/* Copyright (C) 2003, 2004, 2005, 2006, 2008, 2009 Dean Beeler, Jerome Fisher
 * Copyright (C) 2011-2022 Dean Beeler, Jerome Fisher, Sergey V. Mikayev
 *
 *  This program is free software: you can redistribute it and/or modify
 *  it under the terms of the GNU Lesser General Public License as published by
 *  the Free Software Foundation, either version 2.1 of the License, or
 *  (at your option) any later version.
 *
 *  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 Lesser General Public License for more details.
 *
 *  You should have received a copy of the GNU Lesser General Public License
 *  along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */

#include <cstdlib>

#include "internals.h"

#include "TVP.h"
#include "Part.h"
#include "Partial.h"
#include "Poly.h"
#include "Synth.h"
#include "TVA.h"

namespace MT32Emu {

// FIXME: Add Explanation
static Bit16u lowerDurationToDivisor[] = {34078, 37162, 40526, 44194, 48194, 52556, 57312, 62499};

// These values represent unique options with no consistent pattern, so we have to use something like a table in any case.
// The table matches exactly what the manual claims (when divided by 8192):
// -1, -1/2, -1/4, 0, 1/8, 1/4, 3/8, 1/2, 5/8, 3/4, 7/8, 1, 5/4, 3/2, 2, s1, s2
// ...except for the last two entries, which are supposed to be "1 cent above 1" and "2 cents above 1", respectively. They can only be roughly approximated with this integer math.
static Bit16s pitchKeyfollowMult[] = {-8192, -4096, -2048, 0, 1024, 2048, 3072, 4096, 5120, 6144, 7168, 8192, 10240, 12288, 16384, 8198, 8226};

// Note: Keys < 60 use keyToPitchTable[60 - key], keys >= 60 use keyToPitchTable[key - 60].
// FIXME: This table could really be shorter, since we never use e.g. key 127.
static Bit16u keyToPitchTable[] = {
	    0,   341,   683,  1024,  1365,  1707,  2048,  2389,
	 2731,  3072,  3413,  3755,  4096,  4437,  4779,  5120,
	 5461,  5803,  6144,  6485,  6827,  7168,  7509,  7851,
	 8192,  8533,  8875,  9216,  9557,  9899, 10240, 10581,
	10923, 11264, 11605, 11947, 12288, 12629, 12971, 13312,
	13653, 13995, 14336, 14677, 15019, 15360, 15701, 16043,
	16384, 16725, 17067, 17408, 17749, 18091, 18432, 18773,
	19115, 19456, 19797, 20139, 20480, 20821, 21163, 21504,
	21845, 22187, 22528, 22869
};

// We want to do processing 4000 times per second. FIXME: This is pretty arbitrary.
static const int NOMINAL_PROCESS_TIMER_PERIOD_SAMPLES = SAMPLE_RATE / 4000;

// In all hardware units we emulate, the main clock frequency of the MCU is 12MHz.
// However, the MCU used in the 3rd-gen sound modules (like CM-500 and LAPC-N)
// is significantly faster. Importantly, the software timer also works faster,
// yet this fact has been seemingly missed. To be more specific, the software timer
// ticks each 8 "state times", and 1 state time equals to 3 clock periods
// for 8095 and 8098 but 2 clock periods for 80C198. That is, on MT-32 and CM-32L,
// the software timer tick rate is 12,000,000 / 3 / 8 = 500kHz, but on the 3rd-gen
// devices it's 12,000,000 / 2 / 8 = 750kHz instead.

// For 1st- and 2nd-gen devices, the timer ticks at 500kHz. This is how much to increment
// timeElapsed once 16 samples passes. We multiply by 16 to get rid of the fraction
// and deal with just integers.
static const int PROCESS_TIMER_TICKS_PER_SAMPLE_X16_1N2_GEN = (500000 << 4) / SAMPLE_RATE;
// For 3rd-gen devices, the timer ticks at 750kHz. This is how much to increment
// timeElapsed once 16 samples passes. We multiply by 16 to get rid of the fraction
// and deal with just integers.
static const int PROCESS_TIMER_TICKS_PER_SAMPLE_X16_3_GEN = (750000 << 4) / SAMPLE_RATE;

