Re: slab: avoid linear search in kmalloc? (GCC Guru wanted :)

[Momchil Velikov]

>>>>> "Arjan" == Arjan van de Ven <[e-mail]> writes:

Arjan> Momchil Velikov wrote:
>> 
>> >>>>> "Bernd" == Bernd Eckenfels <[e-mail]> writes:
>> 
Bernd> Hello,
Bernd> I noticed that kmalloc and kmem_find_general_cachep are doing a linear
Bernd> search in the cache_sizes array. Isnt it better to speed that up by doing a
Bernd> binary search or a b-tree if like the following patch?
>> 
>> Here is a patch using a gcc extension. gcc generates binary search for the case.

Arjan> the big "case" statement makes you wonder if ffz(~size) would do the
Arjan> same ;)

*Nod*, espesially on architectures, where it is a single instruction.

OTOH, this is possible only if the sizes are powers of two. And they
could be more closely sparsed ...

Regards,
-velco