On 24/11/17 07:25, Jun Zhao wrote: > > > On 2017/11/24 8:51, Mark Thompson wrote: >> Also adds some extra fields to the main context structure that may >> be needed by a hwaccel decoder. >> --- >> The YUVJ formats really mess with this. This patch hacks them out so that >> the hwaccel works, suggestions welcome on what to actually do about them. >> >> >> libavcodec/mjpegdec.c | 70 >> +++++++++++++++++++++++++++++++++++++++++++++++++-- >> libavcodec/mjpegdec.h | 11 ++++++++ >> 2 files changed, 79 insertions(+), 2 deletions(-) >> >> diff --git a/libavcodec/mjpegdec.c b/libavcodec/mjpegdec.c >> index f01d44a169..846dec2f42 100644 >> --- a/libavcodec/mjpegdec.c >> +++ b/libavcodec/mjpegdec.c >> @@ -36,6 +36,7 @@ >> #include "avcodec.h" >> #include "blockdsp.h" >> #include "copy_block.h" >> +#include "hwaccel.h" >> #include "idctdsp.h" >> #include "internal.h" >> #include "jpegtables.h" >> @@ -147,6 +148,7 @@ av_cold int ff_mjpeg_decode_init(AVCodecContext *avctx) >> s->org_height = avctx->coded_height; >> avctx->chroma_sample_location = AVCHROMA_LOC_CENTER; >> avctx->colorspace = AVCOL_SPC_BT470BG; >> + s->hwaccel_pix_fmt = s->hwaccel_sw_pix_fmt = AV_PIX_FMT_NONE; >> >> if ((ret = build_basic_mjpeg_vlc(s)) < 0) >> return ret; >> @@ -279,6 +281,11 @@ int ff_mjpeg_decode_dht(MJpegDecodeContext *s) >> code_max + 1, 0, 0)) < 0) >> return ret; >> } >> + >> + for (i = 0; i < 16; i++) >> + s->raw_huffman_lengths[class][index][i] = bits_table[i + 1]; >> + for (i = 0; i < 256; i++) >> + s->raw_huffman_values[class][index][i] = val_table[i]; >> } >> return 0; >> } >> @@ -636,6 +643,26 @@ unk_pixfmt: >> return AVERROR_BUG; >> } >> >> + if (s->avctx->pix_fmt == AV_PIX_FMT_YUVJ420P) >> + s->avctx->pix_fmt = AV_PIX_FMT_YUV420P; >> + if (s->avctx->pix_fmt == AV_PIX_FMT_YUVJ422P) >> + s->avctx->pix_fmt = AV_PIX_FMT_YUV422P; >> + > Maybe need to give a comments in the code for this workaround, not just > in the commit message.
This hack is not intended to be applied. The note above after the commit message is inviting comment on what to actually do. - Mark _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