TVP::TVP(const Partial *usePartial) :
	partial(usePartial),
	system(&usePartial->getSynth()->mt32ram.system),
	processTimerTicksPerSampleX16(
		partial->getSynth()->controlROMFeatures->quirkFastPitchChanges
		? PROCESS_TIMER_TICKS_PER_SAMPLE_X16_3_GEN
		: PROCESS_TIMER_TICKS_PER_SAMPLE_X16_1N2_GEN)
{}

static Bit16s keyToPitch(unsigned int key) {
	// We're using a table to do: return round_to_nearest_or_even((key - 60) * (4096.0 / 12.0))
	// Banker's rounding is just slightly annoying to do in C++
	int k = int(key);
	Bit16s pitch = keyToPitchTable[abs(k - 60)];
	return key < 60 ? -pitch : pitch;
}

static inline Bit32s coarseToPitch(Bit8u coarse) {
	return (coarse - 36) * 4096 / 12; // One semitone per coarse offset
}

static inline Bit32s fineToPitch(Bit8u fine) {
	return (fine - 50) * 4096 / 1200; // One cent per fine offset
}

static Bit32u calcBasePitch(const Partial *partial, const TimbreParam::PartialParam *partialParam, const MemParams::PatchTemp *patchTemp, unsigned int key, const ControlROMFeatureSet *controlROMFeatures) {
	Bit32s basePitch = keyToPitch(key);
	basePitch = (basePitch * pitchKeyfollowMult[partialParam->wg.pitchKeyfollow]) >> 13; // PORTABILITY NOTE: Assumes arithmetic shift
	basePitch += coarseToPitch(partialParam->wg.pitchCoarse);
	basePitch += fineToPitch(partialParam->wg.pitchFine);
	if (controlROMFeatures->quirkKeyShift) {
		// NOTE:Mok: This is done on MT-32, but not LAPC-I:
		basePitch += coarseToPitch(patchTemp->patch.keyShift + 12);
	}
	basePitch += fineToPitch(patchTemp->patch.fineTune);

	const ControlROMPCMStruct *controlROMPCMStruct = partial->getControlROMPCMStruct();
	if (controlROMPCMStruct != NULL) {
		basePitch += (Bit32s(controlROMPCMStruct->pitchMSB) << 8) | Bit32s(controlROMPCMStruct->pitchLSB);
	} else {
		if ((partialParam->wg.waveform & 1) == 0) {
			basePitch += 37133; // This puts Middle C at around 261.64Hz (assuming no other modifications, masterTune of 64, etc.)
		} else {
			// Sawtooth waves are effectively double the frequency of square waves.
			// Thus we add 4096 less than for square waves here, which results in halving the frequency.
			basePitch += 33037;
		}
	}

	// MT-32 GEN0 does 16-bit calculations here, allowing an integer overflow.
	// This quirk is observable playing the patch defined for timbre "HIT BOTTOM" in Larry 3.
	// Note, the upper bound isn't checked either.
	if (controlROMFeatures->quirkBasePitchOverflow) {
		basePitch = basePitch & 0xffff;
	} else if (basePitch < 0) {
		basePitch = 0;
	} else if (basePitch > 59392) {
		basePitch = 59392;
	}
	return Bit32u(basePitch);
}

static Bit32u calcVeloMult(Bit8u veloSensitivity, unsigned int velocity) {
	if (veloSensitivity == 0) {
		return 21845; // aka floor(4096 / 12 * 64), aka ~64 semitones
	}
	unsigned int reversedVelocity = 127 - velocity;
	unsigned int scaledReversedVelocity;
	if (veloSensitivity > 3) {
		// Note that on CM-32L/LAPC-I veloSensitivity is never > 3, since it's clipped to 3 by the max tables.
		// MT-32 GEN0 has a bug here that leads to unspecified behaviour. We assume it is as follows.
		scaledReversedVelocity = (reversedVelocity << 8) >> ((3 - veloSensitivity) & 0x1f);
	} else {
		scaledReversedVelocity = reversedVelocity << (5 + veloSensitivity);
	}
	// When velocity is 127, the multiplier is 21845, aka ~64 semitones (regardless of veloSensitivity).
	// The lower the velocity, the lower the multiplier. The veloSensitivity determines the amount decreased per velocity value.
	// The minimum multiplier on CM-32L/LAPC-I (with velocity 0, veloSensitivity 3) is 170 (~half a semitone).
	return ((32768 - scaledReversedVelocity) * 21845) >> 15;
}

static Bit32s calcTargetPitchOffsetWithoutLFO(const TimbreParam::PartialParam *partialParam, int levelIndex, unsigned int velocity) {
	int veloMult = calcVeloMult(partialParam->pitchEnv.veloSensitivity, velocity);
	int targetPitchOffsetWithoutLFO = partialParam->pitchEnv.level[levelIndex] - 50;
	targetPitchOffsetWithoutLFO = (targetPitchOffsetWithoutLFO * veloMult) >> (16 - partialParam->pitchEnv.depth); // PORTABILITY NOTE: Assumes arithmetic shift
	return targetPitchOffsetWithoutLFO;
}

void TVP::reset(const Part *usePart, const TimbreParam::PartialParam *usePartialParam) {
	part = usePart;
	partialParam = usePartialParam;
	patchTemp = part->getPatchTemp();

	unsigned int key = partial->getPoly()->getKey();
	unsigned int velocity = partial->getPoly()->getVelocity();

