[llvm] [LoopUnroll] Fix freqs for unconditional latches: N>2, uniform (PR #182405)

[Joel E. Denny via llvm-commits]

jdenny-ornl wrote:

> Hi @jdenny-ornl, it looks like this PR introduces an assertion failure that is reliably triggered by hot loops with realistic branch weights.

Thanks for the report.  The problem is that, for some reason, I computed absolute differences instead of relative differences when checking precision.  I will try to land a fix today.

https://github.com/llvm/llvm-project/pull/182405