As seen in the Templated test from JDK-8378968, we don't Idealize the 64-lane byte case, because the AndL node folds away early, and so the pattern looks different now.
- relates to
-
JDK-8378968 C2 VectorAVX AVX2: wrong result, bad optimization in VectorLongToMaskNode::Ideal
-
- Open
-
-
JDK-8277997 Intrinsic creation for VectorMask.fromLong API
-
- Resolved
-