Perf: typed int **106 / pow(n,106) chained smul omit llvm.pow.f64 (#36386)

[PurHur]

Summary

• Part of Performance: v2.0 targets and the codegen roadmap to reach them — typed code ≥ 3x Zend, untyped ≤ 1.5x slower, web request ≥ php-fpm: inline caches for property/method access, escape analysis for arrays/strings, guarded speculation, per-function O1 by default (lib/JIT.php, lib/JIT/Builtin/Type/Object_.php, lib/JIT/Analyzer.php) #36386 (typed fast paths). Lane overflow: CI: php-llvm-structgep-assert.patch fails to apply — blocks north-star5-verify-fast and all bootstrap gates #36143–VM: 1M-iteration loops exit 255 silently — restore compute-suite headroom on bin/vm.php #36148 / RunForge north star (Sep 2026): trust, build speed, performance — automation tracker #36142 children closed; #36386 % 3 == 2 is lane C.
• Lower compile-time exponent **106 / pow(n,106) via (hundredfifth×n) chained smul.with.overflow in JitPowIntegerEmitHigh — omit llvm.pow.f64. Even: (-1)**106 === 1. Append-only (no new spine TU).
• No new C ABI. php-src: Zend/zend_operators.c pow_function / zend_pow / mul_function; ext/standard/math.c PHP_FUNCTION(pow).

Part of #36386.

Test plan

• vendor/bin/phpunit --filter 'NoThrowPowHundred(fifth|sixth)MulAotTest' → 8/8 (46 assertions)
• Zend vs AOT on test/differential/cases/pow_hundredsixth_smul_36386.php → identical (0/1/1/0 + doubles); IR has 0 llvm.pow.f64, 203 smul.with.overflow
• ./script/aot-smoke.sh → 9/9; size gate OK
• php script/check-size-budgets.php → OK (157)
• make north-star5-verify-fast → OK

Made with Cursor

[PurHur]

Superseded by #37703 (same **106 slice already on master). Closing duplicate.