[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