Quoting Diego Biurrun (2016-06-09 17:11:44)
> From: Alexandra Hájková <alexan...@khirnov.net>
>
> The new bit reader features a simpler API and an implementation without
> stacks of nested macros. Performance is the same or better, especially
> for mezzanine formats and 64-bit platforms.
> ---
> Changelog | 1 +
> libavcodec/bitstream.h | 380
> +++++++++++++++++++++++++++++++++++++++++++++++++
> 2 files changed, 381 insertions(+)
> create mode 100644 libavcodec/bitstream.h
>
> diff --git a/Changelog b/Changelog
> index a2a9d46..6977397 100644
> --- a/Changelog
> +++ b/Changelog
> @@ -57,6 +57,7 @@ version <next>:
> - Generic OpenMAX IL encoder with support for Raspberry Pi
> - MMAL-accelerated MPEG-2 and VC-1 decoding
> - G.729 raw demuxer
> +- new bitstream reading framework
>
>
> version 11:
> diff --git a/libavcodec/bitstream.h b/libavcodec/bitstream.h
> new file mode 100644
> index 0000000..539b721
> --- /dev/null
> +++ b/libavcodec/bitstream.h
> +static inline void refill_64(BitstreamContext *bc)
> +{
> + if (av_unlikely(bc->ptr >= bc->buffer_end || !bc->buffer))
Why should the second condition ever be true?
> +/* Read MPEG-1 dc-style VLC (sign bit + mantissa with no MSB).
> + * If MSB not set it is negative. */
> +static inline int bitstream_read_xbits(BitstreamContext *bc, unsigned length)
> +{
> + int32_t cache = bitstream_peek(bc, 32);
> + int sign = ~cache >> 31;
> + skip_remaining(bc, length);
This used to be bitstream_skip() in the previous versions of the patch.
It is not immediately obvious to me why the change is correct.
Same for read_vlc().
--
Anton Khirnov
_______________________________________________
libav-devel mailing list
libav-devel@libav.org
https://lists.libav.org/mailman/listinfo/libav-devel
