From 58669c12812b71838fcf455f2f7036406876ae6a Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Jeremy=20Lain=C3=A9?= <jeremy.laine@m4x.org>
Date: Thu, 4 Oct 2018 10:54:14 +0200
Subject: [PATCH 1/3] Always use send / receive mechanism for encoding
It has been available since at least FFmpeg 3.1.
Only subtitles require using the legacy APIs, but we don't support
encoding subtitles anyway.
---
av/audio/codeccontext.pyx | 21 -------------------
av/codec/context.pxd | 3 +--
av/codec/context.pyx | 37 +++++-----------------------------
av/video/codeccontext.pyx | 24 ----------------------
include/libavcodec/avcodec.pxd | 14 -------------
5 files changed, 6 insertions(+), 93 deletions(-)
diff --git a/av/audio/codeccontext.pyx b/av/audio/codeccontext.pyx
index b2a0546..0026f8b 100644
--- a/av/audio/codeccontext.pyx
+++ b/av/audio/codeccontext.pyx
@@ -55,27 +55,6 @@ cdef class AudioCodecContext(CodecContext):
return frames
- cdef _encode(self, Frame frame):
- """Encodes a frame of audio, returns a packet if one is ready.
- The output packet does not necessarily contain data for the most recent frame,
- as encoders can delay, split, and combine input frames internally as needed.
- If called with with no args it will flush out the encoder and return the buffered
- packets until there are none left, at which it will return None.
- """
-
- cdef Packet packet = Packet()
- cdef int got_packet = 0
-
- err_check(lib.avcodec_encode_audio2(
- self.ptr,
- &packet.struct,
- frame.ptr if frame is not None else NULL,
- &got_packet,
- ))
-
- if got_packet:
- return packet
-
cdef Frame _alloc_next_frame(self):
return alloc_audio_frame()
diff --git a/av/codec/context.pxd b/av/codec/context.pxd
index 7f62350..4290d6f 100644
--- a/av/codec/context.pxd
+++ b/av/codec/context.pxd
@@ -42,7 +42,7 @@ cdef class CodecContext(object):
cdef _set_default_time_base(self)
# Wraps both versions of the transcode API, returning lists.
- cpdef encode(self, Frame frame=?, unsigned int count=?, bint prefer_send_recv=?)
+ cpdef encode(self, Frame frame=?)
cpdef decode(self, Packet packet=?, unsigned int count=?, bint prefer_send_recv=?)
# Used by both transcode APIs to setup user-land objects.
@@ -53,7 +53,6 @@ cdef class CodecContext(object):
cdef _setup_decoded_frame(self, Frame, Packet)
# Implemented by children for the encode/decode API.
- cdef _encode(self, Frame frame)
cdef _decode(self, lib.AVPacket *packet, int *data_consumed)
# Implemented by base for the generic send/recv API.
diff --git a/av/codec/context.pyx b/av/codec/context.pyx
index f2a4eac..b9869ef 100644
--- a/av/codec/context.pyx
+++ b/av/codec/context.pyx
@@ -308,12 +308,14 @@ cdef class CodecContext(object):
if not res:
return packet
- cpdef encode(self, Frame frame=None, unsigned int count=0, bint prefer_send_recv=True):
+ cpdef encode(self, Frame frame=None):
"""Encode a list of :class:`.Packet` from the given :class:`.Frame`."""
+ if self.ptr.codec_type not in [lib.AVMEDIA_TYPE_VIDEO, lib.AVMEDIA_TYPE_AUDIO]:
+ raise NotImplementedError('Encoding is only supported for audio and video.')
