32 MB in a random number generator (2026-10-18)

Moved on 2026-10-26 from docs/MEMORY.md (“Findings in the games’ code”, “Cuts, ranked” row 1 and the packages table), where the cross-game memory picture stays.

Symptom

The MEM=1 batch of 2026-10-18 (00:40-01:40 UTC, Mesa ps3.299, one run per game) had Yamagi Quake II at 77 MB of Rss + 3.4 MB swapped, with 42.1 MB of anonymous memory (the most of the ten games) and a 23 MB .bss in the client, 18 MB of it resident.

Cause

src/common/shared/rand.c (the KISS PRNG, also in rogue/ and xatrix/) had static uint64_t QARY[0x200000], 16 MiB, and randk_seed() writes every entry. The file is linked into the client (quake2, whose 23 MB .bss was 18 MB resident, 16 of it this table) and into game.so (16 MB .bss, all resident): 32 MB of the game’s 77. A lag of 4096 entries keeps a period far beyond anything a game draws.

Fix

yquake2 0002 YQ2_KISS_QSIZE: the PRNG table allocated at first use, a power of two from 64 to 2^21 entries, 4096 (32 KiB per module) when unset. The .bss is 7.7 MB in quake2 (was 23.7) and 81 KiB in game.so (was 16 MB). YQ2_KISS_QSIZE=2097152 (recipe variant kissbig) gives the old size and sequence (the same million draws, checked on the host), with the 16 MiB from calloc. Package yamagi-quake2 8.70+dfsg-1+ps3.408 had the switch, .415 made 4096 the default. Client-side random effects differ from the old table (the demo itself doesn’t).

Measured on the PS3 (kiss4k, the recipe variant before it became the default):

Numbers: batch 2026-10-18-mesa-407-lut (Mesa ps3.407 NV30_TRANSLATE_LUT, yquake2 ps3.408 kiss4k, luanti ps3.405 trim16 (all with WAKER kick), 2026-10-18; medians over the runs of each game and variant).

Same boot, x2: RSS 125 -> 56 MB, swap used 80 -> 9 MB (about 70 MB less RAM), gameplay stalls 22.6 -> 0.05 s, longest frame 4.5 s -> 91 ms, steady 65.7 -> 68.7 fps, 1% low 20.1 -> 31.1. The two resident copies had pushed the game’s other pages to swap. The upstream offer (a smaller static table) is in upstream.md.