Justin Ruggles <justin.rugg...@gmail.com> writes:
> +/**
> + * @addtogroup lavu_math
> + * @{
> + */
> +
> +#if AV_GCC_VERSION_AT_LEAST(3,4)
> +#ifndef ff_ctz
> +# define ff_ctz(v) (__builtin_ctz(v))
> +#endif
> +#endif
> +
> +#ifndef ff_ctz
> +#define ff_ctz ff_ctz_c
> +static av_always_inline av_const int ff_ctz_c(int v)
> +{
> + return av_log2(v & -v);
> +}
> +#endif
Sorry, I was a bit unclear earlier. We probably don't want to use the
table-based av_log2 for this. There are a bunch of CPUs that don't have
a CTZ instruction but do have a CLZ instruction, and those benefit from
that trick. However, gcc seems to know about that in the cases I
checked. I'd be more worried about gcc messing up completely with
__builtin_ctz() when there is no CTZ or CLZ instruction. Their fallback
implementation of __builtin_clz() at least used to be pretty poor.
--
Måns Rullgård
m...@mansr.com
_______________________________________________
libav-devel mailing list
libav-devel@libav.org
https://lists.libav.org/mailman/listinfo/libav-devel
