llvm-project

9bfb3be8 - [AArch64][GlobalISel] Added support for neon left shift intrinsics on single-element vector types (#170827)

[AArch64][GlobalISel] Added support for neon left shift intrinsics on single-element vector types (#170827) Previously, the left shift family of intrinsics would fail to lower for <1 x i64> / <1 x i32> vector types, as IRTranslation lowers these to scalars. Marking these intrinsics as only using fp operands ensures the RegBankSelect phase places the vector on an fpr bank, instead of a gpr bank.