On Thu, Jul 09, 2015 at 11:14:54AM +0100, Markos Chandras wrote:
> On 07/09/2015 11:03 AM, Ralf Baechle wrote:
> > On Thu, Jul 09, 2015 at 10:40:36AM +0100, Markos Chandras wrote:
> >
> >> index d41e8e284825..abee2bfd10dc 100644
> >> --- a/arch/mips/include/asm/cpu-type.h
> >> +++ b/arch/mips/include/asm/cpu-type.h
> >> @@ -77,6 +77,10 @@ static inline int __pure __get_cpu_type(const int cpu_type)
> >> */
> >> #endif
> >>
> >> +#ifdef CONFIG_SYS_HAS_CPU_MIPS64_R6
> >> + case CPU_I6400:
> >> +#endif
> >> +
> >
> > To ensure best possible optimization you may want to introduce a new
> > CPU type CPU_I6400 in Kconfig then change above code segment to
> >
> > #ifdef CONFIG_SYS_HAS_CPU_I6400
> > case CPU_I6400:
> > #endif
> >
>
> Why? That function uses MIPS32_XX and MIPS64_XX in other places as well.
This is one matters a lot for optimization by GCC.
Currently you're ok for as long as The I6400 stays the sole CPU wrapped
by #ifdef CONFIG_SYS_HAS_CPU_MIPS64_R6. As soon as another CPU becomes
does the same optimization will suffer.
Example:
#ifdef CONFIG_SYS_HAS_CPU_MIPS32_R1
case CPU_4KC:
case CPU_ALCHEMY:
case CPU_PR4450:
#endif
GCC is clever enough these days to figure out that __get_cpu_type() will
only return certain values so it will do some optimization but the best
case of course is if __get_cpu_type() returns one value only.
Ralf
