On Sat, Jul 03, 2010 at 11:31:44PM +0900, Shinya Kuribayashi wrote:
> On 07/03/2010 06:32 AM, David VomLehn wrote:
> > Usually it's better to control things on a feature-by-feature basis rather
> > than rely on things like CPU model. This allows you to easily handle case
> > where, for example, you have a different CPU that normally doesn't have
> > a feature but a particular variant does have it. IIRC, the MIPS family has
> > examples of this. So, I think it's better to go with the:
> > __builtin_constant_p(cpu_has_clo_clz) && cpu_has_clo_clz
> > used in fls().
>
> Ok, now I've come to the same conclusion. Revised patch will be like
> this. Malta is a development platform supporting various types of
> MIPS32/MIPS64 cores, hence use cpu_has_clo_clz directly. The same goes
> to MIPSSim.
>
> Another concern is that, I'm not really sure whether cpu_has_clo_clz is
> acceptable or not for Malta (and MIPSSim). Hopefully Ralf will help us
> make things in the right direction.
My grief with this patch at this moment is:
o The suggestion of using __builtin_ffs or similar is nice but these
functions appear to have introduced in GCC 3.4 but we unfortunately
support GCC >= 3.2.
o no Signed-off-by: line. So can you sen me one? Thanks
Ralf
|