[llvm] [AMDGPU][AsmParser] Do not use predicates for validation of NamedIntOperands. (PR #90251)

[Ivan Kosarev via llvm-commits]

================
@@ -1103,25 +1107,37 @@ let DefaultValue = "0xf" in {
 def DppRowMask : NamedIntOperand<i32, "row_mask">;
 def DppBankMask : NamedIntOperand<i32, "bank_mask">;
 }
-def DppBoundCtrl : NamedIntOperand<i1, "bound_ctrl", 1, "DppBoundCtrl",
-    "[this] (int64_t &BC) -> bool { return convertDppBoundCtrl(BC); }">;
+def DppBoundCtrl : NamedIntOperand<i1, "bound_ctrl"> {
+  let ConvertMethod = "[this] (int64_t &BC) -> bool { return convertDppBoundCtrl(BC); }";
+}
 
 let DecoderMethod = "decodeDpp8FI" in
 def Dpp8FI : NamedIntOperand<i32, "fi", 1, "DppFI">;
 def Dpp16FI : NamedIntOperand<i32, "fi", 1, "DppFI">;
 
 def blgp : CustomOperand<i32, 1, "BLGP">;
-def CBSZ : NamedIntOperand<i32, "cbsz">;
-def ABID : NamedIntOperand<i32, "abid">;
-
+def CBSZ : NamedIntOperand<i32, "cbsz"> {
+  let Validator = "isUInt<3>";
----------------
kosarev wrote:

I wanted to not go that far in this patch and give it some time to settle and see how it works with other kinds of operands, like `NamedBitOperand`, etc. Shouldn't be hard to do that later, if needed?

https://github.com/llvm/llvm-project/pull/90251