RFR(XS): 8209684: Intrinsics that assume some input non null should use GraphKit::must_be_not_null()

Vladimir Kozlov vladimir.kozlov at oracle.com
Mon Aug 20 20:35:46 UTC 2018


Looks good to me too. But I am concern about intrinsics with predicate code (AES) which may not expect that there is an 
other exit.

Tobias, please, run whole testing with it.

Thanks,
Vladimir

On 8/20/18 2:20 AM, Tobias Hartmann wrote:
> Hi Roland,
> 
> looks good to me (I think you can remove the double new line at 5813).
> 
> Best regards,
> Tobias
> 
> On 20.08.2018 10:21, Roland Westrelin wrote:
>>
>> http://cr.openjdk.java.net/~roland/8209684/webrev.00/
>>
>> A number of intrinsics omit null checks on inputs that are known to be
>> not null. They should mark these inputs as not null instead with
>> GraphKit::must_be_not_null().
>>
>> This change is from the shenandoah repo where it's required:
>> shenandoah's barrier read a field from the object. If the object is not
>> registered to be not null then this field load can turn on some path
>> into a load from a null object which causes C2 to crash.
>>
>> Roland.
>>


More information about the hotspot-compiler-dev mailing list