[clang] [llvm] [X86][AVX10.2] Support AVX10.2-BF16 new instructions. (PR #101603)

[Freddy Ye via cfe-commits]

================
@@ -14836,6 +14837,9 @@ Value *CodeGenFunction::EmitX86BuiltinExpr(unsigned BuiltinID,
   case X86::BI__builtin_ia32_vfmaddph512_mask:
   case X86::BI__builtin_ia32_vfmaddph512_maskz:
   case X86::BI__builtin_ia32_vfmaddph512_mask3:
+  case X86::BI__builtin_ia32_vfmaddnepbh128:
+  case X86::BI__builtin_ia32_vfmaddnepbh256:
+  case X86::BI__builtin_ia32_vfmaddnepbh512:
----------------
FreddyLeaf wrote:

this implementation didn't define them as bf16 fma instructions don't have rounding control, safe to implement by select.

https://github.com/llvm/llvm-project/pull/101603