Request for review (M): 6860469: remix_address_expressions sets incorrect control causing crash in split_if_with_block_post
Thomas.Rodriguez at Sun.COM
Wed Jul 15 13:37:23 PDT 2009
http://cr.openjdk.java.net/~never/6860469 is what I'm going to push
with you and hiroshi marked as contributors. I added the copyright
and fixed the test case run running with jtreg. The class needed to
be public and the class name was missing from the @run line.
On Jul 15, 2009, at 12:24 PM, Chuck Rasbold wrote:
> I presume that Sun will add the copyright notice when the code is
> committed, and of course, I'm OK with that.
> IANAL, but I'm reluctant to add the notice in advance, mostly
> because I am not employed by Sun.
> On Wed, Jul 15, 2009 at 12:07 PM, Tom Rodriguez <Thomas.Rodriguez at sun.com
> > wrote:
> The test is missing the copyright notice but otherwise this looks
> good. BTW, I ran a full ctw with this change both as is and as an
> assert that checked for cases where this would produce a different
> control and I only found 4 cases where it ever changed the answer
> and none of those triggered a failure in the way the test case
> does. Hiroshi, thanks for tracking this down and Chuck, thanks for
> distilling a test case.
> On Jul 15, 2009, at 9:27 AM, Chuck Rasbold wrote:
> Fixed 6860469: remix_address_expressions sets incorrect control
> causing crash in split_if_with_block_post
> Consult the control node of both inputs when deciding where to place
> new LShiftI node.
> Previously, we assumed the invar input should set the new node's
> control, and the scale input was irrelevant for control's sake.
> However, sometimes the invar input is a constant, allowing it to be
> higher in the dom tree than the scale.
> Fix provided by Hiroshi Yamauchi (yamauchi at google.com)
More information about the hotspot-compiler-dev