Upstream PR batch 1 U4: SPIR-V post-pass and IR/control-flow compile-time CPU

[RINNECODER]

Upstream PR batch 1, unit U4: SPIR-V post-pass and IR/control-flow compile-time CPU. One commit per upstream PR, each with "Ported from upstream PR #N" in the body.

Ported

• perf(shader): rewrite high-fan-out IR uses in linear time boykopovar/AnyPS5#4035 high-fan-out IR use rewrite in linear time (IrValue.cpp, new shader_ir_use_rewrite_tests)
• perf(shader): index specialization tables by SPIR-V id boykopovar/AnyPS5#4772 specialization tables indexed by SPIR-V id
• fix(shader): honor disabled optimization during specialization boykopovar/AnyPS5#4827 honor APS5_SPIRV_OPT=none during specialization
• perf(shader): discard helper functions unreachable after specialization boykopovar/AnyPS5#5064 discard helper functions unreachable after specialization (applied after perf(shader): index specialization tables by SPIR-V id boykopovar/AnyPS5#4772; the helper pass was adapted to the id-indexed tables)
• fix(shader): drop module constants nothing references after specialization boykopovar/AnyPS5#5315 drop module constants nothing references after specialization (also relaxes one size ratio in FormattedBufferOnce, as upstream does)
• fix(shader): bound SPIR-V scalar replacement boykopovar/AnyPS5#4288 cap SPIR-V scalar replacement (SPIRV-Tools builds only)
• perf(recompiler): iterate structurizer fixpoints in post-order and bound shared merge splitting growth boykopovar/AnyPS5#5028 structurizer post-order fixpoints and bounded shared-merge splitting. Upstream's branch stacks "structurize refused control flow with reach flags" under it, so that change is included in this commit. The debt note went to docs/dev/debt/libSceAgcDriver.md (upstream's debt/shader.md does not exist here).
• perf(shader): emit image_bvh_intersect_ray as one shared function per variant boykopovar/AnyPS5#5455 image_bvh_intersect_ray as one shared function per variant
• perf(shader): avoid redundancy traversal map copies boykopovar/AnyPS5#5511 SPIRV-Tools redundancy-elimination patch (only built with ANYPS5_ENABLE_SPIRV_TOOLS=ON, default OFF); the debt line moved from TechnicalDebt.md to debt/libSceAgcDriver.md

Skipped

None.

Added, not a port

fix(metal): accept a specialized BDA module that dropped one of its two BDA descriptors. With boykopovar#5064 in, a smoke run skipped 1111+ compute dispatches with "BDA ABI requires page-table and fault-buffer descriptor metadata": specialization pruned the unreferenced page-table variable but kept the fault buffer, and MetalShaderBridge.cpp demanded both roles. Bisected to boykopovar#5064 (skips absent at boykopovar#4035+boykopovar#4772, present at boykopovar#5064). The bridge now requires at least one of the two roles; the module's descriptors are still checked against the prepared plan. This is a check in our Metal bridge only, so no upstream comment draft applies.

Evidence

• ShaderDiskCache FormatVersion unchanged (34): the payload layout is the same, and entries are already rejected when the source version differs.
• Tests (x86_64 build, run under Rosetta): shader_ir_use_rewrite, agc_structurizer (+ intersection_off, iteration_selection_off, forced_fallback), agc_shader_control_flow (+ iteration_selection_off, forced_fallback), agc_spirv_module, agc_spirv_bvh_intersect_function, agc_shader_disk_cache, agc_formatted_buffer_once: 12/12 pass. With ANYPS5_ENABLE_SPIRV_TOOLS=ON: agc_spirv_optimizer (default, full, none) pass, also after the perf(shader): avoid redundancy traversal map copies boykopovar/AnyPS5#5511 patch. Metal bridge build (ANYPS5_METAL_SHADER_BRIDGE=ON): metal_utility_tests pass. No new test covers the bridge relaxation; it is exercised by the smoke below.
• Game smoke, one 280 s run per tree (eboot lock, plugged in, cold shader cache): main and this branch both end in the same boot-movie loop with no fault and the same single unimplemented import. main: no skipped dispatches (one rejected PM4 submission). Branch: no skipped dispatches or conversion failures; the Metal pipeline archive write failed once (cannot pack in the GPU archiver), which is environmental. Compared by log only; no screenshots taken.
• Performance: not measured. No compile-time CPU numbers on Metal were taken, so no speedup is claimed.

🤖 Generated with Claude Code

https://claude.ai/code/session_01DgW4n3qYgi5jAAohEmXRQj

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-14T02:22:06.216450Z	170b5c4	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

• Configuration used: defaults
• Review profile: CHILL
• Plan: Advanced
• Run ID: a7431bee-209c-44aa-bf43-10a01e49727e

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}