On 27 March 2017 at 19:08, Paul B Mahol <one...@gmail.com> wrote:
> Signed-off-by: Paul B Mahol <one...@gmail.com>
> ---
> libavcodec/dnxhdenc.c | 227 ++++++++++++++++++++++++++++++
> +++++---------------
> libavcodec/dnxhdenc.h | 6 +-
> 2 files changed, 163 insertions(+), 70 deletions(-)
>
> diff --git a/libavcodec/dnxhdenc.c b/libavcodec/dnxhdenc.c
> index 33f25fb..6716b74 100644
> --- a/libavcodec/dnxhdenc.c
> +++ b/libavcodec/dnxhdenc.c
> @@ -23,6 +23,8 @@
> * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
> 02110-1301 USA
> */
>
> +#define ASSERT_LEVEL 10
>
Debug code?
> +
> #include "libavutil/attributes.h"
> #include "libavutil/internal.h"
> #include "libavutil/opt.h"
> @@ -112,6 +114,37 @@ void dnxhd_10bit_get_pixels_8x4_sym(int16_t
> *av_restrict block,
> memcpy(block + 4 * 8, pixels + 3 * line_size, 8 * sizeof(*block));
> }
>
> +static int dnxhd_10bit_dct_quantize_444(MpegEncContext *ctx, int16_t
> *block,
> + int n, int qscale, int *overflow)
> +{
> + const uint8_t *scantable= ctx->intra_scantable.scantable;
> + const int *qmat = n<4 ? ctx->q_intra_matrix[qscale] :
> ctx->q_chroma_intra_matrix[qscale];
> + int last_non_zero = 0;
> + int i;
> +
> + ctx->fdsp.fdct(block);
> +
> + // Divide by 4 with rounding, to compensate scaling of DCT
> coefficients
> + block[0] = (block[0] + 2) >> 2;
> +
> + for (i = 1; i < 64; ++i) {
> + int j = scantable[i];
> + int sign = FF_SIGNBIT(block[j]);
> + int level = (block[j] ^ sign) - sign;
>
Wait, what? This just makes level == block[j].
> + level = level * qmat[j] >> 16;
> + block[j] = (level ^ sign) - sign;
>
Don't use this to sign-ify something, compilers have long been smart enough
to optimize sign * value.
Just do
block[j] = (block[j] * qmat[j] >> 16);
Right shifting signed stuff is defined
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
