[FFmpeg-devel] [PATCH] nvdec/vc1: add marker insertion logic
averne
averne381 at gmail.com
Tue Apr 8 17:28:32 EEST 2025
Le 08/04/2025 à 11:19, Zhao Zhili a écrit :
> Sorry for the long delay.
No problem, thank you for the review.
>
>> On Mar 5, 2025, at 01:17, averne <averne381 at gmail.com> wrote:
>>
>>
>> ---
>> Insert the relevant marker into the bitstream on
>> slice submission.
>> This is analogous to the logic found in DXVA and
>> D3D hwaccels.
>>
>> Fixes decoding of various VC-1 streams, eg.:
>> https://drive.google.com/file/d/1WJyiRhcdU4FHTW3sVMitS7UdrZM1NBy-/view?usp=sharing
>
> I have tested and the patch works to fix the issue.
>
>>
>> This was investigated using my nvdec tracing tool:
>> https://github.com/averne/NvdecTrace
>>
>> Signed-off-by: averne <averne381 at gmail.com>
>> ---
>> libavcodec/nvdec_vc1.c | 50 ++++++++++++++++++++++++++++++++++++++++++++++++--
>> 1 file changed, 48 insertions(+), 2 deletions(-)
>>
>> diff --git a/libavcodec/nvdec_vc1.c b/libavcodec/nvdec_vc1.c
>> index fbfba1ecb43421573ef8fea1e37a2425c272edc9..2726574a26583b0cfc28bdec5595c15bdc465ff8 100644
>> --- a/libavcodec/nvdec_vc1.c
>> +++ b/libavcodec/nvdec_vc1.c
>> @@ -22,6 +22,7 @@
>>
>> #include "config_components.h"
>>
>> +#include "libavutil/mem.h"
>> #include "avcodec.h"
>> #include "hwaccel_internal.h"
>> #include "internal.h"
>> @@ -107,6 +108,51 @@ static int nvdec_vc1_start_frame(AVCodecContext *avctx, const uint8_t *buffer, u
>> return 0;
>> }
>>
>> +static int nvdec_vc1_decode_slice(AVCodecContext *avctx, const uint8_t *buffer,
>> + uint32_t size)
>> +{
>> + NVDECContext *ctx = avctx->internal->hwaccel_priv_data;
>> + const VC1Context *v = avctx->priv_data;
>> + uint32_t marker;
>> + int marker_size;
>> + void *tmp;
>> +
>> + if (avctx->codec_id != AV_CODEC_ID_VC1) {
>> + marker_size = 0;
>> + } else {
>> + if (ctx->bitstream_len)
>> + marker = VC1_CODE_SLICE;
>> + else if (v->profile == PROFILE_ADVANCED && v->fcm == ILACE_FIELD && v->second_field)
>> + marker = VC1_CODE_FIELD;
>> + else
>> + marker = VC1_CODE_FRAME;
>> +
>> + marker_size = (size >= sizeof(marker) && AV_RB32(buffer) != marker) ? sizeof(marker) : 0;
>> + }
>
> I’m not familiar with VC1 or WMV3, but if the maker only needed for VC1, you can remove the check on
> codec_id, and keep ff_wmv3_nvdec_hwaccel as before (use ff_nvdec_simple_decode_slice).
>
> Otherwise LGTM, thanks.
That's a good point, I will send a revised patch shortly.
>
>> +
>> + tmp = av_fast_realloc(ctx->bitstream_internal, &ctx->bitstream_allocated,
>> + ctx->bitstream_len + size + marker_size);
>> + if (!tmp)
>> + return AVERROR(ENOMEM);
>> + ctx->bitstream = ctx->bitstream_internal = tmp;
>> +
>> + tmp = av_fast_realloc(ctx->slice_offsets, &ctx->slice_offsets_allocated,
>> + (ctx->nb_slices + 1) * sizeof(*ctx->slice_offsets));
>> + if (!tmp)
>> + return AVERROR(ENOMEM);
>> + ctx->slice_offsets = tmp;
>> +
>> + if (marker_size)
>> + AV_WB32(ctx->bitstream_internal + ctx->bitstream_len, marker);
>> +
>> + memcpy(ctx->bitstream_internal + ctx->bitstream_len + marker_size, buffer, size);
>> + ctx->slice_offsets[ctx->nb_slices] = ctx->bitstream_len;
>> + ctx->bitstream_len += size + marker_size;
>> + ctx->nb_slices++;
>> +
>> + return 0;
>> +}
>> +
>> static int nvdec_vc1_frame_params(AVCodecContext *avctx,
>> AVBufferRef *hw_frames_ctx)
>> {
>> @@ -121,7 +167,7 @@ const FFHWAccel ff_vc1_nvdec_hwaccel = {
>> .p.pix_fmt = AV_PIX_FMT_CUDA,
>> .start_frame = nvdec_vc1_start_frame,
>> .end_frame = ff_nvdec_simple_end_frame,
>> - .decode_slice = ff_nvdec_simple_decode_slice,
>> + .decode_slice = nvdec_vc1_decode_slice,
>> .frame_params = nvdec_vc1_frame_params,
>> .init = ff_nvdec_decode_init,
>> .uninit = ff_nvdec_decode_uninit,
>> @@ -136,7 +182,7 @@ const FFHWAccel ff_wmv3_nvdec_hwaccel = {
>> .p.pix_fmt = AV_PIX_FMT_CUDA,
>> .start_frame = nvdec_vc1_start_frame,
>> .end_frame = ff_nvdec_simple_end_frame,
>> - .decode_slice = ff_nvdec_simple_decode_slice,
>> + .decode_slice = nvdec_vc1_decode_slice,
>> .frame_params = nvdec_vc1_frame_params,
>> .init = ff_nvdec_decode_init,
>> .uninit = ff_nvdec_decode_uninit,
>>
>> ---
>> base-commit: f76195ff656d6bea68feee783160652e2b3e3d60
>> change-id: 20250304-nvdec-vc1-marker2-53d6bd30ee99
>>
>> Best regards,
>> --
>> averne <averne381 at gmail.com>
>>
>> _______________________________________________
>> ffmpeg-devel mailing list
>> ffmpeg-devel at ffmpeg.org
>> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>>
>> To unsubscribe, visit link above, or email
>> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".
>
More information about the ffmpeg-devel
mailing list