D29862: LSR: an alternative way to resolve complex solution

[David Green via llvm-commits]

Hello,

Firstly, thanks for doing this work and looking into the regressions. My comment about changing the default was made in haste before I had read the llvm-dev post, I'll leave that kind of thing to you.

I ran the new patch against some downstream benchmarks and, so long as I am looking at this correctly, the results look pretty positive. These were not full performance runs, but I ran some quick runs on M23 and M7's, and there were some good improvements in the places we previously saw regressions. I'm not sure about the hmmer benchmarks, but I would like to see the results on a full run when the patch makes it in.

Cheers
Dave