[FFmpeg-devel] [PATCH 2/2] hwaccel: add VideoToolbox support.

Sebastien Zwickert dilaroga at gmail.com
Sun Sep 23 15:34:57 CEST 2012


Hello,

On Sep 18, 2012, at 5:04 AM, Michael Niedermayer <michaelni at gmx.at> wrote:

> On Sun, Sep 16, 2012 at 08:38:05PM +0200, Sebastien Zwickert wrote:
>> ---
>> Changelog                |    1 +
>> MAINTAINERS              |    1 +
>> configure                |   12 ++
>> libavcodec/Makefile      |    7 +
>> libavcodec/allcodecs.c   |    4 +
>> libavcodec/h264.c        |    1 +
>> libavcodec/mpeg12.c      |    3 +
>> libavcodec/mpegvideo.c   |    1 +
>> libavcodec/vt.c          |  365 ++++++++++++++++++++++++++++++++++++++++++++++
>> libavcodec/vt.h          |  186 +++++++++++++++++++++++
>> libavcodec/vt_h264.c     |   79 ++++++++++
>> libavcodec/vt_internal.h |   42 ++++++
>> libavcodec/vt_mpeg2.c    |   62 ++++++++
>> libavcodec/vt_mpeg4.c    |   77 ++++++++++
>> libavutil/pixdesc.c      |    6 +
>> libavutil/pixfmt.h       |    1 +
>> 16 files changed, 848 insertions(+)
>> create mode 100644 libavcodec/vt.c
>> create mode 100644 libavcodec/vt.h
>> create mode 100644 libavcodec/vt_h264.c
>> create mode 100644 libavcodec/vt_internal.h
>> create mode 100644 libavcodec/vt_mpeg2.c
>> create mode 100644 libavcodec/vt_mpeg4.c
> [...]
>> diff --git a/libavcodec/vt.c b/libavcodec/vt.c
>> new file mode 100644
>> index 0000000..e357b56
>> --- /dev/null
>> +++ b/libavcodec/vt.c
>> @@ -0,0 +1,365 @@
>> +/*
>> + * VideoToolbox hardware acceleration
>> + *
>> + * copyright (c) 2012 Sebastien Zwickert
>> + *
>> + * This file is part of FFmpeg.
>> + *
>> + * FFmpeg is free software; you can redistribute it and/or
>> + * modify it under the terms of the GNU Lesser General Public
>> + * License as published by the Free Software Foundation; either
>> + * version 2.1 of the License, or (at your option) any later version.
>> + *
>> + * FFmpeg is distributed in the hope that it will be useful,
>> + * but WITHOUT ANY WARRANTY; without even the implied warranty of
>> + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
>> + * Lesser General Public License for more details.
>> + *
>> + * You should have received a copy of the GNU Lesser General Public
>> + * License along with FFmpeg; if not, write to the Free Software
>> + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
>> + */
>> +
>> +#include "libavutil/avutil.h"
>> +#include "libavformat/avio.h"
>> +#include "libavformat/movenc.h"
>> +
>> +#include "vt_internal.h"
>> +
>> +static CFDataRef vt_esds_extradata_create(AVCodecContext *avctx)
>> +{
>> +    CFDataRef data  = NULL;
>> +    MOVTrack *track = NULL;
>> +    AVIOContext *pb = NULL;
>> +    uint8_t *buf    = NULL;
>> +    int sb = 8; // skip first atom info : size and fourcc
>> +    int buf_size = 3 + 5 + 13 + 5 + avctx->extradata_size + 3;
>> +    // ES_DescrTag data + DecoderConfigDescrTag + data + DecSpecificInfoTag + size + SLConfigDescriptor
>> +    int padding = 12 + sb;
>> +    int s;
>> +
>> +    if (!(buf = av_mallocz(buf_size + padding)))
>> +        goto end;
>> +
>> +    if (!(track = av_mallocz(sizeof(MOVTrack))))
>> +        goto end;
>> +
> 
>> +    pb = avio_alloc_context(buf, buf_size + padding, 1, NULL, NULL, NULL, NULL);
> 
> the use of libavformat from libavcodec is problematic as it would make
> libavcodec depend on libavformat if VideoToolbox support is enabled.
> 
> I think many people would be opposed to such dependancy.
> 
> the solution here is either to undo the factorization or to change
> the code to use only functions available in libavcodec/libavutil
> and then change the libavformat part to be based on top of this.

Factorization undone.

> 
> 
> [...]
>> +static void vt_decoder_callback(void *vt_hw_ctx,
>> +                                void *sourceFrameRefCon,
>> +                                OSStatus status,
>> +                                VTDecodeInfoFlags flags,
>> +                                CVImageBufferRef image_buffer,
>> +                                CMTime pts,
>> +                                CMTime duration)
>> +{
>> +    struct vt_context *vt_ctx = vt_hw_ctx;
>> +    vt_ctx->cv_buffer = NULL;
>> +
>> +    if (!image_buffer)
>> +        return;
>> +
>> +    if (vt_ctx->cv_pix_fmt != CVPixelBufferGetPixelFormatType(image_buffer))
>> +        return;
> 
> Are these return cases normal occurances or are these error/warning
> conditions ? if later it might make sense to at least produce a
> av_log(X, AV_LOG_DEBUG,...) in the case

These are error, I added some debug logs.

