[llvm] [RISCV] Add branch folding before branch relaxation (PR #134760)

[Philip Reames via llvm-commits]

preames wrote:

> > There is a considerably smaller code size change, but it's still there.
> 
> It's increasing code size right?

Not that I see?  But honestly, I'm rather confused by the LNT numbers.  I'd been assuming baseline was some random old commit, and that the run-over-run was the set to look at, but now I'm noticing that the baseline doesn't seem to correspond to any LLVM commit, and was run "04/21/2025 04:47:47" which doesn't fit my theory of what this was telling me.

Hm, I now see I misread Mikhail's previous comment.  My prior comment was me misreading and looking at the dynamic icount numbers and interpreting them as code size.  (And also assuming DirA is baseline.)  

So at the moment, we have a very slight evidence of dynamic icount improvement, and I can't make heads or tails of anything beyond that.  

Honestly, I'm getting pretty frustrated with the data quality here.

https://github.com/llvm/llvm-project/pull/134760