AVX_Revectorize_v11.5.146

Command line guidance to enable AVX-256 revectorization in V8.

• Packages:

1. x64-revec.release
   $ unzip -q x64-revec_release.zip
   Release version for performance test.

2. x64-revec.optdebug
   $ unzip -q x64-revec_release.zip
   Optimized version with debug check enabled.

• Performance test

1. Default:
   $ d8 --experimental-wasm-revectorize

2. Turbofan only (Skip lazy compilation and liftoff):
   $ d8 --experimental-wasm-revectorize --no-liftoff --no-wasm_lazy_compilation

[Note] By default v8 enables lazy compilation and liftoff baseline compilation before tiering up to turbofan with advanced optimization. The AVX revectorization phase is enbled in Turbofan only. If the test only runs few times, it may not get the chance to enter Turbofan phase and get optimized to AVX-256.

• Trace and debug issues:
  $ d8 --experimental-wasm-revectorize --trace-wasm-revectorize --no-liftoff --no-wasm-lazy-compilation --wasm-num-compilation-tasks=1 > run.log

To debug, please send us the trace and let us know the function #id that does not get revectorized as expected.