[aarch64-port-dev ] RFR: JDK-8185786: AArch64: disable some address reshapings.

[Zhongwei Yao]

Hi, all,

Is this patch OK for JDK10?

On 16 August 2017 at 14:51, Zhongwei Yao <zhongwei.yao at linaro.org> wrote:
> Thanks!
>
> Done:
> http://cr.openjdk.java.net/~njian/8185786/webrev.02/
>
> On 15 August 2017 at 21:50, Andrew Haley <aph at redhat.com> wrote:
>> On 14/08/17 22:55, Zhongwei Yao wrote:
>>> Yeah, it will do the same work. As line 4 has already ensured 'u' is
>>> MemNode, which is required at line 9. But such implication is easy to
>>> be ignored, so I make the "u->is_Mem()" checking to be explicit. What
>>> do you think?
>>
>> So do this:
>>
>>       // Any use that can't embed the address computation?
>>        for (DUIterator_Fast imax, i = addp->fast_outs(imax); i < imax; i++) {
>>          Node* u = addp->fast_out(i);
>>          if (! u->is_Mem())
>>            return;
>>          if (u->is_LoadVector() || u->is_StoreVector() || u->Opcode() == Op_StoreCM) {
>>            return;
>>          }
>>          if (addp2->in(AddPNode::Offset)->Opcode() != Op_ConvI2L) {
>>            int scale = 1 << addp2->in(AddPNode::Offset)->in(2)->get_int();
>>            if (VM_Version::expensive_load(u->as_Mem()->memory_size(), scale)) {
>>              return;
>>            }
>>          }
>>        }
>>
>> --
>> Andrew Haley
>> Java Platform Lead Engineer
>> Red Hat UK Ltd. <https://www.redhat.com>
>> EAC8 43EB D3EF DB98 CC77 2FAD A5CD 6035 332F A671
>
>
>
> --
> Best regards,
> Zhongwei



-- 
Best regards,
Zhongwei