Re: [arm-gnu] NEON usage

[James <jamessteward@xxxxxxxxxxxxxxx>]

On Tue, 2010-05-11 at 05:41 +0200, 42Bastian wrote:
> Am 11.05.2010 05:20, schrieb James:
> 
> > But the ARM NEON Intrinsics here:
> > http://gcc.gnu.org/onlinedocs/gcc/ARM-NEON-Intrinsics.html
> > 
> > seems to suggest that 
> > 
> > int64x2_t vmlal_s32 (int64x2_t, int32x2_t, int32x2_t)
> > 
> > could be used?
> 
> Excuse my ignorance, but aren't intrinsics meant to be used by the
> programmer ?

Yes, as I understand it.  What I was getting at is that the platform has
the ability to perform a 32x32=64 bit computation, but the vectorizer
doesn't seem to know about it.

> So you need to write:
> 
> int32x2_t x,y;
> int64x2_t sum;
> for(i = 0; i < N; i+=2){
>   x = vld1_s32(&x[i]); y = vld1_s32(&y[i]);
>   vmlal_s32(sum,x,y);
> }
> // add upper and lower 64bit
> // shift down

More than likely.  I was hoping to write in portable C and let the
compiler do the NEON dirty work.

Regards,
James.