#include "../../stdafx.h"\r
\r
#include "image_mixer.h"\r
-#include "image_kernel.h"\r
\r
+#include "image_kernel.h"\r
+#include "../write_frame.h"\r
#include "../gpu/ogl_device.h"\r
#include "../gpu/host_buffer.h"\r
#include "../gpu/device_buffer.h"\r
-#include "../write_frame.h"\r
-\r
#include "../../video_channel_context.h"\r
\r
-#include <common/concurrency/executor.h>\r
#include <common/exception/exceptions.h>\r
#include <common/gl/gl_check.h>\r
\r
-#include <core/producer/frame/image_transform.h>\r
+#include <core/producer/frame/frame_transform.h>\r
#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
\r
#include <algorithm>\r
-#include <array>\r
-#include <unordered_map>\r
+#include <deque>\r
+\r
+using namespace boost::assign;\r
\r
namespace caspar { namespace core {\r
+ \r
+typedef std::deque<render_item> layer;\r
+\r
+class image_renderer\r
+{\r
+ video_channel_context& channel_; \r
+ image_kernel kernel_; \r
+ std::shared_ptr<device_buffer> active_buffer_;\r
+public:\r
+ image_renderer(video_channel_context& channel)\r
+ : channel_(channel)\r
+ {\r
+ }\r
+ \r
+ boost::unique_future<safe_ptr<host_buffer>> render(std::deque<layer>&& layers)\r
+ { \r
+ auto layers2 = std::move(layers);\r
+ return channel_.ogl().begin_invoke([=]() mutable\r
+ {\r
+ return do_render(std::move(layers2));\r
+ });\r
+ }\r
+ \r
+private:\r
+ safe_ptr<host_buffer> do_render(std::deque<layer>&& layers)\r
+ {\r
+ std::shared_ptr<device_buffer> layer_key_buffer;\r
+\r
+ auto draw_buffer = create_device_buffer(4);\r
+ \r
+ BOOST_FOREACH(auto& layer, layers)\r
+ draw_layer(std::move(layer), draw_buffer, layer_key_buffer);\r
\r
-struct image_mixer::implementation : boost::noncopyable\r
-{ \r
- typedef std::deque<render_item> layer;\r
+ auto host_buffer = channel_.ogl().create_host_buffer(channel_.get_format_desc().size, host_buffer::read_only);\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
- video_channel_context& channel_;\r
+ void draw_layer(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::vector<image_transform> transform_stack_;\r
- std::vector<video_mode::type> mode_stack_;\r
+ std::pair<int, std::shared_ptr<device_buffer>> local_key_buffer = std::make_pair(0, nullptr); // int is fields flag\r
+ \r
+ if(layer.front().blend_mode != blend_mode::normal && has_overlapping_items(layer))\r
+ {\r
+ auto layer_draw_buffer = create_device_buffer(4); // int is fields flag\r
+ auto layer_blend_mode = layer.front().blend_mode;\r
\r
- std::queue<std::deque<render_item>> layers_; // layer/stream/items\r
- \r
- image_kernel kernel_;\r
+ BOOST_FOREACH(auto& item, layer)\r
+ {\r
+ item.blend_mode = blend_mode::normal; // Disable blending and just merge.\r
+ draw_item(std::move(item), *layer_draw_buffer, local_key_buffer, layer_key_buffer); \r
+ }\r
+ \r
+ render_item item;\r
+ item.pix_desc.pix_fmt = pixel_format::bgra;\r
+ item.pix_desc.planes = list_of(pixel_format_desc::plane(channel_.get_format_desc().width, channel_.get_format_desc().height, 4));\r
+ item.textures = list_of(layer_draw_buffer);\r
+ item.transform = frame_transform();\r
+ item.blend_mode = layer_blend_mode;\r
+\r
+ kernel_.draw(channel_.ogl(), std::move(item), *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::field_mode::progressive);\r
+\r
+ std::swap(local_key_buffer.second, layer_key_buffer);\r
+ }\r
+\r
+ void draw_item(render_item&& item, \r
+ 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.is_key)\r