> 
>> +
>> +    vt_ctx->cv_buffer = CVPixelBufferRetain(image_buffer);
>> +}
>> +
>> +static int vt_session_decode_frame(struct vt_context *vt_ctx)
>> +{
>> +    OSStatus status;
>> +    CMSampleBufferRef sample_buf;
>> +
>> +    sample_buf = vt_sample_buffer_create(vt_ctx->cm_fmt_desc,
>> +                                         vt_ctx->priv_bitstream,
>> +                                         vt_ctx->priv_bitstream_size);
>> +
>> +    if (!sample_buf)
>> +        return -1;
>> +
>> +    status = VTDecompressionSessionDecodeFrame(vt_ctx->session,
>> +                                               sample_buf,
>> +                                               0,               // decodeFlags
>> +                                               NULL,            // sourceFrameRefCon
>> +                                               0);              // infoFlagsOut
>> +    if (status == noErr)
>> +        status = VTDecompressionSessionWaitForAsynchronousFrames(vt_ctx->session);
>> +
>> +    CFRelease(sample_buf);
>> +
>> +    return status;
>> +}
> 
> is a OSStatus guranteed to fit in a int ? it is implicitly cast to
> one here.
> also the error code is propagated down to hwaccel end_frame, later
> needs negative values for errors, are errors in OSStatus guranteed
> to be negative ? (i tried to find some docs but failed)

Yes, OSStatus is a signed 32-bit integer.

> 
> 
> [...]
>> +/**
>> + * This structure is used to provide the necessary configuration and data
>> + * to the VideoToolbox FFmpeg HWAccel implementation.
>> + *
>> + * The application must make it available as AVCodecContext.hwaccel_context.
>> + */
>> +struct vt_context {
>> +    /**
>> +     * VideoToolbox decompression session.
>> +     *
>> +     * - encoding: unused.
>> +     * - decoding: Set/Unset by libavcodec.
>> +     */
>> +    VTDecompressionSessionRef   session;
>> +
> 
> 
>> +    /**
>> +     * The width of encoded video.
>> +     *
>> +     * - encoding: unused.
>> +     * - decoding: Set/Unset by user.
>> +     */
>> +    int                 width;
>> +
>> +    /**
>> +     * The height of encoded video.
>> +     *
>> +     * - encoding: unused.
>> +     * - decoding: Set/Unset by user.
>> +     */
>> +    int                 height;
>> +
> 
> these 2 seem unused

Right, removed.

> 
> [...]
> 
>> +static int vt_mpeg2_end_frame(AVCodecContext *avctx)
>> +{
>> +    return ff_vt_end_frame(avctx);
>> +}
>> +
>> +AVHWAccel ff_mpeg2_vt_hwaccel = {
>> +    .name           = "mpeg2_vt",
>> +    .type           = AVMEDIA_TYPE_VIDEO,
>> +    .id             = AV_CODEC_ID_MPEG2VIDEO,
>> +    .pix_fmt        = PIX_FMT_VT_VLD,
>> +    .start_frame    = vt_mpeg2_start_frame,
>> +    .end_frame      = vt_mpeg2_end_frame,
>> +    .decode_slice   = vt_mpeg2_decode_slice,
>> +};
> 
> the vt_mpeg2_end_frame wraper function seems unneeded ff_vt_end_frame
> could directly be used in the struct
> the same applies to vt_h264_end_frame() and vt_mpeg4_end_frame()
> 
> 
> [...]
> 
>> +static int vt_mpeg4_start_frame(AVCodecContext *avctx,
>> +                                const uint8_t *buffer,
>> +                                uint32_t size)
>> +{
>> +    struct vt_context *vt_ctx = avctx->hwaccel_context;
>> +
>> +    if (!vt_ctx->session)
>> +        return -1;
>> +
>> +    return ff_vt_buffer_copy(vt_ctx, buffer, size);
>> +}
>> +
>> +static int vt_mpeg4_decode_slice(AVCodecContext *avctx,
>> +                                 const uint8_t *buffer,
>> +                                 uint32_t size)
>> +{
>> +    struct vt_context *vt_ctx = avctx->hwaccel_context;
>> +
>> +    if (!vt_ctx->session)
>> +        return -1;
>> +
>> +    return 0;
>> +}
> 
> these 2 look redundant relative to the mpeg2 variants

Factorization done.
Finally I merged the implementation into one file, it makes sense because
most of the code is shared between the decoders.
Also I replaced 'vt' with 'videotoolbox' which is more descriptive.

Best regards,

--
Sebastien Zwickert


More information about the ffmpeg-devel mailing list