winamp/Src/external_dependencies/openmpt-trunk/soundlib/AudioReadTarget.h

143 lines
3.8 KiB
C
Raw Normal View History

2024-09-24 12:54:57 +00:00
/*
* AudioReadTarget.h
* -----------------
* Purpose: Callback class implementations for audio data read via CSoundFile::Read.
* Notes : (currently none)
* Authors: OpenMPT Devs
* The OpenMPT source code is released under the BSD license. Read LICENSE for more details.
*/
#pragma once
#include "openmpt/all/BuildSettings.hpp"
#include "Sndfile.h"
#include "mpt/audio/span.hpp"
#include "openmpt/soundbase/SampleFormat.hpp"
#include "openmpt/soundbase/CopyMix.hpp"
#include "openmpt/soundbase/Dither.hpp"
#include "MixerLoops.h"
#include "Mixer.h"
#include "../common/Dither.h"
#include <type_traits>
OPENMPT_NAMESPACE_BEGIN
template <typename Taudio_span, typename TDithers = DithersOpenMPT>
class AudioTargetBuffer
: public IAudioTarget
{
private:
std::size_t countRendered;
TDithers &dithers;
protected:
Taudio_span outputBuffer;
public:
AudioTargetBuffer(Taudio_span buf, TDithers &dithers_)
: countRendered(0)
, dithers(dithers_)
, outputBuffer(buf)
{
return;
}
std::size_t GetRenderedCount() const { return countRendered; }
public:
void Process(mpt::audio_span_interleaved<MixSampleInt> buffer) override
{
std::visit(
[&](auto &ditherInstance)
{
ConvertBufferMixInternalFixedToBuffer<MixSampleIntTraits::mix_fractional_bits, false>(mpt::make_audio_span_with_offset(outputBuffer, countRendered), buffer, ditherInstance, buffer.size_channels(), buffer.size_frames());
},
dithers.Variant()
);
countRendered += buffer.size_frames();
}
void Process(mpt::audio_span_interleaved<MixSampleFloat> buffer) override
{
std::visit(
[&](auto &ditherInstance)
{
ConvertBufferMixInternalToBuffer<false>(mpt::make_audio_span_with_offset(outputBuffer, countRendered), buffer, ditherInstance, buffer.size_channels(), buffer.size_frames());
},
dithers.Variant()
);
countRendered += buffer.size_frames();
}
};
template <typename Taudio_span, typename TDithers = DithersOpenMPT>
class AudioTargetBufferWithGain
: public AudioTargetBuffer<Taudio_span>
{
private:
using Tbase = AudioTargetBuffer<Taudio_span>;
private:
const float gainFactor;
public:
AudioTargetBufferWithGain(Taudio_span buf, TDithers &dithers, float gainFactor_)
: Tbase(buf, dithers)
, gainFactor(gainFactor_)
{
return;
}
public:
void Process(mpt::audio_span_interleaved<MixSampleInt> buffer) override
{
const std::size_t countRendered_ = Tbase::GetRenderedCount();
if constexpr(!std::is_floating_point<typename Taudio_span::sample_type>::value)
{
int32 gainFactor16_16 = mpt::saturate_round<int32>(gainFactor * (1 << 16));
if(gainFactor16_16 != (1<<16))
{
// only apply gain when != +/- 0dB
// no clipping prevention is done here
for(std::size_t frame = 0; frame < buffer.size_frames(); ++frame)
{
for(std::size_t channel = 0; channel < buffer.size_channels(); ++channel)
{
buffer(channel, frame) = Util::muldiv(buffer(channel, frame), gainFactor16_16, 1 << 16);
}
}
}
}
Tbase::Process(buffer);
if constexpr(std::is_floating_point<typename Taudio_span::sample_type>::value)
{
if(gainFactor != 1.0f)
{
// only apply gain when != +/- 0dB
for(std::size_t frame = 0; frame < buffer.size_frames(); ++frame)
{
for(std::size_t channel = 0; channel < buffer.size_channels(); ++channel)
{
Tbase::outputBuffer(channel, countRendered_ + frame) *= gainFactor;
}
}
}
}
}
void Process(mpt::audio_span_interleaved<MixSampleFloat> buffer) override
{
if(gainFactor != 1.0f)
{
// only apply gain when != +/- 0dB
for(std::size_t frame = 0; frame < buffer.size_frames(); ++frame)
{
for(std::size_t channel = 0; channel < buffer.size_channels(); ++channel)
{
buffer(channel, frame) *= gainFactor;
}
}
}
Tbase::Process(buffer);
}
};
OPENMPT_NAMESPACE_END