On Feb 28, 2014, at 11:29 PM, Martin Buchholz <marti...@google.com> wrote:
> Are word-size reads in ByteBuffer actually intrinsified? I could find no
> evidence for that. Perhaps this is an important missing optimization for
> hotspot to make?
>
> I see DirectByteBuffer, but not garden-variety ByteBuffer, using Unsafe.
> share/classes/java/nio/Direct-X-Buffer.java.template
>
Yes, no choice :-) I see it only does it when unaligned access is supported:
private long getLong(long a) {
if (unaligned) {
long x = unsafe.getLong(a);
return (nativeByteOrder ? x : Bits.swap(x));
}
return Bits.getLong(a, bigEndian);
}
I suppose it could detect if the address + position also aligns, but ideally
that should be hoisted out of any loop so the unaligned pre/post bytes are
handled separately to the aligned sequence viewed as longs, which most likely
means explicit code for the comparison use-case.
Now i am in two minds to whether to add ByteBuffer.compareUnsigned or add
Arrays.compareUnsigned.
An explosion of methods on Arrays for all types (plus to/from versions) would
be annoying.
Paul.
