accumulate hackc Profiles while computing crc
Summary:
I've added `codegen_bytes` to hackc's `compile::Profile` so we can measure
cumulative allocation in hackc's emitter arena.
In order to collect profiles, we need to run the compiler with a NativeEnv.
This changes the way default options are computed and introduces unwanted
changes to the bytecode. After a lot of trial and error I found the right combination
of options that matches what you get with a None NativeEnv.
Also, when attempting to run bc_compare.py on ~/www/flib, I see stack overflows
even though it uses elastic stack. I gave up debugging and hardcoded 32MB
stack for rayon workers in crc.rs.
Reviewed By: aorenste
Differential Revision:
D34241123
fbshipit-source-id:
81f8a0002486b11319a2a3659691c5eff452f64b