ppsspp/Core/HW/SasAudio.cpp

// Copyright (c) 2012- PPSSPP Project.

// This program 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, version 2.0 or later versions.

// This program 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 2.0 for more details.

// A copy of the GPL 2.0 should have been included with the program.
// If not, see http://www.gnu.org/licenses/

// Official git repository and contact information can be found at
// https://github.com/hrydgard/ppsspp and http://www.ppsspp.org/.

#include <algorithm>

#include "Common/Profiler/Profiler.h"

#include "Common/Serialize/SerializeFuncs.h"
#include "Core/MemMapHelpers.h"
#include "Core/HLE/sceAtrac.h"
#include "Core/Config.h"
#include "Core/Reporting.h"
#include "Core/Util/AudioFormat.h"
#include "Core/System.h"
#include "SasAudio.h"

static const u8 f[16][2] = {
	{   0,   0 },
	{  60,   0 },
	{ 115,  52 },
	{  98,  55 },
	{ 122,  60 },
	// Everything past index 4 is the hardware reading off the end of its own table, so this is garbage extra values.
	{   0,   0 },
	{   0,   0 },
	{  52,   0 },
	{  55,   2 },
	{  60, 125 },
	{   0,   0 },
	{   0,  91 },
	{   0,   0 },
	{   2, 216 },
	{ 125,   6 },
	{   0, 151 },
};

void VagDecoder::Start(u32 data, u32 vagSize, bool loopEnabled) {
	loopEnabled_ = loopEnabled;
	loopAtNextBlock_ = false;
	loopStartBlock_ = -1;
	numBlocks_ = vagSize / 16;
	end_ = false;
	data_ = data;
	read_ = data;
	curSample = 28;
	curBlock_ = -1;
	s_1 = 0;	// per block?
	s_2 = 0;
}

void VagDecoder::DecodeBlock(const u8 *&read_pointer) {
	if (curBlock_ == numBlocks_ - 1) {
		end_ = true;
		return;
	}

	_dbg_assert_(curBlock_ < numBlocks_);

	const u8 *readp = read_pointer;
	int predict_nr = *readp++;
	int shift_factor = predict_nr & 0xf;
	predict_nr >>= 4;
	int flags = *readp++;
	if (flags == 7) {
		VERBOSE_LOG(Log::SasMix, "VAG ending block at %d", curBlock_);
		end_ = true;
		return;
	}
	else if (flags == 6) {
		loopStartBlock_ = curBlock_;
	}
	else if (flags == 3) {
		if (loopEnabled_) {
			loopAtNextBlock_ = true;
		}
	}

	// Keep state in locals to avoid bouncing to memory.
	int s1 = s_1;
	int s2 = s_2;

	int coef1 = f[predict_nr][0];
	int coef2 = -f[predict_nr][1];

	// TODO: Unroll once more and interleave the unpacking with the decoding more?
	for (int i = 0; i < 28; i += 2) {
		u8 d = *readp++;
		int sample1 = (short)((d & 0xf) << 12) >> shift_factor;
		int sample2 = (short)((d & 0xf0) << 8) >> shift_factor;
		s2 = clamp_s16(sample1 + ((s1 * coef1 + s2 * coef2) >> 6));
		s1 = clamp_s16(sample2 + ((s2 * coef1 + s1 * coef2) >> 6));
		samples[i] = s2;
		samples[i + 1] = s1;
	}

	s_1 = s1;
	s_2 = s2;
	curSample = 0;
	curBlock_++;

	read_pointer = readp;
}

void VagDecoder::GetSamples(s16 *outSamples, int numSamples) {
	if (end_) {
		memset(outSamples, 0, numSamples * sizeof(s16));
		return;
	}
	if (!Memory::IsValidRange(read_, numBlocks_ * 16)) {
		WARN_LOG_REPORT_ONCE(badvagaddr, Log::SasMix, "Bad VAG samples address? %08x / %d", read_, numBlocks_);
		return;
	}

	const u8 *readp = Memory::GetPointerUnchecked(read_);
	const u8 *origp = readp;

	for (int i = 0; i < numSamples; i++) {
		if (curSample == 28) {
			if (loopAtNextBlock_) {
				VERBOSE_LOG(Log::SasMix, "Looping VAG from block %d/%d to %d", curBlock_, numBlocks_, loopStartBlock_);
				// data_ starts at curBlock = -1.
				read_ = data_ + 16 * loopStartBlock_ + 16;
				readp = Memory::GetPointerUnchecked(read_);
				origp = readp;
				curBlock_ = loopStartBlock_;
				loopAtNextBlock_ = false;
			}
			DecodeBlock(readp);
			if (end_) {
				// Clear the rest of the buffer and return.
				memset(&outSamples[i], 0, (numSamples - i) * sizeof(s16));
				return;
			}
		}
		_dbg_assert_(curSample < 28);
		outSamples[i] = samples[curSample++];
	}

