]> git.sesse.net Git - nageru/blobdiff - mixer.cpp
Refcount the input frames directly instead of trying to free them after-the-fact...
[nageru] / mixer.cpp
index db5f3a322d479a39ea63b53a188d2268807ece4e..2ae8276ae7ef3780885c54ccc5b1f390778390a4 100644 (file)
--- a/mixer.cpp
+++ b/mixer.cpp
@@ -50,32 +50,43 @@ using namespace std::placeholders;
 
 Mixer *global_mixer = nullptr;
 
-Mixer::Mixer(QSurface *surface1, QSurface *surface2, QSurface *surface3, QSurface *surface4)
-       : surface1(surface1), surface2(surface2), surface3(surface3), surface4(surface4)
+Mixer::Mixer(const QSurfaceFormat &format)
+       : mixer_surface(create_surface(format)),
+         h264_encoder_surface(create_surface(format))
 {
        CHECK(init_movit(MOVIT_SHADER_DIR, MOVIT_DEBUG_OFF));
        check_error();
 
-       chain.reset(new EffectChain(WIDTH, HEIGHT));
-       check_error();
+       resource_pool.reset(new ResourcePool);
+       output_channel[OUTPUT_LIVE].parent = this;
+       output_channel[OUTPUT_PREVIEW].parent = this;
 
        ImageFormat inout_format;
        inout_format.color_space = COLORSPACE_sRGB;
        inout_format.gamma_curve = GAMMA_sRGB;
 
-       YCbCrFormat ycbcr_format;
-       ycbcr_format.chroma_subsampling_x = 2;
-       ycbcr_format.chroma_subsampling_y = 1;
-       ycbcr_format.cb_x_position = 0.0;
-       ycbcr_format.cr_x_position = 0.0;
-       ycbcr_format.cb_y_position = 0.5;
-       ycbcr_format.cr_y_position = 0.5;
-       ycbcr_format.luma_coefficients = YCBCR_REC_601;
-       ycbcr_format.full_range = false;
-
-       input[0] = new YCbCrInput(inout_format, ycbcr_format, WIDTH, HEIGHT, YCBCR_INPUT_SPLIT_Y_AND_CBCR);
+       YCbCrFormat input_ycbcr_format;
+       input_ycbcr_format.chroma_subsampling_x = 2;
+       input_ycbcr_format.chroma_subsampling_y = 1;
+       input_ycbcr_format.cb_x_position = 0.0;
+       input_ycbcr_format.cr_x_position = 0.0;
+       input_ycbcr_format.cb_y_position = 0.5;
+       input_ycbcr_format.cr_y_position = 0.5;
+       input_ycbcr_format.luma_coefficients = YCBCR_REC_601;
+       input_ycbcr_format.full_range = false;
+
+       YCbCrFormat output_ycbcr_format;
+       output_ycbcr_format.chroma_subsampling_x = 1;
+       output_ycbcr_format.chroma_subsampling_y = 1;
+       output_ycbcr_format.luma_coefficients = YCBCR_REC_601;
+       output_ycbcr_format.full_range = false;
+
+       // Main chain.
+       chain.reset(new EffectChain(WIDTH, HEIGHT, resource_pool.get()));
+       check_error();
+       input[0] = new YCbCrInput(inout_format, input_ycbcr_format, WIDTH, HEIGHT, YCBCR_INPUT_SPLIT_Y_AND_CBCR);
        chain->add_input(input[0]);
-       input[1] = new YCbCrInput(inout_format, ycbcr_format, WIDTH, HEIGHT, YCBCR_INPUT_SPLIT_Y_AND_CBCR);
+       input[1] = new YCbCrInput(inout_format, input_ycbcr_format, WIDTH, HEIGHT, YCBCR_INPUT_SPLIT_Y_AND_CBCR);
        chain->add_input(input[1]);
        resample_effect = chain->add_effect(new ResampleEffect(), input[0]);
        padding_effect = chain->add_effect(new IntegralPaddingEffect());
@@ -91,28 +102,40 @@ Mixer::Mixer(QSurface *surface1, QSurface *surface2, QSurface *surface3, QSurfac
 
        chain->add_effect(new OverlayEffect(), padding_effect, padding2_effect);
 
