[FFmpeg-devel] [PATCH] lavc/vaapi_encode_h264: add option to insert AUD nal to AVC stream
Jun Zhao
mypopydev at gmail.com
Wed May 3 06:06:20 EEST 2017
On 2017/5/3 4:23, Mark Thompson wrote:
> On 02/05/17 03:49, Jun Zhao wrote:
>> From e39078e9fe02c8d77c5e28436aef4d80a2d7b3a0 Mon Sep 17 00:00:00 2001
>> From: Jun Zhao <jun.zhao at intel.com>
>> Date: Tue, 2 May 2017 10:36:55 +0800
>> Subject: [PATCH] lavc/vaapi_encode_h264: add option to insert AUD nal to AVC
>> stream.
>>
>> Add aud option to support insert AUD nal in AVC stream. Default is
>> disable.
>>
>> Signed-off-by: Jun Zhao <jun.zhao at intel.com>
>> Signed-off-by: Yi A Wang <yi.a.wang at intel.com>
>> ---
>> libavcodec/vaapi_encode.c | 15 ++++++++++++++
>> libavcodec/vaapi_encode.h | 4 ++++
>> libavcodec/vaapi_encode_h264.c | 45 ++++++++++++++++++++++++++++++++++++++++++
>> 3 files changed, 64 insertions(+)
>
> This puts the AUDs in the wrong place?
>
> ./ffmpeg_g -y -vaapi_device /dev/dri/renderD128 -hwaccel vaapi -hwaccel_output_format vaapi -i in.mp4 -an -c:v h264_vaapi -aud 1 -b 5M -maxrate 5M out.h264
>
> IDR frame:
>
> 00000000 00 00 00 01 67 64 00 33 ac 2e c0 78 02 27 e5 c0 |....gd.3...x.'..| <- SPS
> 00000010 44 00 00 03 00 04 00 00 03 01 e3 89 80 00 98 96 |D...............|
> 00000020 00 01 31 2c bd ce 01 e1 10 8a 70 00 00 00 01 68 |..1,......p....h| <- PPS
> 00000030 ee 38 b0 00 00 00 01 06 00 07 80 83 d6 00 00 03 |.8..............| <- SEI
> 00000040 00 40 01 04 00 00 03 00 02 05 79 59 94 8b 28 11 |. at ........yY..(.|
> 00000050 ec 45 af 96 75 19 d4 1f ea a9 4d 4c 61 76 63 35 |.E..u.....MLavc5|
> 00000060 37 2e 39 33 2e 31 30 30 20 2f 20 56 41 41 50 49 |7.93.100 / VAAPI|
> 00000070 20 30 2e 34 30 2e 30 20 2f 20 49 6e 74 65 6c 20 | 0.40.0 / Intel |
> 00000080 69 39 36 35 20 64 72 69 76 65 72 20 66 6f 72 20 |i965 driver for |
> 00000090 49 6e 74 65 6c 28 52 29 20 43 68 65 72 72 79 56 |Intel(R) CherryV|
> 000000a0 69 65 77 20 2d 20 31 2e 38 2e 32 2e 70 72 65 31 |iew - 1.8.2.pre1|
> 000000b0 20 28 31 2e 38 2e 31 2d 35 35 2d 67 39 39 36 65 | (1.8.1-55-g996e|
> 000000c0 63 65 64 29 80 00 00 00 01 09 10 00 00 00 01 65 |ced)...........e| <- AUD, IDR slice
> 000000d0 b8 04 0f fe c5 c0 00 ce c7 8f f2 87 2b 23 ba dd |............+#..|
> 000000e0 b5 a3 38 20 c6 f1 d3 00 00 03 00 00 03 00 00 03 |..8 ............|
> 000000f0 00 00 03 00 00 12 8f f3 09 91 c0 00 00 03 00 05 |................|
> 00000100 14 00 00 03 02 26 00 00 03 01 37 00 00 03 00 d4 |.....&....7.....|
> 00000110 00 00 03 00 c9 00 00 03 00 e2 00 00 03 00 fe 00 |................|
