-/*\r
-* Copyright (c) 2011 Sveriges Television AB <info@casparcg.com>\r
-*\r
-* This file is part of CasparCG (www.casparcg.com).\r
-*\r
-* CasparCG is free software: you can redistribute it and/or modify\r
-* it under the terms of the GNU General Public License as published by\r
-* the Free Software Foundation, either version 3 of the License, or\r
-* (at your option) any later version.\r
-*\r
-* CasparCG is distributed in the hope that it will be useful,\r
-* but WITHOUT ANY WARRANTY; without even the implied warranty of\r
-* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the\r
-* GNU General Public License for more details.\r
-*\r
-* You should have received a copy of the GNU General Public License\r
-* along with CasparCG. If not, see <http://www.gnu.org/licenses/>.\r
-*\r
-* Author: Robert Nagy, ronag89@gmail.com\r
-*/\r
-\r
-#include "../../stdafx.h"\r
-\r
-#include "image_mixer.h"\r
-\r
-#include "../util/data_frame.h"\r
-#include "../util/xmm.h"\r
-\r
-#include <common/assert.h>\r
-#include <common/gl/gl_check.h>\r
-#include <common/concurrency/async.h>\r
-#include <common/memory/memcpy.h>\r
-\r
-#include <core/frame/data_frame.h>\r
-#include <core/frame/frame_transform.h>\r
-#include <core/frame/pixel_format.h>\r
-#include <core/video_format.h>\r
-\r
-#include <modules/ffmpeg/producer/util/util.h>\r
-\r
-#include <asmlib.h>\r
-\r
-#include <gl/glew.h>\r
-\r
-#include <tbb/cache_aligned_allocator.h>\r
-#include <tbb/parallel_for_each.h>\r
-\r
-#include <boost/assign.hpp>\r
-#include <boost/foreach.hpp>\r
-#include <boost/range.hpp>\r
-#include <boost/range/algorithm_ext/erase.hpp>\r
-#include <boost/thread/future.hpp>\r
-\r
-#include <algorithm>\r
-#include <stdint.h>\r
-#include <vector>\r
-\r
-#if defined(_MSC_VER)\r
-#pragma warning (push)\r
-#pragma warning (disable : 4244)\r
-#endif\r
-extern "C" \r
-{\r
- #include <libswscale/swscale.h>\r
- #include <libavcodec/avcodec.h>\r
- #include <libavformat/avformat.h>\r
-}\r
-#if defined(_MSC_VER)\r
-#pragma warning (pop)\r
-#endif\r
-\r
-namespace caspar { namespace accelerator { namespace cpu {\r
- \r
-struct item\r
-{\r
- core::pixel_format_desc pix_desc;\r
- std::vector<spl::shared_ptr<host_buffer>> buffers;\r
- core::image_transform transform;\r
-\r
- item()\r
- : pix_desc(core::pixel_format::invalid)\r
- {\r
- }\r
-};\r
-\r
-bool operator==(const item& lhs, const item& rhs)\r
-{\r
- return lhs.buffers == rhs.buffers && lhs.transform == rhs.transform;\r
-}\r
-\r
-bool operator!=(const item& lhs, const item& rhs)\r
-{\r
- return !(lhs == rhs);\r
-}\r
- \r
-inline xmm::s8_x blend(xmm::s8_x d, xmm::s8_x s)\r
-{ \r
- using namespace xmm;\r
- \r
- // T(S, D) = S * D[A] + 0x80\r
- auto aaaa = s8_x::shuffle(d, s8_x(15, 15, 15, 15, 11, 11, 11, 11, 7, 7, 7, 7, 3, 3, 3, 3));\r
- d = s8_x(u8_x::min(u8_x(d), u8_x(aaaa))); // overflow guard\r
-\r
- auto xaxa = s16_x(aaaa) >> 8; \r
- \r
- auto t1 = s16_x::multiply_low(s16_x(s) & 0x00FF, xaxa) + 0x80; \r
- auto t2 = s16_x::multiply_low(s16_x(s) >> 8 , xaxa) + 0x80;\r
- \r
- // C(S, D) = S + D - (((T >> 8) + T) >> 8);\r
- auto xyxy = s8_x(((t1 >> 8) + t1) >> 8); \r
- auto yxyx = s8_x((t2 >> 8) + t2); \r
- auto argb = s8_x::blend(xyxy, yxyx, s8_x(-1, 0, -1, 0));\r
-\r
- return s8_x(s) + (d - argb);\r
-}\r
- \r
-template<typename temporal, typename alignment>\r
-static void kernel(uint8_t* dest, const uint8_t* source, size_t count)\r
-{ \r
- using namespace xmm;\r
-\r
- for(auto n = 0; n < count; n += 32) \r
- {\r
- auto s0 = s8_x::load<temporal_tag, alignment>(dest+n+0);\r
- auto s1 = s8_x::load<temporal_tag, alignment>(dest+n+16);\r
-\r
- auto d0 = s8_x::load<temporal_tag, alignment>(source+n+0);\r
- auto d1 = s8_x::load<temporal_tag, alignment>(source+n+16);\r
