RFR(S) 8016157: During CTW: C2: assert(!def_outside->member(r)) failed: Use of external LRG overlaps the same LRG defined in this block

[Vladimir Kozlov]

Good.

It looks like a 'band-aid' but I think it is fine for now until Niclas 
clean up the mess in this code.
Note, the fix is general since we observed the failure on SPARC and x86.

Thanks,
Vladimir

On 6/10/13 1:12 AM, Niclas Adlertz wrote:
> Hi all,
>
> In bug 8005956 (https://jbs.oracle.com/bugs/browse/JDK-8005956) this assert triggered when trying to rematerialize reaches at phi nodes. This time it's failing because of rematerialization at uses (reg_split.cpp: 931).  These are the only two places that rematerialization is handled during splitting.
>
> Doing the same fix as for JDK-8005956, i.e. removing rematerialization for nodes with live range input, gave too much performance regression. Instead, this fix invalidates rematerialization of the node type that is causing the problem negD node.
>
> WEBREV: http://cr.openjdk.java.net/~adlertz/JDK-8016157/webrev00/
> JBS: https://jbs.oracle.com/bugs/browse/JDK-8016157
>
> Kind Regards,
> Niclas Adlertz
>
>