The sort for CSE size optimization should be different from exec

[papaslavik]

Should the compare functions be different for the size and execution time optimization?

/cc @briansull @mikedn

[briansull]

Yes, your changes are correct. This will change the order that we examine CSE candidates
when (CodeOptKind() == Compiler::SMALL_CODE) is true.
Which would be in places like static class constructors.

I would like to examine what the impact of this change is on Code size before we commit this.

[briansull]

@russellhadley We should see what the Asm diffs are for this commit

[russellhadley]

https://github.com/dotnet/jitutils has a set of tools that will let you run and analyze the codesize and show the diffs

[papaslavik]

@briansull By the way, I tested on ARM that switching SMALL_CODE to everything while crossgen doesn't affect the size of System.Private.CoreLib.ni.dll

@russellhadley thanks for the link

[papaslavik]

@russellhadley are there also any tests which would detect the difference?

[papaslavik]

@Dmitri-Botcharnikov

[briansull]

I took a look at this and also ran ASM diffs and am satisfied that that this change is fine.

[papaslavik]

Thank you @briansull !

Were there any actual diffs and could you share which tests you tried or at least point to the right direction? Just for a future memory consumption research...