[FFmpeg-cvslog] Remove leal op to fix icl inline asm.

[Reimar Döffinger]

On Mon, Apr 07, 2014 at 01:36:03PM +0200, Matt Oliver wrote:
>          "setae %b0         \n\t"
>          "cmovb %4, %1      \n\t"
> -        "cmovb %3, %2      \n\t"
> -        : "+q"(bit), "+r"(high), "+r"(code_word), "+r"(low_shift)
> -        : "r"(low)
> +        "cmovb %5, %2      \n\t"
> +        : "+q"(bit), "+&r"(c->high), "+&r"(c->code_word)
> +        : "r"(low_shift), "r"(low), "r"(code_word)

While that issue existed before and probably will not cause issues, I
think the "bit" input needs to be +&, too, since %3, 4 and 5 are still
used afterwards.
It could be avoided by moving the setae last, but that might
make performance worse (not sure though).