[llvm] [X86] Lower bf16 round-to-integer on AVX10.2 via vrndscalebf16 (PR #213276)

[Phoebe Wang via llvm-commits]

================
@@ -2635,6 +2639,10 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
     setOperationAction(ISD::FMA, MVT::v32bf16, Legal);
     setOperationAction(ISD::SETCC, MVT::v32bf16, Custom);
     SetFPMinMaxAction(MVT::v32bf16);
+    for (auto VT : {MVT::v8bf16, MVT::v16bf16, MVT::v32bf16})
----------------
phoebewang wrote:

v32bf16 requires `Subtarget.useAVX512Regs()`

https://github.com/llvm/llvm-project/pull/213276