[llvm] [X86][AVX10.2] Support AVX10.2-COMEF new instructions. (PR #108063)

[Phoebe Wang via llvm-commits]

================
@@ -1225,3 +1225,49 @@ defm VFNMADD132NEPBF16 : avx10_fma3p_132_bf16<0x9C, "vfnmadd132nepbf16", X86any_
 defm VFNMSUB132NEPBF16 : avx10_fma3p_132_bf16<0x9E, "vfnmsub132nepbf16", X86any_Fnmsub,
                                               X86Fnmsub, SchedWriteFMA>;
 }
+
+//-------------------------------------------------
+// AVX10  COMEF instructions
+//-------------------------------------------------
+multiclass avx10_com_ef_int<bits<8> Opc, X86VectorVTInfo _, SDNode OpNode,
+                             string OpcodeStr,
+                             Domain d,
+                             X86FoldableSchedWrite sched = WriteFComX> {
+  let ExeDomain = d, mayRaiseFPException = 1 in {
+    def rr_Int : AVX512<Opc, MRMSrcReg, (outs), (ins _.RC:$src1, _.RC:$src2),
----------------
phoebewang wrote:

We also need the `rr` and `rm` definations for optimization like `https://godbolt.org/z/d3sGofdTs`, but I think we can leave it for a second patch. Just add a TODO is ok to me.

https://github.com/llvm/llvm-project/pull/108063