RFR (Tedious) 8204301: Make OrderAccess functions available to hpp rather than inline.hpp files

[coleen.phillimore at oracle.com]

On 6/5/18 8:45 PM, David Holmes wrote:
> Hi Coleen,
>
> On 6/06/2018 7:19 AM, coleen.phillimore at oracle.com wrote:
>> Summary: move orderAccess.inline.hpp into orderAccess.hpp and remove 
>> os.hpp inclusion and conditional os::is_MP() for fence on x86 platforms
>>
>> See discussion in bug.  Left os::is_MP() conditional for arm32. 
>> Tested by Boris U, thanks!
>>
>> open webrev at http://cr.openjdk.java.net/~coleenp/8204301.01/webrev
>> bug link https://bugs.openjdk.java.net/browse/JDK-8204301
>
> That all looks fine to me.
>
> Only observation I have is that I think the compiler_barrier() calls 
> in the x86 fence routines (except perhaps Windows) are redundant. I 
> think the original code should have been:
>
> if (os::is_MP()) {
>   __ asm lock add ...
> else
>   compiler_barrier();
> }
>
> but this is harmless and has no runtime impact.

I see.  I'll leave them so I don't have to generate another webrev.

Thanks for the review and discussion!
Coleen
>
> Thanks,
> David
>
>> Tested on mach5 hs-tier1-2 on Oracle platforms: linux-x64, 
>> windows-x64, macosx-x64 and solaris-sparcv9.  Built on linux-aarch64 
>> and linux-zero.   Boris built on arm32.  There were no actual changes 
>> on s390 or ppc.
>>
>> Thanks,
>> Coleen