-       ycbcr_format.chroma_subsampling_x = 1;
-
        chain->add_output(inout_format, OUTPUT_ALPHA_FORMAT_POSTMULTIPLIED);
-       chain->add_ycbcr_output(inout_format, OUTPUT_ALPHA_FORMAT_POSTMULTIPLIED, ycbcr_format, YCBCR_OUTPUT_SPLIT_Y_AND_CBCR);
+       chain->add_ycbcr_output(inout_format, OUTPUT_ALPHA_FORMAT_POSTMULTIPLIED, output_ycbcr_format, YCBCR_OUTPUT_SPLIT_Y_AND_CBCR);
        chain->set_dither_bits(8);
        chain->set_output_origin(OUTPUT_ORIGIN_TOP_LEFT);
        chain->finalize();
 
-       h264_encoder.reset(new H264Encoder(surface2, WIDTH, HEIGHT, "test.mp4"));
+       // Preview chain (always shows just first input for now).
+       preview_chain.reset(new EffectChain(WIDTH, HEIGHT, resource_pool.get()));
+       check_error();
+       preview_input = new YCbCrInput(inout_format, input_ycbcr_format, WIDTH, HEIGHT, YCBCR_INPUT_SPLIT_Y_AND_CBCR);
+       preview_chain->add_input(preview_input);
+       preview_chain->add_output(inout_format, OUTPUT_ALPHA_FORMAT_POSTMULTIPLIED);
+       preview_chain->set_dither_bits(0);  // Don't bother.
+       preview_chain->set_output_origin(OUTPUT_ORIGIN_TOP_LEFT);
+       preview_chain->finalize();
+
+       h264_encoder.reset(new H264Encoder(h264_encoder_surface, WIDTH, HEIGHT, "test.mp4"));
 
        printf("Configuring first card...\n");
        cards[0].usb = new BMUSBCapture(0x1edb, 0xbd3b);  // 0xbd4f
        cards[0].usb->set_frame_callback(std::bind(&Mixer::bm_frame, this, 0, _1, _2, _3, _4, _5, _6, _7));
-       cards[0].frame_allocator.reset(new PBOFrameAllocator(1280 * 750 * 2 + 44));
+       cards[0].frame_allocator.reset(new PBOFrameAllocator(1280 * 750 * 2 + 44, 1280, 720));
        cards[0].usb->set_video_frame_allocator(cards[0].frame_allocator.get());
        cards[0].usb->configure_card();
+       cards[0].surface = create_surface(format);
+#if NUM_CARDS == 2
+       cards[1].surface = create_surface(format);
+#endif
 
        if (NUM_CARDS == 2) {
                printf("Configuring second card...\n");
                cards[1].usb = new BMUSBCapture(0x1edb, 0xbd4f);
                cards[1].usb->set_frame_callback(std::bind(&Mixer::bm_frame, this, 1, _1, _2, _3, _4, _5, _6, _7));
-               cards[1].frame_allocator.reset(new PBOFrameAllocator(1280 * 750 * 2 + 44));
+               cards[1].frame_allocator.reset(new PBOFrameAllocator(1280 * 750 * 2 + 44, 1280, 720));
                cards[1].usb->set_video_frame_allocator(cards[1].frame_allocator.get());
                cards[1].usb->configure_card();
        }
@@ -128,7 +151,6 @@ Mixer::Mixer(QSurface *surface1, QSurface *surface2, QSurface *surface3, QSurfac
        //chain->enable_phase_timing(true);
 
        // Set up stuff for NV12 conversion.
-       resource_pool = chain->get_resource_pool();
 
