X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=h264encode.cpp;h=0946eb02b06126f0fcfc6c5a197e4d933daef814;hb=9c8a63d449829cbaa449cb177cfb01c7a1d2f358;hp=44867cc6d1eda148f80fb09ece9911bc842b830a;hpb=69d5a2b610c8e05ad5a813c8f13aa919f42c1d76;p=nageru diff --git a/h264encode.cpp b/h264encode.cpp index 44867cc..0946eb0 100644 --- a/h264encode.cpp +++ b/h264encode.cpp @@ -114,7 +114,6 @@ static int frame_width = 176; static int frame_height = 144; static int frame_width_mbaligned; static int frame_height_mbaligned; -static int frame_rate = FPS; static unsigned int frame_bitrate = 0; static unsigned int frame_slices = 1; static double frame_size = 0; @@ -122,7 +121,7 @@ static int initial_qp = 15; //static int initial_qp = 28; static int minimal_qp = 0; static int intra_period = 30; -static int intra_idr_period = FPS; +static int intra_idr_period = MAX_FPS; // About a second; more at lower frame rates. Not ideal. static int ip_period = 3; static int rc_mode = -1; static int rc_default_modes[] = { @@ -155,6 +154,20 @@ typedef struct __bitstream bitstream; using namespace std; +// Supposedly vaRenderPicture() is supposed to destroy the buffer implicitly, +// but if we don't delete it here, we get leaks. The GStreamer implementation +// does the same. +static void render_picture_and_delete(VADisplay dpy, VAContextID context, VABufferID *buffers, int num_buffers) +{ + VAStatus va_status = vaRenderPicture(dpy, context, buffers, num_buffers); + CHECK_VASTATUS(va_status, "vaRenderPicture"); + + for (int i = 0; i < num_buffers; ++i) { + va_status = vaDestroyBuffer(va_dpy, buffers[i]); + CHECK_VASTATUS(va_status, "vaDestroyBuffer"); + } +} + static unsigned int va_swap32(unsigned int val) { @@ -605,7 +618,7 @@ static int build_packed_slice_buffer(unsigned char **header_buffer) // // Getting pts and dts right with variable frame rate (VFR) and B-frames can be a // bit tricky. We assume first of all that the frame rate never goes _above_ -// , which gives us a frame period N. The decoder can always decode +// MAX_FPS, which gives us a frame period N. The decoder can always decode // in at least this speed, as long at dts <= pts (the frame is not attempted // presented before it is decoded). Furthermore, we never have longer chains of // B-frames than a fixed constant C. (In a B-frame chain, we say that the base @@ -698,7 +711,7 @@ void encoding2display_order( *displaying_order = encoding_order; // IDR frames are a special case; I honestly can't find the logic behind // why this is the right thing, but it seems to line up nicely in practice :-) - *pts_lag = TIMEBASE / frame_rate; + *pts_lag = TIMEBASE / MAX_FPS; } else if (((encoding_order_gop - 1) % ip_period) != 0) { /* B frames */ *frame_type = FRAME_B; *displaying_order = encoding_order - 1; @@ -859,7 +872,7 @@ static int process_cmdline(int argc, char *argv[]) } if (frame_bitrate == 0) - frame_bitrate = frame_width * frame_height * 12 * frame_rate / 50; + frame_bitrate = frame_width * frame_height * 12 * MAX_FPS / 50; if (coded_fn == NULL) { struct stat buf; @@ -1371,8 +1384,7 @@ static int render_sequence(void) render_id[0] = seq_param_buf; render_id[1] = rc_param_buf; - va_status = vaRenderPicture(va_dpy, context_id, &render_id[0], 2); - CHECK_VASTATUS(va_status, "vaRenderPicture");; + render_picture_and_delete(va_dpy, context_id, &render_id[0], 2); if (misc_priv_type != 0) { va_status = vaCreateBuffer(va_dpy, context_id, @@ -1385,7 +1397,7 @@ static int render_sequence(void) misc_param_tmp->data[0] = misc_priv_value; vaUnmapBuffer(va_dpy, misc_param_tmpbuf); - va_status = vaRenderPicture(va_dpy, context_id, &misc_param_tmpbuf, 1); + render_picture_and_delete(va_dpy, context_id, &misc_param_tmpbuf, 1); } return 0; @@ -1464,10 +1476,9 @@ static int render_picture(void) va_status = vaCreateBuffer(va_dpy, context_id, VAEncPictureParameterBufferType, sizeof(pic_param), 1, &pic_param, &pic_param_buf); - CHECK_VASTATUS(va_status, "vaCreateBuffer");; + CHECK_VASTATUS(va_status, "vaCreateBuffer"); - va_status = vaRenderPicture(va_dpy, context_id, &pic_param_buf, 1); - CHECK_VASTATUS(va_status, "vaRenderPicture"); + render_picture_and_delete(va_dpy, context_id, &pic_param_buf, 1); return 0; } @@ -1502,8 +1513,7 @@ static int render_packedsequence(void) render_id[0] = packedseq_para_bufid; render_id[1] = packedseq_data_bufid; - va_status = vaRenderPicture(va_dpy, context_id, render_id, 2); - CHECK_VASTATUS(va_status, "vaRenderPicture"); + render_picture_and_delete(va_dpy, context_id, render_id, 2); free(packedseq_buffer); @@ -1540,8 +1550,7 @@ static int render_packedpicture(void) render_id[0] = packedpic_para_bufid; render_id[1] = packedpic_data_bufid; - va_status = vaRenderPicture(va_dpy, context_id, render_id, 