From 639a20bca92ac789407935ecb8590c69475056e9 Mon Sep 17 00:00:00 2001 From: Roz K Date: Sat, 4 Oct 2025 17:39:37 +0200 Subject: [PATCH] decoder multithreading --- pve.py | 19 ++++--- pyav/decoder.py | 5 +- pyav/libav.py | 136 +++++++++++++++++++++++++++++++++++++++++++++++- 3 files changed, 151 insertions(+), 9 deletions(-) diff --git a/pve.py b/pve.py index fa60253..c24f740 100644 --- a/pve.py +++ b/pve.py @@ -14,6 +14,7 @@ # You should have received a copy of the GNU General Public License along with People's Video Editor. # If not, see . +import multiprocessing import cProfile from pyav.demuxer import Demuxer @@ -25,10 +26,15 @@ print(f"nb_streams = {demuxer.nb_streams}") print(f"video codec = {demuxer.video_stream.codec.description()}") print(f"audio codec = {demuxer.audio_stream.codec.description()}") -video_decoder = Decoder(demuxer.video_stream) -audio_decoder = Decoder(demuxer.audio_stream) +cpu_count = multiprocessing.cpu_count() -num_frames = 0 +print(f"using {cpu_count} threads for video decoding") + +video_decoder = Decoder(demuxer.video_stream, cpu_count) +audio_decoder = Decoder(demuxer.audio_stream, 1) + +num_video_frames = 0 +num_audio_frames = 0 with cProfile.Profile() as pr: while True: @@ -36,13 +42,12 @@ with cProfile.Profile() as pr: eof = (packet is None) if eof or demuxer.video_stream.contains(packet): video_frames = video_decoder.decode(packet) - num_frames += len(video_frames) - # print(f"decoded {len(video_frames)} video frames") + num_video_frames += len(video_frames) if eof or demuxer.audio_stream.contains(packet): audio_frames = audio_decoder.decode(packet) - # print(f"decoded {len(audio_frames)} audio frames") + num_audio_frames += len(audio_frames) if eof: break -print(f"num frames: {num_frames}") +print(f"num video frames: {num_video_frames}, audio frames: {num_audio_frames}") pr.print_stats() diff --git a/pyav/decoder.py b/pyav/decoder.py index 54144b3..4517377 100644 --- a/pyav/decoder.py +++ b/pyav/decoder.py @@ -21,7 +21,7 @@ from .frame import Frame class Decoder: __slots__ = '_context' - def __init__(self, stream): + def __init__(self, stream, thread_count): self._context = libav.codec_alloc_context(stream.codec) if not self._context: raise MemoryError @@ -29,6 +29,9 @@ class Decoder: if errcode < 0: libav.codec_free_context(self._context) raise Exception("Failed to set context parameters") + if thread_count > 1: + self._context.contents.thread_count = thread_count + self._context.contents.thread_type = libav.FF_THREAD_FRAME errcode = libav.codec_open(self._context, stream.codec) if errcode < 0: libav.codec_free_context(self._context) diff --git a/pyav/libav.py b/pyav/libav.py index bb8562b..fe47288 100644 --- a/pyav/libav.py +++ b/pyav/libav.py @@ -49,6 +49,9 @@ AVMEDIA_TYPE_ATTACHMENT = 4 AV_NUM_DATA_POINTERS = 8 +FF_THREAD_FRAME = 1 +FF_THREAD_SLICE = 2 + c_uint8_p = ctypes.POINTER(ctypes.c_uint8) c_uint8_pp = ctypes.POINTER(c_uint8_p) @@ -143,8 +146,139 @@ class AVCodec(ctypes.Structure): AVCodec_p = ctypes.POINTER(AVCodec) AVCodec_pp = ctypes.POINTER(AVCodec_p) +class AVChannelLayout_u(ctypes.Union): + _fields_ = [ + ("mask", ctypes.c_uint64), + ("map", ctypes.c_void_p)] # AVChannelCustom + +class AVChannelLayout(ctypes.Structure): + _fields_ = [ + ("order", ctypes.c_int), # AVChannelOrder + ("nb_channels", ctypes.c_int), + ("u", AVChannelLayout_u), + ("opaque", ctypes.c_void_p)] + class AVCodecContext(ctypes.Structure): - pass + _fields_ = [ + ("av_class", ctypes.c_void_p), + ("log_level_offset", ctypes.c_int), + ("codec_type", ctypes.c_int), + ("codec", AVCodec_p), + ("codec_id", ctypes.c_int), + ("codec_tag", ctypes.c_uint), + ("priv_data", ctypes.c_void_p), + ("internal", ctypes.c_void_p), # AVCodecInternal + ("opaque", ctypes.c_void_p), + ("bit_rate", ctypes.c_int64), + ("flags", ctypes.c_int), + ("flags2", ctypes.c_int), + ("extradata", ctypes.POINTER(ctypes.c_uint8)), + ("extradata_size", ctypes.c_int), + ("time_base", AVRational), + ("pkt_timebase", AVRational), + ("framerate", AVRational), + ("delay", ctypes.c_int), + ("width", ctypes.c_int), + ("height", ctypes.c_int), + ("coded_width", ctypes.c_int), + ("coded_height", ctypes.c_int), + ("sample_aspect_ratio", AVRational), + ("pix_fmt", ctypes.c_int), # AVPixelFormat + ("sw_pix_fmt", ctypes.c_int), # AVPixelFormat + ("color_primaries", ctypes.c_int), # AVColorPrimaries + ("color_trc", ctypes.c_int), # AVColorTransferCharacteristic + ("colorspace", ctypes.c_int), # AVColorSpace + ("color_range", ctypes.c_int), # AVColorRange + ("chroma_sample_location", ctypes.c_int), # AVChromaLocation + ("field_order", ctypes.c_int), # AVFieldOrder + ("refs", ctypes.c_int), + ("has_b_frames", ctypes.c_int), + ("slice_flags", ctypes.c_int), + ("draw_horiz_band", ctypes.c_void_p), + ("get_format", ctypes.c_void_p), + ("max_b_frames", ctypes.c_int), + ("b_quant_factor", ctypes.c_float), + ("b_quant_offset", ctypes.c_float), + ("i_quant_factor", ctypes.c_float), + ("i_quant_offset", ctypes.c_float), + ("lumi_masking", ctypes.c_float), + ("temporal_cplx_masking", ctypes.c_float), + ("spatial_cplx_masking", ctypes.c_float), + ("p_masking", ctypes.c_float), + ("dark_masking", ctypes.c_float), + ("nsse_weight", ctypes.c_int), + ("me_cmp", ctypes.c_int), + ("me_sub_cmp", ctypes.c_int), + ("mb_cmp", ctypes.c_int), + ("ildct_cmp", ctypes.c_int), + ("dia_size", ctypes.c_int), + ("last_predictor_count", ctypes.c_int), + ("me_pre_cmp", ctypes.c_int), + ("pre_dia_size", ctypes.c_int), + ("me_subpel_quality", ctypes.c_int), + ("me_range", ctypes.c_int), + ("mb_decision", ctypes.c_int), + ("intra_matrix", ctypes.POINTER(ctypes.c_uint16)), + ("inter_matrix", ctypes.POINTER(ctypes.c_uint16)), + ("chroma_intra_matrix", ctypes.POINTER(ctypes.c_uint16)), + ("intra_dc_precision", ctypes.c_int), + ("mb_lmin", ctypes.c_int), + ("mb_lmax", ctypes.c_int), + ("bidir_refine", ctypes.c_int), + ("keyint_min", ctypes.c_int), + ("gop_size", ctypes.c_int), + ("mv0_threshold", ctypes.c_int), + ("slices", ctypes.c_int), + ("sample_rate", ctypes.c_int), + ("sample_fmt", ctypes.c_int), # AVSampleFormat + ("ch_layout", AVChannelLayout), + ("frame_size", ctypes.c_int), + ("block_align", ctypes.c_int), + ("cutoff", ctypes.c_int), + ("audio_service_type", ctypes.c_int), # AVAudioServiceType + ("request_sample_fmt", ctypes.c_int), # AVSampleFormat + ("initial_padding", ctypes.c_int), + ("trailing_padding", ctypes.c_int), + ("seek_preroll", ctypes.c_int), + ("get_buffer2", ctypes.c_void_p), + ("bit_rate_tolerance", ctypes.c_int), + ("global_quality", ctypes.c_int), + ("compression_level", ctypes.c_int), + ("qcompress", ctypes.c_float), + ("qblur", ctypes.c_float), + ("qmin", ctypes.c_int), + ("qmax", ctypes.c_int), + ("max_qdiff", ctypes.c_int), + ("rc_buffer_size", ctypes.c_int), + ("rc_override_count", ctypes.c_int), + ("rc_override", ctypes.c_void_p), # RcOverride + ("rc_max_rate", ctypes.c_int64), + ("rc_min_rate", ctypes.c_int64), + ("rc_max_available_vbv_use", ctypes.c_float), + ("rc_min_vbv_overflow_use", ctypes.c_float), + ("rc_initial_buffer_occupancy", ctypes.c_int), + ("trellis", ctypes.c_int), + ("stats_out", ctypes.c_char_p), + ("stats_in", ctypes.c_char_p), + ("workaround_bugs", ctypes.c_int), + ("strict_std_compliance", ctypes.c_int), + ("error_concealment", ctypes.c_int), + ("debug", ctypes.c_int), + ("err_recognition", ctypes.c_int), + ("hwaccel", ctypes.c_void_p), # AVHWAccel + ("hwaccel_context", ctypes.c_void_p), + ("hw_frames_ctx", ctypes.c_void_p), # AVBufferRef + ("hw_device_ctx", ctypes.c_void_p), # AVBufferRef + ("hwaccel_flags", ctypes.c_int), + ("extra_hw_frames", ctypes.c_int), + ("error", ctypes.c_uint64 * AV_NUM_DATA_POINTERS), + ("dct_algo", ctypes.c_int), + ("idct_algo", ctypes.c_int), + ("bits_per_coded_sample", ctypes.c_int), + ("bits_per_raw_sample", ctypes.c_int), + ("thread_count", ctypes.c_int), + ("thread_type", ctypes.c_int)] + # ... AVCodecContext_p = ctypes.POINTER(AVCodecContext) AVCodecContext_pp = ctypes.POINTER(AVCodecContext_p)