Unsafe.{get,put}-X-Unaligned performance (original) (raw)

Vitaly Davidovich vitalyd at gmail.com
Thu Mar 12 22:15:53 UTC 2015

Switches currently don't profile well (if at all) - John can shed more light on that as this came up on the compiler list a few weeks ago.

sent from my phone On Mar 12, 2015 6:06 PM, "Peter Levart" <peter.levart at gmail.com> wrote:

On 03/12/2015 10:04 PM, Peter Levart wrote: ... putLongUnaligned in the style of above getLongUnaligned is more tricky with current code structure. But there may be a middle ground (or a sweet spot): public final void putLongUnaligned(Object o, long offset, long x) { if (((int) offset & 1) == 1) { putLongParts(o, offset, (byte) (x >>> 0), (short) (x >>> 8), (short) (x >>> 24), (short) (x >>> 40), (byte) (x >>> 56)); } else if (((int) offset & 2) == 2) { putLongParts(o, offset, (short)(x >>> 0), (int)(x >>> 16), (short)(x >>> 48)); } else if (((int) offset & 4) == 4) { putLongParts(o, offset, (int)(x >> 0), (int)(x >>> 32)); } else { putLong(o, offset, x); } } ...this has the same number of branches, but less instructions. You also need the following two:

At least on Intel (with -XX:-UseUnalignedAccesses) above code (Unaligned2) is not any faster then your code (Unaligned) according to a JMH random-access test. Neither is the reversal of if/else branches (Unaligned1). Unaligned3 is switch-based variant (just get) and is slowest. Your variant seems to be the fastest by a hair: Benchmark Mode Samples Score Score error Units j.t.UnalignedTest.getLongUnaligned avgt 5 16.375 0.837 ns/op j.t.UnalignedTest.getLongUnaligned1 avgt 5 18.340 0.617 ns/op j.t.UnalignedTest.getLongUnaligned2 avgt 5 16.784 0.969 ns/op j.t.UnalignedTest.getLongUnaligned3 avgt 5 19.634 0.871 ns/op j.t.UnalignedTest.putLongUnaligned avgt 5 15.521 0.589 ns/op j.t.UnalignedTest.putLongUnaligned1 avgt 5 16.676 1.042 ns/op j.t.UnalignedTest.putLongUnaligned2 avgt 5 16.394 3.028 ns/op Regards, Peter Peter -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/attachments/20150312/b5bbb402/attachment.html>

More information about the hotspot-compiler-dev mailing list