	if (readp > origp) {
		if (MemBlockInfoDetailed())
			NotifyMemInfo(MemBlockFlags::READ, read_, readp - origp, "SasVagDecoder");
		read_ += readp - origp;
	}
}

void VagDecoder::DoState(PointerWrap &p) {
	auto s = p.Section("VagDecoder", 1, 2);
	if (!s)
		return;

	if (s >= 2) {
		DoArray(p, samples, ARRAY_SIZE(samples));
	} else {
		int samplesOld[ARRAY_SIZE(samples)];
		DoArray(p, samplesOld, ARRAY_SIZE(samples));
		for (size_t i = 0; i < ARRAY_SIZE(samples); ++i) {
			samples[i] = samplesOld[i];
		}
	}
	Do(p, curSample);

	Do(p, data_);
	Do(p, read_);
	Do(p, curBlock_);
	Do(p, loopStartBlock_);
	Do(p, numBlocks_);

	Do(p, s_1);
	Do(p, s_2);

	Do(p, loopEnabled_);
	Do(p, loopAtNextBlock_);
	Do(p, end_);
}

// The context pointer is assumed to be valid.
int SasAtrac3::SetContext(u32 contextAddr) {
	contextAddr_ = contextAddr;
	// Note: On hardware, atracID_ is also stored in the loopNum member of the context.
	// But we don't actually mirror our struct to memory, so it doesn't really matter.
	atracID_ = AtracSasBindContextAndGetID(contextAddr);
	if (!sampleQueue_)
		sampleQueue_ = new BufferQueue();
	sampleQueue_->clear();
	end_ = false;
	return 0;
}

void SasAtrac3::GetNextSamples(s16 *outbuf, int wantedSamples) {
	if (atracID_ < 0) {
		end_ = true;
		return;
	}

	if (!buf_) {
		buf_ = new s16[0x800];
	}

	int finish = 0;
	int wantedbytes = wantedSamples * sizeof(s16);
	while (!finish && sampleQueue_->getQueueSize() < wantedbytes) {
		int numSamples = 0;
		AtracSasDecodeData(atracID_, (u8*)buf_, &numSamples, &finish);
		if (numSamples > 0)
			sampleQueue_->push((u8*)buf_, numSamples * sizeof(s16));
		else
			finish = 1;
	}
	sampleQueue_->pop_front((u8*)outbuf, wantedbytes);
	end_ = finish == 1;
}

int SasAtrac3::Concatenate(u32 bufPtr, u32 addbytes) {
	if (atracID_ >= 0) {
		AtracSasAddStreamData(atracID_, bufPtr, addbytes);
	}
	return 0;
}

void SasAtrac3::DoState(PointerWrap &p) {
	auto s = p.Section("SasAtrac3", 1, 2);
	if (!s)
		return;

	Do(p, contextAddr_);
	Do(p, atracID_);
	if (p.mode == p.MODE_READ && atracID_ >= 0 && !sampleQueue_) {
		sampleQueue_ = new BufferQueue();
	}
	if (s >= 2) {
		Do(p, end_);
	}
}

// http://code.google.com/p/jpcsp/source/browse/trunk/src/jpcsp/HLE/modules150/sceSasCore.java

static int simpleRate(int n) {
	n &= 0x7F;
	if (n == 0x7F) {
		return 0;
	}
	int rate = ((7 - (n & 0x3)) << 26) >> (n >> 2);
	if (rate == 0) {
		return 1;
	}
	return rate;
}

static int exponentRate(int n) {
	n &= 0x7F;
	if (n == 0x7F) {
		return 0;
	}
	int rate = ((7 - (n & 0x3)) << 24) >> (n >> 2);
	if (rate == 0) {
		return 1;
	}
	return rate;
}

static int getAttackRate(int bitfield1) {
	return simpleRate(bitfield1 >> 8);
}

static int getAttackType(int bitfield1) {
	return (bitfield1 & 0x8000) == 0 ? PSP_SAS_ADSR_CURVE_MODE_LINEAR_INCREASE : PSP_SAS_ADSR_CURVE_MODE_LINEAR_BENT;
}

static int getDecayRate(int bitfield1) {
	int n = (bitfield1 >> 4) & 0x000F;
	if (n == 0)
		return 0x7FFFFFFF;
	return 0x80000000 >> n;
}

static int getSustainType(int bitfield2) {
	return (bitfield2 >> 14) & 3;
}

static int getSustainRate(int bitfield2) {
	if (getSustainType(bitfield2) == PSP_SAS_ADSR_CURVE_MODE_EXPONENT_DECREASE) {
		return exponentRate(bitfield2 >> 6);
	} else {
		return simpleRate(bitfield2 >> 6);
	}
}

static int getReleaseType(int bitfield2) {
	return (bitfield2 & 0x0020) == 0 ? PSP_SAS_ADSR_CURVE_MODE_LINEAR_DECREASE : PSP_SAS_ADSR_CURVE_MODE_EXPONENT_DECREASE;
}