	// FIXME: We're using a per-TVP timer instead of a system-wide one for convenience.
	timeElapsed = 0;
	processTimerIncrement = 0;

	basePitch = calcBasePitch(partial, partialParam, patchTemp, key, partial->getSynth()->controlROMFeatures);
	currentPitchOffset = calcTargetPitchOffsetWithoutLFO(partialParam, 0, velocity);
	targetPitchOffsetWithoutLFO = currentPitchOffset;
	phase = 0;

	if (partialParam->pitchEnv.timeKeyfollow) {
		timeKeyfollowSubtraction = Bit32s(key - 60) >> (5 - partialParam->pitchEnv.timeKeyfollow); // PORTABILITY NOTE: Assumes arithmetic shift
	} else {
		timeKeyfollowSubtraction = 0;
	}
	lfoPitchOffset = 0;
	counter = 0;
	pitch = basePitch;

	// These don't really need to be initialised, but it aids debugging.
	pitchOffsetChangePerBigTick = 0;
	targetPitchOffsetReachedBigTick = 0;
	shifts = 0;
}

Bit32u TVP::getBasePitch() const {
	return basePitch;
}

void TVP::updatePitch() {
	Bit32s newPitch = basePitch + currentPitchOffset;
	if (!partial->isPCM() || (partial->getControlROMPCMStruct()->len & 0x01) == 0) { // FIXME: Use !partial->pcmWaveEntry->unaffectedByMasterTune instead
		// FIXME: There are various bugs not yet emulated
		// 171 is ~half a semitone.
		newPitch += partial->getSynth()->getMasterTunePitchDelta();
	}
	if ((partialParam->wg.pitchBenderEnabled & 1) != 0) {
		newPitch += part->getPitchBend();
	}

	// MT-32 GEN0 does 16-bit calculations here, allowing an integer overflow.
	// This quirk is exploited e.g. in Colonel's Bequest timbres "Lightning" and "SwmpBackgr".
	if (partial->getSynth()->controlROMFeatures->quirkPitchEnvelopeOverflow) {
		newPitch = newPitch & 0xffff;
	} else if (newPitch < 0) {
		newPitch = 0;
	}
	// This check is present in every unit.
	if (newPitch > 59392) {
		newPitch = 59392;
	}
	pitch = Bit16u(newPitch);

	// FIXME: We're doing this here because that's what the CM-32L does - we should probably move this somewhere more appropriate in future.
	partial->getTVA()->recalcSustain();
}

void TVP::targetPitchOffsetReached() {
	currentPitchOffset = targetPitchOffsetWithoutLFO + lfoPitchOffset;

	switch (phase) {
	case 3:
	case 4:
	{
		int newLFOPitchOffset = (part->getModulation() * partialParam->pitchLFO.modSensitivity) >> 7;
		newLFOPitchOffset = (newLFOPitchOffset + partialParam->pitchLFO.depth) << 1;
		if (pitchOffsetChangePerBigTick > 0) {
			// Go in the opposite direction to last time
			newLFOPitchOffset = -newLFOPitchOffset;
		}
		lfoPitchOffset = newLFOPitchOffset;
		int targetPitchOffset = targetPitchOffsetWithoutLFO + lfoPitchOffset;
		setupPitchChange(targetPitchOffset, 101 - partialParam->pitchLFO.rate);
		updatePitch();
		break;
	}
	case 6:
		updatePitch();
		break;
	default:
		nextPhase();
	}
}

void TVP::nextPhase() {
	phase++;
	int envIndex = phase == 6 ? 4 : phase;

	targetPitchOffsetWithoutLFO = calcTargetPitchOffsetWithoutLFO(partialParam, envIndex, partial->getPoly()->getVelocity()); // pitch we'll reach at the end

	int changeDuration = partialParam->pitchEnv.time[envIndex - 1];
	changeDuration -= timeKeyfollowSubtraction;
	if (changeDuration > 0) {
		setupPitchChange(targetPitchOffsetWithoutLFO, changeDuration); // changeDuration between 0 and 112 now
		updatePitch();
	} else {
		targetPitchOffsetReached();
	}
}

