RFR(XXS): 8214434: Disabling ZOptimizeLoadBarriers hits assert
per.liden at oracle.com
Wed Nov 28 15:18:44 UTC 2018
Looks good to me too.
On 11/28/18 3:14 PM, Erik Osterlund wrote:
> You could also make the asserts conditional so we keep on sanity checking in the common case. I’ll leave the choice to you. Either way, looks good, and dont need snother webrev.
>> On 28 Nov 2018, at 14:54, Nils Eliasson <nils.eliasson at oracle.com> wrote:
>> This patch removes two asserts that was moved here with change "8208601: Introduce native oop barriers in C2 for OopHandle". These asserts are not correct when ZOptimizeLoadBarriers is disabled. The gvn.transform on the Proj-node may return a constant based on type analysis. When ZOptimizeLoadBarriers is enabled - the loadbarrier substitution will happen already at the call to gvn.transform(barrier), so the path with the asserts are not taken.
>> This code is going to change with a future patch of mine so choose just to remove them for now. I will consider adding a test or remove the flag in that change.
>> Bug: https://bugs.openjdk.java.net/browse/JDK-8214434
>> Webrev: http://cr.openjdk.java.net/~neliasso/8214434/webrev.02/
More information about the hotspot-compiler-dev