[Clank700]

What

57 NTSC-U (SCUS_974.65) functions move from unmatched to matched, one C file each; one of them (func_L00_00528810) replaces its near miss in nonmatching/. US matched functions 5,435 → 5,492 (414,340 → 418,508 bytes), PAL 6,032 → 6,083 (497,616 → 501,564 bytes). config/us/map.tsv, config/pal/map.tsv, both progress reports and the README table are regenerated with the repository tools; only the rows of the new sources changed.

Every function here is new on both discs: none has a src/pal/ source on main, so no existing source or its credit moves (compared function by function in progress/report.json and progress/pal/report.json, per unit: no function name appears in two units with different percentages, and none of the existing functions gets worse or disappears).

This branch is rebased on current main (c78d917, after #90) and mergeable.

Compiler: 29 of the 57 files carry their own compiler flags

The default route (SN GCC 2.95.3, -O2 -G8 -fopt-stack) builds 28 of the files. The other 29 only match with a per-file header, which is how this repository already handles it (tools/build.sh reads /* cflags: ... */ from the first lines; nothing in the build scripts changes):

• 23 files: @ee29, the Sony EE GCC 2.9 route. Not all of RAC4 was built with one compiler: besides the SN compiler, part of the library-like and glue code comes out of Sony's ee-gcc 2.9 (2.9-ee-991111). This repository already supports that whole-file route (commit ddfe94d, tools/cc.sh, docs/TOOLCHAIN.md "Sony EE GCC 2.9 route"); the token @ee29 in the first-line cflags selects it and defaults to -O2 -G0 (one file adds -O2 -G0 explicitly). These 23 functions do not match under the SN route at all, and match byte for byte under the Sony route. Supporting evidence from our own objects (not from the disc): all 23 start with sd saves of callee-saved registers, 21 of them at ascending 16-byte spacing (offsets 0, 0x10, 0x20, ...), which is the Sony 2.9 prologue style; SN 2.95.3 with -fopt-stack saves with 8-byte slots.
• 6 files: -fno-schedule-insns variants (5 × -fno-schedule-insns, 1 × -fno-schedule-insns -fno-schedule-insns2), which the repository already uses in other sources; they turn off one or both instruction-scheduling passes of the SN compiler.

How we checked that the flag is actually needed: we compiled the 29 files again without their cflags line (default SN route) and ran the same strict comparison; all 29 fail (bytes differ at every one of their 40 placements). With their own header all 29 are exact.

A caveat we want to be honest about: matching under an alternative whole-file profile shows the code is compatible with that profile; it does not prove the historical flag set. The per-file header only states what reproduces the bytes.

How the C was found

Reconstructed from the retail code of this repository (our assembly, m2c) and made to match by compiling with the file's own flags (above). Plain C only: no inline assembly, register declarations or do { } while (0) shaping, and no source adapted from another project.

Checks (clean build of every source of this exact head, 4349a29, on top of main c78d917)

• Build: all 6,298 sources of the head (outside src/libgcc and src/libm) compiled with each file's own flags, every compile's exit code logged: 0 non-zero. 6,292 objects come from a clean full build of the previous head of this branch (the sources of the head are identical to that tree except the 6 PAL files that main added since, which were compiled fresh); build/libgcc and build/libm are copied from an earlier full build, since src/ changes do not affect them.
• tools/audit_matches.py: exit 0, 4,032 exact, 3 not exact (the same 3 as on main: func_00161890, func_00547A98, func_L01_004F3100).
• tools/audit_relocs.py (with the overlay ELFs): exit 1, 80 mismatches in 43 functions, 3 unresolved. This is not new: the same lines are reported on main without this change (diffed line by line against a build of 60818dd; only the count of checked functions differs), and none of the 57 functions appears in them. CONTRIBUTING asks for 0, so we do not call this gate clean; the existing entries (54 call targets, 25 GPREL16, 1 HI16/LO16, in sources this PR does not touch) need a separate fix.
• tools/match_map.py --check and tools/progress.py --check, for US and --version pal: all exit 0.
• Beyond the repository tools, a per-placement check of all 57 functions (every catalogued level placement, 173 in all: code bytes with relocation fields masked, and every call target resolved at that placement's own level): 57 of 57 exact at every placement; negative controls (bytes of another function, call targets shifted by +4) are all rejected.
• Our local rebuild of config/*/map.tsv also flips some unrelated rows (libm rows, because build/libm was not rebuilt here); those rows were restored to main's values, so the map diff contains only the new functions.
• No overlap with the open PRs Match 45 NTSC-U helpers with pinned compiler profiles #71, Match 48 PAL resident functions and retain one nonmatching reconstruction #93, Match 26 NTSC-U functions #94, Match 50 PAL engine and menu functions #95, Match 165 PAL functions #96 and Match 198 NTSC-U functions converted from their PAL matches #97, by file, by function name or by PAL placement. Match 45 NTSC-U helpers with pinned compiler profiles #71 also edits the maps, reports and README, so whichever of the two lands second needs tools/match_map.py --write / tools/progress.py run again.

