\r
std::deque<std::deque<render_item>> layers_; // layer/stream/items\r
\r
- image_kernel kernel_;\r
- \r
- std::shared_ptr<device_buffer> draw_buffer_;\r
- std::shared_ptr<device_buffer> write_buffer_;\r
-\r
- std::shared_ptr<device_buffer> local_key_buffer_;\r
- std::shared_ptr<device_buffer> layer_key_buffer_;\r
+ image_kernel kernel_; \r
\r
- std::shared_ptr<host_buffer> empty_buffer_;\r
- \r
+ std::shared_ptr<device_buffer> active_buffer_;\r
public:\r
implementation(video_channel_context& video_channel) \r
: channel_(video_channel)\r
, transform_stack_(1)\r
, mode_stack_(1, video_mode::progressive)\r
{\r
- initialize_buffers();\r
}\r
\r
~implementation()\r
{\r
channel_.ogl().gc();\r
}\r
-\r
- void initialize_buffers()\r
- {\r
- write_buffer_ = channel_.ogl().create_device_buffer(channel_.get_format_desc().width, channel_.get_format_desc().height, 4);\r
- layer_key_buffer_ = channel_.ogl().create_device_buffer(channel_.get_format_desc().width, channel_.get_format_desc().height, 1);\r
- draw_buffer_ = channel_.ogl().create_device_buffer(channel_.get_format_desc().width, channel_.get_format_desc().height, 4);\r
- local_key_buffer_ = channel_.ogl().create_device_buffer(channel_.get_format_desc().width, channel_.get_format_desc().height, 1);\r
-\r
- empty_buffer_ = channel_.ogl().create_host_buffer(channel_.get_format_desc().size, host_buffer::write_only);\r
- memset(empty_buffer_->data(), 0, empty_buffer_->size());\r
-\r
- channel_.ogl().gc();\r
- }\r
-\r
+ \r
void begin(core::basic_frame& frame)\r
{\r
transform_stack_.push_back(transform_stack_.back()*frame.get_image_transform());\r
boost::unique_future<safe_ptr<host_buffer>> render()\r
{ \r
auto layers = std::move(layers_);\r
- return channel_.ogl().begin_invoke([=]()mutable{return render(std::move(layers));});\r
+ return channel_.ogl().begin_invoke([=]() mutable\r
+ {\r
+ return render(std::move(layers));\r
+ });\r
}\r
\r
safe_ptr<host_buffer> render(std::deque<layer>&& layers)\r
{\r
- if(layers.empty())\r
- return make_safe(empty_buffer_);\r
+ std::shared_ptr<device_buffer> layer_key_buffer;\r
\r
- if(channel_.get_format_desc().width != write_buffer_->width() || channel_.get_format_desc().height != write_buffer_->height())\r
- initialize_buffers();\r
- \r
- channel_.ogl().clear(*layer_key_buffer_);\r
- channel_.ogl().clear(*local_key_buffer_);\r
- channel_.ogl().clear(*draw_buffer_);\r
- \r
- bool layer_key = false;\r
- \r
+ auto draw_buffer = create_device_buffer(4);\r
+ \r
BOOST_FOREACH(auto& layer, layers)\r
- draw(std::move(layer), layer_key);\r
-\r
- std::swap(draw_buffer_, write_buffer_);\r
+ draw(std::move(layer), draw_buffer, layer_key_buffer);\r
\r
auto host_buffer = channel_.ogl().create_host_buffer(channel_.get_format_desc().size, host_buffer::read_only);\r
- \r
- channel_.ogl().attach(*write_buffer_);\r
- host_buffer->begin_read(write_buffer_->width(), write_buffer_->height(), format(write_buffer_->stride()));\r
- \r
- GL(glFlush());\r
+ channel_.ogl().attach(*draw_buffer);\r
+ host_buffer->begin_read(draw_buffer->width(), draw_buffer->height(), format(draw_buffer->stride()));\r
