#include "audio_mixer.h"\r
\r
#include <core/mixer/write_frame.h>\r
-#include <core/producer/frame/audio_transform.h>\r
+#include <core/producer/frame/frame_transform.h>\r
+\r
+#include <tbb/parallel_for.h>\r
+\r
+#include <safeint.h>\r
+\r
+#include <stack>\r
+#include <deque>\r
\r
namespace caspar { namespace core {\r
+\r
+struct audio_item\r
+{\r
+ const void* tag;\r
+ frame_transform transform;\r
+ std::vector<int32_t> audio_data;\r
+};\r
\r
struct audio_mixer::implementation\r
{\r
- std::deque<std::vector<int16_t>> audio_data_;\r
- std::stack<core::audio_transform> transform_stack_;\r
-\r
- std::map<int, core::audio_transform> prev_audio_transforms_;\r
- std::map<int, core::audio_transform> next_audio_transforms_;\r
+ std::stack<core::frame_transform> transform_stack_;\r
+ std::map<const void*, core::frame_transform> prev_frame_transforms_;\r
+ const core::video_format_desc format_desc_;\r
+ std::vector<audio_item> items;\r
\r
public:\r
- implementation()\r
+ implementation(const core::video_format_desc& format_desc)\r
+ : format_desc_(format_desc)\r
{\r
- transform_stack_.push(core::audio_transform());\r
- audio_data_.push_back(std::vector<int16_t>()); // One frame delay\r
+ transform_stack_.push(core::frame_transform());\r
}\r
\r
void begin(core::basic_frame& frame)\r
{\r
- transform_stack_.push(transform_stack_.top()*frame.get_audio_transform());\r
+ transform_stack_.push(transform_stack_.top()*frame.get_frame_transform());\r
}\r
\r
void visit(const core::write_frame& frame)\r
{\r
- if(!transform_stack_.top().get_has_audio())\r
+ // We only care about the last field.\r
+ if(format_desc_.field_mode == field_mode::upper && transform_stack_.top().field_mode == field_mode::upper)\r
return;\r
\r
- const auto& audio_data = frame.audio_data();\r
- const auto tag = frame.tag(); // Get the identifier for the audio-stream.\r
-\r
- if(audio_data_.back().empty())\r
- audio_data_.back().resize(audio_data.size(), 0);\r
- \r
- const auto next = transform_stack_.top();\r
- auto prev = next;\r
+ if(format_desc_.field_mode == field_mode::lower && transform_stack_.top().field_mode == field_mode::lower)\r
+ return;\r
\r
- const auto it = prev_audio_transforms_.find(tag);\r
- if(it != prev_audio_transforms_.end())\r
- prev = it->second;\r
- \r
- next_audio_transforms_[tag] = next; // Store all active tags, inactive tags will be removed in end_pass.\r
- \r
- if(next.get_gain() < 0.001 && prev.get_gain() < 0.001)\r
+ // Skip empty audio.\r
+ if(transform_stack_.top().volume < 0.002 || frame.audio_data().empty())\r
return;\r
- \r
- static const int BASE = 1<<15;\r
\r
- const auto next_gain = static_cast<int>(next.get_gain()*BASE);\r
- const auto prev_gain = static_cast<int>(prev.get_gain()*BASE);\r
- \r
- const int n_samples = audio_data_.back().size();\r
-\r
- tbb::parallel_for\r
- (\r
- tbb::blocked_range<size_t>(0, audio_data.size()),\r
- [&](const tbb::blocked_range<size_t>& r)\r
- {\r
- for(size_t n = r.begin(); n < r.end(); ++n)\r
- {\r
- const int sample_gain = (prev_gain - (prev_gain * n)/n_samples) + (next_gain * n)/n_samples;\r
- const int sample = (static_cast<int>(audio_data[n])*sample_gain)/BASE;\r
- audio_data_.back()[n] = static_cast<int16_t>((static_cast<int>(audio_data_.back()[n]) + sample) & 0xFFFF);\r
- }\r
- }\r
- );\r
+ audio_item item;\r
