#include <core/producer/frame/pixel_format.h>\r
#include <core/video_format.h>\r
\r
+#include <gl/glew.h>\r
+\r
#include <boost/foreach.hpp>\r
#include <boost/range.hpp>\r
+#include <boost/range/algorithm/find.hpp>\r
\r
#include <algorithm>\r
#include <array>\r
+#include <deque>\r
#include <unordered_map>\r
\r
namespace caspar { namespace core {\r
\r
struct image_mixer::implementation : boost::noncopyable\r
{ \r
- typedef std::deque<render_item> layer;\r
+ typedef std::deque<render_item> layer;\r
\r
- video_channel_context& channel_;\r
+ video_channel_context& channel_;\r
\r
- std::vector<image_transform> transform_stack_;\r
- std::vector<video_mode::type> mode_stack_;\r
+ std::vector<image_transform> transform_stack_;\r
+ std::vector<video_mode::type> mode_stack_;\r
\r
- std::deque<std::deque<render_item>> layers_; // layer/stream/items\r
+ std::deque<std::deque<render_item>> layers_; // layer/stream/items\r
\r
- image_kernel kernel_;\r
+ image_kernel kernel_;\r
\r
- std::array<std::shared_ptr<device_buffer>,2> draw_buffer_;\r
- std::shared_ptr<device_buffer> write_buffer_;\r
-\r
- std::array<std::shared_ptr<device_buffer>,2> local_key_buffer_;\r
- std::shared_ptr<device_buffer> layer_key_buffer_;\r
- \r
+ std::shared_ptr<device_buffer> draw_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_[0] = channel_.ogl().create_device_buffer(channel_.get_format_desc().width, channel_.get_format_desc().height, 4);\r
- draw_buffer_[1] = channel_.ogl().create_device_buffer(channel_.get_format_desc().width, channel_.get_format_desc().height, 4);\r
- local_key_buffer_[0] = channel_.ogl().create_device_buffer(channel_.get_format_desc().width, channel_.get_format_desc().height, 1);\r
- local_key_buffer_[1] = channel_.ogl().create_device_buffer(channel_.get_format_desc().width, channel_.get_format_desc().height, 1);\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
\r
auto& layer = layers_.back();\r
\r
- auto it = boost::range::find(layer, item);\r
- if(it == layer.end())\r
+ if(boost::range::find(layer, item) == layer.end())\r
layer.push_back(item);\r
}\r
\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(channel_.get_format_desc().width != write_buffer_->width() || channel_.get_format_desc().height != write_buffer_->height())\r
- initialize_buffers();\r
- \r
- layer_key_buffer_->clear();\r
- draw_buffer_[0]->clear();\r
- draw_buffer_[1]->clear();\r
- local_key_buffer_[0]->clear();\r
- local_key_buffer_[1]->clear();\r
-\r
- bool layer_key = false;\r
+ std::shared_ptr<device_buffer> layer_key_buffer;\r
\r
+ draw_buffer_ = channel_.ogl().create_device_buffer(channel_.get_format_desc().width, channel_.get_format_desc().height, 4); \r
+ channel_.ogl().clear(*draw_buffer_);\r
+ \r
BOOST_FOREACH(auto& layer, layers)\r
- draw(std::move(layer), layer_key);\r
-\r
- std::swap(draw_buffer_[0], write_buffer_);\r
- \r
+ draw(std::move(layer), layer_key_buffer);\r
+ \r
auto host_buffer = channel_.ogl().create_host_buffer(channel_.get_format_desc().size, host_buffer::read_only);\r
- host_buffer->begin_read(*write_buffer_);\r
+ channel_.ogl().attach(*draw_buffer_);\r
+ host_buffer->begin_read(draw_buffer_->width(), draw_buffer_->height(), format(draw_buffer_->stride()));\r
\r
GL(glFlush());\r
-\r
+ \r
return host_buffer;\r
}\r
\r
- void draw(layer&& layer, bool& layer_key)\r
- { \r
- bool local_key = false;\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, std::shared_ptr<device_buffer>& layer_key_buffer)\r
+ { \r
+ if(layer.empty())\r
+ return;\r
\r
- local_key_buffer_[0]->clear();\r
+ 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 = channel_.ogl().create_device_buffer(channel_.get_format_desc().width, channel_.get_format_desc().height, 4); \r
+ channel_.ogl().clear(*local_draw_buffer);\r
\r
- BOOST_FOREACH(auto& item, layer)\r
- draw(std::move(item), local_key, layer_key);\r
- \r
- layer_key = local_key;\r
+ BOOST_FOREACH(auto& item, layer) \r
+ draw_item(std::move(item), local_draw_buffer, local_key_buffer, layer_key_buffer); \r
+ \r
+ kernel_.draw(channel_.ogl(), create_render_item(local_draw_buffer, layer.front().transform.get_blend_mode()), make_safe(draw_buffer_), nullptr, nullptr);\r
+ }\r
+ else // fast path\r
+ {\r
+ BOOST_FOREACH(auto& item, layer) \r
+ draw_item(std::move(item), make_safe(draw_buffer_), local_key_buffer, layer_key_buffer); \r
+ } \r
\r
- std::swap(local_key_buffer_[0], layer_key_buffer_);\r
+ std::swap(local_key_buffer, layer_key_buffer);\r
}\r
\r
- void draw(render_item&& item, bool& local_key, bool& layer_key)\r
+ void draw_item(render_item&& item, const safe_ptr<device_buffer>& draw_buffer, std::shared_ptr<device_buffer>& local_key_buffer, 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)\r
+ {\r
+ local_key_buffer = channel_.ogl().create_device_buffer(channel_.get_format_desc().width, channel_.get_format_desc().height, 1);\r
+ channel_.ogl().clear(*local_key_buffer);\r
+ }\r
+\r
+ kernel_.draw(channel_.ogl(), std::move(item), make_safe(local_key_buffer), nullptr, nullptr);\r
}\r
else\r
{\r
- draw(draw_buffer_, std::move(item), local_key ? local_key_buffer_[0] : nullptr, layer_key ? layer_key_buffer_ : nullptr); \r
- local_key_buffer_[0]->clear();\r
- local_key = false;\r
+ kernel_.draw(channel_.ogl(), std::move(item), draw_buffer, local_key_buffer, layer_key_buffer);\r
+ local_key_buffer.reset();\r
}\r
- channel_.ogl().yield(); // Return resources to pool as early as possible.\r
}\r
- \r
- void draw(std::array<std::shared_ptr<device_buffer>,2>& targets, render_item&& item, const std::shared_ptr<device_buffer>& local_key, const std::shared_ptr<device_buffer>& layer_key)\r
+ \r
+ render_item create_render_item(const safe_ptr<device_buffer>& buffer, image_transform::blend_mode::type blend_mode)\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
+ CASPAR_ASSERT(buffer->stride() == 4 && "Only used for bgra textures");\r
\r
- targets[1]->attach();\r
- \r
- kernel_.draw(item, make_safe(targets[0]), local_key, layer_key);\r
- \r
- targets[0]->bind();\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
- glCopyTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, 0, 0, targets[0]->width(), targets[0]->height());\r
- \r
- std::swap(targets[0], targets[1]);\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
+ // TODO: Optimize\r
+ bool has_overlapping_items(const layer& layer, image_transform::blend_mode::type blend_mode)\r
+ {\r
+ if(layer.empty())\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
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