X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=h264encode.cpp;h=4d7882a090f9f5d5c27953655b64c4d4d80023e9;hb=408b03dbd380f3b2ea89fc20915c38bd4e4b9750;hp=3d8b463173aab3b5d2f3486d6b02688dc2db8f57;hpb=5a94066158c0277c91d44afb980b83ed8ba2898e;p=nageru diff --git a/h264encode.cpp b/h264encode.cpp index 3d8b463..4d7882a 100644 --- a/h264encode.cpp +++ b/h264encode.cpp @@ -33,6 +33,8 @@ #include "httpd.h" #include "timebase.h" +using namespace std; + class QOpenGLContext; class QSurface; @@ -108,21 +110,17 @@ static int h264_packedheader = 0; /* support pack header? */ static int h264_maxref = (1<<16|1); static int h264_entropy_mode = 1; /* cabac */ -static char *coded_fn = NULL; - 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; 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[] = { @@ -619,7 +617,7 @@ 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 @@ -712,7 +710,7 @@ 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; @@ -873,7 +871,7 @@ 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; @@ -973,28 +971,23 @@ static int init_va(void) } else { switch (h264_profile) { case VAProfileH264Baseline: - printf("Use profile VAProfileH264Baseline\n"); ip_period = 1; constraint_set_flag |= (1 << 0); /* Annex A.2.1 */ h264_entropy_mode = 0; break; case VAProfileH264ConstrainedBaseline: - printf("Use profile VAProfileH264ConstrainedBaseline\n"); constraint_set_flag |= (1 << 0 | 1 << 1); /* Annex A.2.2 */ ip_period = 1; break; case VAProfileH264Main: - printf("Use profile VAProfileH264Main\n"); constraint_set_flag |= (1 << 1); /* Annex A.2.2 */ break; case VAProfileH264High: constraint_set_flag |= (1 << 3); /* Annex A.2.4 */ - printf("Use profile VAProfileH264High\n"); break; default: - printf("unknow profile. Set to Baseline"); h264_profile = VAProfileH264Baseline; ip_period = 1; constraint_set_flag |= (1 << 0); /* Annex A.2.1 */ @@ -1024,23 +1017,6 @@ static int init_va(void) if (attrib[VAConfigAttribRateControl].value != VA_ATTRIB_NOT_SUPPORTED) { int tmp = attrib[VAConfigAttribRateControl].value; - printf("Support rate control mode (0x%x):", tmp); - - if (tmp & VA_RC_NONE) - printf("NONE "); - if (tmp & VA_RC_CBR) - printf("CBR "); - if (tmp & VA_RC_VBR) - printf("VBR "); - if (tmp & VA_RC_VCM) - printf("VCM "); - if (tmp & VA_RC_CQP) - printf("CQP "); - if (tmp & VA_RC_VBR_CONSTRAINED) - printf("VBR_CONSTRAINED "); - - printf("\n"); - if (rc_mode == -1 || !(rc_mode & tmp)) { if (rc_mode != -1) { printf("Warning: Don't support the specified RateControl mode: %s!!!, switch to ", rc_to_string(rc_mode)); @@ -1052,8 +1028,6 @@ static int init_va(void) break; } } - - printf("RateControl mode: %s\n", rc_to_string(rc_mode)); } config_attrib[config_attrib_num].type = VAConfigAttribRateControl; @@ -1065,29 +1039,23 @@ static int init_va(void) if (attrib[VAConfigAttribEncPackedHeaders].value != VA_ATTRIB_NOT_SUPPORTED) { int tmp = attrib[VAConfigAttribEncPackedHeaders].value; - printf("Support VAConfigAttribEncPackedHeaders\n"); - h264_packedheader = 1; config_attrib[config_attrib_num].type = VAConfigAttribEncPackedHeaders; config_attrib[config_attrib_num].value = VA_ENC_PACKED_HEADER_NONE; if (tmp & VA_ENC_PACKED_HEADER_SEQUENCE) { - printf("Support packed sequence headers\n"); config_attrib[config_attrib_num].value |= VA_ENC_PACKED_HEADER_SEQUENCE; } if (tmp & VA_ENC_PACKED_HEADER_PICTURE) { - printf("Support packed picture headers\n"); config_attrib[config_attrib_num].value |= VA_ENC_PACKED_HEADER_PICTURE; } if (tmp & VA_ENC_PACKED_HEADER_SLICE) { - printf("Support packed slice headers\n"); config_attrib[config_attrib_num].value |= VA_ENC_PACKED_HEADER_SLICE; } if (tmp & VA_ENC_PACKED_HEADER_MISC) { - printf("Support packed misc headers\n"); config_attrib[config_attrib_num].value |= VA_ENC_PACKED_HEADER_MISC; } @@ -1096,19 +1064,6 @@ static int init_va(void) } if (attrib[VAConfigAttribEncInterlaced].value != VA_ATTRIB_NOT_SUPPORTED) { - int tmp = attrib[VAConfigAttribEncInterlaced].value; - - printf("Support VAConfigAttribEncInterlaced\n"); - - if (tmp & VA_ENC_INTERLACED_FRAME) - printf("support VA_ENC_INTERLACED_FRAME\n"); - if (tmp & VA_ENC_INTERLACED_FIELD) - printf("Support VA_ENC_INTERLACED_FIELD\n"); - if (tmp & VA_ENC_INTERLACED_MBAFF) - printf("Support VA_ENC_INTERLACED_MBAFF\n"); - if (tmp & VA_ENC_INTERLACED_PAFF) - printf("Support VA_ENC_INTERLACED_PAFF\n"); - config_attrib[config_attrib_num].type = VAConfigAttribEncInterlaced; config_attrib[config_attrib_num].value = VA_ENC_PACKED_HEADER_NONE; config_attrib_num++; @@ -1116,28 +1071,6 @@ static int init_va(void) if (attrib[VAConfigAttribEncMaxRefFrames].value != VA_ATTRIB_NOT_SUPPORTED) { h264_maxref = attrib[VAConfigAttribEncMaxRefFrames].value; - - printf("Support %d RefPicList0 and %d RefPicList1\n", - h264_maxref & 0xffff, (h264_maxref >> 16) & 0xffff ); - } - - if (attrib[VAConfigAttribEncMaxSlices].value != VA_ATTRIB_NOT_SUPPORTED) - printf("Support %d slices\n", attrib[VAConfigAttribEncMaxSlices].value); - - if (attrib[VAConfigAttribEncSliceStructure].value != VA_ATTRIB_NOT_SUPPORTED) { - int tmp = attrib[VAConfigAttribEncSliceStructure].value; - - printf("Support VAConfigAttribEncSliceStructure\n"); - - if (tmp & VA_ENC_SLICE_STRUCTURE_ARBITRARY_ROWS) - printf("Support VA_ENC_SLICE_STRUCTURE_ARBITRARY_ROWS\n"); - if (tmp & VA_ENC_SLICE_STRUCTURE_POWER_OF_TWO_ROWS) - printf("Support VA_ENC_SLICE_STRUCTURE_POWER_OF_TWO_ROWS\n"); - if (tmp & VA_ENC_SLICE_STRUCTURE_ARBITRARY_MACROBLOCKS) - printf("Support VA_ENC_SLICE_STRUCTURE_ARBITRARY_MACROBLOCKS\n"); - } - if (attrib[VAConfigAttribEncMacroblockInfo].value != VA_ATTRIB_NOT_SUPPORTED) { - printf("Support VAConfigAttribEncMacroblockInfo\n"); } free(entrypoints); @@ -1653,7 +1586,7 @@ static int render_slice(void) -int H264Encoder::save_codeddata(storage_task task) +void H264Encoder::save_codeddata(storage_task task) { VACodedBufferSegment *buf_list = NULL; VAStatus va_status; @@ -1661,7 +1594,7 @@ 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"); @@ -1690,13 +1623,13 @@ 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; + 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; @@ -1710,7 +1643,11 @@ 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; @@ -1734,6 +1671,7 @@ 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 @@ -1755,15 +1693,13 @@ int H264Encoder::save_codeddata(storage_task task) printf("%08lld", encode_order); printf("(%06d bytes coded)", coded_size); #endif - - return 0; } // this is weird. but it seems to put a new frame onto the queue void H264Encoder::storage_task_enqueue(storage_task task) { - std::unique_lock lock(storage_task_queue_mutex); + unique_lock lock(storage_task_queue_mutex); storage_task_queue.push(move(task)); srcsurface_status[task.display_order % SURFACE_NUM] = SRC_SURFACE_IN_ENCODING; storage_task_queue_changed.notify_all(); @@ -1775,7 +1711,7 @@ void H264Encoder::storage_task_thread() storage_task current; { // wait until there's an encoded