RFR(S): 8203215: IdealLoopTree::split_outer_loop leaves phi-nodes with only one input

Nils Eliasson nils.eliasson at oracle.com
Wed May 16 13:07:33 UTC 2018

On 2018-05-16 09:14, Roland Westrelin wrote:
> Hi Nils,
>> Webrev: http://cr.openjdk.java.net/~neliasso/8203215/webrev.01/
> replace_input_of(n,..) calls rehash_node_delayed(n) so:
>   1806       igvn.rehash_node_delayed(old_phi);
> is redundant?

Yes, good observation.

> Also, it seems risky to me to perform an igvn with major_progress
> cleared (as it can cause some premature clean up). So, maybe move:
>   2645   C->clear_major_progress();
> down after the igvn?

How about this?


I stepped through it breaking on major_progress(). No code touches 
major_progress until the newly added igvn.optimize()-call. (Except for 
the stop_early/optimize/return cleanup @2705.)  So this new patch will 
have major_progress the same.

// Nils

> Roland.

More information about the hotspot-compiler-dev mailing list