[Cbe-oss-dev] [PATCH 5/6] spufs: add placement computation for scheduling of affinity contexts

André Detsch adetsch at br.ibm.com
Tue Feb 13 07:46:12 EST 2007


Arnd Bergmann wrote:
>> --- linux-2.6.20.orig/arch/powerpc/platforms/cell/spufs/spufs.h
>> +++ linux-2.6.20/arch/powerpc/platforms/cell/spufs/spufs.h
>> @@ -85,6 +85,7 @@ struct spu_context {
>>  
>>  	struct list_head aff_list;
>>  	int aff_flags;
>> +	int aff_rel_displ;
>>  };
>>  
>>  /* Flag bits for spu_context aff_flags */
>> @@ -102,6 +103,8 @@ struct spu_gang {
>>  	struct list_head aff_list_head;
>>  	struct mutex aff_mutex;
>>  	int aff_flags;
>> +	struct spu *aff_ref_point_location;
>> +	atomic_t aff_sched_count;
>>  };
> 
> The aff_sched_count member seems to be completely unused. Can you
> remove it?

It is used, on the beginning of this same patch.

> Your way off abbrev variab seems strng. I have no idea what
> aff_rel_displ even means. aff_ref_point_location sounds like
> it could better be called aff_reference_point or just
> aff_reference.

:)
aff_ref_point_location is the location (spu) on which the reference
context will be placed. 'aff_ref_spu' sound a better name.

aff_rel_displ means affinity relative displacement. Tells how many spus 
I need to walk through, starting from the reference one. I'll try to 
think on a better name.




More information about the cbe-oss-dev mailing list