frame - std::unique_lock lock(storage_task_queue_mutex); + unique_lock lock(storage_task_queue_mutex); storage_task_queue_changed.wait(lock, [this]{ return storage_thread_should_quit || !storage_task_queue.empty(); }); if (storage_thread_should_quit) return; current = move(storage_task_queue.front()); @@ -1790,7 +1726,7 @@ void H264Encoder::storage_task_thread() save_codeddata(move(current)); { - std::unique_lock lock(storage_task_queue_mutex); + unique_lock lock(storage_task_queue_mutex); srcsurface_status[current.display_order % SURFACE_NUM] = SRC_SURFACE_FREE; storage_task_queue_changed.notify_all(); } @@ -1823,27 +1759,6 @@ static int deinit_va() } -static int print_input() -{ - printf("\n\nINPUT:Try to encode H264...\n"); - 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); - printf("INPUT: IDRPeriod : %d\n", intra_idr_period); - printf("INPUT: IpPeriod : %d\n", ip_period); - printf("INPUT: Initial QP : %d\n", initial_qp); - printf("INPUT: Min QP : %d\n", minimal_qp); - printf("INPUT: Coded Clip : %s\n", coded_fn); - - printf("\n\n"); /* return back to startpoint */ - - return 0; -} - H264Encoder::H264Encoder(QSurface *surface, int width, int height, HTTPD *httpd) : current_storage_frame(0), surface(surface), httpd(httpd) { @@ -1867,7 +1782,7 @@ H264Encoder::H264Encoder(QSurface *surface, int width, int height, HTTPD *httpd) frame_bitrate = 15000000; // / 60; current_frame_encoding = 0; - print_input(); + //print_input(); init_va(); setup_encode(); @@ -1879,9 +1794,9 @@ H264Encoder::H264Encoder(QSurface *surface, int width, int height, HTTPD *httpd) memset(&pic_param, 0, sizeof(pic_param)); memset(&slice_param, 0, sizeof(slice_param)); - storage_thread = std::thread(&H264Encoder::storage_task_thread, this); + storage_thread = thread(&H264Encoder::storage_task_thread, this); - copy_thread = std::thread([this]{ + copy_thread = thread([this]{ //SDL_GL_MakeCurrent(window, context); QOpenGLContext *context = create_context(this->surface); eglBindAPI(EGL_OPENGL_API); @@ -1904,7 +1819,7 @@ 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(); @@ -1917,7 +1832,7 @@ bool H264Encoder::begin_frame(GLuint *y_tex, GLuint *cbcr_tex) { { // Wait until this frame slot is done encoding. - std::unique_lock lock(storage_task_queue_mutex); + unique_lock lock(storage_task_queue_mutex); storage_task_queue_changed.wait(lock, [this]{ return storage_thread_should_quit || (srcsurface_status[current_storage_frame % SURFACE_NUM] == SRC_SURFACE_FREE); }); if (storage_thread_should_quit) return false; } @@ -1977,24 +1892,23 @@ bool H264Encoder::begin_frame(GLuint *y_tex, GLuint *cbcr_tex) return true; } -void H264Encoder::add_audio(int64_t pts, std::vector audio) +void H264Encoder::add_audio(int64_t pts, 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) +void H264Encoder::end_frame(RefCountedGLsync fence, int64_t pts, const vector &input_frames) { { unique_lock lock(frame_queue_mutex); 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() @@ -2014,69 +1928,76 @@ void H264Encoder::copy_thread_func() { unique_lock lock(frame_queue_mutex); frame_queue_nonempty.wait(lock, [this]{ return copy_thread_should_quit || pending_video_frames.count(current_frame_display) != 0; }); - if (copy_thread_should_quit) return; - frame = move(pending_video_frames[current_frame_display]); - pending_video_frames.erase(current_frame_display); - } - - // Wait for the GPU to be done with the frame. - glClientWaitSync(frame.fence.get(), 0, 