OpenArena opengl1: batching, and why the lightmap atlas went off (2026-10-17/18)

Symptom. Game-side batching had tripled X-Moto, so the same was tried for OpenArena’s fixed-function renderer: ioquake3 0019-0022 (the lightmap atlas r_mergeLightmaps, a GL state cache, no glLockArraysEXT around batches, the sky from vertex arrays). On llvmpipe at GL 2.1 the bench demo went from 3022 GL calls and 198 draws a frame to 377 and 79, with the pixels within 1-4 levels. On the RSX it measured slower: steady 51.8 fps with all four on against 56.4 with all off (batch 2026-10-17-ioquake3-362-batching).

Evidence. The frame cost itself didn’t change: smooth fps was 63.7-65.7 either way. What changed was the hitches. With 0023’s atlas variants (batch 2026-10-17-ioquake3-378-atlas) the stall seconds were 34.1 with the base atlas, 23.9 with the atlas off, 35.6 with mergelm2 and 36.2 with lmsize256: any atlas page cost about 10 s of extra hitches a run, about 5% of the steady fps.

Cause. The atlas is built at the load and its pages are big uploads that the RSX’s memory pressure makes expensive to move; OpenArena on the PPE is not limited by its GL calls here, so the saved calls bought nothing (the same lesson as ClassiCube: batching pays only where the PPE is the limit).

Fix. 0024 makes r_mergeLightmaps 0 the default. The rest of the batching (state cache, no lock arrays, sky arrays) is neutral on the RSX and stays on, since it costs nothing and cuts the CPU.