+
self.open(strict=False)
- cdef bint is_flushing = frame is None
frames = self._prepare_frames_for_encode(frame)
# Assert the frames are in our time base.
@@ -323,36 +325,10 @@ cdef class CodecContext(object):
frame._rebase_time(self.ptr.time_base)
res = []
-
- if (
- prefer_send_recv and
- lib.PYAV_HAVE_AVCODEC_SEND_PACKET and
- (
- self.ptr.codec_type == lib.AVMEDIA_TYPE_VIDEO or
- self.ptr.codec_type == lib.AVMEDIA_TYPE_AUDIO
- )
- ):
- for frame in frames:
- for packet in self._send_frame_and_recv(frame):
- self._setup_encoded_packet(packet)
- res.append(packet)
- return res
-
-
for frame in frames:
- packet = self._encode(frame)
- if packet:
+ for packet in self._send_frame_and_recv(frame):
self._setup_encoded_packet(packet)
res.append(packet)
-
- while is_flushing and (not count or count > len(res)):
- packet = self._encode(None)
- if packet:
- self._setup_encoded_packet(packet)
- res.append(packet)
- else:
- break
-
return res
cdef _setup_encoded_packet(self, Packet packet):
@@ -365,9 +341,6 @@ cdef class CodecContext(object):
# are off!
packet._time_base = self.ptr.time_base
- cdef _encode(self, Frame frame):
- raise NotImplementedError('Base CodecContext cannot encode frames.')
-
cpdef decode(self, Packet packet=None, unsigned int count=0, bint prefer_send_recv=True):
"""Decode a list of :class:`.Frame` from the given :class:`.Packet`.
diff --git a/av/video/codeccontext.pyx b/av/video/codeccontext.pyx
index 1d804ec..e70326f 100644
--- a/av/video/codeccontext.pyx
+++ b/av/video/codeccontext.pyx
@@ -60,30 +60,6 @@ cdef class VideoCodecContext(CodecContext):
return [vframe]
- cdef _encode(self, Frame frame):
- """Encodes a frame of video, returns a packet if one is ready.
-
- The output packet does not necessarily contain data for the most recent frame,
- as encoders can delay, split, and combine input frames internally as needed.
- If called with with no args it will flush out the encoder and return the buffered
- packets until there are none left, at which it will return None.
-
- """
-
- cdef VideoFrame vframe = frame
-
-
- cdef Packet packet = Packet()
- cdef int got_packet
-
- if vframe is not None:
- ret = err_check(lib.avcodec_encode_video2(self.ptr, &packet.struct, vframe.ptr, &got_packet))
- else:
- ret = err_check(lib.avcodec_encode_video2(self.ptr, &packet.struct, NULL, &got_packet))
-
- if got_packet:
- return packet
-
cdef Frame _alloc_next_frame(self):
return alloc_video_frame()
diff --git a/include/libavcodec/avcodec.pxd b/include/libavcodec/avcodec.pxd
index d8d913e..f452e4a 100644
--- a/include/libavcodec/avcodec.pxd
+++ b/include/libavcodec/avcodec.pxd
@@ -286,20 +286,6 @@ cdef extern from "libavcodec/avcodec.pyav.h" nogil:
AVPacket *packet,
)
- cdef int avcodec_encode_audio2(
- AVCodecContext *ctx,
- AVPacket *avpkt,
- AVFrame *frame,
- int *got_packet_ptr
- )
-
- cdef int avcodec_encode_video2(
- AVCodecContext *ctx,
- AVPacket *avpkt,
- AVFrame *frame,
- int *got_packet_ptr
- )
-
cdef int avcodec_fill_audio_frame(
AVFrame *frame,
int nb_channels,
--
GitLab
From 85b6562c2bb7f640676a0f1b660665af1b4c67a0 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Jeremy=20Lain=C3=A9?= <jeremy.laine@m4x.org>
Date: Thu, 4 Oct 2018 11:58:50 +0200
Subject: [PATCH 2/3] Always use send / recv mechanism for decoding
It has been available since at least FFmpeg 3.1.
For subtitle, we mimick the API.