0); - - // Release back any input frames we needed to render this frame. - frame.input_frames.clear(); - - // Unmap the image. - GLSurface *surf = &gl_surfaces[current_frame_display % SURFACE_NUM]; - eglDestroyImageKHR(eglGetCurrentDisplay(), surf->y_egl_image); - eglDestroyImageKHR(eglGetCurrentDisplay(), surf->cbcr_egl_image); - VAStatus va_status = vaReleaseBufferHandle(va_dpy, surf->surface_image.buf); - CHECK_VASTATUS(va_status, "vaReleaseBufferHandle"); - va_status = vaDestroyImage(va_dpy, surf->surface_image.image_id); - CHECK_VASTATUS(va_status, "vaDestroyImage"); - - VASurfaceID surface = surf->src_surface; - - // Schedule the frame for encoding. - va_status = vaBeginPicture(va_dpy, context_id, surface); - CHECK_VASTATUS(va_status, "vaBeginPicture"); - - if (current_frame_type == FRAME_IDR) { - render_sequence(); - render_picture(); - if (h264_packedheader) { - render_packedsequence(); - render_packedpicture(); + if (copy_thread_should_quit) { + return; + } else { + frame = move(pending_video_frames[current_frame_display]); + pending_video_frames.erase(current_frame_display); } - } else { - //render_sequence(); - render_picture(); } - render_slice(); - - va_status = vaEndPicture(va_dpy, context_id); - CHECK_VASTATUS(va_status, "vaEndPicture"); - // Determine the pts and dts of this frame. - int64_t pts = frame.pts; + // Determine the dts of this frame. 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; + dts = frame.pts - pts_lag; } last_dts = dts; - // so now the data is done encoding (well, async job kicked off)... - // we send that to the storage thread - storage_task tmp; - tmp.display_order = current_frame_display; - tmp.frame_type = current_frame_type; - tmp.pts = pts; - tmp.dts = dts; - storage_task_enqueue(move(tmp)); - - update_ReferenceFrames(); + encode_frame(frame, frame.pts, dts); ++current_frame_encoding; } } + +void H264Encoder::encode_frame(H264Encoder::PendingFrame frame, int64_t pts, int64_t dts) +{ + // Wait for the GPU to be done with the frame. + glClientWaitSync(frame.fence.get(), 0, 0); + + // Release back any input frames we needed to render this frame. + frame.input_frames.clear(); + + // Unmap the image. + GLSurface *surf = &gl_surfaces[current_frame_display % SURFACE_NUM]; + eglDestroyImageKHR(eglGetCurrentDisplay(), surf->y_egl_image); + eglDestroyImageKHR(eglGetCurrentDisplay(), surf->cbcr_egl_image); + VAStatus va_status = vaReleaseBufferHandle(va_dpy, surf->surface_image.buf); + CHECK_VASTATUS(va_status, "vaReleaseBufferHandle"); + va_status = vaDestroyImage(va_dpy, surf->surface_image.image_id); + CHECK_VASTATUS(va_status, "vaDestroyImage"); + + VASurfaceID surface = surf->src_surface; + + // Schedule the frame for encoding. + va_status = vaBeginPicture(va_dpy, context_id, surface); + CHECK_VASTATUS(va_status, "vaBeginPicture"); + + if (current_frame_type == FRAME_IDR) { + render_sequence(); + render_picture(); + if (h264_packedheader) { + render_packedsequence(); + render_packedpicture(); + } + } else { + //render_sequence(); + render_picture(); + } + render_slice(); + + va_status = vaEndPicture(va_dpy, context_id); + CHECK_VASTATUS(va_status, "vaEndPicture"); + + // so now the data is done encoding (well, async job kicked off)... + // we send that to the storage thread + storage_task tmp; + tmp.display_order = current_frame_display; + tmp.frame_type = current_frame_type; + tmp.pts = pts; + tmp.dts = dts; + storage_task_enqueue(move(tmp)); + + update_ReferenceFrames(); +}