+ item.tag = frame.tag();\r
+ item.transform = transform_stack_.top();\r
+ item.audio_data = std::vector<int32_t>(frame.audio_data().begin(), frame.audio_data().end());\r
+\r
+ items.push_back(item); \r
}\r
\r
- void begin(const core::audio_transform& transform)\r
+ void begin(const core::frame_transform& transform)\r
{\r
transform_stack_.push(transform_stack_.top()*transform);\r
}\r
transform_stack_.pop();\r
}\r
\r
- std::vector<int16_t> mix()\r
+ audio_buffer mix()\r
{\r
- prev_audio_transforms_ = std::move(next_audio_transforms_); \r
- auto result = std::move(audio_data_.front());\r
- audio_data_.pop_front();\r
- audio_data_.push_back(std::vector<int16_t>());\r
+ auto result = audio_buffer(format_desc_.audio_samples_per_frame, 0);\r
+\r
+ std::map<const void*, core::frame_transform> next_frame_transforms;\r
+\r
+ BOOST_FOREACH(auto& item, items)\r
+ { \r
+ const auto next = item.transform;\r
+ auto prev = next;\r
+\r
+ const auto it = prev_frame_transforms_.find(item.tag);\r
+ if(it != prev_frame_transforms_.end())\r
+ prev = it->second;\r
+ \r
+ next_frame_transforms[item.tag] = next; // Store all active tags, inactive tags will be removed at the end.\r
+ \r
+ if(next.volume < 0.001 && prev.volume < 0.001)\r
+ continue;\r
+ \r
+ if(static_cast<size_t>(item.audio_data.size()) != format_desc_.audio_samples_per_frame)\r
+ continue;\r
+\r
+ CASPAR_ASSERT(format_desc_.audio_channels == 2);\r
+ \r
+ const float prev_volume = static_cast<float>(prev.volume);\r
+ const float next_volume = static_cast<float>(next.volume);\r
+ const float delta = 1.0f/static_cast<float>(format_desc_.audio_samples_per_frame/2);\r
+ \r
+ tbb::parallel_for\r
+ (\r
+ tbb::blocked_range<size_t>(0, format_desc_.audio_samples_per_frame/2),\r
+ [&](const tbb::blocked_range<size_t>& r)\r
+ {\r
+ for(size_t n = r.begin(); n < r.end(); ++n)\r
+ {\r
+ const float alpha = n * delta;\r
+ const float volume = prev_volume * (1.0f - alpha) + next_volume * alpha;\r
+\r
+ auto sample_epi32 = _mm_loadl_epi64(reinterpret_cast<__m128i*>(&item.audio_data[n*2]));\r
+\r
+ auto sample_ps = _mm_cvtepi32_ps(sample_epi32); \r
+ sample_ps = _mm_mul_ps(sample_ps, _mm_set1_ps(volume)); \r
+\r
+ auto res_sample_epi32 = _mm_loadl_epi64(reinterpret_cast<__m128i*>(&result[n*2]));\r
+ auto res_sample_ps = _mm_cvtepi32_ps(res_sample_epi32); \r
+\r
+ res_sample_ps = _mm_add_ps(sample_ps, res_sample_ps);\r
+ res_sample_epi32 = _mm_cvtps_epi32(res_sample_ps);\r
+ \r
+ _mm_storel_epi64(reinterpret_cast<__m128i*>(&result[n*2]), res_sample_epi32);\r
+ }\r
+ }\r
+ );\r
+ }\r
+\r
+ items.clear();\r
+ prev_frame_transforms_ = std::move(next_frame_transforms); \r
+\r
+ result.resize(format_desc_.audio_samples_per_frame);\r
return std::move(result);\r
}\r
};\r
\r
-audio_mixer::audio_mixer() : impl_(new implementation()){}\r
+audio_mixer::audio_mixer(const core::video_format_desc& format_desc) : impl_(new implementation(format_desc)){}\r
void audio_mixer::begin(core::basic_frame& frame){impl_->begin(frame);}\r
void audio_mixer::visit(core::write_frame& frame){impl_->visit(frame);}\r
void audio_mixer::end(){impl_->end();}\r
-std::vector<int16_t> audio_mixer::mix(){return impl_->mix();}\r
+audio_buffer audio_mixer::mix(){return impl_->mix();}\r
audio_mixer& audio_mixer::operator=(audio_mixer&& other)\r
{\r
impl_ = std::move(other.impl_);\r