fix(recompiler): fill missing pixel interpolators as DEFAULT - #6604

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.

Some comparisons could not be completed because a pull request conflicts with the base branch. Review these shared files manually before concluding that there is no overlap.

Intel and AMD Windows data: the new test passes on both GPUs with this change and fails on both without it.

Windows 11 Pro 26300, WinLibs GCC 15.2.0 r7, Release, every execution test built, ANYPS5_NO_SHADER_CACHE=1, serial CTest, main 47a266cf9. Intel UHD Graphics 770 (8086:a780, 32-wide subgroups), driver 32.0.101.7088. AMD Radeon RX 9070 XT (1002:7550, 64-wide subgroups), driver 32.0.31041.1004, with VkCache emptied first.

Full CTest (--parallel 4) with six open PRs merged into main (#6638, #6604, #6014, #6773, #6772, #6725): the same failures as main alone on both GPUs, nothing new. main alone: UHD 770 1392 of 1398, RX 9070 XT 1396 of 1398. The failures are image tables (#6647) and sample_multisample_target (#6971) on Intel, and quad_pixel_mask (#6255) and layered_volume_vertex_no_gpl (#4768) on AMD.

AI-assisted: yes (Claude Code)

Windows NVIDIA run of the GPU tests, since the description lists the collector test only.

Windows 11 Pro 26200, NVIDIA GeForce RTX 4080 (driver 617.42), AMD Ryzen 7 5800X, WinLibs GCC 15.2.0 r7, Ninja, Release, bundled dependencies, BUILD_TESTING=ON, ANYPS5_NO_SHADER_CACHE=1. Head c32c26edb merged onto main f51724078; full build succeeds.

• interpolation_reference_range, agc_interpolant_mapping, agc_driver_interpolation_modes, agc_driver_interpolation_f16: pass, with ANYPS5_REQUIRE_VULKAN=1.
• Full ctest -j 8: 1398/1398. Unchanged main on the same machine: 1398/1398.

No title run.

AI-assisted: yes (runs done with Claude Code).

Tested on Windows 11, MinGW-w64 GCC 15.2.0, NVIDIA GeForce RTX 5090 (driver 617.42), Ryzen 9 9950X3D. Head c32c26edb merged onto current main (a373e85); the merge is clean.

• interpolation_reference_range passes with the change.
• On main with only the updated test file it fails with attribute 0 with no SPI inputs becomes a default: shader info collection failed: interpolation parameter reference is out of range (attribute 0 component 0 mode 0, 0 inputs) (and the attribute 5 / 2 inputs case), so the test does cover the new behaviour.

Two things I could not settle from the diff:

1. The description says the title result is a retest after deploy ("PPSA03541 stacked: past collector OOR for ~5h, then aborted on undefined attribute; this follow-up addresses that emit path"). Nothing here shows that UFC 5 gets past this shader with this change, so the "unblocks title" label is not backed by a run yet. I have no PPSA03541 dump to check.
2. EnsurePixelInterpolatorCoverage modifies inputInfo.pixel through const_cast<ShaderPixelInputInfo&>, an object the caller passes as const. If that struct is hashed, serialized or used as part of a cache identity before Collect runs, filling in inputNum and interpolatorSettings afterwards could make the key differ from what was compiled. Is it only built per compile, or can it be shared? A copy passed down would avoid the question.

I am not approving until the title run is shown.

AI-assisted: yes. Written with AI assistance and reviewed by me.

Tested on a PC with a real title, EA SPORTS UFC 5 (PPSA03541), as a controlled A/B. Both runs used the same base, upstream main 819c547, plus #6549, #6563 (resolved against main) and #6572, which UFC 5 also needs. The only difference was this PR.

Without this PR: the game aborts at +205 s of the run. The log shows Shader 0x114073b400 stage 5 ... preparing it at use, then presentation failed: shader 0x114073b400 stage 5: ShaderRecompiler::Recompile: shader info collection failed: interpolation parameter reference is out of range (attribute 0 component 0 mode 0, 0 inputs), then SIGABRT (exit 0xC0000409).

With this PR: the same shader 0x114073b400 stage 5 is prepared at use (+424.8 s), and the game does not abort. It kept running for the rest of a 1000 s session, with no abort, fatal or driver-failure line in the log.

Caveats: this is one run each, and the abort and shader prepare times differ because I played to the match by hand. The game is still very slow after this point (about 0.2 to 2 fps), which is a separate problem this PR does not touch. I did not run the unit tests as part of this check.

AI-assisted: yes. Written with AI assistance and reviewed by me.

Thanks, the fix is scoped well: it only changes shaders that fail to compile on main today. Two things before merge:

1. Please remove the game title, the PPSA id and the shader address from the docs/dev/debt/shader.md line. Debt entries describe the behaviour, not a specific game. Title-specific notes belong in the Game bring-up discussion. Something like "a pixel shader can be prepared with NUM_INTERP 0 while it still issues v_interp on attribute 0" keeps the useful part.

2. The emitter change in c32c26e (the default for an attribute past inputNum or without a SPIR-V binding) isn't covered by a test. Nothing fails if it's reverted. Could you add a case that runs SPIR-V emission of GetInterpolationParameter for an attribute past inputNum (and one with no binding) and checks the P0 default and zero deltas? In InterpolationReferenceRange.cpp, it would also help to assert the outcome (the filled slots read as DEFAULT and no parameter input is added) rather than only the absence of one error message.

Minor, no change needed: InterpolateHostP1/P2 are lowered before info collection, so those two cases in EnsurePixelInterpolatorCoverage never fire. And a 16-bit high-half read past NUM_INTERP still fails during emission (as it does on main).

We merged current main and reworded the debt entry so it names no title; no code change.