- \r
- auto argb0 = blend(d0, s0);\r
- auto argb1 = blend(d1, s1);\r
-\r
- s8_x::store<temporal, alignment>(argb0, dest+n+0 );\r
- s8_x::store<temporal, alignment>(argb1, dest+n+16);\r
- } \r
-}\r
-\r
-template<typename temporal>\r
-static void kernel(uint8_t* dest, const uint8_t* source, size_t count)\r
-{ \r
- using namespace xmm;\r
-\r
- if(reinterpret_cast<int>(dest) % 16 != 0 || reinterpret_cast<int>(source) % 16 != 0)\r
- kernel<temporal_tag, unaligned_tag>(dest, source, count);\r
- else\r
- kernel<temporal_tag, aligned_tag>(dest, source, count);\r
-}\r
-\r
-class image_renderer\r
-{\r
- std::pair<std::vector<item>, boost::shared_future<boost::iterator_range<const uint8_t*>>> last_image_;\r
- tbb::concurrent_unordered_map<int, tbb::concurrent_bounded_queue<std::shared_ptr<SwsContext>>> sws_contexts_;\r
-public: \r
- boost::shared_future<boost::iterator_range<const uint8_t*>> operator()(std::vector<item> items, const core::video_format_desc& format_desc)\r
- { \r
- if(last_image_.first == items && last_image_.second.has_value())\r
- return last_image_.second;\r
-\r
- auto image = render(items, format_desc);\r
- last_image_ = std::make_pair(std::move(items), image);\r
- return image;\r
- }\r
-\r
-private:\r
- boost::shared_future<boost::iterator_range<const uint8_t*>> render(std::vector<item> items, const core::video_format_desc& format_desc)\r
- {\r
- convert(items, format_desc.width, format_desc.height); \r
- \r
- auto result = spl::make_shared<host_buffer>(format_desc.size, 0);\r
- if(format_desc.field_mode != core::field_mode::progressive)\r
- { \r
- draw(items, result->data(), format_desc.width, format_desc.height, core::field_mode::upper);\r
- draw(items, result->data(), format_desc.width, format_desc.height, core::field_mode::lower);\r
- }\r
- else\r
- {\r
- draw(items, result->data(), format_desc.width, format_desc.height, core::field_mode::progressive);\r
- }\r
- \r
- return async(launch_policy::deferred, [=]\r
- {\r
- return boost::iterator_range<const uint8_t*>(result->data(), result->data() + format_desc.size);\r
- }); \r
- }\r
-\r
- void draw(std::vector<item> items, uint8_t* dest, int width, int height, core::field_mode field_mode)\r
- { \r
- BOOST_FOREACH(auto& item, items)\r
- item.transform.field_mode &= field_mode;\r
- \r
- // Remove empty items.\r
- boost::range::remove_erase_if(items, [&](const item& item)\r
- {\r
- return item.transform.field_mode == core::field_mode::empty;\r
- });\r
- \r
- // Remove first field stills.\r
- boost::range::remove_erase_if(items, [&](const item& item)\r
- {\r
- return item.transform.is_still && item.transform.field_mode == field_mode; // only us last field for stills.\r
- });\r
-\r
- if(items.empty())\r
- return;\r
- \r
- auto start = field_mode == core::field_mode::lower ? 1 : 0;\r
- auto step = field_mode == core::field_mode::progressive ? 1 : 2;\r
- \r
- // TODO: Add support for fill translations.\r
- // TODO: Add support for mask rect.\r
- // TODO: Add support for opacity.\r
- // TODO: Add support for mix transition.\r
- // TODO: Add support for push transition.\r
- // TODO: Add support for wipe transition.\r
- // TODO: Add support for slide transition.\r
- tbb::parallel_for(tbb::blocked_range<int>(0, height/step), [&](const tbb::blocked_range<int>& r)\r
- {\r
- for(auto i = r.begin(); i != r.end(); ++i)\r
- {\r
- auto y = i*step+start;\r
-\r
- for(std::size_t n = 0; n < items.size()-1; ++n)\r
- kernel<xmm::temporal_tag>(dest + y*width*4, items[n].buffers.at(0)->data() + y*width*4, width*4);\r
- \r
- std::size_t n = items.size()-1; \r
- kernel<xmm::nontemporal_tag>(dest + y*width*4, items[n].buffers.at(0)->data() + y*width*4, width*4);\r
- }\r
-\r
- _mm_mfence();\r
