[mlir][vector] Limit vector.mask op to single vector result

[springerm]

Summary

Multiple vector results require multiple masks and passthru values.

Depends On: D141683

[nicolasvasilache]

test plz

[springerm]

In D141786#4055766, @nicolasvasilache wrote:

test plz

We don't have an ops that has multiple vector results. An unregistered op doesn't work here either because it doesn't implement the MaskableOpInterface, so we fail before reaching this check. I would leave this without a test instead of adding a bunch of C++ boilerplate to create a new dummy op.

[dcaballe]

LGTM. This is preserving existing behavior so it should be good to go. Thanks!

mlir/lib/Dialect/Vector/IR/VectorOps.cpp

LogicalResult MaskOp::verify() {
   if (maskableOp->getNumResults() != getNumResults())
     return emitOpError("expects number of results to match maskable operation "
                        "number of results");
 
   if (!llvm::equal(maskableOp->getResultTypes(), getResultTypes()))
     return emitOpError(
         "expects result type to match maskable operation result type");
 
+  if (llvm::count_if(maskableOp->getResultTypes(),
+                     [](Type t) { return t.isa<VectorType>(); }) > 1)
+    return emitOpError("multiple vector results not supported");
+
   // Mask checks.
   Type expectedMaskType = maskableOp.getExpectedMaskType();
   if (getMask().getType() != expectedMaskType)
     return emitOpError("expects a ")
            << expectedMaskType << " mask for the maskable operation";
 
   // Passthru checks.
   Value passthru = getPassthru();