---
av/audio/codeccontext.pyx | 17 --------
av/codec/context.pxd | 5 +--
av/codec/context.pyx | 69 +++----------------------------
av/packet.pyx | 6 +--
av/stream.pyx | 4 +-
av/subtitles/codeccontext.pyx | 14 +++++--
av/video/codeccontext.pyx | 39 -----------------
include/libav.pxd | 2 -
include/libavcodec/avcodec.pxd | 14 -------
include/libavcodec/avcodec.pyav.h | 11 -----
setup.py | 1 -
tests/test_subtitles.py | 10 +----
12 files changed, 24 insertions(+), 168 deletions(-)
diff --git a/av/audio/codeccontext.pyx b/av/audio/codeccontext.pyx
index 0026f8b..aa4c15c 100644
--- a/av/audio/codeccontext.pyx
+++ b/av/audio/codeccontext.pyx
@@ -58,23 +58,6 @@ cdef class AudioCodecContext(CodecContext):
cdef Frame _alloc_next_frame(self):
return alloc_audio_frame()
- cdef _decode(self, lib.AVPacket *packet, int *data_consumed):
-
- if not self.next_frame:
- self.next_frame = alloc_audio_frame()
-
- cdef int completed_frame = 0
- data_consumed[0] = err_check(lib.avcodec_decode_audio4(self.ptr, self.next_frame.ptr, &completed_frame, packet))
- if not completed_frame:
- return
-
- cdef AudioFrame frame = self.next_frame
- self.next_frame = None
-
- frame._init_user_attributes()
-
- return frame
-
cdef _setup_decoded_frame(self, Frame frame, Packet packet):
CodecContext._setup_decoded_frame(self, frame, packet)
cdef AudioFrame aframe = frame
diff --git a/av/codec/context.pxd b/av/codec/context.pxd
index 4290d6f..6bc9d7c 100644
--- a/av/codec/context.pxd
+++ b/av/codec/context.pxd
@@ -43,7 +43,7 @@ cdef class CodecContext(object):
# Wraps both versions of the transcode API, returning lists.
cpdef encode(self, Frame frame=?)
- cpdef decode(self, Packet packet=?, unsigned int count=?, bint prefer_send_recv=?)
+ cpdef decode(self, Packet packet=?)
# Used by both transcode APIs to setup user-land objects.
# TODO: Remove the `Packet` from `_setup_decoded_frame` (because flushing
@@ -52,9 +52,6 @@ cdef class CodecContext(object):
cdef _setup_encoded_packet(self, Packet)
cdef _setup_decoded_frame(self, Frame, Packet)
- # Implemented by children for the encode/decode API.
- cdef _decode(self, lib.AVPacket *packet, int *data_consumed)
-
# Implemented by base for the generic send/recv API.
# Note that the user cannot send without recieving. This is because
# _prepare_frames_for_encode may expand a frame into multiple (e.g. when
diff --git a/av/codec/context.pyx b/av/codec/context.pyx
index b9869ef..25c8f92 100644
--- a/av/codec/context.pyx
+++ b/av/codec/context.pyx
@@ -341,7 +341,7 @@ cdef class CodecContext(object):
# are off!
packet._time_base = self.ptr.time_base
- cpdef decode(self, Packet packet=None, unsigned int count=0, bint prefer_send_recv=True):
+ cpdef decode(self, Packet packet=None):
"""Decode a list of :class:`.Frame` from the given :class:`.Packet`.
If the packet is None, the buffers will be flushed. This is useful if
@@ -355,66 +355,12 @@ cdef class CodecContext(object):
self.open(strict=False)
- if (
- prefer_send_recv and
- lib.PYAV_HAVE_AVCODEC_SEND_PACKET and
- (
- self.ptr.codec_type == lib.AVMEDIA_TYPE_VIDEO or
- self.ptr.codec_type == lib.AVMEDIA_TYPE_AUDIO
- )
- ):
- res = []
- for frame in self._send_packet_and_recv(packet):
+ res = []
+ for frame in self._send_packet_and_recv(packet):
+ if isinstance(frame, Frame):
self._setup_decoded_frame(frame, packet)
- res.append(frame)
- return res
-
- if packet is None:
- packet = Packet() # Makes our control flow easier.