        // Cb/Cr shader.
        string cbcr_vert_shader = read_file("vs-cbcr.130.vert");
@@ -179,7 +201,8 @@ void Mixer::bm_frame(int card_index, uint16_t timecode,
                std::unique_lock<std::mutex> lock(bmusb_mutex);
                card->new_data_ready_changed.wait(lock, [card]{ return !card->new_data_ready; });
        }
-       GLuint pbo = (GLint)(intptr_t)video_frame.userdata;
+       const PBOFrameAllocator::Userdata *userdata = (const PBOFrameAllocator::Userdata *)video_frame.userdata;
+       GLuint pbo = userdata->pbo;
        check_error();
        glBindBuffer(GL_PIXEL_UNPACK_BUFFER_ARB, pbo);
        check_error();
@@ -187,18 +210,30 @@ void Mixer::bm_frame(int card_index, uint16_t timecode,
        check_error();
        //glMemoryBarrier(GL_CLIENT_MAPPED_BUFFER_BARRIER_BIT);
        //check_error();
+
+       // Upload the textures.
+       glBindTexture(GL_TEXTURE_2D, userdata->tex_y);
+       check_error();
+       glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, 1280, 720, GL_RED, GL_UNSIGNED_BYTE, BUFFER_OFFSET((1280 * 750 * 2 + 44) / 2 + 1280 * 25 + 22));
+       check_error();
+       glBindTexture(GL_TEXTURE_2D, userdata->tex_cbcr);
+       check_error();
+       glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, 1280/2, 720, GL_RG, GL_UNSIGNED_BYTE, BUFFER_OFFSET(1280 * 25 + 22));
+       check_error();
+       glBindTexture(GL_TEXTURE_2D, 0);
+       check_error();
        GLsync fence = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, /*flags=*/0);              
        check_error();
        assert(fence != nullptr);
        {
                std::unique_lock<std::mutex> lock(bmusb_mutex);
                card->new_data_ready = true;
-               card->new_frame = video_frame;
+               card->new_frame = RefCountedFrame(video_frame);
                card->new_data_ready_fence = fence;
                card->new_data_ready_changed.notify_all();
        }
 
-       // Video frame will be released later.
+       // Video frame will be released when last user of card->new_frame goes out of scope.
         card->usb->get_audio_frame_allocator()->release_frame(audio_frame);
 }
        
@@ -266,14 +301,9 @@ void Mixer::place_rectangle(Effect *resample_effect, Effect *padding_effect, flo
        
 void Mixer::thread_func()
 {
-       cards[0].surface = surface3;
-#if NUM_CARDS == 2
-       cards[1].surface = surface4;
-#endif
-
        eglBindAPI(EGL_OPENGL_API);
        QOpenGLContext *context = create_context();
-       if (!make_current(context, surface1)) {
+       if (!make_current(context, mixer_surface)) {
                printf("oops\n");
                exit(1);
        }
@@ -281,10 +311,6 @@ void Mixer::thread_func()
        struct timespec start, now;
        clock_gettime(CLOCK_MONOTONIC, &start);
 
-       GLuint vao;
-       glGenVertexArrays(1, &vao);
-       check_error();
-
        while (!should_quit) {
                ++frame;
 
@@ -367,20 +393,11 @@ void Mixer::thread_func()
                        }
                }
 
-               vector<FrameAllocator::Frame> input_frames_to_release;
-       
                for (int card_index = 0; card_index < NUM_CARDS; ++card_index) {
                        CaptureCard *card = &card_copy[card_index];
                        if (!card->new_data_ready)
                                continue;
 
-                       // Now we're done with the previous frame, so we can definitely
-                       // release it when this is done rendering. (Actually, we could do
-                       // it one frame earlier, but before we have a new one, there's no
-                       // knowing when the current one is released.)
-                       if (bmusb_current_rendering_frame[card_index].owner != nullptr) {
-                               input_frames_to_release.push_back(bmusb_current_rendering_frame[card_index]);
-                       }
                        bmusb_current_rendering_frame[card_index] = card->new_frame;
                        check_error();
 
