[x264-devel] commit: Convert to a unified "pixel" type for pixel data (Oskar Arvidsson )

[Håkan Hjort]

On Thu, Jun 3, 2010 at 19:15, avcoder <ffmpeg at gmail.com> wrote:
> It seems that the following code breaks on non-16 bytes stack such as MSVC++
> +static void ALWAYS_INLINE pixel_memset( pixel *dst, int value, int size )
> +{
> +    for( int i = 0; i < size; i++ )
> +        dst[i] = value;
> +}
>
> because: <compiled by gcc 4.4.4>
>
-snip
>
> please check:
>
>  372:   0f 57 c9                xorps  %xmm1,%xmm1
>  375:   0f b6 f8                movzbl %al,%edi
>  378:   89 7c 24 30             mov    %edi,0x30(%esp)
>  37c:   89 fd                   mov    %edi,%ebp
>  37e:   0f 29 4c 24 10          movaps %xmm1,0x10(%esp)
>
>
> %esp is not aligned on 16
>

This either due to something else (specifying an alignment) or a GCC
bug, I think.


> I just use the following patch:
>
> diff --git a/common/frame.c b/common/frame.c
> index 9d630f8..2209683 100644
> --- a/common/frame.c
> +++ b/common/frame.c
> @@ -255,8 +255,7 @@ int x264_frame_copy_picture( x264_t *h,
> x264_frame_t *dst, x264_picture_t *src )
>
>  static void ALWAYS_INLINE pixel_memset( pixel *dst, int value, int size )
>  {
> -    for( int i = 0; i < size; i++ )
> -        dst[i] = value;
> +    memset(dst, value, size);
>  }
>

That won't be correct when pixel is some other size than 'sizeof char'.
Which, I guess, is the hole point of the initial patch.

(Or well, it will only for values where all char sized sub parts
should get the same bit pattern.)

Regards,
Håkan