[x264-devel] [PATCH 1/2] ppc: Use vec_splats in mc

Luca Barbato lu_zero at gentoo.org
Sat Sep 8 22:50:57 CEST 2018


On 08/09/2018 22:12, Henrik Gramner wrote:
> On Sat, Sep 8, 2018 at 9:01 PM, Luca Barbato <lu_zero at gentoo.org> wrote:
>>  - lowres init :         [FAILED]
>>
>> Looks like something broke indeed.
> 
> Hmm, this is weird. The changes I did to lowres is:
> 
> -            VEC_STORE8(lv, dst0 + 16 * x);
> -            VEC_STORE8(hv, dsth + 16 * x);
> +            VEC_STORE8( lv, dst0 + 16 * x );
> +            VEC_STORE8( hv, dsth + 16 * x );
> 
> -            VEC_STORE8(lv, dstv + 16 * x);
> -            VEC_STORE8(hv, dstc + 16 * x);
> +            VEC_STORE8( lv, dstv + 16 * x );
> +            VEC_STORE8( hv, dstc + 16 * x );
> 
> I didn't touch the VEC_STORE8 macro.
> 

Interestingly enough you did drop a patch that adds a power9 VEC_STORE8
and the xxpermdi code that acts as fallback for when vec_xst_len is not
enabled does misbehave in this specific case.

https://github.com/lu-zero/x264/commit/2181395c97a4aa246156998f757c1658b8bb47db

this fixes the fallback and this

https://github.com/lu-zero/x264/commit/32878f9d2d9388ca6368efccfacee30cd7edd6a3

is the patch you forgot :)

lu


More information about the x264-devel mailing list