fix(agc): order raw shader reads after queued code writes

[JJenkx]

What

A raw shader read behind a queued write to the same code can return the old program. Queue-0 draw commits now run through DrawPipeline, but ReadRawShader only flushes work already recorded by the GPU recorder. A pending pipeline commit is not yet visible to that flush. Both a cached-code comparison and a first-time code read can therefore get ahead of the write.

Drain overlapping queued writes before comparing a cached raw snapshot and before reading each new code span. The drain runs outside the raw-shader cache mutex. Reads outside the queue-0 pipeline, and reads with no overlapping pending write, retain the existing path. No shader instruction semantics, registration policy, or cache format changes.

The CPU-only agc_driver_queued_raw_shader regression exercises cached and fresh reads through the real pipeline committer and guest-memory write path. The queued writer waits until the read returns or its bounded delay expires; the read must observe the replacement program. The cases also run separately as agc_driver_queued_raw_shader_tests cached and ... fresh. No PS5 assets or GPU are required by this regression.

This is independent of #6188 and needs no commits from it. #5751 changes registration/range bookkeeping in the same source file; this PR changes ordering of code reads and does not replace that work.

Tested

Based on main 341613ab62b979e429daa1be5ac2fe524a737df6, public head 6cd8b7240. Linux x86-64 GCC Release, renderer SPIRV-Tools enabled: full runtime libraries and all selected executables build. All 10 selected CTest cases pass without skips on both Radeon RX 6650 XT / RADV Mesa 26.2.4 and lavapipe (6.47 s / 6.58 s). Besides the new CPU regression, these cover draw-cache reuse with GPL on/off, validation and inline modes, draw-write ranges, pipelined DMA/fills and predication. Conventions and whitespace pass.

Negative controls replace only ShaderRegistry.cpp with main's version. Both selectors exit 1 with cached raw shader read ignored a queued write to its code / fresh raw shader read ignored a queued write to its code. The executable SHA-256 stays 4f87301733885662f3e6422eb053b95aab8d53372269a63dcb1c18a6f4189bbd through these controls and the restored-source build. The fix is restored and the full libraries rebuilt before the final passing tests.

The local validation tree differs from the public head only by a private SPIRV-Tools teardown-repair gitlink, excluded from this PR. These local results do not independently validate the official dependency pin. Contributor-reported official-pin and full-suite results are recorded below; Windows validation remains pending. No title unblock or FPS improvement is claimed.

Contributor-reported validation by Dextroid17: head 6cd8b7240 cleanly merged onto main 84b0d2f24, CachyOS Linux 7.2.9, GCC 16.2.1, Release, lavapipe, ANYPS5_REQUIRE_VULKAN=1, ctest -j16.

• With SPIRV-Tools ON and the unchanged official pin 7f2d9ee92, cached and fresh selectors each pass 25/25 with this fix and fail 25/25 with only ShaderRegistry.cpp restored from main; restoring the fix and rebuilding passes another 25/25 each.
• The ON full suite reports 1368/1369 in each of three runs, with 62 tests skipped on lavapipe. agc_compute_pipeline_replay fails with replay did not export a pipeline cache header; the contributor reproduces that failure with main's ShaderRegistry.cpp as well. These are not all-green full-suite results.
• With SPIRV-Tools OFF, the full suite reports 1361/1361 in each of three runs. The queued raw shader regression passes 50/50 with the fix and fails 50/50 with main's ShaderRegistry.cpp. Conventions pass on the merged tree.

These results were reported by the contributor, whose testing was AI-assisted with Claude Code; they were not reproduced locally. No Windows testing was reported.