\r
+ active_buffer_ = draw_buffer;\r
+\r
+ channel_.ogl().flush(); // NOTE: This is important, otherwise fences will deadlock.\r
+ \r
return host_buffer;\r
}\r
\r
- void draw(layer&& layer, bool& layer_key)\r
- { \r
- bool local_key = false;\r
- \r
- channel_.ogl().clear(*local_key_buffer_);\r
-\r
- BOOST_FOREACH(auto& item, layer)\r
- draw(std::move(item), local_key, layer_key);\r
- \r
- layer_key = local_key;\r
+ // TODO: We might have more overlaps for opacity transitions\r
+ // TODO: What about blending modes, are they ok? Maybe only overlap detection is required for opacity?\r
+ void draw(layer&& layer, const safe_ptr<device_buffer>& draw_buffer, std::shared_ptr<device_buffer>& layer_key_buffer)\r
+ { \r
+ if(layer.empty())\r
+ return;\r
\r
- std::swap(local_key_buffer_, layer_key_buffer_);\r
+ std::pair<int, std::shared_ptr<device_buffer>> local_key_buffer;\r
+ \r
+ //if(has_overlapping_items(layer, layer.front().transform.get_blend_mode()))\r
+ //{\r
+ // auto local_draw_buffer = create_device_buffer(4); \r
+\r
+ // auto local_blend_mode = layer.front().transform.get_blend_mode();\r
+\r
+ // int fields = 0;\r
+ // BOOST_FOREACH(auto& item, layer)\r
+ // {\r
+ // if(fields & item.mode)\r
+ // item.transform.set_blend_mode(image_transform::blend_mode::normal); // Disable blending, it will be used when merging back into render stack.\r
+ // else\r
+ // {\r
+ // item.transform.set_blend_mode(image_transform::blend_mode::replace); // Target field is empty, no blending, just copy\r
+ // fields |= item.mode;\r
+ // }\r
+\r
+ // draw_item(std::move(item), local_draw_buffer, local_key_buffer, layer_key_buffer); \r
+ // }\r
+\r
+ // kernel_.draw(channel_.ogl(), create_render_item(local_draw_buffer, local_blend_mode), draw_buffer, nullptr, nullptr);\r
+ //}\r
+ //else // fast path\r
+ //{\r
+ BOOST_FOREACH(auto& item, layer) \r
+ draw_item(std::move(item), draw_buffer, local_key_buffer, layer_key_buffer); \r
+ //} \r
+\r
+ CASPAR_ASSERT(local_key_buffer.first == 0 || local_key_buffer.first == core::video_mode::progressive);\r
+\r
+ std::swap(local_key_buffer.second, layer_key_buffer);\r
}\r
\r
- void draw(render_item&& item, bool& local_key, bool& layer_key)\r
+ void draw_item(render_item&& item, \r
+ const safe_ptr<device_buffer>& draw_buffer, \r
+ std::pair<int, std::shared_ptr<device_buffer>>& local_key_buffer, \r
+ std::shared_ptr<device_buffer>& layer_key_buffer)\r
{ \r
if(item.transform.get_is_key())\r
{\r
- draw(local_key_buffer_, std::move(item), nullptr, nullptr);\r
- local_key = true;\r
+ if(!local_key_buffer.second)\r
+ {\r
+ local_key_buffer.first = 0;\r
+ local_key_buffer.second = create_device_buffer(1);\r
+ }\r
+\r
+ // No transparency for key\r
+ item.transform.set_opacity(1.0);\r
+ item.transform.set_blend_mode(image_transform::blend_mode::normal);\r
+\r
+ local_key_buffer.first |= item.mode;\r
+ kernel_.draw(channel_.ogl(), std::move(item), make_safe(local_key_buffer.second), nullptr, nullptr);\r
}\r
else\r
{\r
