[performance] remove use of snprintf in hot path; it's pretty slow on iOS

include/mbgl/map/tile_data.hpp

@@ -66,6 +66,7 @@ public:
 
 public:
     const Tile::ID id;
+    const std::string name;
     std::atomic<State> state;
 
 public:

src/map/tile_data.cpp

@@ -11,12 +11,12 @@ using namespace mbgl;
 
 TileData::TileData(Tile::ID const& id_, const SourceInfo& source_)
     : id(id_),
+      name(util::sprintf<32>("%d/%d/%d", id.z, id.x, id.y)),
       state(State::initial),
       source(source_),
       debugBucket(debugFontBuffer) {
     // Initialize tile debug coordinates
-    const std::string str = util::sprintf<32>("%d/%d/%d", id_.z, id_.x, id_.y);
-    debugFontBuffer.addText(str.c_str(), 50, 200, 5);
+    debugFontBuffer.addText(name.c_str(), 50, 200, 5);
 }
 
 TileData::~TileData() {

src/renderer/painter.cpp

@@ -370,7 +370,7 @@ void Painter::renderLayer(util::ptr<StyleLayer> layer_desc, const Tile::ID* id,
 void Painter::renderTileLayer(const Tile& tile, util::ptr<StyleLayer> layer_desc, const mat4 &matrix) {
     assert(tile.data);
     if (tile.data->hasData(*layer_desc) || layer_desc->type == StyleLayerType::Raster) {
-        gl::group group(util::sprintf<32>("render %d/%d/%d\n", tile.id.z, tile.id.y, tile.id.z));
+        gl::group group(std::string { "render " } + tile.data->name);
         prepareTile(tile);
         tile.data->render(*this, layer_desc, matrix);
     }