- });\r
- }\r
- \r
- void convert(std::vector<item>& source_items, int width, int height)\r
- {\r
- std::set<std::vector<spl::shared_ptr<host_buffer>>> buffers;\r
-\r
- BOOST_FOREACH(auto& item, source_items)\r
- buffers.insert(item.buffers);\r
- \r
- auto dest_items = source_items;\r
-\r
- tbb::parallel_for_each(buffers.begin(), buffers.end(), [&](const std::vector<spl::shared_ptr<host_buffer>>& buffers)\r
- { \r
- auto pix_desc = std::find_if(source_items.begin(), source_items.end(), [&](const item& item){return item.buffers == buffers;})->pix_desc;\r
-\r
- if(pix_desc.format == core::pixel_format::bgra && \r
- pix_desc.planes.at(0).width == width &&\r
- pix_desc.planes.at(0).height == height)\r
- return;\r
-\r
- auto input_av_frame = ffmpeg::make_av_frame(buffers, pix_desc);\r
- \r
- int key = ((input_av_frame->width << 22) & 0xFFC00000) | ((input_av_frame->height << 6) & 0x003FC000) | ((input_av_frame->format << 7) & 0x00007F00);\r
- \r
- auto& pool = sws_contexts_[key];\r
-\r
- std::shared_ptr<SwsContext> sws_context;\r
- if(!pool.try_pop(sws_context))\r
- {\r
- double param;\r
- sws_context.reset(sws_getContext(input_av_frame->width, input_av_frame->height, static_cast<PixelFormat>(input_av_frame->format), width, height, PIX_FMT_BGRA, SWS_BILINEAR, nullptr, nullptr, ¶m), sws_freeContext);\r
- }\r
- \r
- if(!sws_context) \r
- BOOST_THROW_EXCEPTION(operation_failed() << msg_info("Could not create software scaling context.") << boost::errinfo_api_function("sws_getContext")); \r
- \r
- auto dest_frame = spl::make_shared<host_buffer>(width*height*4);\r
-\r
- {\r
- spl::shared_ptr<AVFrame> dest_av_frame(avcodec_alloc_frame(), av_free); \r
- avcodec_get_frame_defaults(dest_av_frame.get()); \r
- avpicture_fill(reinterpret_cast<AVPicture*>(dest_av_frame.get()), dest_frame->data(), PIX_FMT_BGRA, width, height);\r
- \r
- sws_scale(sws_context.get(), input_av_frame->data, input_av_frame->linesize, 0, input_av_frame->height, dest_av_frame->data, dest_av_frame->linesize); \r
- pool.push(sws_context);\r
- }\r
- \r
- for(std::size_t n = 0; n < source_items.size(); ++n)\r
- {\r
- if(source_items[n].buffers == buffers)\r
- {\r
- dest_items[n].buffers = boost::assign::list_of(dest_frame);\r
- dest_items[n].pix_desc = core::pixel_format_desc(core::pixel_format::bgra);\r
- dest_items[n].pix_desc.planes = boost::assign::list_of(core::pixel_format_desc::plane(width, height, 4));\r
- dest_items[n].transform = source_items[n].transform;\r
- }\r
- }\r
- }); \r
-\r
- source_items = std::move(dest_items);\r
- }\r
-};\r
- \r
-struct image_mixer::impl : boost::noncopyable\r
-{ \r
- image_renderer renderer_;\r
- std::vector<core::image_transform> transform_stack_;\r
- std::vector<item> items_; // layer/stream/items\r
-public:\r
- impl() \r
- : transform_stack_(1) \r
- {\r
- CASPAR_LOG(info) << L"Initialized Streaming SIMD Extensions Accelerated CPU Image Mixer";\r
- }\r
-\r
- void begin_layer(core::blend_mode blend_mode)\r
- {\r
- }\r
- \r
- void push(const core::frame_transform& transform)\r
- {\r
- transform_stack_.push_back(transform_stack_.back()*transform.image_transform);\r
- }\r
- \r
- void visit(const core::data_frame& frame2)\r
- { \r
- auto frame = dynamic_cast<const cpu::data_frame*>(&frame2);\r
- if(frame == nullptr)\r
- return;\r
-\r
- if(frame->pixel_format_desc().format == core::pixel_format::invalid)\r
- return;\r
-\r
- if(frame->buffers().empty())\r
- return;\r
-\r
- if(transform_stack_.back().field_mode == core::field_mode::empty)\r
- return;\r
-\r
- item item;\r
- item.pix_desc = frame->pixel_format_desc();\r
- item.buffers = frame->buffers(); \r
- item.transform = transform_stack_.back();\r
-\r
- items_.push_back(item);\r
- }\r
-\r
