RFR [8]: Fix x86_32 build

[Aleksey Shipilev]

On 07/18/2018 12:08 PM, Roman Kennke wrote:
> Am 18.07.2018 um 11:43 schrieb Aleksey Shipilev:
>> sh/jdk8u build on x86_32 fails because there is no Atomic::load for intptr_t in jdk8:
>>
>> diff -r 3b776ba499bb src/share/vm/gc_implementation/shenandoah/shenandoahPacer.cpp
>> --- a/src/share/vm/gc_implementation/shenandoah/shenandoahPacer.cpp     Mon Jul 16 11:57:03 2018 -0400
>> +++ b/src/share/vm/gc_implementation/shenandoah/shenandoahPacer.cpp     Wed Jul 18 11:43:11 2018 +0200
>> @@ -210,11 +210,11 @@
>>    intptr_t tax = MAX2<intptr_t>(1, (intptr_t)(words * OrderAccess::load_acquire(&_tax_rate)));
>>    Atomic::add(tax, &_budget);
>>  }
>>
>>  intptr_t ShenandoahPacer::epoch() {
>> -  return Atomic::load(&_epoch);
>> +  return OrderAccess::load_acquire(&_epoch);
>>  }
>>
>>  void ShenandoahPacer::pace_for_alloc(size_t words) {
>>    assert(ShenandoahPacing, "Only be here when pacing is enabled");
>>
>>
>> Testing: x86_32 build, tier1_gc_shenandoah
> 
> Ok. Ordinary load doesn't do it? intptr_t on x86_32 is 32bit and atomic,
> or is it not?

It is shared between threads, so it is better to be somewhat recent. Other Pacer variables like that
are using release/acquire already.

-Aleksey