[oliverames]

What and why

On an iPhone the launcher shows Donkey Kong Bananza with the game's own upscaler and Metal 4 pipelines off, while every iPhone run of mine used docked, DM_UPSCALER=metalfx and DM_PSO_MTL4=1; main's note also records a community run on the defaults (issue #1183, 7ffa63cca). This changes the launcher's defaults for this game to match, and nothing else.

@seedds: this is a default for a Switch 2 game running on iPhone, and whether a game ships a non-default launcher setting is your call. It changes what the launcher shows first and what Reset to tested defaults sets. It does not remove the choice: a saved choice, DM_UPSCALER and Documents/env.txt still win, and a start without the launcher keeps the host's own default. It is not a fix for anything: sustained warm performance and the fur's shimmer are not established (below). If you would rather keep the game's upscaler as the default, pso_mtl4 on and upscaler_default are independent lines in game.tsv and either can be dropped.

Before and after (the iOS launcher, no saved choice)

Read with dm/games/settings.py for 040074A01BF12000, GFXAPI=nvn, SETTINGS_PLATFORM=ios, before and after the change:

Setting	Variable	Before	After	From
Console mode	DM_OPMODE	Docked 1	Docked 1	op_mode, console is the default of every game (unchanged)
Game's upscaler	DM_UPSCALER	game	metalfx	new upscaler_default metalfx
Upscaling	DM_PRESENT_UPSCALE	off (Standard)	off (Standard)	settings.tsv (unchanged)
Metal 4 pipelines	DM_PSO_MTL4	0	1	pso_mtl4 on
Frame buffering	DM_PRESENT_SNAP	0	0	present_snap, off is the default (unchanged)

So two values change. On the Mac settings window nothing changes: Metal 4 pipelines are already on for every game there, and its Game's upscaler default stays game (output of the same script for SETTINGS_PLATFORM=mac: DM_UPSCALER=game, DM_PSO_MTL4=1).

The profile key

DM_UPSCALER had no profile key: its default was the literal game in dm/games/settings.tsv. The smallest generic key is upscaler_default (game, the default, or metalfx): game.py validates it and writes DM_GAME_UPSCALER_DEFAULT into the profile header, and the DM_UPSCALER row's default becomes DM_GAME_UPSCALER_DEFAULT|game (the existing <ios>|<mac> form, as DM_GAME_PSO_MTL4|1 does), so only the iOS launcher follows it. No host code changed: the macro is read by settings.py only. The other four settings needed no key (see the table). The key is a row in dm/games/README.md.

One consequence of pso_mtl4 on, from dm/ios/tools/build_ipa.sh and scripts/build_ipa_from_nsp.py: an IPA built for this game now builds the pipeline archive made offline for g18p (A19 / A19 Pro, the pso_gpu default), as for any profile with pso_mtl4 on. The pulling libraries (_bp) are in every iOS build since #1758, whatever the key says, so they're no longer part of this change. I did not build an IPA.

Checked

Python 3.14 venv outside the repo (lz4, capstone>=5,<6, fonttools, and .github/requirements/lint.txt), DM_GAME=040074A01BF12000, on macOS, at head 984af1e4 on origin/main 777a5f6c. On Nov 9 I rebased onto main 6d9edb07 (head d739252a, one commit) and ran the five tests named below again; what changed is marked "After the rebase":

Tests:

• python3 dm/games/settings.py equivalent (settings.build) before and after: the table above.
• New and changed tests pass: dm/tools/tests/test_settings.py (3 tests; the new one asserts all five defaults on iOS and Mac), test_game_profile.py (8; the new one checks the key's header macro, its default and its rejection), test_launcher_settings.py (16), test_docs.py (11, links and anchors), test_env_switches.py (1).
• After the rebase: the same five pass. Run directly they now report test_settings.py 4, test_game_profile.py 4, test_launcher_settings.py 17, test_docs.py 6 and test_env_switches.py 1, since main's own tests have changed. The 145-file loop below was not run again.
• The repository's tool, iOS and CPU-only gfx tests, the same loop as the tests workflow's "Run the test files" step: all 145 files under dm/tools/tests, dm/ios/tests, dm/gfx/tests ran, none failed. Ten of them report skips (test_astral_fps, test_image_atomic_bounds, test_txq_type, test_dmsh_ampere_alu, test_dmsh_ampere_mem, test_dmsh_border, test_dmsh_control_flow, test_dmsh_geometry, test_pso_archive, test_transform_feedback): tests that need dotnet or spirv-val, which this Mac run lacked. Skipped is not passed.
• python3 dm/tools/docs_symptoms.py, docs_patches.py, docs_gallery.py: no diff.
• python3 dm/tools/lint_python.py --against origin/main: 389 findings, 389 in the baseline, 0 new. After the rebase: 339 findings, 336 in the baseline, 3 new, all three in dm/gfx/tools/dmsh_build.py on main and none from this change. Main has since fixed those three (3ade2f2ef).
• game.py header 040074A01BF12000 writes DM_GAME_PSO_MTL4 1 and DM_GAME_UPSCALER_DEFAULT "metalfx", and the header compiles (clang -fsyntax-only) with both.

