fix(agc): prepare a rect-list draw's missing rectangle at draw from its prepared stages - #1928

#1825 and #1732 change the same lines in Draw.cpp. Both call ResolvePreparedGraphics(..., 17u, ...) before the PreparedRectangle lookup, but only when the vertex program is unregistered (programs[0].snapshot->header.empty()). This PR covers a different trigger: the vertex program is registered and the fragment variant was prepared at the draw (the null pixel program from #1902, or a shader deferred by #1836).

So the changes don't duplicate each other, but they will conflict. Whichever lands first, I'm happy to rebase so both cases go through one helper (DrawRectangle here, which builds the rectangle when the lookup misses), or to drop that part if the other PR's version covers it once merged.

AI-assisted: yes

I verified the primitive-type question against the current decoder: Graphics::DecodeState sets rectList only for primitive values 7 or 17. ResolvePreparedGraphics uses the value only to set rectangleRequested when it is either 7 or 17; rectangle generation does not distinguish them. So passing 17 from this fallback is equivalent for any draw that reached the rect-list path.

Tested this with Hades II (PPSA36082), together with a fix for the null program miss (the registration fix I described on #1902). The fallback kicks in once while the level loads, for the null fragment program, and the game goes on to gameplay without problems.

Uncharted: Legacy of Thieves Collection (PPSA05684) stops here once its fragment variants are prepared at the draw (locally: the null pixel program and its own shaders, see #1902):

AGC driver: prepared rectangle artifacts are missing

I fixed it locally the same way (build the rectangle pair from the prepared stages at the draw and keep it with the snapshot) to reach the title's main menu. Separately, the fault buffer of the built control/evaluation shaders could collide with another stage's binding group; that fix is #2198. I have not run this branch itself with the title.

Two more titles hit the case this PR fixes, on Linux with an RX 9070 XT (RADV).

• Ratchet & Clank: Rift Apart (PPSA01474), on main 43972e6 + feat(libkernel): implement mlock and munlock #1673, feat(libSceVideoOut): register the VRR status event and accept the VRR status call #1339, fix(agc): defer shaders that cannot be prepared at registration to their first use #1836, fix(agc): accept bound copies of a registered shader header that differ only in user_data #1837, fix(agc): skip registration-time preparation for unsupported shader types #1893, perf(recompiler): compute post-dominators visiting successors first #2105, fix(libkernel): start the pthread object with the thread id #2184, fix(agc): accept disabled conservative rasterization with uncertainty-region bits #2226, feat(agc-driver): draw DB_RENDER_CONTROL depth clears #2219, fix(agc): prepare the null pixel program at draw when its variant was not prepared #1902:
  • Without this PR, the run stops about 20 s in on prepared rectangle artifacts are missing.
  • With it, the log shows Rect-list draw of shader 0x116643b200 has no rectangle for fragment shader 0x7f33…1700; preparing it at draw four times. The fragment shader is the null pixel program that fix(agc): prepare the null pixel program at draw when its variant was not prepared #1902 prepares at draw.
  • The run goes on to an unrelated stop (color views of several array slices are unsupported, Layered rendering: a draw into all slices of a 3D color target aborts with "color views of several array slices are unsupported" #2102).
• Astro Bot (PPSA21567) hits the same miss after fix(agc): prepare the null pixel program at draw when its variant was not prepared #1902. I reported it on fix(agc): prepare the null pixel program at draw when its variant was not prepared #1902 before I found this PR: the null pixel variant 66d was prepared at draw, with only (30a, 30b) and (30a, 30c) rectangles registered.

Tested and written by Claude (AI assistant) on a contributor's machine.

Astro's Playroom (PPSA01325) hits this on Windows with NVIDIA, right after #1902.

Setup: main 745b775 + #1836 + #1837 + #1395 (rebased locally) + #1902, plus a local workaround for vertex fetch exceeds descriptor record count (no PR yet).

• Without this PR, a rect-list draw stops with AGC driver: prepared rectangle artifacts are missing.
• With this PR (def0be1) it logs Rect-list draw of shader 0x400201900 has no rectangle for fragment shader 0x7ff9c095b700; preparing it at draw and the draw goes through.
• The next stop is a compute shader reading through a null SRT pointer: compute shader 0x5001c6a00: AGC driver: guest memory is not readable at 0x28 (range 0x28+0x4). That is the feat(agc): implement image and sampler tables indexed at runtime offsets #757-feat(agc): implement deferred faults for reads through null SRT pointers #759 stack.

Windows 11, RTX 3060 Laptop (NVIDIA 616.92), WinLibs GCC 15.2.

🤖 Generated with Claude Code

Tested with Sonic Origins (PPSA05324) on Linux, R9 380 (RADV). Its depth/stencil clears are rect-list draws without a pixel program, so they stopped at prepared rectangle artifacts are missing once #1902 prepared the null pixel variants. With this PR, the log shows three Rect-list draw of shader ... has no rectangle for fragment shader ...; preparing it at draw lines, the clears run, and the title presents its SEGA logo. Testing branch: main d70b899 + #2022, #2219, #1902, #1837, #425 and #1544.

Bring-up thread: #2021

Tested on Windows with Marvel's Wolverine (PPSA03671), after #1902.

Setup: Windows 11 Pro (10.0.26200), RTX 3070, GCC 15.2.0 UCRT (WinLibs), Release. This PR (def0be1) merged onto main 54f28f2 with #1836, #1893, #1709, #1837, #1940, #2022, #2219, #1902, and a local change that lets sceKernelSignalSema accept a count of 0. It merged without conflicts.

Without this PR: the first presentation fails with AGC driver: prepared rectangle artifacts are missing.

With this PR: the rect-list draws pass. The next stop is a compute shader deferred at registration: GetImageResource dword 0 is not a valid runtime value; chain: ReadConstBuffer@pc=0x00001140(GetBufferResource(ReadConst(GetSrtResource, Immediate), ..., probably #757's case.

Limits: one title, one run each way. No frame is presented yet.

Checked the lookup move is behaviour-preserving, the fallback only runs on a missing rectangle (any other failure still throws), and primitive types 7/17 take the same rect-list path. Build + ctest of libSceAgcDriver green (273/273, agc_driver_prepared_graphics passes), CI green on this head, and several title runs in the thread confirm it. Merging.

Note: #2261 touches the same lines in Draw.cpp; whichever lands second needs a rebase.