RFR(M): 8200303: C2 should leverage profiling for lookupswitch/tableswitch

Vladimir Kozlov vladimir.kozlov at oracle.com
Mon Apr 9 16:28:46 UTC 2018


I included jdk tier1 testing in this round and one test 
java/lang/Character/CheckScript.java failed. It passed with 


On 4/9/18 7:42 AM, Vladimir Kozlov wrote:
> And since everything passed we can push webrev.02.
> Let me re-test it with current jdk/hs.
> Thanks,
> Vladimir
> On 4/9/18 7:40 AM, Vladimir Kozlov wrote:
>> I did say:
>>  > Please, file a bug.
>> I never asked to fix it in current changes. We may never fix it due to 
>> limited resources :( but we will have record. It is definitely not 
>> high priority.
>> Thanks,
>> Vladimir
>> On 4/9/18 4:02 AM, Roland Westrelin wrote:
>>>> But in your case I thouight just add MachJumpNode class and its supers
>>>> MachConstantNode and MachTypeNode.
>>> MachJumpNode in the SA would need code to read the probabilities from
>>> the MachJumpNode in the VM process (that is, read the array address and
>>> then each elements, one per existing projection of the
>>> MachJumpNode). That doesn't sound straightforward (and high priority
>>> either). I'd like to file a bug for that and take care of it later. What
>>> do you think?
>>> Roland.

More information about the hotspot-compiler-dev mailing list