@@ -391,14 +408,19 @@ void Mixer::thread_func()
                        check_error();
                        glDeleteSync(card->new_data_ready_fence);
                        check_error();
-                       GLint input_tex_pbo = (GLint)(intptr_t)card->new_frame.userdata;
-                       input[card_index]->set_pixel_data(0, (unsigned char *)BUFFER_OFFSET((1280 * 750 * 2 + 44) / 2 + 1280 * 25 + 22), input_tex_pbo);
-                       input[card_index]->set_pixel_data(1, (unsigned char *)BUFFER_OFFSET(1280 * 25 + 22), input_tex_pbo);
+                       const PBOFrameAllocator::Userdata *userdata = (const PBOFrameAllocator::Userdata *)card->new_frame->userdata;
+                       input[card_index]->set_texture_num(0, userdata->tex_y);
+                       input[card_index]->set_texture_num(1, userdata->tex_cbcr);
 
                        if (NUM_CARDS == 1) {
                                // Set to the other one, too.
-                               input[1]->set_pixel_data(0, (unsigned char *)BUFFER_OFFSET((1280 * 750 * 2 + 44) / 2 + 1280 * 25 + 22), input_tex_pbo);
-                               input[1]->set_pixel_data(1, (unsigned char *)BUFFER_OFFSET(1280 * 25 + 22), input_tex_pbo);
+                               input[1]->set_texture_num(0, userdata->tex_y);
+                               input[1]->set_texture_num(1, userdata->tex_cbcr);
+                       }
+
+                       if (card_index == 0) {
+                               preview_input->set_texture_num(0, userdata->tex_y);
+                               preview_input->set_texture_num(1, userdata->tex_cbcr);
                        }
                }
 
@@ -406,83 +428,36 @@ void Mixer::thread_func()
                bool got_frame = h264_encoder->begin_frame(&y_tex, &cbcr_tex);
                assert(got_frame);
 
-               GLuint chroma_tex = resource_pool->create_2d_texture(GL_RG8, WIDTH, HEIGHT);
-
-               // Render chain.
-               GLuint rgba_tex = resource_pool->create_2d_texture(GL_RGBA8, WIDTH, HEIGHT);
-               GLuint ycbcr_fbo = resource_pool->create_fbo(y_tex, chroma_tex, rgba_tex);
-               chain->render_to_fbo(ycbcr_fbo, WIDTH, HEIGHT);
-               resource_pool->release_fbo(ycbcr_fbo);
-
-               // Set up for extraction.
-               float vertices[] = {
-                       0.0f, 2.0f,
-                       0.0f, 0.0f,
-                       2.0f, 0.0f
-               };
-
-               glBindVertexArray(vao);
-               check_error();
-
-               // Extract Cb/Cr.
-               GLuint cbcr_fbo = resource_pool->create_fbo(cbcr_tex);
-               glBindFramebuffer(GL_FRAMEBUFFER, cbcr_fbo);
-               glViewport(0, 0, WIDTH/2, HEIGHT/2);
-               check_error();
-
-               glUseProgram(cbcr_program_num);
-               check_error();
-
-               glActiveTexture(GL_TEXTURE0);
-               check_error();
-               glBindTexture(GL_TEXTURE_2D, chroma_tex);
-               check_error();
-               glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
-               check_error();
-               glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
-               check_error();
-               glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
-               check_error();
+               // Render main chain.
+               GLuint cbcr_full_tex = resource_pool->create_2d_texture(GL_RG8, WIDTH, HEIGHT);
+               GLuint rgba_tex = resource_pool->create_2d_texture(GL_RGB565, WIDTH, HEIGHT);  // Saves texture bandwidth, although dithering gets messed up.
+               GLuint fbo = resource_pool->create_fbo(y_tex, cbcr_full_tex, rgba_tex);
+               chain->render_to_fbo(fbo, WIDTH, HEIGHT);
+               resource_pool->release_fbo(fbo);
 
-               float chroma_offset_0[] = { -0.5f / WIDTH, 0.0f };
-               set_uniform_vec2(cbcr_program_num, "foo", "chroma_offset_0", chroma_offset_0);
+               subsample_chroma(cbcr_full_tex, cbcr_tex);
+               resource_pool->release_2d_texture(cbcr_full_tex);
 