static int getReleaseRate(int bitfield2) {
	int n = bitfield2 & 0x001F;
	if (n == 31) {
		return 0;
	}
	if (getReleaseType(bitfield2) == PSP_SAS_ADSR_CURVE_MODE_LINEAR_DECREASE) {
		if (n == 30) {
			return 0x40000000;
		} else if (n == 29) {
			return 1;
		}
		return 0x10000000 >> n;
	}
	if (n == 0)
		return 0x7FFFFFFF;
	return 0x80000000 >> n;
}

static int getSustainLevel(int bitfield1) {
	return ((bitfield1 & 0x000F) + 1) << 26;
}

void ADSREnvelope::SetEnvelope(int flag, int a, int d, int s, int r) {
	if ((flag & 0x1) != 0)
		attackType = (SasADSRCurveMode)a;
	if ((flag & 0x2) != 0)
		decayType = (SasADSRCurveMode)d;
	if ((flag & 0x4) != 0)
		sustainType = (SasADSRCurveMode)s;
	if ((flag & 0x8) != 0)
		releaseType = (SasADSRCurveMode)r;

	if (PSP_CoreParameter().compat.flags().RockmanDash2SoundFix && sustainType == PSP_SAS_ADSR_CURVE_MODE_LINEAR_INCREASE) {
		sustainType = PSP_SAS_ADSR_CURVE_MODE_LINEAR_DECREASE;
	}
}

void ADSREnvelope::SetRate(int flag, int a, int d, int s, int r) {
	if ((flag & 0x1) != 0)
		attackRate = a;
	if ((flag & 0x2) != 0)
		decayRate = d;
	if ((flag & 0x4) != 0)
		sustainRate = s;
	if ((flag & 0x8) != 0)
		releaseRate = r;
}

void ADSREnvelope::SetSimpleEnvelope(u32 ADSREnv1, u32 ADSREnv2) {
	attackRate 		= getAttackRate(ADSREnv1);
	attackType 		= (SasADSRCurveMode)getAttackType(ADSREnv1);
	decayRate 		= getDecayRate(ADSREnv1);
	decayType 		= PSP_SAS_ADSR_CURVE_MODE_EXPONENT_DECREASE;
	sustainRate 	= getSustainRate(ADSREnv2);
	sustainType 	= (SasADSRCurveMode)getSustainType(ADSREnv2);
	releaseRate 	= getReleaseRate(ADSREnv2);
	releaseType 	= (SasADSRCurveMode)getReleaseType(ADSREnv2);
	sustainLevel 	= getSustainLevel(ADSREnv1);

	if (PSP_CoreParameter().compat.flags().RockmanDash2SoundFix && sustainType == PSP_SAS_ADSR_CURVE_MODE_LINEAR_INCREASE) {
		sustainType = PSP_SAS_ADSR_CURVE_MODE_LINEAR_DECREASE;
	}

	if (attackRate < 0 || decayRate < 0 || sustainRate < 0 || releaseRate < 0) {
		ERROR_LOG_REPORT(Log::SasMix, "Simple ADSR resulted in invalid rates: %04x, %04x", ADSREnv1, ADSREnv2);
	}
}

SasInstance::SasInstance() {
	memset(&waveformEffect, 0, sizeof(waveformEffect));
	waveformEffect.type = PSP_SAS_EFFECT_TYPE_OFF;
	waveformEffect.isDryOn = 1;
	memset(mixTemp_, 0, sizeof(mixTemp_));  // just to avoid a static analysis warning.
}

SasInstance::~SasInstance() {
	ClearGrainSize();
}

void SasInstance::GetDebugText(char *text, size_t bufsize) {
	char voiceBuf[4096];
	voiceBuf[0] = '\0';
	char *p = voiceBuf;
	for (int i = 0; i < maxVoices; i++) {
		if (voices[i].playing) {
			uint32_t readAddr = voices[i].GetReadAddress();
			const char *indicator = "";
			switch (voices[i].type) {
			case VOICETYPE_VAG:
				if (readAddr < voices[i].vagAddr || readAddr > voices[i].vagAddr + voices[i].vagSize) {
					indicator = " (BAD!)";
				}
				break;
			default:
				break;
			}
			p += snprintf(p, sizeof(voiceBuf) - (p - voiceBuf), " %d: Pitch %04x L/R,FX: %d,%d|%d,%d VAG: %08x:%d:%08x%s Height:%d%%\n", i,
				voices[i].pitch, voices[i].volumeLeft, voices[i].volumeRight, voices[i].effectLeft, voices[i].effectRight,
				voices[i].vagAddr, voices[i].vagSize, voices[i].GetReadAddress(), indicator, (int)((int64_t)voices[i].envelope.GetHeight() * 100 / PSP_SAS_ENVELOPE_HEIGHT_MAX));
			p += snprintf(p, sizeof(voiceBuf) - (p - voiceBuf), "  - ADSR: %s/%s/%s/%s\n",
				ADSRCurveModeAsString(voices[i].envelope.attackType),
				ADSRCurveModeAsString(voices[i].envelope.decayType),
				ADSRCurveModeAsString(voices[i].envelope.sustainType),
				ADSRCurveModeAsString(voices[i].envelope.releaseType)
			);
		}
	}

