[Mlir-commits] [mlir] [MLIR][XeGPU] Add truncf and dpas_mx ops (PR #180059)

[Sang Ik Lee]

================
@@ -1187,6 +1187,32 @@ LogicalResult StoreMatrixOp::verify() {
                                getLayoutAttr(), [&]() { return emitError(); });
 }
 
+//===----------------------------------------------------------------------===//
+// XeGPU_TruncfOp
+//===----------------------------------------------------------------------===//
+
+LogicalResult TruncfOp::verify() {
+  auto sourceVecType = dyn_cast<VectorType>(getSource().getType());
+  auto resultVecType = dyn_cast<VectorType>(getResult().getType());
+
+  if (sourceVecType.getElementTypeBitWidth() <=
+      resultVecType.getElementTypeBitWidth())
+    return emitOpError("input type must be wider than result type.");
+
+  return success();
+}
+
+//===----------------------------------------------------------------------===//
+// XeGPU_DpasMxOp
+//===----------------------------------------------------------------------===//
+
+LogicalResult DpasMxOp::verify() {
----------------
silee2 wrote:

There is a constraint set for A and B in XeGPUOps.td
`AllElementTypesMatch<["a", "b"]>`

https://github.com/llvm/llvm-project/pull/180059