movit_texel_subpixel_precision /= 2.0;
resource_pool.reset(new ResourcePool);
- theme.reset(new Theme(global_flags.theme_filename, global_flags.theme_dirs, resource_pool.get(), num_cards));
for (unsigned i = 0; i < NUM_OUTPUTS; ++i) {
output_channel[i].parent = this;
output_channel[i].channel = i;
ycbcr_format.luma_coefficients = YCBCR_REC_601;
}
ycbcr_format.full_range = false;
- ycbcr_format.num_levels = 256;
+ ycbcr_format.num_levels = 1 << global_flags.x264_bit_depth;
ycbcr_format.cb_x_position = 0.0f;
ycbcr_format.cr_x_position = 0.0f;
ycbcr_format.cb_y_position = 0.5f;
// Display chain; shows the live output produced by the main chain (or rather, a copy of it).
display_chain.reset(new EffectChain(global_flags.width, global_flags.height, resource_pool.get()));
check_error();
- display_input = new YCbCrInput(inout_format, ycbcr_format, global_flags.width, global_flags.height, YCBCR_INPUT_SPLIT_Y_AND_CBCR);
+ GLenum type = global_flags.x264_bit_depth > 8 ? GL_UNSIGNED_SHORT : GL_UNSIGNED_BYTE;
+ display_input = new YCbCrInput(inout_format, ycbcr_format, global_flags.width, global_flags.height, YCBCR_INPUT_SPLIT_Y_AND_CBCR, type);
display_chain->add_input(display_input);
display_chain->add_output(inout_format, OUTPUT_ALPHA_FORMAT_POSTMULTIPLIED);
display_chain->set_dither_bits(0); // Don't bother.
video_encoder.reset(new VideoEncoder(resource_pool.get(), h264_encoder_surface, global_flags.va_display, global_flags.width, global_flags.height, &httpd, global_disk_space_estimator));
+ // Must be instantiated after VideoEncoder has initialized global_flags.use_zerocopy.
+ theme.reset(new Theme(global_flags.theme_filename, global_flags.theme_dirs, resource_pool.get(), num_cards));
+
// Start listening for clients only once VideoEncoder has written its header, if any.
httpd.start(9095);
v210_converter->precompile_shader(3840);
v210_converter->precompile_shader(4096);
}
+ if (global_flags.ten_bit_output) {
+ if (!v210Converter::has_hardware_support()) {
+ fprintf(stderr, "ERROR: --ten-bit-output requires support for OpenGL compute shaders\n");
+ fprintf(stderr, " (OpenGL 4.3, or GL_ARB_compute_shader + GL_ARB_shader_image_load_store).\n");
+ exit(1);
+ }
+ }
timecode_renderer.reset(new TimecodeRenderer(resource_pool.get(), global_flags.width, global_flags.height));
display_timecode_in_stream = global_flags.display_timecode_in_stream;
++dropped_frames;
}
+#if 0
if (dropped_frames > 0) {
fprintf(stderr, "Card %u dropped %u frame(s) to keep latency down.\n",
card_index, dropped_frames);
}
+#endif
}
output_ycbcr_format.chroma_subsampling_y = 1;
output_ycbcr_format.luma_coefficients = ycbcr_output_coefficients;
output_ycbcr_format.full_range = false;
- output_ycbcr_format.num_levels = 256;
+ output_ycbcr_format.num_levels = 1 << global_flags.x264_bit_depth;
chain->change_ycbcr_output_format(output_ycbcr_format);
- const int64_t av_delay = lrint(global_flags.audio_queue_length_ms * 0.001 * TIMEBASE); // Corresponds to the delay in ResamplingQueue.
- GLuint y_tex, cbcr_tex;
- bool got_frame = video_encoder->begin_frame(pts_int + av_delay, duration, ycbcr_output_coefficients, theme_main_chain.input_frames, &y_tex, &cbcr_tex);
- assert(got_frame);
-
- // Render main chain. We take an extra copy of the created outputs,
+ // Render main chain. If we're using zerocopy Quick Sync encoding
+ // (the default case), we take an extra copy of the created outputs,
// so that we can display it back to the screen later (it's less memory
// bandwidth than writing and reading back an RGBA texture, even at 16-bit).
// Ideally, we'd like to avoid taking copies and just use the main textures
- // for display as well, but if they're used for zero-copy Quick Sync encoding
- // (the default case), they're just views into VA-API memory and must be
+ // for display as well, but they're just views into VA-API memory and must be
// unmapped during encoding, so we can't use them for display, unfortunately.
