Below code snippet is in test/micro/org/openjdk/bench/jdk/incubator/vector/MaskFromLongBenchmark.java

    @Benchmark
    public int microMaskFromLong_Byte64() {
        VectorMask mask = VectorMask.fromLong(ByteVector.SPECIES_64, val);
        return mask.laneIsSet(1) ? 1 : 0;
    }

After inlining laneIsSet, the VectorMaskToLong-VectorLongToMask pair would be optimized to nothing, thus this benchmark is actually to be zero cost.

Implementation of laneIsSet:
    @Override
    @ForceInline
    public boolean laneIsSet(int i) {
        int length = length();
        Objects.checkIndex(i, length);
        if (length <= Long.SIZE) {
            return ((toLong() >>> i) & 1L) == 1;
        } else {
            return getBits()[i];
        }
    }