	snprintf(text, bufsize,
		"SR: %d Mode: %s Grain: %d\n"
		"Effect: Type: %s Dry: %d Wet: %d L: %d R: %d Delay: %d Feedback: %d\n"
		"\n%s\n",
		sampleRate, outputMode == PSP_SAS_OUTPUTMODE_RAW ? "Raw" : "Mixed", grainSize,
		SasReverb::GetPresetName(waveformEffect.type), waveformEffect.isDryOn, waveformEffect.isWetOn, waveformEffect.leftVol, waveformEffect.rightVol, waveformEffect.delay, waveformEffect.feedback,
		voiceBuf);
}

void SasInstance::ClearGrainSize() {
	delete[] mixBuffer;
	delete[] sendBuffer;
	delete[] sendBufferDownsampled;
	delete[] sendBufferProcessed;
	mixBuffer = nullptr;
	sendBuffer = nullptr;
	sendBufferDownsampled = nullptr;
	sendBufferProcessed = nullptr;
}

void SasInstance::SetGrainSize(int newGrainSize) {
	grainSize = newGrainSize;

	// If you change the sizes here, don't forget DoState().
	delete[] mixBuffer;
	delete[] sendBuffer;
	delete[] sendBufferDownsampled;
	delete[] sendBufferProcessed;

	mixBuffer = new s32[grainSize * 2];
	sendBuffer = new s32[grainSize * 2];
	sendBufferDownsampled = new s16[grainSize];
	sendBufferProcessed = new s16[grainSize * 2];
	memset(mixBuffer, 0, sizeof(int) * grainSize * 2);
	memset(sendBuffer, 0, sizeof(int) * grainSize * 2);
	memset(sendBufferDownsampled, 0, sizeof(s16) * grainSize);
	memset(sendBufferProcessed, 0, sizeof(s16) * grainSize * 2);
}

int SasInstance::EstimateMixUs() {
	int voicesPlayingCount = 0;

	for (int v = 0; v < PSP_SAS_VOICES_MAX; v++) {
		SasVoice &voice = voices[v];
		if (!voice.playing || voice.paused)
			continue;
		voicesPlayingCount++;
	}

	// Each voice costs extra time, and each byte of grain costs extra time.
	int cycles = 20 + voicesPlayingCount * 68 + (grainSize * 60) / 100;
	// Cap to 1200 to fix FFT, see issue #9956.
	return std::min(cycles, 1200);
}

void SasVoice::ReadSamples(s16 *output, int numSamples) {
	// Read N samples into the resample buffer. Could do either PCM or VAG here.
	switch (type) {
	case VOICETYPE_VAG:
		vag.GetSamples(output, numSamples);
		break;
	case VOICETYPE_PCM:
		{
			int needed = numSamples;
			s16 *out = output;
			while (needed > 0) {
				u32 size = std::min(pcmSize - pcmIndex, needed);
				if (!on) {
					pcmIndex = 0;
					break;
				}
				Memory::Memcpy(out, pcmAddr + pcmIndex * sizeof(s16), size * sizeof(s16), "SasVoicePCM");
				pcmIndex += size;
				needed -= size;
				out += size;
				if (pcmIndex >= pcmSize) {
					if (!loop) {
						// All out, quit.  We'll end in HaveSamplesEnded().
						break;
					}
					pcmIndex = pcmLoopPos;
				}
			}
			if (needed > 0) {
				memset(out, 0, needed * sizeof(s16));
			}
		}
		break;
	case VOICETYPE_ATRAC3:
		atrac3.GetNextSamples(output, numSamples);
		break;
	default:
		memset(output, 0, numSamples * sizeof(s16));
		break;
	}
}

bool SasVoice::HaveSamplesEnded() const {
	switch (type) {
	case VOICETYPE_VAG:
		return vag.End();

	case VOICETYPE_PCM:
		return pcmIndex >= pcmSize;

	case VOICETYPE_ATRAC3:
		return atrac3.End();

	default:
		return false;
	}
}

void SasInstance::MixVoice(SasVoice &voice) {
	switch (voice.type) {
	case VOICETYPE_VAG:
		if (voice.type == VOICETYPE_VAG && !voice.vagAddr)
			break;
		// else fallthrough! Don't change the check above.
		[[fallthrough]];
	case VOICETYPE_PCM:
		if (voice.type == VOICETYPE_PCM && !voice.pcmAddr)
			break;
		// else fallthrough! Don't change the check above.
		[[fallthrough]];
	default:
		// This feels a bit hacky.  The first 32 samples after a keyon are 0s.
		int delay = 0;
		if (voice.envelope.NeedsKeyOn()) {
			const bool ignorePitch = voice.type == VOICETYPE_PCM && voice.pitch > PSP_SAS_PITCH_BASE;
			delay = ignorePitch ? 32 : (32 * (u32)voice.pitch) >> PSP_SAS_PITCH_BASE_SHIFT;
			// VAG seems to have an extra sample delay (not shared by PCM.)
			if (voice.type == VOICETYPE_VAG)
				++delay;
		}

