[llvm] [LLVM][AArch64] Correctly lower funnel shifts by zero. (PR #140058)

[Paul Walker via llvm-commits]

paulwalker-arm wrote:

Updated to normalise the shift amounts for both `fshl` and `fshr`, with new tests that trigger isel failures prior to this PR.

I did try marking `ISD::FSHL` as needing expansion, but that triggered other asserts and worse code generation so I'd rather get the bug fix out of the way and revisit whether the custom lowering code can be removed at a later date.

https://github.com/llvm/llvm-project/pull/140058