RFR(M): 7197327: 40% regression on 8 b41 comp 8 b40 on specjvm2008.mpegaudio on oob

Roland Westrelin roland.westrelin at oracle.com
Tue Jan 29 07:48:51 PST 2013


Here is a new webrev:

http://cr.openjdk.java.net/~roland/7197327/webrev.01/

> What is effect on ref workload?

I haven't done a refworkload on this latest version but on a previous one, refworkload shown nothing of interests except the improvement on specjvm2008's mpegaudio. I'll redo refworkload before the push.

> c1 = get_control(n1);
> c2 = get_control(n2);
> 
> if (is_dominator(c1, c2)) {
>   c2 = c1;
> } else if (is_dominator(c2, c1)) {
>   c1 = c2;

The 2 tests below cannot work because they don't guarantee that c2 is in the other If branch or that c2 is not on the other branch but inside an imbricated If:

> } else if (c1->is_Proj() && c1->in(0)->is_If() &&
>            is_dominator(c1->in(0), c2)) {
>   c1 = c2 = c1->in(0);
> } else if (c2->is_Proj() && c2->in(0)->is_If() &&
>            is_dominator(c2->in(0), c1)) {
>   c1 = c2 = c2->in(0);
> }
> if (n1->in(0) != c1) {
>   n1->set_req(0, c1);
> }
> if (n2->in(0) != c2) {
>   n2->set_req(0, c2);
> }

Roland.


More information about the hotspot-compiler-dev mailing list