[llvm] [AArch64] Only fold into ands directly if AND is one-use (PR #167371)

[David Green via llvm-commits]

davemgreen wrote:

This looks like two patches - one for GISel and one for SDAG. I'm not sure we shouldn't be doing _more_ optimization in instruction selection, not less, but they can have upsides and downsides. GISel is still missing a lot of combines, and many of the changes seem to be on the GISel side. I ran some benchmarks for sdag and the perf was worse (but only a little, probably noise) and codesize was larger overall (although not a lot).

https://github.com/llvm/llvm-project/pull/167371