review (S) for 6953576: bottom_type for matched AddPNodes doesn't always agree with ideal
tom.rodriguez at oracle.com
Tue May 18 15:14:59 PDT 2010
We talked about it but I never officially sent it out for review. It's unchanged from what you saw earlier. Thanks!
On May 18, 2010, at 2:59 PM, Vladimir Kozlov wrote:
> Did not I review it already?
> Anyway it looks good.
> Tom Rodriguez wrote:
>> 6953576: bottom_type for matched AddPNodes doesn't always agree with ideal
>> 6715633 added an assert to make sure that the adr_type of nodes
>> doesn't change during matching. If the address pieces are matched
>> separately this assert can fail because the logic in
>> AddPNode::mach_bottom_type isn't the equivalent to
>> AddPNode::bottom_type. The fix is to simply capture the bottom_type
>> when transforming the nodes by making AddP matches inherit from
>> MachTypeNode. This was originally discovered in the ia64 port but I
>> reproduced it by disabling the use of complex addressing modes to
>> force the matching AddPs separately.
More information about the hotspot-compiler-dev