> 00000120 00 03 01 32 00 00 03 02 1a 00 00 04 04 00 00 06 |...2............|
> 00000130 28 00 00 0e c0 00 00 03 00 00 03 00 00 03 00 00 |(...............|
> 00000140 03 00 00 03 00 00 03 00 00 03 00 00 03 00 00 03 |................|
>
> Normal frame:
>
> 00000250 00 03 00 00 03 00 00 03 00 00 03 00 00 03 00 00 |................|
> 00000260 03 00 00 03 00 00 03 00 01 01 00 00 00 01 06 01 |................| <- SEI
> 00000270 04 00 00 03 02 06 80 00 00 00 01 09 30 00 00 00 |............0...| <- AUD
> 00000280 01 21 e0 26 3f 00 00 03 00 00 03 00 00 03 00 00 |.!.&?...........| <- Non-IDR slice
> 00000290 03 00 00 03 00 00 03 00 00 03 00 00 03 00 00 03 |................|
>
> (Braswell N3700, i965 driver from git as of an hour ago.)
I have verified in Skylake, i965 driver fix this issue in GEN9 with the commit fd52137
in master branch, but don't fix in Braswell GEN8, open a issue to i965 driver
https://github.com/01org/intel-vaapi-driver/issues/155
>
>>
>> diff --git a/libavcodec/vaapi_encode.c b/libavcodec/vaapi_encode.c
>> index 7e9c00f51d..77a10f98a7 100644
>> --- a/libavcodec/vaapi_encode.c
>> +++ b/libavcodec/vaapi_encode.c
>> @@ -236,6 +236,21 @@ static int vaapi_encode_issue(AVCodecContext *avctx,
>> goto fail;
>> }
>>
>> + if (ctx->va_packed_headers & VA_ENC_PACKED_HEADER_RAW_DATA &&
>> + ctx->codec->write_aud_header) {
>> + bit_len = 8 * sizeof(data);
>> + err = ctx->codec->write_aud_header(avctx, pic, data, &bit_len);
>> + if (err < 0) {
>> + av_log(avctx, AV_LOG_ERROR, "Failed to write aud "
>> + "header %d: %d.\n", err);
>> + goto fail;
>> + }
>> + err = vaapi_encode_make_packed_header(avctx, pic, VAEncPackedHeaderRawData,
>> + data, bit_len);
>> + if (err < 0)
>> + goto fail;
>> + }
>> +
>> if (pic->type == PICTURE_TYPE_IDR) {
>> if (ctx->va_packed_headers & VA_ENC_PACKED_HEADER_SEQUENCE &&
>> ctx->codec->write_sequence_header) {
>> diff --git a/libavcodec/vaapi_encode.h b/libavcodec/vaapi_encode.h
>> index 0edf27e4cb..09a5d87f7d 100644
>> --- a/libavcodec/vaapi_encode.h
>> +++ b/libavcodec/vaapi_encode.h
>> @@ -267,6 +267,10 @@ typedef struct VAAPIEncodeType {
>> VAAPIEncodePicture *pic,
>> int index, int *type,
>> char *data, size_t *data_len);
>> + // Write an AU packed header, called by AVC encoder to insert AUD
>> + int (*write_aud_header)(AVCodecContext *avctx,
>> + VAAPIEncodePicture *pic,
>> + char *data, size_t *data_len);
>> } VAAPIEncodeType;
>
> You don't need a separate function - this is what write_extra_header is for.
