RFR (S) CR 8014886: @Contended fields can overrun oop maps

[Aleksey Shipilev]

Thanks Vladimir!

The updated webrev is here:
  http://cr.openjdk.java.net/~shade/8014886/webrev.02/

On 05/21/2013 07:58 AM, Vladimir Kozlov wrote:
> Is not has_nonstatic_fields also wrong?

Yes, it is. I haven't fixed this here for three reasons:
 a) this is not about oop maps
 b) tracked down the impact through the HS code, seems minimal
 c) because of (b): I'm still trying to get proper regression test for
has_nonstatic_fields ready

I will post another CR for fixing that one up.

> Rename: max_oop_maps -> max_nonstatic_oop_maps.

Done.

> For FieldsAllocationStyle=2 (default) we do special oop fields layout
> for super and subclass to have only one oopmap. You need to add non-oop
> fields (class in your test has only oop fields) to test different layouts.

Done. I had also the super-class to get the oops aligned in the oop chunk.

> Next added mess to already complicated code:
> 
> 3188   unsigned int nonstatic_double_count =
> fac->count[NONSTATIC_DOUBLE] - fac_contended.count[NONSTATIC_DOUBLE];
> 3189   unsigned int nonstatic_word_count   = fac->count[NONSTATIC_WORD]
>   - fac_contended.count[NONSTATIC_WORD];
> 3190   unsigned int nonstatic_short_count  = fac->count[NONSTATIC_SHORT]
>  - fac_contended.count[NONSTATIC_SHORT];
> 3191   unsigned int nonstatic_byte_count   = fac->count[NONSTATIC_BYTE]
>   - fac_contended.count[NONSTATIC_BYTE];
> 3192   unsigned int nonstatic_oop_count    = fac->count[NONSTATIC_OOP]  
> - fac_contended.count[NONSTATIC_OOP];

> At least add comment that these counts are for not contended fields.

Already taken care of in the baseline: there is a terse comment. I will
clean up that mess during further cleanup work. The regression tests
like this one will help to build enough safety net for the refactoring.

-Aleksey.