[oliverames]

Not fully tested; opt-in. Filed so the change is reviewable while the measurement is open.

What and why

Mario Kart World 1.8.0 (#1383's build) on a physical iPhone runs at 60 fps for about a minute, then, in some runs, the process leaves the fast cores with thermal state nominal: mean clock 3.8 to 1.65 GHz and 55% to 12% of CPU time on performance cores, with the main thread's real-time declaration still in effect. cheemu21/plumber-racer-2 reports a slow state of the same kind on 1.0.0. iOS has no core affinity and every guest thread is already user-interactive, so the one lever left is a time-constraint declaration, which is what the scheduler places on a fast core and what the performance controller raises the clock for.

With DM_PCORE=1, the three busiest guest threads without a declaration (60 ms/s or more in the CPU sampler's 300-frame window) get a 60 Hz declaration of 1.5 times their measured per-frame CPU (1 to 6 ms), once each (pcore_declare in hle_thread.c, log line DM_PCORE: thread ... declared). The main and acquire threads keep their own declarations. Off by default: with the variable unset the code runs as before.

Checked

• Tests: hle_thread.c compiles on this base with a generated profile header (clang syntax-only; this caught a missing <mach/mach_time.h>, now included); dm/tools/tests/test_docs.py passes; the new switch is in docs/env_switches.tsv with its meaning (env_switches.py --check still reports DM_NVN_BC_ALIAS, which is missing from the table on main too).
• Games run, with title ID and version: Mario Kart World (0400C3F00006E000) 1.8.0, built from WIP: Mario Kart World 1.8.0 on iOS - rendering, memory and presentation #1383's tree, on the phone: the declarations were set (6.00 / 4.96 / 2.37 ms for the three busiest threads).
• Devices and OS: iPhone 19,4 (2 Performance + 4 Efficiency cores), iOS 27.3; Apple-silicon Mac, macOS 27.3
• scripts/regress.py check --games: not run: off by default, so every game runs as before without the variable.
• Not checked: whether it keeps the game on the fast cores. The one run with it started hot (thermal 2, 1.5 to 2.4 GHz before the declarations took effect) and logged GPU error discards, so no effect is established either way; later HDR-off runs had a healthy CPU with or without it. No unit test (nothing observable on the Mac), no Mac run. A cooled-phone A/B is the next step.

Front ends

None: no front end has to act on this.

Docs

• The mechanism is in docs/development.md and the switch in docs/env_switches.tsv
• A chain tool changed: dmsh_chain.py --header was run and dm/gfx/nvn/dmsh_chain.h is part of the change. Not applicable: no chain tool changed.
• No game files, keys or data extracted from a game in the change or the attachments

[oliverames]

Combined into #1713 with the other two pacing switches; same commit, same tests. Closing this one.