LRB midpath code quality

[Roman Kennke]

In case you want to see or experiment with it:

http://cr.openjdk.java.net/~rkennke/streamline-lrb-midpath/streamline-lrb-midpath.patch

hotspot_gc_shenandoah seems fine without the cloned null-check. Will
also run all of specjvm before RFRing.

Roland, WDYT?

Roman

> On 3/6/19 10:34 AM, Roman Kennke wrote:
>>>>> Well yes, an implicit null-check would be good. It would never fire,
>>>>> because if the value is null, it would blow up earlier in the cset
>>>>> check. Or am I missing something?
>>>>
>>>> I suppose PhaseCFG::implicit_null_check() doesn't recognize the barrier
>>>> load as a candidate for implicit null checks for some reason.
>>>
>>> Yes, maybe.
>>>
>>> But can you explain to me why it should be needed at all? If a
>>> dominating (implicit?) null-check before the in-cset-check should catch
>>> all nulls on that path already? Why does it matter?
>>
>> Also, I don't think we want to generate an NPE/SEGV on that path. We
>> want to skip a bunch of stuff if the value is NULL.
> 
> We might just back-branch to the access, no? And let NPE be handled there?
> 
> In here:
>   https://paste.fedoraproject.org/paste/nMZb19apbJq0uOB9CboUOA
> 
> This branch:
> 
> 0x00007f46f04b9272: test   %rax,%rax
> 0x00007f46f04b9275: je     0x00007f46f04b9294
> 
> ...can just go to where every other branch is going:
> 
> 0x00007f46f04b9247: movl   $0x2a,0x20(%rax)
>    ; implicit exception: dispatches to 0x00007f46f04b9294
> 
> -Aleksey
>