-               GLuint position_vbo = fill_vertex_attribute(cbcr_program_num, "position", 2, GL_FLOAT, sizeof(vertices), vertices);
-               GLuint texcoord_vbo = fill_vertex_attribute(cbcr_program_num, "texcoord", 2, GL_FLOAT, sizeof(vertices), vertices);  // Same as vertices.
-
-               glDrawArrays(GL_TRIANGLES, 0, 3);
-               check_error();
+               // Render preview chain.
+               GLuint preview_rgba_tex = resource_pool->create_2d_texture(GL_RGB565, output_channel[OUTPUT_PREVIEW].width, output_channel[OUTPUT_PREVIEW].height);  // Saves texture bandwidth, although dithering gets messed up.
+               fbo = resource_pool->create_fbo(preview_rgba_tex);
+               preview_chain->render_to_fbo(fbo, output_channel[OUTPUT_PREVIEW].width, output_channel[OUTPUT_PREVIEW].height);
+               resource_pool->release_fbo(fbo);
 
                RefCountedGLsync fence(GL_SYNC_GPU_COMMANDS_COMPLETE, /*flags=*/0);
                check_error();
 
-               cleanup_vertex_attribute(cbcr_program_num, "position", position_vbo);
-               cleanup_vertex_attribute(cbcr_program_num, "texcoord", texcoord_vbo);
-
-               glUseProgram(0);
-               check_error();
-
-               resource_pool->release_fbo(cbcr_fbo);
-               resource_pool->release_2d_texture(chroma_tex);
-
-               h264_encoder->end_frame(fence, input_frames_to_release);
-
-               // Store this frame for display. Remove the ready frame if any
-               // (it was seemingly never used).
-               {
-                       std::unique_lock<std::mutex> lock(display_frame_mutex);
-                       if (has_ready_display_frame) {
-                               resource_pool->release_2d_texture(ready_display_frame.texnum);
-                               ready_display_frame.ready_fence.reset();
-                       }
-                       ready_display_frame.texnum = rgba_tex;
-                       ready_display_frame.ready_fence = fence;
-                       has_ready_display_frame = true;
+               // Make sure the H.264 gets a reference to all the
+               // input frames needed, so that they are not released back
+               // until the rendering is done.
+               vector<RefCountedFrame> input_frames;
+               for (int card_index = 0; card_index < NUM_CARDS; ++card_index) {
+                       input_frames.push_back(bmusb_current_rendering_frame[card_index]);
                }
+               h264_encoder->end_frame(fence, input_frames);
 
-               if (has_new_frame_ready_callback) {
-                       new_frame_ready_callback();
-               }
+               output_channel[OUTPUT_LIVE].output_frame(rgba_tex, fence);
+               output_channel[OUTPUT_PREVIEW].output_frame(preview_rgba_tex, fence);
 
                clock_gettime(CLOCK_MONOTONIC, &now);
                double elapsed = now.tv_sec - start.tv_sec +
@@ -504,38 +479,67 @@ void Mixer::thread_func()
                }
                check_error();
        }
-       glDeleteVertexArrays(1, &vao);
 }
 
-bool Mixer::get_display_frame(DisplayFrame *frame)
+void Mixer::subsample_chroma(GLuint src_tex, GLuint dst_tex)
 {
-       std::unique_lock<std::mutex> lock(display_frame_mutex);
-       if (!has_current_display_frame && !has_ready_display_frame) {
-               return false;
-       }
+       GLuint vao;
+       glGenVertexArrays(1, &vao);
+       check_error();
 
-       if (has_current_display_frame && has_ready_display_frame) {
-               // We have a new ready frame. Toss the current one.
-               resource_pool->release_2d_texture(current_display_frame.texnum);
-               current_display_frame.ready_fence.reset();
-               has_current_display_frame = false;
-       }
-       if (has_ready_display_frame) {
-               assert(!has_current_display_frame);
-               current_display_frame = ready_display_frame;
-               ready_display_frame.ready_fence.reset();  // Drop the refcount.
-               has_current_display_frame = true;
-               has_ready_display_frame = false;
-       }
+       float vertices[] = {
+               0.0f, 2.0f,
+               0.0f, 0.0f,
+               2.0f, 0.0f
+       };
 