+ {\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
+ local_key_buffer.first |= item.transform.field_mode; // Add field to flag.\r
+ kernel_.draw(channel_.ogl(), std::move(item), *local_key_buffer.second, nullptr, nullptr);\r
+ }\r
+ else\r
+ {\r
+ kernel_.draw(channel_.ogl(), std::move(item), draw_buffer, local_key_buffer.second, layer_key_buffer);\r
+ local_key_buffer.first ^= item.transform.field_mode; // Remove field from flag.\r
+ \r
+ if(local_key_buffer.first == 0) // If all fields from key has been used, reset it\r
+ {\r
+ local_key_buffer.first = 0;\r
+ local_key_buffer.second.reset();\r
+ }\r
+ }\r
+ }\r
+\r
+ bool has_overlapping_items(const layer& layer)\r
+ { \r
+ auto upper_count = boost::range::count_if(layer, [&](const render_item& item)\r
+ {\r
+ return !item.transform.is_key && (item.transform.field_mode & field_mode::upper);\r
+ });\r
+\r
+ auto lower_count = boost::range::count_if(layer, [&](const render_item& item)\r
+ {\r
+ return !item.transform.is_key && (item.transform.field_mode & field_mode::lower);\r
+ });\r
+\r
+ return upper_count > 1 || lower_count > 1;\r
+ } \r
\r
- std::array<std::shared_ptr<device_buffer>,2> draw_buffer_;\r
- std::shared_ptr<device_buffer> write_buffer_;\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
- std::array<std::shared_ptr<device_buffer>,2> stream_key_buffer_;\r
- std::shared_ptr<device_buffer> layer_key_buffer_;\r
- \r
+ \r
+struct image_mixer::implementation : boost::noncopyable\r
+{ \r
+ ogl_device& ogl_;\r
+ image_renderer renderer_;\r
+ std::vector<frame_transform> transform_stack_;\r
+ blend_mode::type active_blend_mode_;\r
+ std::deque<std::deque<render_item>> layers_; // layer/stream/items\r
public:\r
implementation(video_channel_context& video_channel) \r
- : channel_(video_channel)\r
+ : ogl_(video_channel.ogl())\r
+ , renderer_(video_channel)\r
, transform_stack_(1)\r
- , mode_stack_(1, video_mode::progressive)\r
- {\r
- initialize_buffers();\r
- }\r
-\r
- ~implementation()\r
+ , active_blend_mode_(blend_mode::normal) \r
{\r
- channel_.ogl().gc();\r
}\r
\r
- void initialize_buffers()\r
+ void begin_layer(blend_mode::type blend_mode)\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
- stream_key_buffer_[0] = channel_.ogl().create_device_buffer(channel_.get_format_desc().width, channel_.get_format_desc().height, 1);\r
- stream_key_buffer_[1] = channel_.ogl().create_device_buffer(channel_.get_format_desc().width, channel_.get_format_desc().height, 1);\r
- channel_.ogl().gc();\r
+ active_blend_mode_ = blend_mode;\r
+ layers_ += layer();\r
}\r
-\r
+ \r
void begin(core::basic_frame& frame)\r
{\r
- transform_stack_.push_back(transform_stack_.back()*frame.get_image_transform());\r
- mode_stack_.push_back(frame.get_mode() == video_mode::progressive ? mode_stack_.back() : frame.get_mode());\r
+ transform_stack_.push_back(transform_stack_.back()*frame.get_frame_transform());\r
}\r
\r
void visit(core::write_frame& frame)\r
{ \r
- CASPAR_ASSERT(!layers_.empty());\r
-\r
- // Check if frame has been discarded by interlacing\r
- if(boost::range::find(mode_stack_, video_mode::upper) != mode_stack_.end() && boost::range::find(mode_stack_, video_mode::lower) != mode_stack_.end())\r
+ if(transform_stack_.back().field_mode == field_mode::empty)\r
return;\r
\r
- core::render_item item = {frame.get_pixel_format_desc(), frame.get_textures(), transform_stack_.back(), mode_stack_.back(), frame.tag()}; \r
