[Mlir-commits] [mlir] [MLIR][NVVM] Update convert Ops to use builtin types (PR #159704)

[Durgadoss R]

================
@@ -189,6 +189,18 @@ LogicalResult ConvertFloatToTF32Op::verify() {
   return success();
 }
 
+LogicalResult ConvertF32x2ToF6x2Op::verify() {
+  mlir::MLIRContext *ctx = getContext();
+
+  if (!llvm::isa<mlir::Float6E2M3FNType, mlir::Float6E3M2FNType>(getDstTy())) {
+    return emitOpError("Only ")
+           << mlir::Float6E2M3FNType::get(ctx) << " and "
+           << mlir::Float6E3M2FNType::get(ctx)
+           << " types are supported for conversions from f32x2 to f6x2.";
----------------
durga4github wrote:

[Not for this PR]:
I am wondering if there is a way to specify this type constraint in the Op itself (in tablegen).
That way, the default case checks can all happen within the td file.


https://github.com/llvm/llvm-project/pull/159704