[core] add algorithm for computing masks for raster tiles

This commit is contained in:
Konstantin Käfer
2017-07-05 15:38:50 +02:00
parent 541712aac7
commit 4dff51719b
9 changed files with 323 additions and 0 deletions

View File

@@ -0,0 +1,128 @@
#pragma once
#include <mbgl/renderer/tile_mask.hpp>
#include <vector>
#include <functional>
#include <algorithm>
namespace mbgl {
namespace algorithm {
namespace {
template <typename Renderable>
void computeTileMasks(
const CanonicalTileID& root,
const UnwrappedTileID ref,
typename std::vector<std::reference_wrapper<Renderable>>::const_iterator it,
const typename std::vector<std::reference_wrapper<Renderable>>::const_iterator end,
TileMask& mask) {
// If the reference or any of its children is found in the list, we need to recurse.
for (; it != end; ++it) {
auto& renderable = it->get();
if (!renderable.used) {
continue;
}
if (ref == renderable.id) {
// The current tile is masked out, so we don't need to add them to the mask set.
return;
} else if (renderable.id.isChildOf(ref)) {
// There's at least one child tile that is masked out, so recursively descend.
for (const auto& child : ref.children()) {
computeTileMasks<Renderable>(root, child, it, end, mask);
}
return;
}
}
// We couldn't find a child, so it's definitely a masked part.
// Compute the difference between the root tile ID and the reference tile ID, since TileMask
// elements are always relative (see below for explanation).
const uint8_t diffZ = ref.canonical.z - root.z;
mask.emplace(diffZ, ref.canonical.x - (root.x << diffZ), ref.canonical.y - (root.y << diffZ));
}
} // namespace
// Updates the TileMasks for all renderables. Renderables are objects that have an UnwrappedTileID
// property indicating where they should be rendered on the screen. A TileMask describes all regions
// within that tile that are *not* covered by other Renderables.
// Example: Renderables in our list are 2/1/3, 3/3/6, and 4/5/13. The schematic for creating the
// TileMask for 2/1/3 looks like this:
//
// ┌────────┬────────┬─────────────────┐
// │ │ │#################│
// │ 4/4/12 │ 4/5/12 │#################│
// │ │ │#################│
// ├──────3/2/6──────┤#####3/3/6#######│
// │ │########│#################│
// │ 4/4/13 │#4/5/13#│#################│
// │ │########│#################│
// ├────────┴──────2/1/3───────────────┤
// │ │ │
// │ │ │
// │ │ │
// │ 3/2/7 │ 3/3/7 │
// │ │ │
// │ │ │
// │ │ │
// └─────────────────┴─────────────────┘
//
// The TileMask for 2/1/3 thus consists of the tiles 4/4/12, 4/5/12, 4/4/13, 3/2/7, and 3/3/7,
// but it does *not* include 4/5/13, and 3/3/6, since these are other Renderables.
// A TileMask always contains TileIDs *relative* to the tile it is generated for, so 2/1/3 is
// "subtracted" from these TileIDs. The final TileMask for 2/1/3 will thus be:
//
// ┌────────┬────────┬─────────────────┐
// │ │ │#################│
// │ 2/0/0 │ 2/1/0 │#################│
// │ │ │#################│
// ├────────┼────────┤#################│
// │ │########│#################│
// │ 2/0/1 │########│#################│
// │ │########│#################│
// ├────────┴────────┼─────────────────┤
// │ │ │
// │ │ │
// │ │ │
// │ 1/0/1 │ 1/1/1 │
// │ │ │
// │ │ │
// │ │ │
// └─────────────────┴─────────────────┘
//
// Only other Renderables that are *children* of the Renderable we are generating the mask for will
// be considered. For example, adding a Renderable with TileID 4/8/13 won't affect the TileMask for
// 2/1/3, since it is not a descendant of it.
template <typename Renderable>
void updateTileMasks(std::vector<std::reference_wrapper<Renderable>> renderables) {
std::sort(renderables.begin(), renderables.end(),
[](const Renderable& a, const Renderable& b) { return a.id < b.id; });
TileMask mask;
const auto end = renderables.end();
for (auto it = renderables.begin(); it != end; it++) {
auto& renderable = it->get();
if (!renderable.used) {
continue;
}
// Try to add all remaining ids as children. We sorted the tile list
// by z earlier, so all preceding items cannot be children of the current
// tile. We also compute the lower bound of the next wrap, because items of the next wrap
// can never be children of the current wrap.
auto child_it = std::next(it);
const auto children_end = std::lower_bound(
child_it, end,
UnwrappedTileID{ static_cast<int16_t>(renderable.id.wrap + 1), { 0, 0, 0 } },
[](auto& a, auto& b) { return a.get().id < b; });
mask.clear();
computeTileMasks<Renderable>(renderable.id.canonical, renderable.id, child_it, children_end,
mask);
renderable.setMask(std::move(mask));
}
}
} // namespace algorithm
} // namespace mbgl