// Shifts val to the left until bit 31 is 1 and returns the number of shifts
static Bit8u normalise(Bit32u &val) {
	Bit8u leftShifts;
	for (leftShifts = 0; leftShifts < 31; leftShifts++) {
		if ((val & 0x80000000) != 0) {
			break;
		}
		val = val << 1;
	}
	return leftShifts;
}

void TVP::setupPitchChange(int targetPitchOffset, Bit8u changeDuration) {
	bool negativeDelta = targetPitchOffset < currentPitchOffset;
	Bit32s pitchOffsetDelta = targetPitchOffset - currentPitchOffset;
	if (pitchOffsetDelta > 32767 || pitchOffsetDelta < -32768) {
		pitchOffsetDelta = 32767;
	}
	if (negativeDelta) {
		pitchOffsetDelta = -pitchOffsetDelta;
	}
	// We want to maximise the number of bits of the Bit16s "pitchOffsetChangePerBigTick" we use in order to get the best possible precision later
	Bit32u absPitchOffsetDelta = (pitchOffsetDelta & 0xFFFF) << 16;
	Bit8u normalisationShifts = normalise(absPitchOffsetDelta); // FIXME: Double-check: normalisationShifts is usually between 0 and 15 here, unless the delta is 0, in which case it's 31
	absPitchOffsetDelta = absPitchOffsetDelta >> 1; // Make room for the sign bit

	changeDuration--; // changeDuration's now between 0 and 111
	unsigned int upperDuration = changeDuration >> 3; // upperDuration's now between 0 and 13
	shifts = normalisationShifts + upperDuration + 2;
	Bit16u divisor = lowerDurationToDivisor[changeDuration & 7];
	Bit16s newPitchOffsetChangePerBigTick = ((absPitchOffsetDelta & 0xFFFF0000) / divisor) >> 1; // Result now fits within 15 bits. FIXME: Check nothing's getting sign-extended incorrectly
	if (negativeDelta) {
		newPitchOffsetChangePerBigTick = -newPitchOffsetChangePerBigTick;
	}
	pitchOffsetChangePerBigTick = newPitchOffsetChangePerBigTick;

	int currentBigTick = timeElapsed >> 8;
	int durationInBigTicks = divisor >> (12 - upperDuration);
	if (durationInBigTicks > 32767) {
		durationInBigTicks = 32767;
	}
	// The result of the addition may exceed 16 bits, but wrapping is fine and intended here.
	targetPitchOffsetReachedBigTick = currentBigTick + durationInBigTicks;
}

void TVP::startDecay() {
	phase = 5;
	lfoPitchOffset = 0;
	targetPitchOffsetReachedBigTick = timeElapsed >> 8; // FIXME: Afaict there's no good reason for this - check
}

Bit16u TVP::nextPitch() {
	// We emulate MCU software timer using these counter and processTimerIncrement variables.
	// The value of NOMINAL_PROCESS_TIMER_PERIOD_SAMPLES approximates the period in samples
	// between subsequent firings of the timer that normally occur.
	// However, accurate emulation is quite complicated because the timer is not guaranteed to fire in time.
	// This makes pitch variations on real unit non-deterministic and dependent on various factors.
	if (counter == 0) {
		timeElapsed = (timeElapsed + processTimerIncrement) & 0x00FFFFFF;
		// This roughly emulates pitch deviations observed on real units when playing a single partial that uses TVP/LFO.
		counter = NOMINAL_PROCESS_TIMER_PERIOD_SAMPLES + (rand() & 3);
		processTimerIncrement = (processTimerTicksPerSampleX16 * counter) >> 4;
		process();
	}
	counter--;
	return pitch;
}

void TVP::process() {
	if (phase == 0) {
		targetPitchOffsetReached();
		return;
	}
	if (phase == 5) {
		nextPhase();
		return;
	}
	if (phase > 7) {
		updatePitch();
		return;
	}

	Bit16s negativeBigTicksRemaining = (timeElapsed >> 8) - targetPitchOffsetReachedBigTick;
	if (negativeBigTicksRemaining >= 0) {
		// We've reached the time for a phase change
		targetPitchOffsetReached();
		return;
	}
	// FIXME: Write explanation for this stuff
	// NOTE: Value of shifts may happily exceed the maximum of 31 specified for the 8095 MCU.
	// We assume the device performs a shift with the rightmost 5 bits of the counter regardless of argument size,
	// since shift instructions of any size have the same maximum.
	int rightShifts = shifts;
	if (rightShifts > 13) {
		rightShifts -= 13;
		negativeBigTicksRemaining = negativeBigTicksRemaining >> (rightShifts & 0x1F); // PORTABILITY NOTE: Assumes arithmetic shift
		rightShifts = 13;
	}
	int newResult = (negativeBigTicksRemaining * pitchOffsetChangePerBigTick) >> (rightShifts & 0x1F); // PORTABILITY NOTE: Assumes arithmetic shift
	newResult += targetPitchOffsetWithoutLFO + lfoPitchOffset;
	currentPitchOffset = newResult;
	updatePitch();
}

} // namespace MT32Emu