🤖 Generated with Claude Code

[bytematchdev]

ByteMatch progress report

SCES_532.85

Code matched	8.63% (+0.07%) · 501.56 KB of 5.81 MB
Fuzzy match	9.12% (+0.07%)
Functions	6,083 (+51) / 18,113
Units complete	3,270 / 3,329

✅ Newly matched (51)

• func_L00_0077D4D0 (196 bytes)
• func_L00_00528D10 (164 bytes)
• func_L00_00507D18 (144 bytes)
• func_L00_007722B8 (120 bytes)
• func_L01_00575660 (116 bytes)
• func_L00_004FF728 (112 bytes)
• func_L01_0048BCD0 (104 bytes)
• func_L00_005000C0 (104 bytes)
• func_L00_00506868 (104 bytes)
• func_L00_0052FBC0 (104 bytes)
• func_L00_00541A28 (104 bytes)
• func_L00_00541AE0 (104 bytes)
• func_L00_00780CD0 (100 bytes)
• func_L00_004FFFC0 (96 bytes)
• func_L00_00540838 (96 bytes)
• func_L00_00521408 (92 bytes)
• func_L00_00508270 (88 bytes)
• func_L00_00517468 (88 bytes)
• func_L00_00714F98 (88 bytes)
• func_L06_004C8B58 (84 bytes)
• func_L00_00519B58 (84 bytes)
• func_L41_003E33D0 (84 bytes)
• func_L00_005082C8 (80 bytes)
• func_L00_00517340 (80 bytes)
• func_L00_00520AA8 (80 bytes)
• func_L00_0074DFE8 (80 bytes)
• func_L00_00719490 (76 bytes)
• func_L00_00773630 (76 bytes)
• func_L00_005303C8 (72 bytes)
• func_L00_00538690 (72 bytes)
• …and 21 more

SCUS_974.65

Code matched	7.24% (+0.07%) · 418.51 KB of 5.78 MB
Fuzzy match	7.80% (+0.07%)
Functions	5,492 (+57) / 18,085
Units complete	2,839 / 2,899

✅ Newly matched (57)

• func_L00_007780F8 (196 bytes)
• func_L00_00528810 (164 bytes)
• func_L00_00507818 (144 bytes)
• func_L00_0076D2A8 (120 bytes)
• func_L01_00572A68 (116 bytes)
• func_L00_004FF228 (112 bytes)
• func_L01_0048A628 (104 bytes)
• func_L00_004FFBC0 (104 bytes)
• func_L00_00506368 (104 bytes)
• func_L00_0052F6C0 (104 bytes)
• func_L00_00541528 (104 bytes)
• func_L00_005415E0 (104 bytes)
• func_L00_0077B8F0 (100 bytes)
• func_L00_004FFAC0 (96 bytes)
• func_L00_00540338 (96 bytes)
• func_L00_00520F08 (92 bytes)
• func_L00_00507D70 (88 bytes)
• func_L00_00516F68 (88 bytes)
• func_L00_00711178 (88 bytes)
• func_L06_004C7320 (84 bytes)
• func_L00_00519658 (84 bytes)
• func_L41_003E2B98 (84 bytes)
• func_L00_00507DC8 (80 bytes)
• func_L00_00516E40 (80 bytes)
• func_L00_005205A8 (80 bytes)
• func_L00_007493B8 (80 bytes)
• func_L00_00715648 (76 bytes)
• func_L00_0076E620 (76 bytes)
• func_L00_0052FEC8 (72 bytes)
• func_L00_00538190 (72 bytes)
• …and 27 more

_{Commit 4349a29 · Ratchet: Deadlocked on ByteMatch}