>
>
>> diff --git a/libavcodec/vaapi_encode_h264.c b/libavcodec/vaapi_encode_h264.c
>> index 47d0c9496a..7fa9ca70e0 100644
>> --- a/libavcodec/vaapi_encode_h264.c
>> +++ b/libavcodec/vaapi_encode_h264.c
>> @@ -168,6 +168,7 @@ typedef struct VAAPIEncodeH264Options {
>> int qp;
>> int quality;
>> int low_power;
>> + int aud;
>> } VAAPIEncodeH264Options;
>>
>>
>> @@ -750,6 +751,41 @@ static int vaapi_encode_h264_write_slice_header(AVCodecContext *avctx,
>> tmp, header_len);
>> }
>>
>> +static int vaapi_encode_h264_write_aud_header(AVCodecContext *avctx,
>> + VAAPIEncodePicture *pic,
>> + char *data, size_t *data_len)
>> +{
>> + VAAPIEncodeContext *ctx = avctx->priv_data;
>> + PutBitContext pbc;
>> + char tmp[256];
>> + size_t header_len;
>> + int primary_pic_type;
>> +
>> + init_put_bits(&pbc, tmp, sizeof(tmp));
>> + vaapi_encode_h264_write_nal_header(&pbc, H264_NAL_AUD, 0);
>> + switch (pic->type) {
>> + case PICTURE_TYPE_IDR:
>> + case PICTURE_TYPE_I:
>> + primary_pic_type = 0;
>> + break;
>> + case PICTURE_TYPE_P:
>> + primary_pic_type = 1;
>> + break;
>> + case PICTURE_TYPE_B:
>> + primary_pic_type = 2;
>> + break;
>> + default:
>> + av_assert0(0 && "unknown pic type");
>> + break;
>> + }
>> + write_u(&pbc, 3, primary_pic_type, primary_pic_type);
>> + vaapi_encode_h264_write_trailing_rbsp(&pbc);
>> + header_len = put_bits_count(&pbc);
>> + flush_put_bits(&pbc);
>> + return ff_vaapi_encode_h26x_nal_unit_to_byte_stream(data, data_len,
>> + tmp, header_len);
>> +}
>
> Mild preference for this being derived in init_picture_params rather than inside the writing function?
I can change it with this way, but I don't think this way have more benefit. BTW, we also need to
support insert AUD in HEVC encoder.
>
>> +
>> static int vaapi_encode_h264_write_extra_header(AVCodecContext *avctx,
>> VAAPIEncodePicture *pic,
>> int index, int *type,
>> @@ -1180,6 +1216,8 @@ static const VAAPIEncodeType vaapi_encode_type_h264 = {
>> .write_slice_header = &vaapi_encode_h264_write_slice_header,
>>
>> .write_extra_header = &vaapi_encode_h264_write_extra_header,
>> +
>> + .write_aud_header = &vaapi_encode_h264_write_aud_header,
>> };
>>
>> static av_cold int vaapi_encode_h264_init(AVCodecContext *avctx)
>> @@ -1265,6 +1303,11 @@ static av_cold int vaapi_encode_h264_init(AVCodecContext *avctx)
>> VA_ENC_PACKED_HEADER_SLICE | // Slice headers.
>> VA_ENC_PACKED_HEADER_MISC; // SEI.
>>
>> + if (opt->aud == 1) {
>> + ctx->va_packed_headers |=
>> + VA_ENC_PACKED_HEADER_RAW_DATA;
>> + }
>> +
>> ctx->surface_width = FFALIGN(avctx->width, 16);
>> ctx->surface_height = FFALIGN(avctx->height, 16);
>>
>> @@ -1282,6 +1325,8 @@ static const AVOption vaapi_encode_h264_options[] = {
>> { "low_power", "Use low-power encoding mode (experimental: only supported "
>> "on some platforms, does not support all features)",
>> OFFSET(low_power), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, FLAGS },
>> + { "aud", "Use access unit delimiters",
>> + OFFSET(aud), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, FLAGS },
>> { NULL },
>> };
>>
>> --
>> 2.11.0
>>
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>
More information about the ffmpeg-devel
mailing list