feat(agc): implement packet and range predication and the predication sizes

[CnRJay]

What

Five libSceAgc exports threw NotImplemented.

• sceAgcSetPacketPredication(packet, predication) sets (1) or clears (0) the PM4 predicate bit (bit 0) of a type-3 packet header and leaves the payload alone. A null packet, a non-type-3 header or a value above 1 throws.
• sceAgcSetRangePredication(start, end, predication) does the same for every type-3 packet in [start, end). It skips type-2 filler dwords. A range that ends inside a packet or holds another packet type throws.
• sceAgcDcbSetZPassPredicationEnableGetSize, sceAgcDcbSetPredicationDisableGetSize and sceAgcDcbSetBoolPredicationEnableGetSize return 16 bytes.
  • Their builders are inline in the SDK and not exported. Z-pass, boolean and disable are all PM4 SET_PREDICATION operations, so the size is that of the packet sceAgcDcbSetPredication writes.
  • This is derived, not observed, so it is recorded under Unknown function info in TechnicalDebt.

The executor still rejects predicated packets ("PM4 header flags ... are not implemented") and SET_PREDICATION itself ("GPU query predication is not implemented"). A title that predicates commands still stops at execution with a clear error, rather than running them unconditionally.

Tested

Windows 11, MinGW GCC 15.2.0 r7. New agc_predication_tests:

• Sizes: the three size queries against the bytes sceAgcDcbSetPredication writes.
• Single packet: setting and clearing the bit on a packet, and rejecting bad arguments without modifying anything.
• Range: a range with a filler dword between packets, where payloads and fillers stay untouched. Also an exclusive end, an empty range, a range ending mid-packet, a reversed range, an invalid value and a non-type-2/3 packet.

Full ctest: 217/218 pass. agc_shader_memory fails without this change too; #574 fixes it.

Checklist

• Based on current main; no other open PR implements the same functions
• One topic per PR; follow-ups go in a new PR
• No comments in code except technical debt
• Unimplemented paths throw (NotImplemented_nid_no_patch); silent stubs are listed in TechnicalDebt
• No notes, investigation .md files or images added to the repository (attach them to this PR)
• New third-party code is a submodule built from source
• Generic behaviour, not specific to one title
• Depends on: none
• AI-assisted: yes

🤖 Generated with Claude Code

[oneandonlydean]

Thanks. The five NIDs recompute to these names, the predicate bit and the size from the count field are right for a type-3 header, and it matches BryKytyPS5's AgcSetPacketPredication/AgcSetRangePredication, except that you throw for an invalid value or a reversed range, which is fine. The 16-byte sizes with the TechnicalDebt line look good to me too.

One gap in the range walk: the one-dword pad 0xffff1000 (a type-3 NOP with count 0x3fff, which the CP treats as a single dword; main's Commands.cpp test calls it the empty payload marker). Here it's read as a 16385-dword packet. In a short range that throws, but in a range longer than that it silently skips the next 16384 dwords, and those packets keep their old predicate bit. BryKytyPS5 treats (header & 0x3fffff00) == 0x3fff1000 as one dword. Could you do the same (or throw on it), with a test row next to the type-2 filler?

agc_predication_tests passes here and fails on main's code ("sceAgcDcbSetZPassPredicationEnableGetSize not implemented"). It conflicts with #338, #407, #545 and #550 in TechnicalDebt.md (neighbouring lines), so whichever lands later needs a one-line rebase. PPSA21564 doesn't import any of these functions. The CI run still needs workflow approval.

[github-actions]

Progress report (7de111f → 73a751a)

📈 System libraries: 80.85% (+0.15%, +5 implemented functions)

✅ 5 implemented

Library	Name
libSceAgc	sceAgcDcbSetBoolPredicationEnableGetSize
libSceAgc	sceAgcDcbSetPredicationDisableGetSize
libSceAgc	sceAgcDcbSetZPassPredicationEnableGetSize
libSceAgc	sceAgcSetPacketPredication
libSceAgc	sceAgcSetRangePredication

[oneandonlydean]

One more thing for when you push the pad change: #338 and #613 have landed since, and both add lines at the end of "Unknown function info" in docs/dev/TechnicalDebt.md, where this PR adds its predication sizes line. Could you rebase onto main at the same time, keeping main's lines and adding yours after them?

[github-actions]

Other open pull requests touch the same code. Merging one will break the other, or both implement the same thing. If this one builds on them, list them in Depends on.

Pull request	Conflicting files (lines on main)	Same exports	Same new files
#778	docs/dev/TechnicalDebt.md L122

[oneandonlydean]

Thanks for the pad change and the rebase. d4f7dda steps over 0xffff10xx as one dword, like a type-2 filler, and leaves it unpredicated. The test now puts the pad between two packets and checks that both still get the bit and that the pad is unchanged, which throws on the previous head. That matches the one-dword size #772 uses for sceAgcGetPacketSize.