RFR: Remove C2 write-barrier from .ad files

[Aleksey Shipilev]

On 07/09/2018 06:36 PM, Roman Kennke wrote:
> Am 09.07.2018 um 18:34 schrieb Aleksey Shipilev:
>> On 07/09/2018 06:07 PM, Roman Kennke wrote:
>>> I propose to get rid of write-barriers in C2 .ad files, and instead
>>> always expand WBs in ideal graph. We've been doing this for quite a
>>> while now by default, with seemingly no ill effects.
>>>
>>> In next step, this will make it easier to move
>>> MacroAssembler::shenandoah_write_barrier() away into gc/shenandoah
>>> subfolder.
>>>
>>> http://cr.openjdk.java.net/~rkennke/remove-c2-wb/webrev.01/
>>
>> That makes sense, looks good. Would it make sense to leave the ShenandoahWriteBarrier matcher with
>> ShouldNotReachHere()? Or are we crashing the compilation reliably (e.g. in release builds?) when
>> Ideal is not matched?
> 
> I'm pretty sure that C2 blows up with 'bad AD file' if it can't match a
> ShenandoahWB. (Had that problem today when I tried traversal+LVB and got
> a ShenandoahEnqueueBarrier that slipped through...)

Okay then. Check with Roland if this is okay.

-Aleksey