Request for reviews (S): 6730276: JDI_REGRESSION tests fail with "Error: count must be non-zero" error on x86

[Vladimir Kozlov]

I modified assembler code to check for 0 count and renamed conjoint_bytes to conjoint_jbytes. And I added verification code for these routines.

Updated webrev:

http://cr.openjdk.java.net/~kvn/6730276/webrev.01

Vladimir

Vladimir Kozlov wrote:
> Yes, some methods have zero count check but some does not.
> And I don't want to relay on current code in copy.hpp.
> It seems I have to go through all implementations, verify
> zero count check and add comments.
> 
> Which are most used copy methods?
> I will look that they will not have duplicated checks.
> 
> Thanks,
> Vladimir
> 
> John Coomes wrote:
>> Paul Hohensee (paul.hohensee at oracle.com) wrote:
>>> Actually, gc copying is done by pd_disjoint_words, which checks for a
>>> zero length already.
>>
>> Vladimir is adding an extra check to that path, in
>> Copy::disjoint_words* :-(.
>>
>> -John
>>
>>> On 6/9/10 3:13 PM, John Coomes wrote:
>>>> Vladimir Kozlov (vladimir.kozlov at oracle.com) wrote:
>>>>   
>>>>> http://cr.openjdk.java.net/~kvn/6730276/webrev
>>>>>
>>>>> Fixed 6730276: JDI_REGRESSION tests fail with "Error: count must be 
>>>>> non-zero" error on x86
>>>>>
>>>>> Replace assert_non_zero with if (count == 0) return.
>>>>>      
>>>> The code change is fine, as long as there's no performance hit in GC.
>>>> The heaviest user of these routines is GC, and we call only when
>>>> there's something to copy.
>>>>
>>>> -John
>>>>
>>>>    
>>