-\r
- auto& layer = layers_.back();\r
+ core::render_item item;\r
+ item.pix_desc = frame.get_pixel_format_desc();\r
+ item.textures = frame.get_textures();\r
+ item.transform = transform_stack_.back();\r
+ item.blend_mode = active_blend_mode_; \r
\r
- auto it = boost::range::find(layer, item);\r
- if(it == layer.end())\r
- layer.push_back(item);\r
+ layers_.back() += item;\r
}\r
\r
void end()\r
{\r
transform_stack_.pop_back();\r
- mode_stack_.pop_back();\r
- }\r
-\r
- void begin_layer()\r
- {\r
- layers_.push(layer());\r
}\r
\r
void end_layer()\r
- {\r
- }\r
- \r
- boost::unique_future<safe_ptr<host_buffer>> render()\r
{ \r
- auto layers = std::move(layers_);\r
-\r
- return channel_.ogl().begin_invoke([=]() mutable -> safe_ptr<host_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
- return do_render(std::move(layers));\r
- });\r
}\r
\r
- safe_ptr<host_buffer> do_render(std::queue<layer>&& layers)\r
+ boost::unique_future<safe_ptr<host_buffer>> render()\r
{\r
- auto read_buffer = channel_.ogl().create_host_buffer(channel_.get_format_desc().size, host_buffer::read_only);\r
-\r
- layer_key_buffer_->clear();\r
- draw_buffer_[0]->clear();\r
- draw_buffer_[1]->clear();\r
- stream_key_buffer_[0]->clear();\r
- stream_key_buffer_[1]->clear();\r
-\r
- bool local_key = false;\r
- bool layer_key = false;\r
-\r
- while(!layers.empty())\r
- { \r
- stream_key_buffer_[0]->clear();\r
-\r
- auto layer = std::move(layers.front());\r
- layers.pop();\r
- \r
- while(!layer.empty())\r
- {\r
- auto item = std::move(layer.front());\r
- layer.pop_front();\r
- \r
- if(item.transform.get_is_key())\r
- {\r
- render_item(stream_key_buffer_, std::move(item), nullptr, nullptr);\r
- local_key = true;\r
- }\r
- else\r
- {\r
- render_item(draw_buffer_, std::move(item), local_key ? stream_key_buffer_[0] : nullptr, layer_key ? layer_key_buffer_ : nullptr); \r
- stream_key_buffer_[0]->clear();\r
- local_key = false;\r
- }\r
- channel_.ogl().yield(); // Return resources to pool as early as possible.\r
- }\r
-\r
- layer_key = local_key;\r
- local_key = false;\r
- std::swap(stream_key_buffer_[0], layer_key_buffer_);\r
- }\r
-\r
- std::swap(draw_buffer_[0], write_buffer_);\r
-\r
- // device -> host. \r
- read_buffer->read(*write_buffer_);\r
-\r
- return read_buffer;\r
+ return renderer_.render(std::move(layers_));\r
}\r
- \r
- void render_item(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
- BOOST_FOREACH(auto& texture, item.textures)\r
- {\r
- if(!texture->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
\r
- targets[1]->attach();\r
- \r
- kernel_.draw(item, make_safe(targets[0]), local_key, layer_key);\r
- \r
- targets[0]->bind();\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
- \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
+ return make_safe<write_frame>(ogl_, tag, desc);\r
}\r
};\r
\r
void image_mixer::end(){impl_->end();}\r
boost::unique_future<safe_ptr<host_buffer>> image_mixer::render(){return impl_->render();}\r
safe_ptr<write_frame> image_mixer::create_frame(const void* tag, const core::pixel_format_desc& desc){return impl_->create_frame(tag, desc);}\r
-void image_mixer::begin_layer(){impl_->begin_layer();}\r
+void image_mixer::begin_layer(blend_mode::type blend_mode){impl_->begin_layer(blend_mode);}\r
void image_mixer::end_layer(){impl_->end_layer();}\r
image_mixer& image_mixer::operator=(image_mixer&& other)\r
{\r