-       *frame = current_display_frame;
-       return true;
+       glBindVertexArray(vao);
+       check_error();
+
+       // Extract Cb/Cr.
+       GLuint fbo = resource_pool->create_fbo(dst_tex);
+       glBindFramebuffer(GL_FRAMEBUFFER, fbo);
+       glViewport(0, 0, WIDTH/2, HEIGHT/2);
+       check_error();
+
+       glUseProgram(cbcr_program_num);
+       check_error();
+
+       glActiveTexture(GL_TEXTURE0);
+       check_error();
+       glBindTexture(GL_TEXTURE_2D, src_tex);
+       check_error();
+       glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
+       check_error();
+       glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
+       check_error();
+       glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
+       check_error();
+
+       float chroma_offset_0[] = { -0.5f / WIDTH, 0.0f };
+       set_uniform_vec2(cbcr_program_num, "foo", "chroma_offset_0", chroma_offset_0);
+
+       GLuint position_vbo = fill_vertex_attribute(cbcr_program_num, "position", 2, GL_FLOAT, sizeof(vertices), vertices);
+       GLuint texcoord_vbo = fill_vertex_attribute(cbcr_program_num, "texcoord", 2, GL_FLOAT, sizeof(vertices), vertices);  // Same as vertices.
+
+       glDrawArrays(GL_TRIANGLES, 0, 3);
+       check_error();
+
+       cleanup_vertex_attribute(cbcr_program_num, "position", position_vbo);
+       cleanup_vertex_attribute(cbcr_program_num, "texcoord", texcoord_vbo);
+
+       glUseProgram(0);
+       check_error();
+
+       resource_pool->release_fbo(fbo);
+       glDeleteVertexArrays(1, &vao);
 }
 
-void Mixer::set_frame_ready_fallback(new_frame_ready_callback_t callback)
+void Mixer::release_display_frame(DisplayFrame *frame)
 {
-       new_frame_ready_callback = callback;
-       has_new_frame_ready_callback = true;
+       resource_pool->release_2d_texture(frame->texnum);
+       frame->texnum = 0;
+       frame->ready_fence.reset();
 }
 
 void Mixer::start()
@@ -553,3 +557,58 @@ void Mixer::cut(Source source)
 {
        current_source = source;
 }
+
+void Mixer::OutputChannel::output_frame(GLuint tex, RefCountedGLsync fence)
+{
+       // Store this frame for display. Remove the ready frame if any
+       // (it was seemingly never used).
+       {
+               std::unique_lock<std::mutex> lock(frame_mutex);
+               if (has_ready_frame) {
+                       parent->release_display_frame(&ready_frame);
+               }
+               ready_frame.texnum = tex;
+               ready_frame.ready_fence = fence;
+               has_ready_frame = true;
+       }
+
+       if (has_new_frame_ready_callback) {
+               new_frame_ready_callback();
+       }
+}
+
+bool Mixer::OutputChannel::get_display_frame(DisplayFrame *frame)
+{
+       std::unique_lock<std::mutex> lock(frame_mutex);
+       if (!has_current_frame && !has_ready_frame) {
+               return false;
+       }
+
+       if (has_current_frame && has_ready_frame) {
+               // We have a new ready frame. Toss the current one.
+               parent->release_display_frame(&current_frame);
+               has_current_frame = false;
+       }
+       if (has_ready_frame) {
+               assert(!has_current_frame);
+               current_frame = ready_frame;
+               ready_frame.ready_fence.reset();  // Drop the refcount.
+               has_current_frame = true;
+               has_ready_frame = false;
+       }
+
+       *frame = current_frame;
+       return true;
+}
+
+void Mixer::OutputChannel::set_frame_ready_callback(Mixer::new_frame_ready_callback_t callback)
+{
+       new_frame_ready_callback = callback;
+       has_new_frame_ready_callback = true;
+}
+
+void Mixer::OutputChannel::set_size(int width, int height)
+{
+       this->width = width;
+       this->height = height;
+}