		// Resample to the correct pitch, writing exactly "grainSize" samples. We need a buffer that can
		// fit 4x that, as the max pitch is 0x4000.
		// TODO: Special case no-resample case (and 2x and 0.5x) for speed, it's not uncommon

		// Two passes: First read, then resample.
		mixTemp_[0] = voice.resampleHist[0];
		mixTemp_[1] = voice.resampleHist[1];

		int voicePitch = voice.pitch;
		u32 sampleFrac = voice.sampleFrac;
		int samplesToRead = (sampleFrac + voicePitch * std::max(0, grainSize - delay)) >> PSP_SAS_PITCH_BASE_SHIFT;
		if (samplesToRead > ARRAY_SIZE(mixTemp_) - 2) {
			ERROR_LOG(Log::sceSas, "Too many samples to read (%d)! This shouldn't happen.", samplesToRead);
			samplesToRead = ARRAY_SIZE(mixTemp_) - 2;
		}
		int readPos = 2;
		if (voice.envelope.NeedsKeyOn()) {
			readPos = 0;
			samplesToRead += 2;
		}
		voice.ReadSamples(&mixTemp_[readPos], samplesToRead);
		int tempPos = readPos + samplesToRead;

		for (int i = 0; i < delay; ++i) {
			// Walk the curve.  This means we'll reach ATTACK already, likely.
			// This matches the results of tests (but maybe we can just remove the STATE_KEYON_STEP hack.)
			voice.envelope.Step();
		}

		const bool needsInterp = voicePitch != PSP_SAS_PITCH_BASE || (sampleFrac & PSP_SAS_PITCH_MASK) != 0;
		for (int i = delay; i < grainSize; i++) {
			const int16_t *s = mixTemp_ + (sampleFrac >> PSP_SAS_PITCH_BASE_SHIFT);

			// Two-tap linear interpolation. The hardware does the same, unlike the PSX there's no bicubic lookup table etc.
			int sample = s[0];
			if (needsInterp) {
				int f = sampleFrac & PSP_SAS_PITCH_MASK;
				sample = s[0] - (((s[0] - s[1]) * f) >> PSP_SAS_PITCH_BASE_SHIFT);
			}
			sampleFrac += voicePitch;

			// The maximum envelope height (PSP_SAS_ENVELOPE_HEIGHT_MAX) is (1 << 30) - 1.
			// Reduce it to 14 bits, by shifting off 15.  Round up by adding (1 << 14) first.
			int envelopeValue = voice.envelope.GetHeight();
			voice.envelope.Step();
			envelopeValue = (envelopeValue + (1 << 14)) >> 15;

			// We just scale by the envelope before we scale by volumes.
			// Again, we round up by adding (1 << 14) first (*after* multiplying.)
			sample = ((sample * envelopeValue) + (1 << 14)) >> 15;

			// We mix into this 32-bit temp buffer and clip in a second loop
			// Ideally, the shift right should be there too but for now I'm concerned about
			// not overflowing.
			mixBuffer[i * 2] += (sample * voice.volumeLeft) >> 12;
			mixBuffer[i * 2 + 1] += (sample * voice.volumeRight) >> 12;
			sendBuffer[i * 2] += sample * voice.effectLeft >> 12;
			sendBuffer[i * 2 + 1] += sample * voice.effectRight >> 12;
		}

		voice.resampleHist[0] = mixTemp_[tempPos - 2];
		voice.resampleHist[1] = mixTemp_[tempPos - 1];

		voice.sampleFrac = sampleFrac - (tempPos - 2) * PSP_SAS_PITCH_BASE;

		if (voice.HaveSamplesEnded())
			voice.envelope.End();
		if (voice.envelope.HasEnded()) {
			// NOTICE_LOG(Log::SasMix, "Hit end of envelope");
			voice.playing = false;
			voice.on = false;
		}
	}
}

void SasInstance::Mix(u32 outAddr, u32 inAddr, int leftVol, int rightVol, bool mute) {
	for (int v = 0; v < PSP_SAS_VOICES_MAX; v++) {
		SasVoice &voice = voices[v];
		if (!voice.playing || voice.paused)
			continue;
		MixVoice(voice);
	}

	// Apply mute if needed (note: we try to keep everything else identical to the non-muted case).
	if (mute) {
		memset(mixBuffer, 0, grainSize * sizeof(int) * 2);
		memset(sendBuffer, 0, grainSize * sizeof(int) * 2);
	}

	// Then mix the send buffer in with the rest.

