RFR: 8143219: AArch64 broken by 8141132: JEP 254: Compact Strings
tobias.hartmann at oracle.com
Thu Nov 19 16:48:07 UTC 2015
On 19.11.2015 16:13, Andrew Haley wrote:
> On 11/19/2015 03:08 PM, Tobias Hartmann wrote:
>> On 19.11.2015 12:56, Andrew Haley wrote:
>>> Two things happened, one simple and one not-so-simple.
>>> Firstly, the string comparators now take a byte count (because it's a
>>> byte array now) rather than a char count.
>> Looks good.
>>> Secondly, string_indexOf needs to have a guard to make sure that
>>> substr.count <= string.count. This is guaranteed by
>>> LibraryCallKit::inline_string_indexOf but not by
>>> LibraryCallKit::inline_string_indexOfI. This is a subtle change, and
>>> was quite tricky to figure out. I have fixed it here in the AArch64
>>> back end.
>> I think this check should be done in inline_string_indexOfI()
>> because other intrinsics may also depend on it (see
>> macroAssembler_x86.cpp). Same applies to the "substr count == 0"
>> check. One could probably factor out the checks and use them for
>> both inline_string_indexOf* methods.
>> I'm moving some more runtime checks from the Java code into the
>> intrinsics (see JDK-8142303 which is currently out for review on
>> If you want, I can take care of these checks as well.
> OK, that would be cool. In the meantime it's a bit difficult for
> AArch64 because nothing works without these fixes. I wonder if I
> should push this workaround and back it out later.
I've sent an updated version of my fix for JDK-8142303. Maybe you could verify that this solves the problem.
More information about the hotspot-dev