diff --git a/src/dusk/audio/DuskDsp.cpp b/src/dusk/audio/DuskDsp.cpp index 4009b37e39..0122adbc43 100644 --- a/src/dusk/audio/DuskDsp.cpp +++ b/src/dusk/audio/DuskDsp.cpp @@ -101,21 +101,6 @@ static u32 ConvertSamplesToDataLength(const JASDsp::TChannel& channel, u32 sampl return (samples / channel.mSamplesPerBlock) * BlockBytes(channel); } -/** - * Render the audio data contributed by a single DSP channel. Reads & decodes new input samples. - */ -static void RenderChannel( - JASDsp::TChannel& channel, - ChannelAuxData& channelAux, - OutputSubframe& subframe); - -static void RenderOutputChannel( - const JASDsp::TChannel& sourceChannel, - ChannelAuxData& aux, - OutputChannel outputChannel, - const std::span inputSamples, - OutputSubframe& fullOutputSubframe); - /** * Converts a pitch value on a DSP channel to a sample rate. */ @@ -203,16 +188,14 @@ static void GenerateEvolvingHarmonic() { } } - static void RenderOscChannel( JASDsp::TChannel& channel, ChannelAuxData& channelAux, - OutputSubframe& subframe) { + DspSubframe& buf) { if (channel.mResetFlag) ResetChannel(channel, channelAux); const u32 pitch = channel.mPitch; - DspSubframe buf = {}; const auto oscType = static_cast(channel.mBytesPerBlock); switch (oscType) { @@ -270,142 +253,6 @@ static void RenderOscChannel( DuskLog.error("RenderOscChannel: unimplemented oscillator type {}", channel.mBytesPerBlock); break; } - - auto samples = std::span(buf).subspan(0, DSP_SUBFRAME_SIZE); - RenderOutputChannel(channel, channelAux, OutputChannel::LEFT, samples, subframe); - RenderOutputChannel(channel, channelAux, OutputChannel::RIGHT, samples, subframe); -} - - -void dusk::audio::DspRender(OutputSubframe& subframe) { - ZoneScoped; - if (DumpAudio != sDumpWasActive) { - sDumpWasActive = DumpAudio; - if (DumpAudio) { - OpenChannelDumpFiles(); - } else { - CloseChannelDumpFiles(); - } - } - - GenerateEvolvingHarmonic(); - - std::span channels(JASDsp::CH_BUF, DSP_CHANNELS); - - DspSubframe reverbInputL = {}; - DspSubframe reverbInputR = {}; - bool anyReverbInput = false; - - DspSubframe surroundBus = {}; - bool anySurroundInput = false; - - for (int i = 0; i < channels.size(); i++) { - auto& channel = channels[i]; - auto& channelAux = ChannelAux[i]; - - if (!channel.mIsActive) { - continue; - } - else if (channel.mPauseFlag) { - // Not really sure what the practical difference between pause and - // deactivation is. Either avoids clearing state or allows the DSP to avoid popping? - continue; - } - else if (channel.mForcedStop) { - channel.mIsFinished = true; - continue; - } - - OutputSubframe channelSubframe = {}; - if (channel.mWaveAramAddress == 0) { - RenderOscChannel(channel, channelAux, channelSubframe); - } else { - ValidateChannel(channel); - RenderChannel(channel, channelAux, channelSubframe); - } - - if (EnableReverb) { - // scale the input to the reverb rather than using wet/dry on the output. - // this way the reverb's internal buffers accumulate energy proportional to mAutoMixerFxMix, - // so any tail always decays at the correct level regardless of mAutoMixerFxMix changes - // prevents transients when the next sound starts playing with a different reverb level - // 600.0f was pulled out of my ass and just sounds good enough for console - f32 inputGain = (channel.mAutoMixerFxMix >> 8) / 600.0f; - if (inputGain > 0) { - anyReverbInput = true; - for (int j = 0; j < DSP_SUBFRAME_SIZE; j++) { - reverbInputL[j] += channelSubframe.channels[0][j] * inputGain; - reverbInputR[j] += channelSubframe.channels[1][j] * inputGain; - } - } - } - - if (EnableHrtf && channel.mAutoMixerBeenSet) { - f32 dolby = (channel.mAutoMixerPanDolby & 0xFF) / 127.0f; - if (dolby > 0.0f) { - anySurroundInput = true; - f32 extract = dolby * HRTF_EXTRACT_MAX; - f32 frontScale = 1.0f - extract; - for (int j = 0; j < DSP_SUBFRAME_SIZE; j++) { - f32 mono = (channelSubframe.channels[0][j] + channelSubframe.channels[1][j]) * 0.5f; - surroundBus[j] += mono * extract; - channelSubframe.channels[0][j] *= frontScale; - channelSubframe.channels[1][j] *= frontScale; - } - } - } - - if (DumpAudio && sChannelDumpFiles[i]) { - f32 interleaved[DSP_SUBFRAME_SIZE * 2]; - for (int j = 0; j < DSP_SUBFRAME_SIZE; j++) { - interleaved[j * 2 + 0] = channelSubframe.channels[0][j]; - interleaved[j * 2 + 1] = channelSubframe.channels[1][j]; - } - fwrite(interleaved, sizeof(f32), DSP_SUBFRAME_SIZE * 2, sChannelDumpFiles[i]); - } - - for (int o = 0; o < subframe.channels.size(); o++) { - MixSubframe(subframe.channels[o], channelSubframe.channels[o]); - } - } - - if (EnableReverb && (anyReverbInput || ReverbHasTail)) { - // Equivalent to -80 dBFS: rms = 1e-4, rms^2 = 1e-8, sumSq = 2 * N * 1e-8 - constexpr f32 REVERB_ENERGY_EPSILON = 2.0f * DSP_SUBFRAME_SIZE * 1e-8f; - f32 wetEnergy = SharedReverb.processmix( - reverbInputL.data(), reverbInputR.data(), - subframe.channels[0].data(), subframe.channels[1].data(), - DSP_SUBFRAME_SIZE, 1, 1.0f - ); - ReverbHasTail = wetEnergy >= REVERB_ENERGY_EPSILON; - } - - if (EnableHrtf && anySurroundInput) { - // Two-pole LPF: -12 dB/oct above 3 kHz - for (int j = 0; j < DSP_SUBFRAME_SIZE; j++) { - sHrtfLp1 = (1.0f - HRTF_LP_K) * sHrtfLp1 + HRTF_LP_K * surroundBus[j]; - sHrtfLp2 = (1.0f - HRTF_LP_K) * sHrtfLp2 + HRTF_LP_K * sHrtfLp1; - surroundBus[j] = sHrtfLp2; - } - - // Mix into L and R - // L gets the filtered signal directly; R gets it allpass for mild decorrelation - for (int j = 0; j < DSP_SUBFRAME_SIZE; j++) { - f32 s = surroundBus[j]; - - subframe.channels[0][j] += s * HrtfGain; - - f32 r = -HRTF_ALLPASS_G * s + sHrtfApIn1 + HRTF_ALLPASS_G * sHrtfApOut1; - sHrtfApIn1 = s; - sHrtfApOut1 = r; - subframe.channels[1][j] += r * HrtfGain; - } - } - - for (auto& channel : subframe.channels) { - ApplyVolume(channel, channel, PrevMasterVolume, MasterVolume); - } - PrevMasterVolume = MasterVolume; } /** @@ -539,128 +386,12 @@ static void FillDecodeBuf(JASDsp::TChannel& channel, ChannelAuxData& aux, int ne } /** - * Get the expected BusConnect value needed to define the given output channel in a DSP channel. - */ -constexpr u16 GetBusConnect(const OutputChannel channel) { - switch (channel) { - // TODO: This is a guess for now. - case OutputChannel::LEFT: - return 0x0D00; - case OutputChannel::RIGHT: - return 0x0D60; - default: - CRASH("Invalid output channel!"); - } -} - -/** - * For a DSP channel the JASDsp::OutputChannelConfig value targeting the given output channel. - * Returns null if the DSP channel does not output to this output channel. - */ -static const JASDsp::OutputChannelConfig* GetOutputConfig( - const JASDsp::TChannel& sourceChannel, - OutputChannel channel) { - - auto busConnect = GetBusConnect(channel); - for (const auto& mOutputChannel : sourceChannel.mOutputChannels) { - auto config = &mOutputChannel; - if (config->mBusConnect == busConnect) { - return config; - } - } - - return nullptr; -} - -struct VolumeValue { - f32 Target; - f32 Init; -}; - -/** - * Get the volume that the given DSP channel should render to the given output channel at. - */ -static VolumeValue GetVolumeForOutputChannel( - const JASDsp::TChannel& sourceChannel, - OutputChannel outputChannel) { - - u16 volume; - u16 initVolume; - f32 panValue = 1; - if (sourceChannel.mAutoMixerBeenSet) { - volume = sourceChannel.mAutoMixerVolume; - initVolume = sourceChannel.mAutoMixerInitVolume; - - auto autoMixerPan = static_cast(sourceChannel.mAutoMixerPanDolby >> 8) / 127; - - switch (outputChannel) { - case OutputChannel::LEFT: - panValue = 1 - autoMixerPan; - break; - case OutputChannel::RIGHT: - panValue = autoMixerPan; - break; - default: - CRASH("Unhandled output channel: OutputChannel"); - } - - } else { - auto config = GetOutputConfig(sourceChannel, outputChannel); - if (config == nullptr) { - return {0, 0}; - } - - volume = config->mTargetVolume; - initVolume = config->mCurrentVolume; - } - - // TODO: interpolate to avoid popping. - f32 targetRatio = VolumeFromU16(volume); - targetRatio *= panValue; - - f32 initRatio = VolumeFromU16(initVolume); - initRatio *= panValue; - - return {targetRatio, initRatio}; -} - -/** - * Given decoded & resampled input samples, render a DSP channel to a given output channel. - */ -static void RenderOutputChannel( - const JASDsp::TChannel& sourceChannel, - ChannelAuxData& aux, - OutputChannel outputChannel, - const std::span inputSamples, - OutputSubframe& fullOutputSubframe) { - - auto& outputSubframe = fullOutputSubframe[outputChannel]; - assert(inputSamples.size() <= outputSubframe.size()); - - auto volume = GetVolumeForOutputChannel(sourceChannel, outputChannel); - - f32 targetVolume = volume.Target; - auto& prevVolume = aux.PrevVolume(outputChannel); - if (std::isnan(prevVolume)) { - // Initialize previous volume to new volume on first render. - prevVolume = volume.Init; - } - - if (prevVolume == 0 && targetVolume == 0) { - return; - } - - ApplyVolume(outputSubframe, inputSamples, prevVolume, targetVolume); - prevVolume = targetVolume; -} - -/** - * Fetch, decode, resample, output + * Render the audio data contributed by a single DSP channel. Reads & decodes new input samples. */ static void RenderChannel( JASDsp::TChannel& channel, ChannelAuxData& channelAux, - OutputSubframe& subframe) { + DspSubframe& buf) { if (channel.mResetFlag) { ResetChannel(channel, channelAux); @@ -679,7 +410,6 @@ static void RenderChannel( channel.mIsFinished = true; } - DspSubframe audioLoadBuffer = {}; f32 pos = channelAux.resamplePos; s16 prev = channelAux.resamplePrev; s16 next = channelAux.decodeBufCount > 0 ? channelAux.decodeBuf[0] : prev; @@ -687,7 +417,7 @@ static void RenderChannel( // linear resampling and f32 conversion for (int i = 0; i < DSP_SUBFRAME_SIZE; i++) { - audioLoadBuffer[i] = (prev + pos * (next - prev)) / 32768.0f; + buf[i] = (prev + pos * (next - prev)) / 32768.0f; pos += step; while (pos >= 1.0f) { pos -= 1.0f; @@ -705,7 +435,7 @@ static void RenderChannel( // IIR part 1, low-pass: out[n] = (in[n] - in[n-1]) * (coeff/128) + out[n-1] if (s16 coeff = channel.iir_filter_params[4]; coeff != 0) { - for (f32& sample : audioLoadBuffer) { + for (f32& sample : buf) { f32 out = std::clamp( (sample - channelAux.prev_lp_in) * ((f32)coeff / 128.0f) + channelAux.prev_lp_out, -1.0f, 1.0f ); @@ -717,7 +447,7 @@ static void RenderChannel( // IIR part 2, biquad: out[n] = (b1*in[n-1] + b2*in[n-2] + a1*out[n-1] + a2*out[n-2]) / 32768 if ((channel.mFilterMode & 0x20) != 0) { - for (f32& sample : audioLoadBuffer) { + for (f32& sample : buf) { f32 out = std::clamp(( channel.iir_filter_params[0] * channelAux.biq_in1 + // b1 channel.iir_filter_params[1] * channelAux.biq_in2 + // b2 @@ -740,28 +470,16 @@ static void RenderChannel( } channelAux.decodeBufCount = std::max(0, remainingDecodeBuf); - - auto hasReadSamples = std::span(audioLoadBuffer).subspan(0, DSP_SUBFRAME_SIZE); - - static_assert(OutputSubframe::NUM_CHANNELS == 2, "Keep RenderChannel