-
- cdef int data_consumed = 0
- cdef list decoded_objs = []
-
- cdef uint8_t *original_data = packet.struct.data
- cdef int original_size = packet.struct.size
-
- cdef bint is_flushing = not (packet.struct.data and packet.struct.size)
-
- # Keep decoding while there is data in this packet.
- while is_flushing or packet.struct.size > 0:
-
- if is_flushing:
- packet.struct.data = NULL
- packet.struct.size = 0
-
- decoded = self._decode(&packet.struct, &data_consumed)
- packet.struct.data += data_consumed
- packet.struct.size -= data_consumed
-
- if decoded:
-
- if isinstance(decoded, Frame):
- self._setup_decoded_frame(decoded, packet)
- decoded_objs.append(decoded)
-
- # Sometimes we will error if we try to flush the stream
- # (e.g. MJPEG webcam streams), and so we must be able to
- # bail after the first, even though buffers may build up.
- if count and len(decoded_objs) >= count:
- break
-
- # Sometimes there are no frames, and no data is consumed, and this
- # is ok. However, no more frames are going to be pulled out of here.
- # (It is possible for data to not be consumed as long as there are
- # frames, e.g. during flushing.)
- elif not data_consumed:
- break
-
- # Restore the packet.
- packet.struct.data = original_data
- packet.struct.size = original_size
-
- return decoded_objs
+ res.append(frame)
+ return res
cdef _setup_decoded_frame(self, Frame frame, Packet packet):
@@ -427,9 +373,6 @@ cdef class CodecContext(object):
frame.index = self.ptr.frame_number - 1
- cdef _decode(self, lib.AVPacket *packet, int *data_consumed):
- raise NotImplementedError('Base CodecContext cannot decode packets.')
-
property name:
def __get__(self):
return self.codec.name
diff --git a/av/packet.pyx b/av/packet.pyx
index ad1fabf..3f2377c 100644
--- a/av/packet.pyx
+++ b/av/packet.pyx
@@ -101,15 +101,15 @@ cdef class Packet(Buffer):
self._time_base = dst
- def decode(self, count=0):
+ def decode(self):
"""Decode the data in this packet into a list of Frames."""
- return self._stream.decode(self, count)
+ return self._stream.decode(self)
def decode_one(self):
"""Decode the first frame from this packet.
Returns ``None`` if there is no frame."""
- res = self._stream.decode(self, count=1)
+ res = self._stream.decode(self)
return res[0] if res else None
property stream_index:
diff --git a/av/stream.pyx b/av/stream.pyx
index 359848b..eeed9d4 100644
--- a/av/stream.pyx
+++ b/av/stream.pyx
@@ -124,8 +124,8 @@ cdef class Stream(object):
packet.struct.stream_index = self._stream.index
return packets
- def decode(self, packet=None, count=0):
- return self.codec_context.decode(packet, count)
+ def decode(self, packet=None):
+ return self.codec_context.decode(packet)
def seek(self, offset, whence='time', backward=True, any_frame=False):
"""seek(offset, whence='time', backward=True, any_frame=False)
diff --git a/av/subtitles/codeccontext.pyx b/av/subtitles/codeccontext.pyx
index 64c4e21..ef2dff7 100644
--- a/av/subtitles/codeccontext.pyx
+++ b/av/subtitles/codeccontext.pyx
@@ -1,17 +1,23 @@
cimport libav as lib
from av.frame cimport Frame
+from av.packet cimport Packet
from av.subtitles.subtitle cimport SubtitleProxy, SubtitleSet
from av.utils cimport err_check
cdef class SubtitleCodecContext(CodecContext):
- cdef _decode(self, lib.AVPacket *packet, int *data_consumed):
-
+ cdef _send_packet_and_recv(self, Packet packet):
cdef SubtitleProxy proxy = SubtitleProxy()
cdef int got_frame = 0
- data_consumed[0] = err_check(lib.avcodec_decode_subtitle2(self.ptr, &proxy.struct, &got_frame, packet))
+ err_check(lib.avcodec_decode_subtitle2(
+ self.ptr,
+ &proxy.struct,
+ &got_frame,
+ &packet.struct if packet else NULL))
if got_frame:
- return SubtitleSet(proxy)
+ return [SubtitleSet(proxy)]
+ else:
+ return []
diff --git a/av/video/codeccontext.pyx b/av/video/codeccontext.pyx
index e70326f..ceb12a1 100644
--- a/av/video/codeccontext.pyx
+++ b/av/video/codeccontext.pyx
@@ -68,45 +68,6 @@ cdef class VideoCodecContext(CodecContext):
cdef VideoFrame vframe = frame
vframe._init_user_attributes()
- cdef _decode(self, lib.AVPacket *packet, int *data_consumed):
-
- # Create a frame if we don't have one ready.