- void pop()\r
- {\r
- transform_stack_.pop_back();\r
- }\r
-\r
- void end_layer()\r
- { \r
- }\r
- \r
- boost::shared_future<boost::iterator_range<const uint8_t*>> render(const core::video_format_desc& format_desc)\r
- {\r
- return renderer_(std::move(items_), format_desc);\r
- }\r
- \r
- virtual spl::shared_ptr<core::data_frame> create_frame(const void* tag, const core::pixel_format_desc& desc, double frame_rate, core::field_mode field_mode)\r
- {\r
- return spl::make_shared<cpu::data_frame>(tag, desc, frame_rate, field_mode);\r
- }\r
-};\r
-\r
-image_mixer::image_mixer() : impl_(new impl()){}\r
-void image_mixer::push(const core::frame_transform& transform){impl_->push(transform);}\r
-void image_mixer::visit(const core::data_frame& frame){impl_->visit(frame);}\r
-void image_mixer::pop(){impl_->pop();}\r
-boost::shared_future<boost::iterator_range<const uint8_t*>> image_mixer::operator()(const core::video_format_desc& format_desc){return impl_->render(format_desc);}\r
-void image_mixer::begin_layer(core::blend_mode blend_mode){impl_->begin_layer(blend_mode);}\r
-void image_mixer::end_layer(){impl_->end_layer();}\r
-spl::shared_ptr<core::data_frame> image_mixer::create_frame(const void* tag, const core::pixel_format_desc& desc, double frame_rate, core::field_mode field_mode) {return impl_->create_frame(tag, desc, frame_rate, field_mode);}\r
-\r
-}}}
\ No newline at end of file
+/*
+* Copyright (c) 2011 Sveriges Television AB <info@casparcg.com>
+*
+* This file is part of CasparCG (www.casparcg.com).
+*
+* CasparCG is free software: you can redistribute it and/or modify
+* it under the terms of the GNU General Public License as published by
+* the Free Software Foundation, either version 3 of the License, or
+* (at your option) any later version.
+*
+* CasparCG is distributed in the hope that it will be useful,
+* but WITHOUT ANY WARRANTY; without even the implied warranty of
+* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+* GNU General Public License for more details.
+*
+* You should have received a copy of the GNU General Public License
+* along with CasparCG. If not, see <http://www.gnu.org/licenses/>.
+*
+* Author: Robert Nagy, ronag89@gmail.com
+*/
+
+#include "../../StdAfx.h"
+
+#include "image_mixer.h"
+
+#include "../util/xmm.h"
+
+#include <common/assert.h>
+#include <common/gl/gl_check.h>
+#include <common/future.h>
+#include <common/array.h>
+
+#include <core/frame/frame.h>
+#include <core/frame/frame_transform.h>
+#include <core/frame/pixel_format.h>
+#include <core/video_format.h>
+
+#include <modules/ffmpeg/producer/util/util.h>
+
+#include <asmlib.h>
+
+#include <GL/glew.h>
+
+#include <tbb/parallel_for.h>
+#include <tbb/parallel_for_each.h>
+#include <tbb/concurrent_queue.h>
+
+#include <boost/range/algorithm_ext/erase.hpp>
+#include <boost/thread/future.hpp>
+
+#include <algorithm>
+#include <cstdint>
+#include <vector>
+#include <set>
+#include <array>
+
+#if defined(_MSC_VER)
+#pragma warning (push)
+#pragma warning (disable : 4244)
+#endif
+extern "C"
+{
+ #include <libswscale/swscale.h>
+ #include <libavcodec/avcodec.h>
+ #include <libavformat/avformat.h>
+}
+#if defined(_MSC_VER)
+#pragma warning (pop)
+#endif
+
+namespace caspar { namespace accelerator { namespace cpu {
+
+struct item
+{
+ core::pixel_format_desc pix_desc = core::pixel_format::invalid;
+ std::array<const uint8_t*, 4> data;
+ core::image_transform transform;
+
+ item()
+ {
+ data.fill(0);
+ }
+};
+
+bool operator==(const item& lhs, const item& rhs)
+{
+ return lhs.data == rhs.data && lhs.transform == rhs.transform;
+}
+
+bool operator!=(const item& lhs, const item& rhs)
+{
+ return !(lhs == rhs);
+}
+
+// 100% accurate blending with correct rounding.
