[MPlayer-users] mplayer compiling issue on ARM

[Reimar Döffinger]

On Fri, Aug 29, 2014 at 09:20:41AM +0200, X3fil wrote:
> Il 28/08/2014 20:36, Reimar Döffinger ha scritto:
> >On Thu, Aug 28, 2014 at 11:02:55AM +0200, X3fil wrote:
> >>I'm trying to compile mplayer. I got this error:
> >>./libavutil/cpu_internal.h:25:6: error: 'HAVE_SETEND' undeclared (first use in this function)
> >>      (HAVE_ ## cpuext ## suffix && ((flags) & AV_CPU_FLAG_ ## cpuext))
> >>
> >>Any suggestion?
> >
> >Yes, should be fixed by SVN r37258.
> >Also, since I don't know how much experience you have compiling for ARM:
> >ensure that NEON support gets enabled by configure.
> >I usually tend to do that the lazy and hackish way, by using
> >--cc='gcc -mfpu=neon' as extra configure option.
> >_
> 
> I've solved it defining HAVE_SETEND to 1 into config.h. I don't know it's the right definition. :)
> BTW, is there a way to check out if it is compiled with neon?
> 
> The config.h contains:
> 
> #define HAVE_NEON 0

That means it's disabled. That's likely fairly bad for the CPU you are
using.

> Also I don't think it is. But from cli it is checkable?
> And how to use/invoke it if it's enabled?

It's just a compilation option.
It enables SIMD instructions (on x86 SIMD is called e.g. SSE, SSE2 etc).

> My processor is an ARM Cortex A7.

Yes, I know. It's a ARMv7 with NEON support, so you should be
compiling specifically for it or you'll waste a lot of performance.
Unfortunately I think -march=native still doesn't exist/work for ARM,
making it quite a pain :(
Maybe I should consider doing some Linux-specific hacks parsing
/proc, or alternatively just make ARMv7+NEON the default and
leave anyone using something older to configure it...