Phoebe Wang fd3e7e3a1e [X86] Adding lowerings for vector ISD::LRINT and ISD::LLRINT (#90065) ...

- [V]CVTP[D,S]2DQ supports `f64/f32` -> `i32` conversions that can be
mapped to `llvm.lrint.vNi32.vNf64/32` since SSE2. AVX and AVX512 added
256-bit and 512-bit support;
- VCVTP[D,S]2QQ supports `f64/f32` -> `i64` conversions that can be
mapped to `llvm.l[l]rint.vNi64.vNf64/32` since AVX512DQ. All 128-bit,
256-bit (require AVX512VL) and 512-bit are supported.

2024-05-03 09:31:27 +08:00

26 KiB

Raw Blame History

View Raw