+inline xmm::s8_x blend(xmm::s8_x d, xmm::s8_x s)
+{
+ using namespace xmm;
+
+ // C(S, D) = S + D - (((T >> 8) + T) >> 8);
+ // T(S, D) = S * D[A] + 0x80
+
+ auto aaaa = s8_x::shuffle(d, s8_x(15, 15, 15, 15, 11, 11, 11, 11, 7, 7, 7, 7, 3, 3, 3, 3));
+ d = s8_x(u8_x::min(u8_x(d), u8_x(aaaa))); // Overflow guard. Some source files have color values which incorrectly exceed pre-multiplied alpha values, e.g. red(255) > alpha(254).
+
+ auto xaxa = s16_x(aaaa) >> 8;
+
+ auto t1 = s16_x::multiply_low(s16_x(s) & 0x00FF, xaxa) + 0x80;
+ auto t2 = s16_x::multiply_low(s16_x(s) >> 8 , xaxa) + 0x80;
+
+ auto xyxy = s8_x(((t1 >> 8) + t1) >> 8);
+ auto yxyx = s8_x((t2 >> 8) + t2);
+ auto argb = s8_x::blend(xyxy, yxyx, s8_x(-1, 0, -1, 0));
+
+ return s8_x(s) + (d - argb);
+}
+
+template<typename temporal, typename alignment>
+static void kernel(uint8_t* dest, const uint8_t* source, size_t count)
+{
+ using namespace xmm;
+
+ for(auto n = 0; n < count; n += 32)
+ {
+ auto s0 = s8_x::load<temporal_tag, alignment>(dest+n+0);
+ auto s1 = s8_x::load<temporal_tag, alignment>(dest+n+16);
+
+ auto d0 = s8_x::load<temporal_tag, alignment>(source+n+0);
+ auto d1 = s8_x::load<temporal_tag, alignment>(source+n+16);
+
+ auto argb0 = blend(d0, s0);
+ auto argb1 = blend(d1, s1);
+
+ s8_x::store<temporal, alignment>(argb0, dest+n+0 );
+ s8_x::store<temporal, alignment>(argb1, dest+n+16);
+ }
+}
+
+template<typename temporal>
+static void kernel(uint8_t* dest, const uint8_t* source, size_t count)
+{
+ using namespace xmm;
+
+ if(reinterpret_cast<std::uint64_t>(dest) % 16 != 0 || reinterpret_cast<std::uint64_t>(source) % 16 != 0)
+ kernel<temporal_tag, unaligned_tag>(dest, source, count);
+ else
+ kernel<temporal_tag, aligned_tag>(dest, source, count);
+}
+
+class image_renderer
+{
+ tbb::concurrent_unordered_map<int64_t, tbb::concurrent_bounded_queue<std::shared_ptr<SwsContext>>> sws_devices_;
+ tbb::concurrent_bounded_queue<spl::shared_ptr<buffer>> temp_buffers_;
+ core::video_format_desc format_desc_;
+public:
+ std::future<array<const std::uint8_t>> operator()(std::vector<item> items, const core::video_format_desc& format_desc)
+ {
+ if (format_desc != format_desc_)
+ {
+ format_desc_ = format_desc;
+ sws_devices_.clear();
+ }
+
+ convert(items, format_desc.width, format_desc.height);
+
+ // Remove first field stills.
+ boost::range::remove_erase_if(items, [&](const item& item)
+ {
+ return item.transform.is_still && item.transform.field_mode == format_desc.field_mode; // only us last field for stills.