in sync!"); - - RenderOutputChannel(channel, channelAux, OutputChannel::LEFT, hasReadSamples, subframe); - RenderOutputChannel(channel, channelAux, OutputChannel::RIGHT, hasReadSamples, subframe); } -void dusk::audio::DspInit() { - SharedReverb.setwet(1.0f); - SharedReverb.setdry(0.0f); - SharedReverb.setroomsize(0.5f); - SharedReverb.setdamp(0.7f); - SharedReverb.setwidth(1.0f); - SharedReverb.setmode(0.0f); - SharedReverb.mute(); -} +struct VolumeValue { + f32 Target; + f32 Init; +}; -void dusk::audio::ApplyVolume( +static void ApplyVolume( std::span dst, - const std::span src, + std::span src, const f32 startVolume, const f32 endVolume) { assert(dst.size() >= src.size()); @@ -777,3 +495,208 @@ void dusk::audio::ApplyVolume( } } } + +static void ApplyPanning( + const JASDsp::TChannel& voice, + ChannelAuxData& aux, + const DspSubframe& input, + OutputSubframe& output) +{ + std::array volumes = {}; + + if (voice.mAutoMixerBeenSet) { + const auto volume = VolumeFromU16(voice.mAutoMixerVolume); + const auto initVolume = VolumeFromU16(voice.mAutoMixerInitVolume); + + const auto right = static_cast(voice.mAutoMixerPanDolby >> 8) / 127.0f; + const auto left = 1.0f - right; + + volumes[0] = {left * volume, left * initVolume}; + volumes[1] = {right * volume, right * initVolume}; + } else { + for (const auto& outChannel : voice.mOutputChannels) { + std::optional ch; + + switch (outChannel.mBusConnect) { + case 0x0D00: + ch = OutputChannel::LEFT; + break; + case 0x0D60: + ch = OutputChannel::RIGHT; + break; + default: + break; + } + + if (ch) { + auto& v = volumes[static_cast(*ch)]; + v.Target = VolumeFromU16(outChannel.mTargetVolume); + v.Init = VolumeFromU16(outChannel.mCurrentVolume); + } + } + } + + for (size_t i = 0; i < OutputSubframe::NUM_CHANNELS; i++) { + const auto ch = static_cast(i); + const auto& volume = volumes[i]; + + const f32 targetVolume = volume.Target; + auto& prevVolume = aux.PrevVolume(ch); + + if (std::isnan(prevVolume)) { + // Initialize previous volume to new volume on first render. + prevVolume = volume.Init; + } + + if (prevVolume == 0 && targetVolume == 0) { + continue; + } + + ApplyVolume(output[ch], input, prevVolume, targetVolume); + prevVolume = targetVolume; + } +} + +void dusk::audio::DspInit() { + SharedReverb.setwet(1.0f); + SharedReverb.setdry(0.0f); + SharedReverb.setroomsize(0.5f); + SharedReverb.setdamp(0.7f); + SharedReverb.setwidth(1.0f); + SharedReverb.setmode(0.0f); + SharedReverb.mute(); +} + +void dusk::audio::DspRender(OutputSubframe& subframe) { + ZoneScoped; + if (DumpAudio != sDumpWasActive) { + sDumpWasActive = DumpAudio; + if (DumpAudio) { + OpenChannelDumpFiles(); + } else { + CloseChannelDumpFiles(); + } + } + + GenerateEvolvingHarmonic(); + + std::span voices(JASDsp::CH_BUF, DSP_CHANNELS); + + DspSubframe reverbInputL = {}; + DspSubframe reverbInputR = {}; + bool anyReverbInput = false; + + DspSubframe surroundBus = {}; + bool anySurroundInput = false; + + for (int i = 0; i < voices.size(); i++) { + auto& voice = voices[i]; + auto& aux = ChannelAux[i]; + + if (!voice.mIsActive) { + continue; + } + else if (voice.mPauseFlag) { + // Not really sure what the practical difference between pause and + // deactivation is. Either avoids clearing state or allows the DSP to avoid popping? + continue; + } + else if (voice.mForcedStop) { + voice.mIsFinished = true; + continue; + } + + DspSubframe monoBuf = {}; + if (voice.mWaveAramAddress == 0) { + RenderOscChannel(voice, aux, monoBuf); + } else { + ValidateChannel(voice); + RenderChannel(voice, aux, monoBuf); + } + + OutputSubframe buf = {}; + ApplyPanning(voice, aux, monoBuf, buf); + + if (EnableReverb) { + // scale the input to the reverb rather than using wet/dry on the output. + // this way the reverb's internal buffers accumulate energy proportional to mAutoMixerFxMix, + // so any tail always decays at the correct level regardless of mAutoMixerFxMix changes + // prevents transients when the next sound starts playing with a different reverb level + // 600.0f was pulled out of my ass and just sounds good enough for console + f32 inputGain = (voice.mAutoMixerFxMix >> 8) / 600.0f; + if (inputGain > 0) { + anyReverbInput = true; + for (int j = 0; j < DSP_SUBFRAME_SIZE; j++) { + reverbInputL[j] += buf.channels[0][j] * inputGain; + reverbInputR[j] += buf.channels[1][j] * inputGain; + } + } + } + + if (EnableHrtf && voice.mAutoMixerBeenSet) { + f32 dolby = (voice.mAutoMixerPanDolby & 0xFF) / 127.0f; + if (dolby > 0.0f) { + anySurroundInput = true; + f32 extract = dolby * HRTF_EXTRACT_MAX; + f32 frontScale = 1.0f - extract; + for (int j = 0; j < DSP_SUBFRAME_SIZE; j++) { + f32 mono = (buf.channels[0][j] + buf.channels[1][j]) * 0.5f; + surroundBus[j] += mono * extract; + buf.channels[0][j] *= frontScale; + buf.channels[1][j] *= frontScale; + } + } + } + + if (DumpAudio && sChannelDumpFiles[i]) { + f32 interleaved[DSP_SUBFRAME_SIZE * 2]; + for (int j = 0; j < DSP_SUBFRAME_SIZE; j++) { + interleaved[j * 2 + 0] = buf.channels[0][j]; + interleaved[j * 2 + 1] = buf.channels[1][j]; + } + fwrite(interleaved, sizeof(f32), DSP_SUBFRAME_SIZE * 2, sChannelDumpFiles[i]); + } + + for (int o = 0; o < subframe.channels.size(); o++) { + MixSubframe(subframe.channels[o], buf.channels[o]); + } + } + + if (EnableReverb && (anyReverbInput || ReverbHasTail)) { + // Equivalent to -80 dBFS: rms = 1e-4, rms^2 = 1e-8, sumSq = 2 * N * 1e-8 + constexpr f32 REVERB_ENERGY_EPSILON = 2.0f * DSP_SUBFRAME_SIZE * 1e-8f; + f32 wetEnergy = SharedReverb.processmix( + reverbInputL.data(), reverbInputR.data(), + subframe.channels[0].data(), subframe.channels[1].data(), + DSP_SUBFRAME_SIZE, 1, 1.0f + ); + ReverbHasTail = wetEnergy >= REVERB_ENERGY_EPSILON; + } + + if (EnableHrtf && anySurroundInput) { + // Two-pole LPF: -12 dB/oct above 3 kHz + for (int j = 0; j < DSP_SUBFRAME_SIZE; j++) { + sHrtfLp1 = (1.0f - HRTF_LP_K) * sHrtfLp1 + HRTF_LP_K * surroundBus[j]; + sHrtfLp2 = (1.0f - HRTF_LP_K) * sHrtfLp2 + HRTF_LP_K * sHrtfLp1; + surroundBus[j] = sHrtfLp2; + } + + // Mix into L and R + // L gets the filtered signal directly; R gets it allpass for mild decorrelation + for (int j = 0; j < DSP_SUBFRAME_SIZE; j++) { + f32 s = surroundBus[j]; + + subframe.channels[0][j] += s * HrtfGain; + + f32 r = -HRTF_ALLPASS_G * s + sHrtfApIn1 + HRTF_ALLPASS_G * sHrtfApOut1; + sHrtfApIn1 = s; + sHrtfApOut1 = r; + subframe.channels[1][j] += r * HrtfGain; + } + } + + for (auto& channel : subframe.channels) { + ApplyVolume(channel, channel, PrevMasterVolume, MasterVolume); + } + PrevMasterVolume = MasterVolume; +} diff --git a/src/dusk/audio/DuskDsp.hpp b/src/dusk/audio/DuskDsp.hpp index cfcbfe3f46..a2504ab586 100644 --- a/src/dusk/audio/DuskDsp.hpp +++ b/src/dusk/audio/DuskDsp.hpp @@ -5,9 +5,6 @@ #include #include -#include "SDL3/SDL_audio.h" -#include - // ReSharper disable once CppUnusedIncludeDirective #include "global.h" @@ -123,12 +120,6 @@ namespace dusk::audio { return channel.mBytesPerBlock; } - /** - * Apply a volume level to audio data. - * Interpolates across the two provided volume levels to avoid clicking. - */ - void ApplyVolume(std::span dst, std::span src, f32 startVolume, f32 endVolume); - extern f32 MasterVolume; extern f32 PrevMasterVolume; extern bool EnableReverb;