2); - CHECK_VASTATUS(va_status, "vaRenderPicture"); + render_picture_and_delete(va_dpy, context_id, render_id, 2); free(packedpic_buffer); @@ -1577,8 +1586,7 @@ static void render_packedslice() render_id[0] = packedslice_para_bufid; render_id[1] = packedslice_data_bufid; - va_status = vaRenderPicture(va_dpy, context_id, render_id, 2); - CHECK_VASTATUS(va_status, "vaRenderPicture"); + render_picture_and_delete(va_dpy, context_id, render_id, 2); free(packedslice_buffer); } @@ -1635,17 +1643,16 @@ static int render_slice(void) va_status = vaCreateBuffer(va_dpy, context_id, VAEncSliceParameterBufferType, sizeof(slice_param), 1, &slice_param, &slice_param_buf); - CHECK_VASTATUS(va_status, "vaCreateBuffer");; + CHECK_VASTATUS(va_status, "vaCreateBuffer"); + + render_picture_and_delete(va_dpy, context_id, &slice_param_buf, 1); - va_status = vaRenderPicture(va_dpy, context_id, &slice_param_buf, 1); - CHECK_VASTATUS(va_status, "vaRenderPicture"); - return 0; } -int H264Encoder::save_codeddata(storage_task task) +void H264Encoder::save_codeddata(storage_task task) { VACodedBufferSegment *buf_list = NULL; VAStatus va_status; @@ -1653,7 +1660,7 @@ int H264Encoder::save_codeddata(storage_task task) string data; - const int64_t global_delay = (ip_period - 1) * (TIMEBASE / frame_rate); // So we never get negative dts. + const int64_t global_delay = (ip_period - 1) * (TIMEBASE / MAX_FPS); // So we never get negative dts. va_status = vaMapBuffer(va_dpy, gl_surfaces[task.display_order % SURFACE_NUM].coded_buf, (void **)(&buf_list)); CHECK_VASTATUS(va_status, "vaMapBuffer"); @@ -1682,13 +1689,13 @@ int H264Encoder::save_codeddata(storage_task task) httpd->add_packet(pkt, task.pts + global_delay, task.dts + global_delay); } // Encode and add all audio frames up to and including the pts of this video frame. - // (They can never be queued to us after the video frame they belong to, only before.) for ( ;; ) { int64_t audio_pts; std::vector audio; { unique_lock lock(frame_queue_mutex); - if (pending_audio_frames.empty()) break; + frame_queue_nonempty.wait(lock, [this]{ return copy_thread_should_quit || !pending_audio_frames.empty(); }); + if (copy_thread_should_quit) return; auto it = pending_audio_frames.begin(); if (it->first > task.pts) break; audio_pts = it->first; @@ -1702,7 +1709,11 @@ int H264Encoder::save_codeddata(storage_task task) frame->channel_layout = AV_CH_LAYOUT_STEREO; unique_ptr int_samples(new int32_t[audio.size()]); - avcodec_fill_audio_frame(frame, 2, AV_SAMPLE_FMT_S32, (const uint8_t*)int_samples.get(), audio.size() * sizeof(int32_t), 0); + int ret = avcodec_fill_audio_frame(frame, 2, AV_SAMPLE_FMT_S32, (const uint8_t*)int_samples.get(), audio.size() * sizeof(int32_t), 1); + if (ret < 0) { + fprintf(stderr, "avcodec_fill_audio_frame() failed with %d\n", ret); + exit(1); + } for (int i = 0; i < frame->nb_samples * 2; ++i) { if (audio[i] >= 1.0f) { int_samples[i] = 2147483647; @@ -1726,6 +1737,7 @@ int H264Encoder::save_codeddata(storage_task task) // TODO: Delayed frames. avcodec_free_frame(&frame); av_free_packet(&pkt); + if (audio_pts == task.pts) break; } #if 0 @@ -1747,8 +1759,6 @@ int H264Encoder::save_codeddata(storage_task task) printf("%08lld", encode_order); printf("(%06d bytes coded)", coded_size); #endif - - return 0; } @@ -1821,7 +1831,6 @@ static int print_input() if (rc_mode != -1) printf("INPUT: RateControl : %s\n", rc_to_string(rc_mode)); printf("INPUT: Resolution : %dx%dframes\n", frame_width, frame_height); - printf("INPUT: FrameRate : %d\n", frame_rate); printf("INPUT: Bitrate : %d\n", frame_bitrate); printf("INPUT: Slieces : %d\n", frame_slices); printf("INPUT: IntraPeriod : %d\n", intra_period); @@ -1896,7 +1905,7 @@ H264Encoder::~H264Encoder() { unique_lock lock(frame_queue_mutex); copy_thread_should_quit = true; - frame_queue_nonempty.notify_one(); + frame_queue_nonempty.notify_all(); } storage_thread.join(); copy_thread.join(); @@ -1975,10 +1984,9 @@ void H264Encoder::add_audio(int64_t pts, std::vector audio) unique_lock lock(frame_queue_mutex); pending_audio_frames[pts] = move(audio); } - frame_queue_nonempty.notify_one(); + frame_queue_nonempty.notify_all(); } - void H264Encoder::end_frame(RefCountedGLsync fence, int64_t pts, const std::vector &input_frames) { { @@ -1986,7 +1994,7 @@ void H264Encoder::end_frame(RefCountedGLsync fence, int64_t pts, const std::vect pending_video_frames[current_storage_frame] = PendingFrame{ fence, input_frames, pts }; ++current_storage_frame; } - frame_queue_nonempty.notify_one(); + frame_queue_nonempty.notify_all(); } void H264Encoder::copy_thread_func() @@ -2053,7 +2061,7 @@ void H264Encoder::copy_thread_func() int64_t dts; if (pts_lag == -1) { assert(last_dts != -1); - dts = last_dts + (TIMEBASE / frame_rate); + dts = last_dts + (TIMEBASE / MAX_FPS); } else { dts = pts - pts_lag; }