Le sextidi 16 nivôse, an CCXXIII, Philip Langdale a écrit : > The encoder writes the display aspect ratio into the output frames, which a > good player will respect, so let's make sure it's correct. >
> The 1.02 scale factor is black magic, but produces correct results. I don't
> know what nvenc is doing.
Can you elaborate on your testing procedure? IMHO, if the input has sample
aspect ratio 1/1, then the resulting bitstream should have the same, as
shown by ffprobe. Is it achieved?
>
> Signed-off-by: Philip Langdale <phil...@overt.org>
> ---
> libavcodec/libnvenc.c | 2 ++
> libavcodec/libnvenc.h | 2 ++
> libavcodec/nvencoder.c | 8 ++++----
> 3 files changed, 8 insertions(+), 4 deletions(-)
>
> diff --git a/libavcodec/libnvenc.c b/libavcodec/libnvenc.c
> index d81e04a..6ba150e 100644
> --- a/libavcodec/libnvenc.c
> +++ b/libavcodec/libnvenc.c
> @@ -129,6 +129,8 @@ static av_cold int ff_libnvenc_init(AVCodecContext *avctx)
> // Basic
> nvenc_ctx->nvenc_cfg.width = avctx->width;
> nvenc_ctx->nvenc_cfg.height = avctx->height;
> + nvenc_ctx->nvenc_cfg.darWidth = 1024 * avctx->width *
> avctx->sample_aspect_ratio.num / avctx->sample_aspect_ratio.den;
> + nvenc_ctx->nvenc_cfg.darHeight = 1024 * avctx->height * 1.02;
Can you explain the 1024?
Anyway, I would suggest to use lavu's rational functions:
AVRational dar = av_mul_q(avctx->sample_aspect_ratio,
av_make_q(avctx->width, avctx->height));
av_reduce(&dar.num, &dar.den, dar.num, dar.den, MAX);
nvenc_ctx->nvenc_cfg.darWidth = 1024 * dar.num;
nvenc_ctx->nvenc_cfg.darHeight = 1045 * dar.den;
(including the *1.02 in the 1024 constant to avoid floating point
operations, but I am very suspicious about that 1.02)
Regards,
--
Nicolas George
signature.asc
Description: Digital signature
_______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