Games run, with title ID and version: Donkey Kong Bananza, title ID 040074A01BF12000, version not stated. The phone runs are listed below and are from earlier local builds, not from this head.

Devices and OS: iPhone 17 Pro Max and iPhone Duo as listed below; iOS version not stated; macOS version not stated (Python checks ran on macOS).

scripts/regress.py check --games: not run, because no host, shader or tool code changed (see Not verified).

Measurements and runs behind the defaults:

All local, kept by the author, under anygame/logs/dkb-sharpness-2026-11-03/ and anygame/WORKLOG.md (not in this repository). The measurements come from local builds that carried changes not on main.

Claim	Source
iPhone 17 Pro Max, docked + DM_UPSCALER=metalfx + DM_PSO_MTL4=1: MetalFX 1676x943 -> 2304x1296	border-summary.json (upscaler: size at frame 4847 ... 1676x943 -> 2304x1296), current-metalfx-log.txt
Same phone, chain 215a6d278114 build: six complete 300-frame windows (frames 9600-11400) at 33.34 ms on-screen p99, 0 dropped, thermal 0, footprint 5119-5129 MiB	border-summary.json, border-final-memstat.tsv
Same phone, other docked runs at thermal 1: some windows at 50 ms (three of five in one rerun, one drop)	border-restored-summary.json, PR #1204's text
Same phone, handheld: 1397x786 -> 1920x1080 (a smaller scene), six windows at 33.34 ms	border-handheld-summary.json
iPhone Duo, same save: the game's own upscaler shimmers the same as MetalFX, per the player. Game's own: GPU p50 about 29.6 ms (30 fps cadence, thermal 2, frame 3900). MetalFX: GPU p50 about 16.4 ms (frames 1800-3600, thermal 2, 33.34 ms p99, 11 to 42 dropped frames per 300, cause not established). Not the same scene: no performance verdict	camera-constants/ab/ (logA-final.txt, logB-final.txt), WORKLOG "upscaler A/B on the Duo"
The player found docked MetalFX sharper than the game's own upscaler (subjective), and still saw jagged edges	WORKLOG, PR #1204's text

The game note (dm/games/040074A01BF12000/README.md, new "iPhone launcher defaults" section) states the same before/after and limits. It does not link PR #1204's section, so the two PRs merge in any order; both touch that README in different places.

Not verified:

• No exact-head iPhone run: no run used a build of main, or of this head. The phone numbers above are from earlier local builds, at different thermal states, so they are not a controlled comparison.
• Sustained 30 fps once the phone is warm is not established. Neither is any shimmer improvement: on the Duo the game's own upscaler and MetalFX shimmer the same.
• I did not build an IPA, so the pso_mtl4 on build path for this game (the g18p archive) is read from the scripts, not run here. make -C dm/gfx/tests GPU tests and regress.py were not run: no host, shader or tool code changed.
• The launcher screen itself was not seen showing these values; the evidence is the generated settings list.
• Hosted checks: see the PR's checks. Main's clang-tidy (lint-c) is red for reasons unrelated to this change: lint-c passes on main again: the JIT request table, DM_PROGRAMS, the web page's scheme handler and the Vulkan occlusion query no longer reach NULL (baseline 241 to 240) #1207 was closed on Nov 5, and the cause now is dm/gfx/tests/t_pso_archive.m (open PR t_pso_archive's cold-quit marker write is nil-safe: lint-c's clang-tidy ratchet is green on main again #2080).
• The default is a maintainer call (above).

Front ends

None: no front end has to act on this.

Docs

• The finding is in its game note or its docs/reference/<subsystem>.md entry, and python3 dm/tools/docs_symptoms.py was run (the symptom index)
• 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]

Closing this. Which launcher settings a game starts on is your call, and the evidence here is one subjective comparison on two iPhones. It also conflicts with main again now that render_cond and gpu_timestamps are gone from the profile. The measurements are still in the description if they're ever useful. Thanks!