	// Alright, all voices mixed. Let's convert and clip, and at the same time, wipe mixBuffer for next time. Could also dither.
	s16 *outp = (s16 *)Memory::GetPointerWriteRangeOrException(outAddr, 4 * grainSize);
	const s16 *inp = inAddr ? (const s16 *)Memory::GetPointerRangeOrException(inAddr, 4 * grainSize) : 0;
	if (!outp) {
		WARN_LOG_REPORT(Log::sceSas, "Bad SAS Mix output address: %08x, grain=%d", outAddr, grainSize);
	} else if (outputMode == PSP_SAS_OUTPUTMODE_MIXED) {
		// Okay, apply effects processing to the Send buffer.
		WriteMixedOutput(outp, inp, leftVol, rightVol);
		if (MemBlockInfoDetailed()) {
			if (inp)
				NotifyMemInfo(MemBlockFlags::READ, inAddr, grainSize * sizeof(u16) * 2, "SasMix");
			NotifyMemInfo(MemBlockFlags::WRITE, outAddr, grainSize * sizeof(u16) * 2, "SasMix");
		}
	} else {
		s16 *outpL = outp + grainSize * 0;
		s16 *outpR = outp + grainSize * 1;
		s16 *outpSendL = outp + grainSize * 2;
		s16 *outpSendR = outp + grainSize * 3;
		WARN_LOG_REPORT_ONCE(sasraw, Log::SasMix, "sceSasCore: raw outputMode");
		for (int i = 0; i < grainSize * 2; i += 2) {
			*outpL++ = clamp_s16(mixBuffer[i + 0]);
			*outpR++ = clamp_s16(mixBuffer[i + 1]);
			*outpSendL++ = clamp_s16(sendBuffer[i + 0]);
			*outpSendR++ = clamp_s16(sendBuffer[i + 1]);
		}
		NotifyMemInfo(MemBlockFlags::WRITE, outAddr, grainSize * sizeof(u16) * 4, "SasMix");
	}
	memset(mixBuffer, 0, grainSize * sizeof(int) * 2);
	memset(sendBuffer, 0, grainSize * sizeof(int) * 2);
}

// Note: leftVol/rightVol here are how much to scale the inp content by, not the mixBuffer.
void SasInstance::WriteMixedOutput(s16 *outp, const s16 *inp, int leftVol, int rightVol) {
	const bool dry = waveformEffect.isDryOn != 0;
	const bool wet = waveformEffect.isWetOn != 0;
	if (wet) {
		ApplyWaveformEffect();
	}

	if (inp) {
		for (int i = 0; i < grainSize * 2; i += 2) {
			int sampleL = ((*inp++) * leftVol >> 12);
			int sampleR = ((*inp++) * rightVol >> 12);
			if (dry) {
				sampleL += mixBuffer[i + 0];
				sampleR += mixBuffer[i + 1];
			}
			if (wet) {
				sampleL += sendBufferProcessed[i + 0];
				sampleR += sendBufferProcessed[i + 1];
			}
			*outp++ = clamp_s16(sampleL);
			*outp++ = clamp_s16(sampleR);
		}
	} else {
		// These are the optimal cases.
		if (dry && wet) {
			for (int i = 0; i < grainSize * 2; i += 2) {
				*outp++ = clamp_s16(mixBuffer[i + 0] + sendBufferProcessed[i + 0]);
				*outp++ = clamp_s16(mixBuffer[i + 1] + sendBufferProcessed[i + 1]);
			}
		} else if (dry) {
			for (int i = 0; i < grainSize * 2; i += 2) {
				*outp++ = clamp_s16(mixBuffer[i + 0]);
				*outp++ = clamp_s16(mixBuffer[i + 1]);
			}
		} else {
			// This is another uncommon case, dry must be off but let's keep it for clarity.
			for (int i = 0; i < grainSize * 2; i += 2) {
				int sampleL = 0;
				int sampleR = 0;
				if (dry) {
					sampleL += mixBuffer[i + 0];
					sampleR += mixBuffer[i + 1];
				}
				if (wet) {
					sampleL += sendBufferProcessed[i + 0];
					sampleR += sendBufferProcessed[i + 1];
				}
				*outp++ = clamp_s16(sampleL);
				*outp++ = clamp_s16(sampleR);
			}
		}
	}
}

void SasInstance::SetWaveformEffectType(int type) {
	if (type != waveformEffect.type) {
		waveformEffect.type = type;
		reverb_.SetPreset(type);
	}
}

void SasInstance::SetWaveformEffectParams(int delay, int feedback) {
	waveformEffect.delay = delay;
	waveformEffect.feedback = feedback;
	// Echo and Delay compute most of their parameters from these; the rest ignore them.
	reverb_.SetParams(delay, feedback);
}

