[llvm] AMDGPU: VALU data fast-forwarding needs no s_delay (PR #205481)

[Jay Foad via llvm-commits]

https://github.com/jayfoad commented:

Just wanted to highlight inline some of the changes due to not ignoring implicit operands. Some of these are good and a few are not. Some we might need to check with hardware experts.

I don't think these need to block the patch, but it would be good to at least record some of the unwanted cases with TODO comments or similar.

https://github.com/llvm/llvm-project/pull/205481