- draw(draw_buffer_, std::move(item), local_key ? local_key_buffer_ : nullptr, layer_key ? layer_key_buffer_ : nullptr); \r
- channel_.ogl().clear(*local_key_buffer_);\r
- local_key = false;\r
+ kernel_.draw(channel_.ogl(), std::move(item), draw_buffer, local_key_buffer.second, layer_key_buffer);\r
+ local_key_buffer.first ^= item.mode;\r
+ \r
+ if(local_key_buffer.first == 0)\r
+ {\r
+ local_key_buffer.first = 0;\r
+ local_key_buffer.second.reset();\r
+ }\r
}\r
}\r
- \r
- void draw(std::shared_ptr<device_buffer>& target, render_item&& item, const std::shared_ptr<device_buffer>& local_key, const std::shared_ptr<device_buffer>& layer_key)\r
- {\r
- if(!std::all_of(item.textures.begin(), item.textures.end(), std::mem_fn(&device_buffer::ready)))\r
- {\r
- CASPAR_LOG(warning) << L"[image_mixer] Performance warning. Host to device transfer not complete, GPU will be stalled";\r
- channel_.ogl().yield(); // Try to give it some more time.\r
- } \r
\r
- kernel_.draw(channel_.ogl(), std::move(item), make_safe(target), local_key, layer_key);\r
+ //// TODO: Optimize\r
+ //bool has_overlapping_items(const layer& layer, image_transform::blend_mode::type blend_mode)\r
+ //{\r
+ // if(layer.size() < 2)\r
+ // return false; \r
+ // \r
+ // implementation::layer fill;\r
+\r
+ // std::copy_if(layer.begin(), layer.end(), std::back_inserter(fill), [&](const render_item& item)\r
+ // {\r
+ // return !item.transform.get_is_key();\r
+ // });\r
+ // \r
+ // if(blend_mode == image_transform::blend_mode::normal) // Only overlap if opacity\r
+ // {\r
+ // return std::any_of(fill.begin(), fill.end(), [&](const render_item& item)\r
+ // {\r
+ // return item.transform.get_opacity() < 1.0 - 0.001;\r
+ // });\r
+ // }\r
+\r
+ // // Simple solution, just check if we have differnt video streams / tags.\r
+ // return std::any_of(fill.begin(), fill.end(), [&](const render_item& item)\r
+ // {\r
+ // return item.tag != fill.front().tag;\r
+ // });\r
+ //} \r
+ // \r
+ //render_item create_render_item(const safe_ptr<device_buffer>& buffer, image_transform::blend_mode::type blend_mode)\r
+ //{\r
+ // CASPAR_ASSERT(buffer->stride() == 4 && "Only used for bgra textures");\r
+\r
+ // pixel_format_desc desc;\r
+ // desc.pix_fmt = pixel_format::bgra;\r
+ // desc.planes.push_back(pixel_format_desc::plane(channel_.get_format_desc().width, channel_.get_format_desc().height, 4));\r
+\r
+ // std::vector<safe_ptr<device_buffer>> textures;\r
+ // textures.push_back(buffer);\r
+ // \r
+ // image_transform transform;\r
+ // transform.set_blend_mode(blend_mode);\r
+\r
+ // return render_item(desc, std::move(textures), transform, video_mode::progressive, nullptr); \r
+ //}\r
+\r
+ safe_ptr<device_buffer> create_device_buffer(size_t stride)\r
+ {\r
+ auto buffer = channel_.ogl().create_device_buffer(channel_.get_format_desc().width, channel_.get_format_desc().height, stride);\r
+ channel_.ogl().clear(*buffer);\r
+ return buffer;\r
}\r
- \r
+\r
safe_ptr<write_frame> create_frame(const void* tag, const core::pixel_format_desc& desc)\r
{\r
return make_safe<write_frame>(channel_.ogl(), tag, desc);\r
projects / casparcg / blobdiff