- if not self.next_frame:
- self.next_frame = alloc_video_frame()
-
- # Decode video into the frame.
- cdef int completed_frame = 0
-
- cdef int result
-
- with nogil:
- result = lib.avcodec_decode_video2(self.ptr, self.next_frame.ptr, &completed_frame, packet)
- data_consumed[0] = err_check(result)
-
- if not completed_frame:
- return
-
- # Check if the frame size has changed so that we can always have a
- # SwsContext that is ready to go.
- if self.last_w != self.ptr.width or self.last_h != self.ptr.height:
- self.last_w = self.ptr.width
- self.last_h = self.ptr.height
- # TODO codec-ctx: Stream would calculate self.buffer_size here.
- self.reformatter = VideoReformatter()
-
- # We are ready to send this one off into the world!
- cdef VideoFrame frame = self.next_frame
- self.next_frame = None
-
- # Share our SwsContext with the frames. Most of the time they will end
- # up using the same settings as each other, so it makes sense to cache
- # it like this.
- # TODO codec-ctx: Stream did this.
- #frame.reformatter = self.reformatter
-
- return frame
-
-
cdef _build_format(self):
self._format = get_video_format(<lib.AVPixelFormat>self.ptr.pix_fmt, self.ptr.width, self.ptr.height)
diff --git a/include/libav.pxd b/include/libav.pxd
index 791ffef..2fb8466 100644
--- a/include/libav.pxd
+++ b/include/libav.pxd
@@ -6,8 +6,6 @@ cdef extern from "pyav/config.h" nogil:
char* PYAV_VERSION_STR
char* PYAV_COMMIT_STR
- int PYAV_HAVE_AVCODEC_SEND_PACKET
-
include "libavutil/avutil.pxd"
include "libavutil/channel_layout.pxd"
diff --git a/include/libavcodec/avcodec.pxd b/include/libavcodec/avcodec.pxd
index f452e4a..1a421a8 100644
--- a/include/libavcodec/avcodec.pxd
+++ b/include/libavcodec/avcodec.pxd
@@ -272,20 +272,6 @@ cdef extern from "libavcodec/avcodec.pyav.h" nogil:
void (*destruct)(AVPacket*)
- cdef int avcodec_decode_video2(
- AVCodecContext *ctx,
- AVFrame *frame,
- int *got_frame,
- AVPacket *packet,
- )
-
- cdef int avcodec_decode_audio4(
- AVCodecContext *ctx,
- AVFrame *frame,
- int *got_frame,
- AVPacket *packet,
- )
-
cdef int avcodec_fill_audio_frame(
AVFrame *frame,
int nb_channels,
diff --git a/include/libavcodec/avcodec.pyav.h b/include/libavcodec/avcodec.pyav.h
index 7f6d493..a4aca92 100644
--- a/include/libavcodec/avcodec.pyav.h
+++ b/include/libavcodec/avcodec.pyav.h
@@ -9,17 +9,6 @@
#endif
-#if !PYAV_HAVE_AVCODEC_SEND_PACKET
-
- // Stub these out so that we don't fail to compile.