Checklist

• Based on current main; no other open PR implements the same ordering fix
• One topic per PR
• No new code comments or investigation files
• Generic behavior; no title-specific paths or silent stubs
• No game assets, binaries, raw captures or new third-party code committed
• Depends on: none (perf(agc): pipeline queue-0 draw commits behind the packet worker #1201/perf(agc): pipelined draws on by default #1202 are merged)
• Unblocks: none
• GPUs tested: AMD Radeon RX 6650 XT, RADV Mesa 26.2.4, Linux 7.2.9; lavapipe Mesa 26.2.4, Linux (the new regression itself is CPU-only)
• AI-assisted: yes

[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
PR 6234	core/libs/prx/libSceAgcDriver/CMakeLists.txt L1832
PR 6291	core/libs/prx/libSceAgcDriver/CMakeLists.txt L1832
PR 6606	core/libs/prx/libSceAgcDriver/CMakeLists.txt L1832
PR 6946	core/libs/prx/libSceAgcDriver/CMakeLists.txt L1832
PR 6947	core/libs/prx/libSceAgcDriver/CMakeLists.txt L1832

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.

Pull request	Shared files not compared
PR 1341	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 1368	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 1395	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 2434	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 4650	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 4891	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 5021	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 5115	core/libs/prx/libSceAgcDriver/Execution/src/Driver/Shaders/ShaderRegistry.cpp
PR 5371	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 5426	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 5499	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 5925	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6044	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6185	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6186	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6188	core/libs/prx/libSceAgcDriver/CMakeLists.txt core/libs/prx/libSceAgcDriver/Execution/src/Driver/Shaders/ShaderRegistry.cpp
PR 6192	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6280	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6300	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6320	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6360	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6412	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6481	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6486	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6490	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6499	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6501	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6503	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6535	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6683	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6741	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6747	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6763	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6788	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6808	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6867	core/libs/prx/libSceAgcDriver/Execution/src/Driver/Shaders/ShaderRegistry.cpp
PR 6868	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6905	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6937	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6941	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6963	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6977	core/libs/prx/libSceAgcDriver/Execution/src/Driver/Shaders/ShaderRegistry.cpp
PR 7201	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 7223	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 7332	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 7642	core/libs/prx/libSceAgcDriver/CMakeLists.txt core/libs/prx/libSceAgcDriver/Execution/src/Driver/Shaders/ShaderRegistry.cpp

[Dextroid17]

I ran this on Linux with the official SPIRV-Tools pin and the full test suite, the two things your Tested section lists as not covered.

Setup: head 6cd8b7240 merged onto main 84b0d2f24 (clean merge; the branch is 1,020 commits behind), CachyOS Linux 7.2.9, GCC 16.2.1, Release, -DANYPS5_ENABLE_SPIRV_TOOLS=ON as in build.yml, with 3rdparty/SPIRV-Tools at the pinned 7f2d9ee92 and no local changes. Tests ran on lavapipe with ANYPS5_REQUIRE_VULKAN=1, ctest -j16.

SPIRV-Tools ON, official pin                          with this PR   ShaderRegistry.cpp from main
agc_driver_queued_raw_shader_tests cached (25 runs)   25/25 pass     0/25 ("cached raw shader read ignored a queued write to its code")
agc_driver_queued_raw_shader_tests fresh  (25 runs)   25/25 pass     0/25 ("fresh raw shader read ignored a queued write to its code")
after restoring the file and rebuilding               25/25 each
full ctest, 3 runs                                    1368/1369 each (the one failure is local, see below), 62 skipped on lavapipe

The one failure, agc_compute_pipeline_replay (replay did not export a pipeline cache header), also fails on my machine with ShaderRegistry.cpp from main, and it passes on CI's main build, so it is something local to my setup and not this change.

With SPIRV-Tools OFF (the CMake default), full ctest passed 1361/1361 in each of 3 runs, and agc_driver_queued_raw_shader passed 50/50 through ctest with the fix and 0/50 with ShaderRegistry.cpp from main. tools/check_conventions.py passes on the merge.

I did not test Windows.

Written with AI assistance (Claude Code). I ran the builds and tests above on my machine.