[llvm] [AArch64] Lower aarch64.neon.fcvtzs.i16.f16 to FP_TO_SINT_SAT (PR #154344)

[Eli Friedman via llvm-commits]

efriedma-quic wrote:

> Adding a pseudo-op in AArch64InstrInfo.td to handle these conversions then expanding that pseudo into the 2 instructions

That's one way.

To make what's happening a bit more transparent to other optimizations, probably better to make a SelectionDAG node AArch64ISD::FCVTZS_HALF that returns an f32 containing the integer bit-pattern, BITCAST that to i32, the truncate to i16.

https://github.com/llvm/llvm-project/pull/154344