canonical.x) < | ||
| std::tie(b.id.canonical.z, b.id.canonical.y, b.id.wrap, b.id.canonical.x); | ||
| }); |
There was a problem hiding this comment.
This sort function sorts differently from the standard UnwrappedTileID's operator< function: It sorts by (z, y, wrap, x), while UnwrappedTileID::operator< sorts by (wrap, z, x, y). I'm not sure why we sort this way here, but might be worth investigating how a different sort order affects feature querying.
There was a problem hiding this comment.
Good point, thanks for pointing that out!
| @@ -43,7 +43,7 @@ RenderLayer::RenderTiles RenderLayer::filterRenderTiles(RenderTiles tiles) const | |||
| } | |||
|
|
|||
| void RenderLayer::sortRenderTiles(const TransformState&) { | |||
| std::sort(renderTiles.begin(), renderTiles.end(), [](const auto& a, const auto& b) { return a.get().id < b.get().id; }); | |||
| // no-op | |||
There was a problem hiding this comment.
If this function is a no-op, we can remove it entirely.
There was a problem hiding this comment.
RenderSymbolLayer still overrides it, however this is something I'd like to change in the following PR.
src/mbgl/renderer/tile_pyramid.cpp
Outdated
| @@ -53,7 +53,7 @@ void TilePyramid::finishRender(PaintParameters& parameters) { | |||
| } | |||
|
|
|||
| std::vector<std::reference_wrapper<RenderTile>> TilePyramid::getRenderTiles() { | |||
| return { renderTiles.begin(), renderTiles.end() }; | |||
| return sortedTiles; | |||
There was a problem hiding this comment.
Instead of maintaining two vectors in parallel, can we just sort renderTiles, and create the reference-wrapped copy of the vector in this function?
There was a problem hiding this comment.
RenderTile does not have a default constructor, so std::swap cannot handle it, therefore standard algorithms look not applicable there :-/
pozdnyakov
force-pushed
the
mikhail_render_layer_sort
branch
from
1ae224e to
1c18585
Compare
|
@kkaefer Thanks for your comments! Could you take another look? |
Thus we obviate unneeded extra sorting of render tiles at each render layer.
pozdnyakov
force-pushed
the
mikhail_render_layer_sort
branch
from
1c18585 to
f564baa
Compare
pozdnyakov
changed the title
Thus we obviate unneeded extra sorting of render tiles at each render layer.