Massive performance degradation with DrawEntireTerrain=Yes

[xezon]

Reported by 2 users. Probably comes from the recent heightmap changes. Needs fixing.

[sailro]

I looked into this together with the related water-at-zoom problem, and it turned out to be two distinct issues that people had been conflating through the DrawEntireTerrain=Yes workaround. I opened two complementary PRs.

Background: why people often set DrawEntireTerrain=Yes (I was doing this)

In the default path (DrawEntireTerrain=No) the terrain is only drawn within a window of tiles around the view center. At high camera zoom that window stops covering the visible ground, and on the many maps that keep a map-sized "Default Water" polygon, that water then renders behind the terrain window and the map looks flooded. The long-standing community workaround was to set DrawEntireTerrain=Yes so the whole map is always drawn — which hid the water bug but is what makes this issue (#2743) so slow.

So there are really two problems:

1. the rendering bug that drives people to the workaround, and
2. the performance cost of the workaround itself.

PR #2785 — fixes the rendering bug (default DrawEntireTerrain=No path)

[#2785] sizes the normal draw window to the actual visible footprint instead of a fixed tile count: it projects the four view corners onto the ground (the same way updateCenter() already positions the window), grows the draw size just enough to span them, bounded by the map extent and snapped to whole vertex-buffer tiles. At normal zoom it is a no-op; the window only grows as you zoom out, on all maps.

Effect on this issue: it removes the need for DrawEntireTerrain=Yes in the common case (no more water bleed-through at zoom without it). It deliberately does not touch the DrawEntireTerrain path, so on its own it does not fix #2743 — it just makes the slow workaround unnecessary for most users.

PR #2786 — fixes this issue directly (DrawEntireTerrain=Yes path)

[#2786] fixes the actual performance degradation. Root cause is two competing sources of truth for the terrain draw size:

• WorldHeightMap::createDrawArea() force-overrides the stored draw size to the full map when DrawEntireTerrain=Yes (or the stretch size when StretchTerrain=Yes).
• W3DView::updateTerrain() requests the normal draw size through HeightMapRenderObjClass::setTerrainDrawSize() every frame.

setTerrainDrawSize() early-returns when the requested size equals the stored size, but the requested normal size can never match the stored full-map size. So every frame it re-ran initHeightData() + a full-map updateBlock(), and set m_needFullUpdate, making updateCenter() do a second full-map update — roughly two full terrain rebuilds (vertices + lighting) per frame. This regressed in #2677, which added the per-frame setTerrainDrawSize() call (W3DView.cpp, commit ccd8766) alongside the full-map override in createDrawArea() (commit c47b01a).

The fix is two parts:

1. Resolve the same StretchTerrain / DrawEntireTerrain overrides inside setTerrainDrawSize() (same precedence as createDrawArea()), so the unchanged-check matches and the terrain is no longer rebuilt every frame.
2. That per-frame rebuild was, however, the only thing repainting the terrain when the vertex grid already covers the whole map — updateCenter() early-returns in that case (small maps, StretchTerrain, or DrawEntireTerrain) and never reached its m_needFullUpdate handling. So updateCenter() now honors a pending m_needFullUpdate in that early-return path with a single full updateBlock, ensuring lighting / device-reset / late-texture repaints still happen — but only when actually requested instead of every frame. (Without this second part, the first part alone left undrawn/untextured tiles on the shell map.)

Steady-state cost returns to a one-time full update plus the normal full-map draw (i.e. the pre-#2677 cost), and terrain still updates correctly on lighting, device-reset and texture changes.

TL;DR

• fix(terrain): Cover the visible ground at high camera zoom #2785 makes DrawEntireTerrain=Yes unnecessary in the common case (fixes the water-at-zoom bug in the default path).
• fix(heightmap): Prevent per-frame full terrain rebuild with DrawEntireTerrain #2786 makes DrawEntireTerrain=Yes fast (fixes the per-frame full rebuild regressed by fix(heightmap): Fix up view and heightmap to work better with low camera pitch #2677) — i.e. this issue.

Tested in Zero Hour with DrawEntireTerrain=Yes: smooth in-game where the unpatched build was heavily laggy, and the shell map / in-game terrain renders correctly.

[sailro]

I really think #2786 is solving this performance issue (while #2785 extend support for larger resolutions). Happy to make changes or discuss anything.

Thanks!