Request for reviews (S): 7006044: materialize cheap non-oop pointers on 64-bit SPARC

Vladimir Kozlov vladimir.kozlov at oracle.com
Tue Dec 14 12:01:26 PST 2010


Looks good.

Vladimir

Christian Thalinger wrote:
> http://cr.openjdk.java.net/~twisti/7006044/webrev.01/
> 
> 7006044: materialize cheap non-oop pointers on 64-bit SPARC
> Summary: After 6961690 we load non-oop pointers for the constant table which could easily be materialized in a few instructions.
> Reviewed-by:
> 
> After 6961690 we load non-oop pointers for the constant table which
> could easily be materialized in a few instructions.  This happens
> commonly with the init of the header word in an allocation.
> 
> Tested with runThese, DaCapo and CTW on a T2+.
> 
> For a whole CTW run 165k non-oop loads were emitted, of which 47%
> could be materialized.
> 
> Additionally I fixed a bug in MacroAssembler::size_of_sethi.  The
> worst case returned the number of instructions while all other returns
> returned the number of bytes of code used by the instructions.  I
> changed all size_of_* functions to return the number of instructions
> and to better reflect what these functions return I renamed them to
> insts_for_*.


More information about the hotspot-compiler-dev mailing list