feat(agc): accept the LINE_STIPPLE_TEX and POS_FIXED_PT pixel inputs

[TheAggie]

What

DecodePixelStageInfo rejected any SPI_PS_INPUT_ENA/ADDR with LINE_STIPPLE_TEX_ENA (bit 7) or POS_FIXED_PT_ENA (bit 15): unsupported SPI_PS_INPUT_ENA/ADDR bit combination. RADV on the PS5 (mihawk-99/PS5_Mesa) sets both. In Mesa's radv_compute_spi_ps_input (src/amd/vulkan/radv_shader.c):

• LINE_STIPPLE_TEX is the required pixel input of a shader that reads no interpolants: "At least one of PERSP_* (0xF) or LINEAR_* (0x70) or LINE_STIPPLE_TEX must be enabled. LINE_STIPPLE_TEX uses the least number of initialized VGPRs". ccleste's (PPSA59322) present shader is one; with feat(agc): accept graphics programs without an AGC shader header #1732 on current main, every frame stopped there (presentation failed).
• POS_FIXED_PT is enabled when a shader reads integer pixel coordinates (reads_pixel_coord): X in bits 15:0, Y in bits 31:16.

Changes:

• Both bits join the supported inputs in DecodePixelStageInfo. PERSP_PULL_MODEL and SAMPLE_COVERAGE are still rejected.
• ShaderInputInfoBuilder places both VGPRs whenever SPI_PS_INPUT_ADDR has the bit. inputAddr is already part of every cache, draw and request key, so no key or serializer changes are needed.
• The pixel entry fills them:
  • POS_FIXED_PT is uint(FragCoord.x) | uint(FragCoord.y) << 16.
  • LINE_STIPPLE_TEX is 0, because line stipple is not emulated. The TechnicalDebt entry under Functional records this.
• Tests:
  • agc_driver_graphics (PixelInputLayoutTests):
    • ENA/ADDR 0x80 decodes, with LINE_STIPPLE_TEX in v0.
    • With 0x8082, LINE_STIPPLE_TEX takes v2 and POS_FIXED_PT v3, after the perspective-center pair.
    • A shader exporting the POS_FIXED_PT VGPR reads FragCoord, and one exporting LINE_STIPPLE_TEX does not.
    • 0x8 and 0x4000 are still rejected.
  • New GPU test agc_driver_pixel_fixed_point_position:
    • A pixel shader splits the POS_FIXED_PT VGPR into X and Y and converts them to float. It exports them with the LINE_STIPPLE_TEX VGPR into a linear 64x16 32_32_32_32 float target.
    • Every pixel must read back (x, y, 0, 1) for three ADDR layouts: the input alone, after LINE_STIPPLE_TEX, and after the perspective-center pair and LINE_STIPPLE_TEX.
    • It fails (pixel (1, 0) holds (0, 0, 0, 1)) without the POS_FIXED_PT emission.

Tested

Windows 11 Pro (26200), MinGW-w64 GCC 15.2.0 (WinLibs r7), Release, NVIDIA GeForce RTX 5070 Ti; the new GPU test also on lavapipe (Mesa 25.2.8, Windows).

• The full suite (478 tests) passes on this branch. It was run from the generated CTestTestfile.cmake commands because antivirus blocks ctest.exe on this machine.
• ccleste renders its title screen on current main with this, feat(agc): accept graphics programs without an AGC shader header #1732 and my other pending changes on a local branch (RADV_DEBUG=nonggc set).

Not tested on AMD, Intel or Linux hardware.

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

[johnmaia]

Linux run on AMD, since the description lists it as not tested: Ubuntu 24.04.5 (kernel 6.8), GCC 13.3, Release, AMD Radeon RX 7600 (RADV NAVI33, Mesa 25.2.8). This PR's head bda16f01, merged onto main f1451396. Full ctest: 495/495 pass, with the GPU tests running on the RX 7600 (none skipped). That includes the new agc_driver_pixel_fixed_point_position and agc_driver_graphics with this PR's cases.

[adriadam10]

Checked on current main: the PR rebased cleanly except for the TechnicalDebt line, which I moved off the end of the section to avoid the append conflict. Build and ctest of libSceAgc/libSceAgcDriver are green; agc_driver_graphics and agc_driver_pixel_fixed_point_position fail without the change. Also confirmed by @johnmaia on an RX 7600 (RADV). Merging once CI is green.