+ });
+
+ // Stills are progressive
+ for (auto& item : items)
+ {
+ if(item.transform.is_still)
+ item.transform.field_mode = core::field_mode::progressive;
+ }
+
+ auto result = spl::make_shared<buffer>(format_desc.size, 0);
+ if(format_desc.field_mode != core::field_mode::progressive)
+ {
+ draw(items, result->data(), format_desc.width, format_desc.height, core::field_mode::upper);
+ draw(items, result->data(), format_desc.width, format_desc.height, core::field_mode::lower);
+ }
+ else
+ {
+ draw(items, result->data(), format_desc.width, format_desc.height, core::field_mode::progressive);
+ }
+
+ temp_buffers_.clear();
+
+ return make_ready_future(array<const std::uint8_t>(result->data(), format_desc.size, true, result));
+ }
+
+private:
+
+ void draw(std::vector<item> items, uint8_t* dest, std::size_t width, std::size_t height, core::field_mode field_mode)
+ {
+ for (auto& item : items)
+ item.transform.field_mode &= field_mode;
+
+ // Remove empty items.
+ boost::range::remove_erase_if(items, [&](const item& item)
+ {
+ return item.transform.field_mode == core::field_mode::empty;
+ });
+
+ if(items.empty())
+ return;
+
+ auto start = field_mode == core::field_mode::lower ? 1 : 0;
+ auto step = field_mode == core::field_mode::progressive ? 1 : 2;
+
+ // TODO: Add support for fill translations.
+ // TODO: Add support for mask rect.
+ // TODO: Add support for opacity.
+ // TODO: Add support for mix transition.
+ // TODO: Add support for push transition.
+ // TODO: Add support for wipe transition.
+ // TODO: Add support for slide transition.
+ tbb::parallel_for(tbb::blocked_range<std::size_t>(0, height/step), [&](const tbb::blocked_range<std::size_t>& r)
+ {
+ for(auto i = r.begin(); i != r.end(); ++i)
+ {
+ auto y = i*step+start;
+
+ for(std::size_t n = 0; n < items.size()-1; ++n)
+ kernel<xmm::temporal_tag>(dest + y*width*4, items[n].data.at(0) + y*width*4, width*4);
+
+ std::size_t n = items.size()-1;
+ kernel<xmm::nontemporal_tag>(dest + y*width*4, items[n].data.at(0) + y*width*4, width*4);
+ }
+
+ _mm_mfence();
+ });
+ }
+
+ void convert(std::vector<item>& source_items, int width, int height)
+ {
+ std::set<std::array<const uint8_t*, 4>> buffers;
+
+ for (auto& item : source_items)
+ buffers.insert(item.data);
+
+ auto dest_items = source_items;
+
+ tbb::parallel_for_each(buffers.begin(), buffers.end(), [&](const std::array<const uint8_t*, 4>& data)
+ {
+ auto pix_desc = std::find_if(source_items.begin(), source_items.end(), [&](const item& item){return item.data == data;})->pix_desc;
+
+ if(pix_desc.format == core::pixel_format::bgra &&
+ pix_desc.planes.at(0).width == width &&
+ pix_desc.planes.at(0).height == height)
+ return;
+
+ std::array<uint8_t*, 4> data2 = {};
+ for(std::size_t n = 0; n < data.size(); ++n)
+ data2.at(n) = const_cast<uint8_t*>(data[n]);
+
+ auto input_av_frame = ffmpeg::make_av_frame(data2, pix_desc);
+
+
+ int64_t key = ((static_cast<int64_t>(input_av_frame->width) << 32) & 0xFFFF00000000) |
+ ((static_cast<int64_t>(input_av_frame->height) << 16) & 0xFFFF0000) |
+ ((static_cast<int64_t>(input_av_frame->format) << 8) & 0xFF00);
+
+ auto& pool = sws_devices_[key];
+
+ std::shared_ptr<SwsContext> sws_device;
+ if(!pool.try_pop(sws_device))
+ {
+ double param;
+ sws_device.reset(sws_getContext(input_av_frame->width, input_av_frame->height, static_cast<AVPixelFormat>(input_av_frame->format), width, height, AVPixelFormat::AV_PIX_FMT_BGRA, SWS_BILINEAR, nullptr, nullptr, ¶m), sws_freeContext);
+ }
+
+ if(!sws_device)
+ CASPAR_THROW_EXCEPTION(operation_failed() << msg_info("Could not create software scaling device.") << boost::errinfo_api_function("sws_getContext"));
