x86: Cache supportsAVX() checks in binary encoding

[BradleyWood]

A profile discovered significant compute time spent running these checks. I will add performance numbers shortly.

[BradleyWood]

@dsouzai Can you look at this PR from the perspective of CRIU and JIT-server?

FYI, @0xdaryl @vijaysun-omr

[dsouzai]

These checks were originally added because OpenJ9 moved from using TR::CodeGenerator::getX86ProcessorInfo() to using the port library, and so these were sanity checks to ensure that the two sources of information agreed with each other.

Given that it's been a few years, do we even need these checks? @mpirvu what do you think?

[BradleyWood]

Given that it's been a few years, do we even need these checks?

I am working on removing the old api in favour of the portlibrary.

[dsouzai]

The main issue here is that these macros are OpenJ9 specific, so it's going to pollute the OMR code unnecessarily. Additionally, even outside of JITServer and CRIU, comp->target() can be different in a normal and relocatable compilation.

I think some of the answers to these queries should be cached in the codegen object, so that it is only initialized once per compilation.