RFR 8208163 [lworld] performance regression on String(char[]) constructor

[Ioi Lam]

Hi Tobias,

I was finally able to get back on the Valhalla work. I've updated the 
webrev according to your suggestions. What do you think?

http://cr.openjdk.java.net/~iklam/valhalla/8208163_string_charary_ctor_slowdown.v02/

Thanks

- Ioi


On 8/14/18 7:28 AM, Tobias Hartmann wrote:
> Hi Ioi,
>
> On 10.08.2018 23:37, Ioi Lam wrote:
>> https://bugs.openjdk.java.net/browse/JDK-8208163
>> http://cr.openjdk.java.net/~iklam/valhalla/8208163_string_charary_ctor_slowdown.v01/
> It seems that with your change, we now mark all non-array allocations as "unknown value", right?
> This will cause PhaseMacroExpand::expand_allocate_common() to add additional runtime checks. It
> might be better to check for primitive arrays (TypeArrayKlass) explicitly.
>
> Also, your comment is a bit misleading because it suggests that the set_unknown_value() code is only
> used for reflective array creation. However, that code should be invoked whenever we create an array
> and don't know if the element type is a value type.
>
> To improve readability, I think it would also be good to wrap the entire code starting from line
> 3972 into 'if (EnableValhalla)' and not only the else branch.
>
> Thanks,
> Tobias