+
+ auto dest_frame = spl::make_shared<buffer>(width*height*4);
+ temp_buffers_.push(dest_frame);
+
+ {
+ auto dest_av_frame = ffmpeg::create_frame();
+ avpicture_fill(reinterpret_cast<AVPicture*>(dest_av_frame.get()), dest_frame->data(), AVPixelFormat::AV_PIX_FMT_BGRA, width, height);
+
+ sws_scale(sws_device.get(), input_av_frame->data, input_av_frame->linesize, 0, input_av_frame->height, dest_av_frame->data, dest_av_frame->linesize);
+ pool.push(sws_device);
+ }
+
+ for(std::size_t n = 0; n < source_items.size(); ++n)
+ {
+ if(source_items[n].data == data)
+ {
+ dest_items[n].data.fill(0);
+ dest_items[n].data[0] = dest_frame->data();
+ dest_items[n].pix_desc = core::pixel_format_desc(core::pixel_format::bgra);
+ dest_items[n].pix_desc.planes = { core::pixel_format_desc::plane(width, height, 4) };
+ dest_items[n].transform = source_items[n].transform;
+ }
+ }
+ });
+
+ source_items = std::move(dest_items);
+ }
+};
+
+struct image_mixer::impl : boost::noncopyable
+{
+ image_renderer renderer_;
+ std::vector<core::image_transform> transform_stack_;
+ std::vector<item> items_; // layer/stream/items
+public:
+ impl(int channel_id)
+ : transform_stack_(1)
+ {
+ CASPAR_LOG(info) << L"Initialized Streaming SIMD Extensions Accelerated CPU Image Mixer for channel " << channel_id;
+ }
+
+ void push(const core::frame_transform& transform)
+ {
+ transform_stack_.push_back(transform_stack_.back()*transform.image_transform);
+ }
+
+ void visit(const core::const_frame& frame)
+ {
+ if(frame.pixel_format_desc().format == core::pixel_format::invalid)
+ return;
+
+ if(frame.pixel_format_desc().planes.empty())
+ return;
+
+ if(frame.pixel_format_desc().planes.at(0).size < 16)
+ return;
+
+ if(transform_stack_.back().field_mode == core::field_mode::empty)
+ return;
+
+ item item;
+ item.pix_desc = frame.pixel_format_desc();
+ item.transform = transform_stack_.back();
+ for(int n = 0; n < item.pix_desc.planes.size(); ++n)
+ item.data.at(n) = frame.image_data(n).begin();
+
+ items_.push_back(item);
+ }
+
+ void pop()
+ {
+ transform_stack_.pop_back();
+ }
+
+ std::future<array<const std::uint8_t>> render(const core::video_format_desc& format_desc)
+ {
+ return renderer_(std::move(items_), format_desc);
+ }
+
+ core::mutable_frame create_frame(const void* tag, const core::pixel_format_desc& desc, const core::audio_channel_layout& channel_layout)
+ {
+ std::vector<array<std::uint8_t>> buffers;
+ for (auto& plane : desc.planes)
+ {
+ auto buf = spl::make_shared<buffer>(plane.size);
+ buffers.push_back(array<std::uint8_t>(buf->data(), plane.size, true, buf));
+ }
+ return core::mutable_frame(std::move(buffers), core::mutable_audio_buffer(), tag, desc, channel_layout);
+ }
+};
+
+image_mixer::image_mixer(int channel_id) : impl_(new impl(channel_id)){}
+image_mixer::~image_mixer(){}
+void image_mixer::push(const core::frame_transform& transform){impl_->push(transform);}
+void image_mixer::visit(const core::const_frame& frame){impl_->visit(frame);}
+void image_mixer::pop(){impl_->pop();}
+int image_mixer::get_max_frame_size() { return std::numeric_limits<int>::max(); }
+std::future<array<const std::uint8_t>> image_mixer::operator()(const core::video_format_desc& format_desc, bool /* straighten_alpha */){return impl_->render(format_desc);}
+core::mutable_frame image_mixer::create_frame(const void* tag, const core::pixel_format_desc& desc, const core::audio_channel_layout& channel_layout) {return impl_->create_frame(tag, desc, channel_layout);}
+
+}}}
projects / casparcg / blobdiff