- int avcodec_send_packet(AVCodecContext *avctx, AVPacket *packet) { return 0; }
- int avcodec_receive_frame(AVCodecContext *avctx, AVFrame *frame) { return 0; }
- int avcodec_send_frame(AVCodecContext *avctx, AVFrame *frame) { return 0; }
- int avcodec_receive_packet(AVCodecContext *avctx, AVPacket *avpkt) { return 0; }
-
-#endif
-
-
// Some of these properties don't exist in both FFMpeg and LibAV, so we
// signal to our code that they are missing via 0.
#ifndef AV_CODEC_PROP_INTRA_ONLY
diff --git a/setup.py b/setup.py
index 513c1d9..6352f07 100644
--- a/setup.py
+++ b/setup.py
@@ -550,7 +550,6 @@ class ReflectCommand(Command):
'av_calloc',
'avformat_alloc_output_context2',
'avformat_close_input',
- 'avcodec_send_packet',
):
print("looking for %s... " % func_name, end='\n' if self.debug else '')
diff --git a/tests/test_subtitles.py b/tests/test_subtitles.py
index 9e5dfce..f28bcad 100644
--- a/tests/test_subtitles.py
+++ b/tests/test_subtitles.py
@@ -14,10 +14,7 @@ class TestSubtitle(TestCase):
fh = av.open(path)
subs = []
for packet in fh.demux():
- try:
- subs.extend(packet.decode())
- except ValueError:
- raise SkipTest
+ subs.extend(packet.decode())
self.assertEqual(len(subs), 3)
self.assertIsInstance(subs[0][0], AssSubtitle)
@@ -33,10 +30,7 @@ class TestSubtitle(TestCase):
fh = av.open(path)
subs = []
for packet in fh.demux():
- try:
- subs.extend(packet.decode())
- except ValueError:
- raise SkipTest
+ subs.extend(packet.decode())
self.assertEqual(len(subs), 43)
--
GitLab
From a0c51ab3c22a46fbb95363b55ce86ee3ed4b3c89 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Jeremy=20Lain=C3=A9?= <jeremy.laine@m4x.org>
Date: Thu, 4 Oct 2018 15:17:28 +0200
Subject: [PATCH 3/3] Remove refcounted_frames
It is only used by the avcodec_decode_(audio4|video2) compatibility wrappers,
which we don't use.
---
av/codec/context.pyx | 3 ---
include/libavcodec/avcodec.pxd | 1 -
scratchpad/memleak.py | 2 +-
3 files changed, 1 insertion(+), 5 deletions(-)
diff --git a/av/codec/context.pyx b/av/codec/context.pyx
index 25c8f92..e9f1399 100644
--- a/av/codec/context.pyx
+++ b/av/codec/context.pyx
@@ -85,9 +85,6 @@ cdef class CodecContext(object):
raise RuntimeError('Wrapping CodecContext with mismatched codec.')
self.codec = wrap_codec(codec if codec != NULL else self.ptr.codec)
- # Signal that we want to reference count.
- self.ptr.refcounted_frames = 1
-
# Set reasonable threading defaults.
# count == 0 -> use as many threads as there are CPUs.
# type == 2 -> thread within a frame. This does not change the API.
diff --git a/include/libavcodec/avcodec.pxd b/include/libavcodec/avcodec.pxd
index 1a421a8..36e900a 100644
--- a/include/libavcodec/avcodec.pxd
+++ b/include/libavcodec/avcodec.pxd
@@ -121,7 +121,6 @@ cdef extern from "libavcodec/avcodec.pyav.h" nogil:
int flags
int thread_count
int thread_type
- int refcounted_frames
int profile
AVDiscard skip_frame
diff --git a/scratchpad/memleak.py b/scratchpad/memleak.py
index ffab979..20da5d4 100644
--- a/scratchpad/memleak.py
+++ b/scratchpad/memleak.py
@@ -67,7 +67,7 @@ def transcode_level1_to_level3():
def decode_using_pyav():
print('Decoding using PyAV.')
- fh = av.open('ffv1_level3.nut', 'r', options={'refcounted_frames': '1'})
+ fh = av.open('ffv1_level3.nut', 'r')
for s in fh.streams:
#print s, s.thread_type, s.thread_count
#pass
--
GitLab