[EdouardSence]

What

PrepareRegistered decodes the shader, builds its control-flow graph and structurizes it, then throws the graph away. Next, ShaderRecompiler::PrepareShader → PrepareResourceProgram decodes, builds and structurizes the same code again with the same SwappcInfo. Every registered shader therefore goes through GraphBuilder::Build + Structurizer::Structurize once more than needed. mrk1987 counted three Structurize passes per shader on #2105; this removes one of them.

The removed pass only did one thing: throw early. Any exception it raised is raised again from PrepareShader for the same code, so a shader that failed registration before still fails it, with the same message.

No title stops on this; it is a load-time cost. On main the registration path runs for every shader a title registers, so I measured it on a title that registers many.

Tested

Linux (CachyOS, GCC, Release), Ryzen 7 7800X3D, RX 9070 XT, RADV / Mesa 26.2.4.

• ctest on this branch (rebased on main d70b899): 512 of 513 pass. The failure is import_audit (source_commit is None because my build tree is a git worktree), unrelated to this change.
• Rise of the Ronin (PPSA04540), cold shader cache (shader_cache deleted before each run), headless gamescope at 1920×1080. Ronin does not get far on main alone, so these runs use a local stack of open PRs on main 43972e6 (fix(agc): defer shaders that cannot be prepared at registration to their first use #1836's first version, fix(agc): accept bound copies of a registered shader header that differ only in user_data #1837, fix(agc): prepare the null pixel program at draw when its variant was not prepared #1902, fix(agc): prepare a rect-list draw's missing rectangle at draw from its prepared stages #1928, perf(recompiler): compute post-dominators visiting successors first #2105 and others). The only difference between A and B is this commit. Both runs stop at the same later point, prepared shader artifact is missing for the requested static ABI … stage=1, which is unrelated to this PR. I time from launch to that stop:

Run	Before (A)	After (B)
1	146 s	105 s
2	144 s	105 s

The shaders compiled are the same in both: 804–809 cache misses in the first 10 s, about 850 in total.

Overlap

#1836 adds its deferral check in PrepareRegistered right before the removed GraphBuilder::Build line. Whichever merges second needs a small rebase. Nothing else changes.

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 (overlaps fix(agc): defer shaders that cannot be prepared at registration to their first use #1836 and fix(agc): defer a shader whose code does not decode instead of aborting the title #2257 in PrepareRegistered, see Overlap)
• AI-assisted: yes. The investigation, code, measurements and this description were written by Claude (Claude Code) on my machine.

🤖 Generated with Claude Code

[Kikoarl]

Tested on Linux (CachyOS x86_64, Intel RPL-S iGPU / Mesa 26.3.0-devel / Vulkan 1.4.363, GCC 16.2.1) at 9bc97292:

• Build: Built cleanly in RAM (/tmp/anyps5_build) with zero warnings/errors.
• Conventions: python3 tools/check_conventions.py --base origin/main --head HEAD -> PASS (0 errors).
• Verification: Verified elimination of redundant structurize pre-check in ShaderRegistry.cpp. Trims shader registration overhead cleanly.

LGTM!

[popcoder565]

Native Windows run, since no Windows results were posted. Setup: Windows 10 (19045), RTX 3060 Laptop (NVIDIA 616.92), MinGW-w64 GCC 15.2.0 (WinLibs r7), Release. Head 9bc97292, merged onto main f8c2f07 in one build with 9 other PRs that had no Windows run.

• This PR has no test of its own. It builds, and the suite passes apart from the failure below.
• Full ctest in that build: 512 of 513. The one failure is agc_driver_depth_fast_clear_submit from fix(agc): storage access to a depth surface's stencil plane through its stencil aspect #2325, which fails intermittently on its own too. The test comes from feat(agc): depth planes sampled as 2D arrays, cubes and integer views #425, where I reported it.

No title was run.

🤖 Addressed by Claude Code

[oneandonlydean]

Needs a rebase: ShaderRegistry.cpp conflicts with main.

[adriadam10]

Rebased onto current main as requested. The conflict was with #1836: main now decodes the shader before the deferral check (UnresolvableSwappcTarget), so the decode stays. The rebased change only drops the redundant GraphBuilder::Build + Structurize pre-check and the Structurizer.hpp include (GraphBuilder.hpp stays, it declares SwappcInfo). PrepareResourceProgram (Recompiler.cpp:100-117) repeats both with the same SwappcInfo, so the exceptions are unchanged.

Build + ctest of libSceAgcDriver on the rebased head: 304/304. The Ronin timing in the description was taken with the decode also removed, so the saving on main will be somewhat smaller.

Merging once CI is green.