// http://psx.rules.org/spu.txt has some information about setting up the delay time by modifying the delay preset.
// See http://report.ppsspp.org/logs/kind/772 for a list of games that use different types. Maybe can help us figure out
// which is which.
void SasInstance::ApplyWaveformEffect() {
	// First, downsample the send buffer to 22khz. We do this naively for now.
	for (int i = 0; i < grainSize / 2; i++) {
		sendBufferDownsampled[i * 2] = clamp_s16(sendBuffer[i * 4]);
		sendBufferDownsampled[i * 2 + 1] = clamp_s16(sendBuffer[i * 4 + 1]);
	}

	// Volume max is 0x1000, while our factor is up to 0x8000. Shifting left by 3 fixes that.
	reverb_.ProcessReverb(sendBufferProcessed, sendBufferDownsampled, grainSize / 2, (uint16_t)(waveformEffect.leftVol << 3), (uint16_t)(waveformEffect.rightVol << 3));
}

void SasInstance::DoState(PointerWrap &p) {
	auto s = p.Section("SasInstance", 1);
	if (!s)
		return;

	Do(p, grainSize);
	if (p.mode == p.MODE_READ) {
		if (grainSize > 0) {
			SetGrainSize(grainSize);
		} else {
			ClearGrainSize();
		}
	}

	Do(p, maxVoices);
	Do(p, sampleRate);
	Do(p, outputMode);

	// SetGrainSize() / ClearGrainSize() should've made our buffers match.
	if (mixBuffer != NULL && grainSize > 0) {
		DoArray(p, mixBuffer, grainSize * 2);
	}
	if (sendBuffer != NULL && grainSize > 0) {
		DoArray(p, sendBuffer, grainSize * 2);
	}
	if (sendBuffer != NULL && grainSize > 0) {
		// Backwards compat
		int16_t *resampleBuf = new int16_t[grainSize * 4 + 3]();
		DoArray(p, resampleBuf, grainSize * 4 + 3);
		delete[] resampleBuf;
	}

	int n = PSP_SAS_VOICES_MAX;
	Do(p, n);
	if (n != PSP_SAS_VOICES_MAX) {
		ERROR_LOG(Log::SaveState, "Wrong number of SAS voices");
		return;
	}
	DoArray(p, voices, ARRAY_SIZE(voices));
	Do(p, waveformEffect);
	if (p.mode == p.MODE_READ) {
		reverb_.SetPreset(waveformEffect.type);
		// SetPreset() alone would leave Echo/Delay at their defaults, since those two compute
		// their parameters from the delay/feedback we just restored.
		reverb_.SetParams(waveformEffect.delay, waveformEffect.feedback);
	}
}

void SasVoice::Reset() {
	resampleHist[0] = 0;
	resampleHist[1] = 0;
}

void SasVoice::KeyOn() {
	envelope.KeyOn();
	switch (type) {
	case VOICETYPE_VAG:
		if (Memory::IsValidAddress(vagAddr)) {
			vag.Start(vagAddr, vagSize, loop);
		} else {
			ERROR_LOG(Log::SasMix, "Invalid VAG address %08x", vagAddr);
			return;
		}
		break;
	default:
		break;
	}
	playing = true;
	on = true;
	paused = false;
	sampleFrac = 0;
}

void SasVoice::KeyOff() {
	on = false;
	envelope.KeyOff();
}

void SasVoice::DoState(PointerWrap &p) {
	auto s = p.Section("SasVoice", 1, 3);
	if (!s)
		return;

	Do(p, playing);
	Do(p, paused);
	Do(p, on);

	Do(p, type);

	Do(p, vagAddr);
	Do(p, vagSize);
	Do(p, pcmAddr);
	Do(p, pcmSize);
	Do(p, pcmIndex);
	if (s >= 2) {
		Do(p, pcmLoopPos);
	} else {
		pcmLoopPos = 0;
	}
	Do(p, sampleRate);

	Do(p, sampleFrac);
	Do(p, pitch);
	Do(p, loop);
	if (s < 2 && type == VOICETYPE_PCM) {
		// We set loop incorrectly before, and always looped.
		// Let's keep always looping, since it's usually right.
		loop = true;
	}

	Do(p, noiseFreq);

