RFR(M): 8210260: [lworld] Improve acmp implementation
john.r.rose at oracle.com
Fri Sep 7 18:29:21 UTC 2018
On Sep 3, 2018, at 4:16 AM, Tobias Hartmann <tobias.hartmann at oracle.com> wrote:
> Here are the performance numbers that Sergey provided (lower is better):
> http://cr.openjdk.java.net/~skuksenko/valhalla/acmp1/acmp_charts_0831.png <http://cr.openjdk.java.net/~skuksenko/valhalla/acmp1/acmp_charts_0831.png>
> The red line is the performance with this patch. Please note that we are very close to baseline
> performance for the cases with high rates of inequality (which should be common).
Bravo! The net of Sergey's numbers is that Valhalla's value-aware acmp
should not incur detectable performance costs, except perhaps in some program
that spends a large fraction of its time repeatedly verifying that equal references
are equal. That sounds like something we can live with.
More information about the valhalla-dev