seedds dac7ba6338 Function-matched cuts: same-name rules are one patch's implementations; Taiko Risshiden V DX's frame-limiter cut reaches the base game and later builds. nso_match.resolve groups rules by name: exactly one implementation must match (the log names it), a red line only when none or more than one does. Before, each implementation of a changed function logged a red line on every build it did not fit. Taiko's MATCH_CUTS has two implementations of the frame end under one name: 1.1.1's (main+0x443e8c, 364 bytes, cut at +288) and the base game's (v0 main 47d7baf6..., main+0x43e140, 172 bytes: 1.1.1 added the frame-skip branch; the same 20 bytes at +104 = main+0x43e1a8). In v0 N is likewise a global (GOT 0xa1be28, data 0x9ec55c, 1) set to 2 only by CPrivateBattleWindow / CPrivateBattleTutorialWindow, and the start tick is read after the update; nso_fnmap.py finds no exact match for 1.1.1's function in v0, found by disassembly. Checked through patch_nso.py: v0 main takes the cut at 0x43e1a8 (implementation 2, no red line); a 1.1.1 main copy with another build ID takes implementation 1, output byte-identical to the registered cut; 1.1.1 keeps its exact CUTS. A scan of the 42 distinct mains on the drives found this limiter only in Taiko. The base game has not been run. Docs: rewriter.md (same-name implementations, procedure step 2, Found on), the game note. Checked: test_nso_match.py (new test fails on the old matcher), all 19 tool test files, docs tests

2026-10-26 13:24:01 +08:00

..

agents

Add GPU diagnostics and close periodic profiles per frame (#225)

2026-10-26 12:09:07 +08:00

images

Pokémon Shield (01008DB008C2C000, update 1.3.2 = v458752, Game Freak's engine on SDK 7.3.2): game profile, no host change. Sword's twin: the same sdk, subsdk0 (movie) and subsdk1 (GLSLC) builds, only main is its own (a1680262…), and the shader split gives Sword's 41,516 shaders with the same data rows in all three tables; the profile is Sword's with Shield's title, name, IPA name, build row and tables (prewarm list copied). The game note (dm/games/01008DB008C2C000/README.md) has what differs, the Mac route (a fresh save needs DM_KBD for the name; Sword's route out of the house works unchanged) and the shader build with DMSH_PREFILL from Sword's libraries (21,974 Mac / 21,684 iOS filled, 176 / 175 compiled). Plays on the iPhone (IPA 12.4 GB, user) and on the Mac at 30 fps with no red line from the first boot through the opening and Postwick to Hop. README gallery entry with a Mac screenshot, a row in the games table, a pointer from Sword's note, AGENTS.md's count to sixty-seven. Checked: the tool tests pass (111, 14 skipped)

2026-10-26 09:20:23 +08:00

reference

Function-matched cuts: same-name rules are one patch's implementations; Taiko Risshiden V DX's frame-limiter cut reaches the base game and later builds. nso_match.resolve groups rules by name: exactly one implementation must match (the log names it), a red line only when none or more than one does. Before, each implementation of a changed function logged a red line on every build it did not fit. Taiko's MATCH_CUTS has two implementations of the frame end under one name: 1.1.1's (main+0x443e8c, 364 bytes, cut at +288) and the base game's (v0 main 47d7baf6..., main+0x43e140, 172 bytes: 1.1.1 added the frame-skip branch; the same 20 bytes at +104 = main+0x43e1a8). In v0 N is likewise a global (GOT 0xa1be28, data 0x9ec55c, 1) set to 2 only by CPrivateBattleWindow / CPrivateBattleTutorialWindow, and the start tick is read after the update; nso_fnmap.py finds no exact match for 1.1.1's function in v0, found by disassembly. Checked through patch_nso.py: v0 main takes the cut at 0x43e1a8 (implementation 2, no red line); a 1.1.1 main copy with another build ID takes implementation 1, output byte-identical to the registered cut; 1.1.1 keeps its exact CUTS. A scan of the 42 distinct mains on the drives found this limiter only in Taiko. The base game has not been run. Docs: rewriter.md (same-name implementations, procedure step 2, Found on), the game note. Checked: test_nso_match.py (new test fails on the old matcher), all 19 tool test files, docs tests

2026-10-26 13:24:01 +08:00

development.md

Add GPU diagnostics and close periodic profiles per frame (#225)

2026-10-26 12:09:07 +08:00

geometry-metal-research.md

Add compute-backed geometry shader replay

2026-10-23 16:29:24 +08:00

image-atomic-binding-contract.md

Guard Metal image atomics and reject unsupported bindings

2026-10-25 23:37:28 +08:00

settings.md

Add opt-in MetalFX spatial upscaling and a single-screen launcher (#215)

2026-10-26 09:38:05 +08:00