	Do(p, volumeLeft);
	Do(p, volumeRight);
	if (s < 3) {
		// There were extra variables here that were for the same purpose.
		Do(p, effectLeft);
		Do(p, effectRight);
	}
	Do(p, effectLeft);
	Do(p, effectRight);
	DoArray(p, resampleHist, ARRAY_SIZE(resampleHist));

	envelope.DoState(p);
	vag.DoState(p);
	atrac3.DoState(p);
}

void ADSREnvelope::WalkCurve(int type, int rate) {
	s64 expDelta;
	switch (type) {
	case PSP_SAS_ADSR_CURVE_MODE_LINEAR_INCREASE:
		height_ += rate;
		break;

	case PSP_SAS_ADSR_CURVE_MODE_LINEAR_DECREASE:
		height_ -= rate;
		break;

	case PSP_SAS_ADSR_CURVE_MODE_LINEAR_BENT:
		if (height_ <= (s64)PSP_SAS_ENVELOPE_HEIGHT_MAX * 3 / 4) {
			height_ += rate;
		} else {
			height_ += rate / 4;
		}
		break;

	case PSP_SAS_ADSR_CURVE_MODE_EXPONENT_DECREASE:
		expDelta = height_ - PSP_SAS_ENVELOPE_HEIGHT_MAX;
		// Flipping the sign so that we can shift in the top bits.
		expDelta += (-expDelta * rate) >> 32;
		height_ = expDelta + PSP_SAS_ENVELOPE_HEIGHT_MAX - (rate + 3UL) / 4UL;
		break;

	case PSP_SAS_ADSR_CURVE_MODE_EXPONENT_INCREASE:
		expDelta = height_ - PSP_SAS_ENVELOPE_HEIGHT_MAX;
		// Flipping the sign so that we can shift in the top bits.
		expDelta += (-expDelta * rate) >> 32;
		height_ = expDelta + 0x4000 + PSP_SAS_ENVELOPE_HEIGHT_MAX;
		break;

	case PSP_SAS_ADSR_CURVE_MODE_DIRECT:
		height_ = rate;  // Simple :)
		break;
	}
}

void ADSREnvelope::SetState(ADSRState state) {
	if (height_ > PSP_SAS_ENVELOPE_HEIGHT_MAX) {
		height_ = PSP_SAS_ENVELOPE_HEIGHT_MAX;
	}
	// TODO: Also check for height_ < 0 and set to 0?
	state_ = state;
}

inline void ADSREnvelope::Step() {
	switch (state_) {
	case STATE_ATTACK:
		WalkCurve(attackType, attackRate);
		if (height_ >= PSP_SAS_ENVELOPE_HEIGHT_MAX || height_ < 0)
			SetState(STATE_DECAY);
		break;
	case STATE_DECAY:
		WalkCurve(decayType, decayRate);
		if (height_ < sustainLevel)
			SetState(STATE_SUSTAIN);
		break;
	case STATE_SUSTAIN:
		WalkCurve(sustainType, sustainRate);
		if (height_ <= 0) {
			height_ = 0;
			SetState(STATE_RELEASE);
		}
		break;
	case STATE_RELEASE:
		WalkCurve(releaseType, releaseRate);
		if (height_ <= 0) {
			height_ = 0;
			SetState(STATE_OFF);
		}
		break;
	case STATE_OFF:
		// Do nothing
		break;

	case STATE_KEYON:
		height_ = 0;
		SetState(STATE_KEYON_STEP);
		break;
	case STATE_KEYON_STEP:
		// This entire state is pretty much a hack to reproduce PSP behavior.
		// The STATE_KEYON state is a real state, but not sure how it switches.
		// It takes 32 steps at 0 for keyon to "kick in", 31 should shift to 0 anyway.
		height_++;
		if (height_ >= 31) {
			height_ = 0;
			SetState(STATE_ATTACK);
		}
		break;
	}
}

void ADSREnvelope::KeyOn() {
	SetState(STATE_KEYON);
}

void ADSREnvelope::KeyOff() {
	SetState(STATE_RELEASE);
}

void ADSREnvelope::End() {
	SetState(STATE_OFF);
	height_ = 0;
}

void ADSREnvelope::DoState(PointerWrap &p) {
	auto s = p.Section("ADSREnvelope", 1, 2);
	if (!s) {
		return;
	}

	Do(p, attackRate);
	Do(p, decayRate);
	Do(p, sustainRate);
	Do(p, releaseRate);
	Do(p, attackType);
	Do(p, decayType);
	Do(p, sustainType);
	Do(p, sustainLevel);
	Do(p, releaseType);
	if (s < 2) {
		Do(p, state_);
		if (state_ == 4) {
			state_ = STATE_OFF;
		}
		int stepsLegacy;
		Do(p, stepsLegacy);
	} else {
		Do(p, state_);
	}
	Do(p, height_);
}

const char *ADSRCurveModeAsString(SasADSRCurveMode mode) {
	switch (mode) {
	case PSP_SAS_ADSR_CURVE_MODE_DIRECT: return "D";
	case PSP_SAS_ADSR_CURVE_MODE_LINEAR_INCREASE: return "L+";
	case PSP_SAS_ADSR_CURVE_MODE_LINEAR_DECREASE: return "L-";
	case PSP_SAS_ADSR_CURVE_MODE_LINEAR_BENT: return "LB";
	case PSP_SAS_ADSR_CURVE_MODE_EXPONENT_DECREASE: return "E-";
	case PSP_SAS_ADSR_CURVE_MODE_EXPONENT_INCREASE: return "E+";
	default: return "N/A";
	}
}