[akshay4]

What and why

The pool walk makes the smallest textures up front while currentAllocatedSize is under a twentieth of the working set. That counter counts each no-copy buffer in full (50 over one 64 MB region add 3,200 MB, see #1482), so in Ocarina of Time it is over budget from the first load and nothing gets made up front. Those IDs read the dummy until a shader's read is noticed, which can show as textures filling in late.

DM_POOL_BUDGET_MB=<MB> compares against the residency set instead. Off by default, so nothing changes without it.

Entry added under the pool walk in docs/reference/memory.md.

Checked

• Tests: check_t_pool_chunks, check_t_cpu_tex, test_docs.py, test_env_switches.py
• Games run: Ocarina of Time 0400486024DEC000 v131072, same save and inputs, about 2 minutes each:

	Default	DM_POOL_BUDGET_MB=6000
IDs left out	3,274	698
Residency set	3,081 MB	3,285 MB
Frame time p50	14.1 ms	14.3 ms
GPU errors	0	0

• Not the same code: that run was on main 75c1a8c with Metal compiles take -fbracket-depth=1024: two Ocarina of Time compute shaders nest 259 brackets #1357, GPU-written storage is tracked as ranges, so a pool that holds both kinds no longer leaves its CPU-written textures at zero #1390, A GPU memory guard before each residency commit stops the game instead of freezing the Mac #1391, A GPU-only pool's render targets share one placement heap: Ocarina of Time's 3 GB of render targets fit in 0.7 GB #1453 and Shared textures no live object resolves to are retired when GPU memory runs short: Ocarina of Time no longer freezes after the talk with Saria #1455, and read the set's size at its last commit. This branch reads it directly and has only been compiled and unit tested, as I no longer have the game files here.
• Whether late textures look better is not measured, only the count. Some of it may be the game's own streaming.
• Devices and OS: M6, 24 GB, macOS 27.2. Not tried on a smaller Mac or other games, which is why it's off by default and I'm not suggesting a value.
• regress.py not run: nothing changes unless the switch is set.

Docs

• Entry in docs/reference/memory.md, docs_symptoms.py run, row in docs/env_switches.tsv
• No game files, keys or data extracted from a game

[akshay4]

Rebased on main, where pool_walk_limit became pool_room, so pool_have() is used there now. Re-ran t_pool_chunks, t_cpu_tex, test_docs.py and test_env_switches.py; still not run in a game. clang-tidy fails on main too, with the same findings.