- GLuint cbcr_full_tex = resource_pool->create_2d_texture(GL_RG8, global_flags.width, global_flags.height);
- GLuint y_copy_tex = resource_pool->create_2d_texture(GL_R8, global_flags.width, global_flags.height);
- GLuint fbo = resource_pool->create_fbo(y_tex, cbcr_full_tex, y_copy_tex);
+ GLuint y_tex, cbcr_full_tex, cbcr_tex;
+ GLuint y_copy_tex, cbcr_copy_tex = 0;
+ GLuint y_display_tex, cbcr_display_tex;
+ GLenum y_type = (global_flags.x264_bit_depth > 8) ? GL_R16 : GL_R8;
+ GLenum cbcr_type = (global_flags.x264_bit_depth > 8) ? GL_RG16 : GL_RG8;
+ const bool is_zerocopy = video_encoder->is_zerocopy();
+ if (is_zerocopy) {
+ cbcr_full_tex = resource_pool->create_2d_texture(cbcr_type, global_flags.width, global_flags.height);
+ y_copy_tex = resource_pool->create_2d_texture(y_type, global_flags.width, global_flags.height);
+ cbcr_copy_tex = resource_pool->create_2d_texture(cbcr_type, global_flags.width / 2, global_flags.height / 2);
+
+ y_display_tex = y_copy_tex;
+ cbcr_display_tex = cbcr_copy_tex;
+
+ // y_tex and cbcr_tex will be given by VideoEncoder.
+ } else {
+ cbcr_full_tex = resource_pool->create_2d_texture(cbcr_type, global_flags.width, global_flags.height);
+ y_tex = resource_pool->create_2d_texture(y_type, global_flags.width, global_flags.height);
+ cbcr_tex = resource_pool->create_2d_texture(cbcr_type, global_flags.width / 2, global_flags.height / 2);
+
+ y_display_tex = y_tex;
+ cbcr_display_tex = cbcr_tex;
+ }
+
+ const int64_t av_delay = lrint(global_flags.audio_queue_length_ms * 0.001 * TIMEBASE); // Corresponds to the delay in ResamplingQueue.
+ bool got_frame = video_encoder->begin_frame(pts_int + av_delay, duration, ycbcr_output_coefficients, theme_main_chain.input_frames, &y_tex, &cbcr_tex);
+ assert(got_frame);
+
+ GLuint fbo;
+ if (is_zerocopy) {
+ fbo = resource_pool->create_fbo(y_tex, cbcr_full_tex, y_copy_tex);
+ } else {
+ fbo = resource_pool->create_fbo(y_tex, cbcr_full_tex);
+ }
check_error();
chain->render_to_fbo(fbo, global_flags.width, global_flags.height);
resource_pool->release_fbo(fbo);
- GLuint cbcr_copy_tex = resource_pool->create_2d_texture(GL_RG8, global_flags.width / 2, global_flags.height / 2);
- chroma_subsampler->subsample_chroma(cbcr_full_tex, global_flags.width, global_flags.height, cbcr_tex, cbcr_copy_tex);
+ if (is_zerocopy) {
+ chroma_subsampler->subsample_chroma(cbcr_full_tex, global_flags.width, global_flags.height, cbcr_tex, cbcr_copy_tex);
+ } else {
+ chroma_subsampler->subsample_chroma(cbcr_full_tex, global_flags.width, global_flags.height, cbcr_tex);
+ }
if (output_card_index != -1) {
cards[output_card_index].output->send_frame(y_tex, cbcr_full_tex, ycbcr_output_coefficients, theme_main_chain.input_frames, pts_int, duration);
}
resource_pool->release_2d_texture(cbcr_full_tex);
- // Set the right state for the Y' and CbCr copies.
+ // Set the right state for the Y' and CbCr textures we use for display.
glBindFramebuffer(GL_FRAMEBUFFER, 0);
- glBindTexture(GL_TEXTURE_2D, y_copy_tex);
+ glBindTexture(GL_TEXTURE_2D, y_display_tex);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
- glBindTexture(GL_TEXTURE_2D, cbcr_copy_tex);
+ glBindTexture(GL_TEXTURE_2D, cbcr_display_tex);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
RefCountedGLsync fence = video_encoder->end_frame();
// The live frame pieces the Y'CbCr texture copies back into RGB and displays them.
- // It owns y_copy_tex and cbcr_copy_tex now.
+ // It owns y_display_tex and cbcr_display_tex now (whichever textures they are).
DisplayFrame live_frame;
live_frame.chain = display_chain.get();
- live_frame.setup_chain = [this, y_copy_tex, cbcr_copy_tex]{
- display_input->set_texture_num(0, y_copy_tex);
- display_input->set_texture_num(1, cbcr_copy_tex);
+ live_frame.setup_chain = [this, y_display_tex, cbcr_display_tex]{
+ display_input->set_texture_num(0, y_display_tex);
+ display_input->set_texture_num(1, cbcr_display_tex);
};
live_frame.ready_fence = fence;
live_frame.input_frames = {};
- live_frame.temp_textures = { y_copy_tex, cbcr_copy_tex };
+ live_frame.temp_textures = { y_display_tex, cbcr_display_tex };
output_channel[OUTPUT_LIVE].output_frame(live_frame);
// Set up preview and any additional channels.