[Ffmpeg-devel] [PATCH] Altivec vorbis optimizations

[Loren Merritt]

On Fri, 11 Aug 2006, Luca Barbato wrote:
> Untested code, probably some assumption are wrong, if somebody is
> willing to test&debug it I'll be grateful ^^

> static void vector_fmul_add_add_altivec(float *dst, const float *src0,
[...]
>    for (i=0; i<len-3; i+=4,y+=8) {
>        s0 = vec_ld(0,src0+i);
>        s1 = vec_ld(0,src1+i);
>        s2 = vec_ld(0,src2+i);
>        t0 = vec_ld(0, dst+i);
>        t1 = vec_ld(16, dst+i);
>        d = vec_add (vec_madd(s0,s1,s2),s3);
>        vec_stl(vec_mergeh(d, t0), 0, dst+y);
>        vec_stl(vec_mergel(d, t1), 16, dst+y);
>    }

Will this work if dst is unaligned?

--Loren Merritt