[llvm] [AArch64][GlobalISel] Fix lowering of i64->f32 itofp. (PR #132703)

[David Green via llvm-commits]

================
@@ -7609,6 +7609,22 @@ LegalizerHelper::lowerU64ToF64BitFloatOps(MachineInstr &MI) {
   return Legalized;
 }
 
+/// i64->fp16 itofp can be lowered to i64->f64,f64->f32,f32->f16. We cannot
+/// convert fpround f64->f16 without double-rounding, so we manually perform the
+/// lowering here where we know it is valid.
+static LegalizerHelper::LegalizeResult
+loweri64tof16ITOFP(MachineInstr &MI, Register Dst, LLT DstTy, Register Src,
----------------
davemgreen wrote:

Yeah I was wondering which was better. It felt more like a lower than a generic narrow. If a target wanted a single fptrunc (because they had an instruction for it), then they wouldn't want it to be split. So they would opt for the narrow. As far as I understand the narrows don't make extra legalization queries, there isn't a way to communicate two different types to them, and the alternative of a target hook doesn't sound the best. Maybe I'm missing something about how this should work.

https://github.com/llvm/llvm-project/pull/132703