TrillRack.cpp

#include <TrillRackApplication_bsp.h>
#include "TrillRackInterface.h"
#include "GlissModes.h"
#include "GlissProtocol.h"
#include "preset.h"
#include <libraries/Trill/Trill.h> // include this above NeoPixel or "HEX" gets screwed up
#include <libraries/Trill/CentroidDetection.h> // include this above NeoPixel or "HEX" gets screwed up
#include "LedSliders.h"
#include <cmath>
#include <assert.h>
#include <atomic>
#include "../../common_stuff/verificationBlock.h"
#include "../../common_stuff/sysex.h"
#include "../../common_stuff/i2cExternal.h"

constexpr std::array<float,CalibrationData::kNumPoints> CalibrationData::points;

extern std::array<rgb_t, 2> gBalancedLfoColors;
extern bool performanceMode_setup(double);
extern void performanceMode_render(BelaContext*, FrameData*);
extern void menu_render(BelaContext*, FrameData*);
extern float getGnd();
extern bool modeAlt_setup();
extern void triggerInToClock(BelaContext* context);
extern int gMtrClkTrigger;
extern LedSliders ledSliders;
extern LedSliders ledSlidersAlt;
extern ButtonView menuBtn;
extern ButtonView performanceBtn;
extern void ledSlidersFixedButtonsProcess(LedSliders& sl, std::vector<bool>& states, std::vector<size_t>& onsets, std::vector<size_t>& offsets, bool onlyUpdateStates);
std::array<float,kNumOutChannels> gManualAnOut;

#define STM32_NEOPIXEL
// #define PRINT_CPU_TIME
// Gliss revs:
// 1: no logo, exposed copper, non-inverting I/O, only used internally
// 2: logo, exposed copper, inverting I/O, first beta testing
// 3: logo, transparent solder mask, inverting I/O, rc1
#define GLISS_HW_REV 3
//#define TRILL_BAR // whether to use an external Trill Bar

#if GLISS_HW_REV >= 2
TrillRackInterface tri(0, 0, 1, __builtin_ctz(SW0_Pin), __builtin_ctz(SW_LED_A_Pin), __builtin_ctz(SW_LED_B_Pin));
#else
TrillRackInterface tri(0, 0, 1, __builtin_ctz(SW0_Pin), __builtin_ctz(SW_LED_Pin), 6 /* dummy */);
#endif

NeoPixelT<kNumLeds> np;
Trill trill;

std::vector<unsigned int> padsToOrderMap = {
#if GLISS_HW_REV >= 2
	29,
	28,
	27,
	26,
	25,
	24,
	23,
	22,
	21,
	17,
	18,
	19,
	20,
	12,
	10,
	9,
	3,
	4,
	5,
	6,
	7,
	8,
	11,
	13,
	14,
	15,
#else
#ifdef OLD
	0,
	1,
	2,
	3,
	4,
	5,
	6,
	7,
	8,
	9,
	10,
	11,
	12,
	13,
	28,
	27,
	26,
	25,
	15,
	17,
	18,
	19,
	20,
	24,
#else // OLD
	29,
	28,
	27,
	26,
	25,
	24,
	23,
	22,
	21,
	17,
	18,
	19,
	14,
	13,
	11,
	3,
	4,
	5,
	6,
	8,
	7,
	9,
	10,
	12,
	15,
	16,
#endif // OLD
#endif
};
CentroidDetectionScaled globalSlider;
int gAlt = 0;

#ifdef STM32_NEOPIXEL
static Stm32NeoPixelT<uint32_t, kNumLeds> snp(&neoPixelHtim, neoPixelHtim_TIM_CHANNEL_x, 0.66 * neoPixelHtim_COUNTER_PERIOD, 0.33 * neoPixelHtim_COUNTER_PERIOD);
#endif // STM32_NEOPIXEL

#ifdef STM32_NEOPIXEL
extern "C" {
void tr_snpDone(void);
};

void tr_snpDone()
{
	snp.done();
}
#endif // STM32_NEOPIXEL

#ifdef PRINT_CPU_TIME
// https://community.st.com/t5/stm32-mcus-products/how-do-you-measure-the-execution-cpu-cycles-for-a-section-of/td-p/213709
//****************************************************************************

volatile unsigned int *DWT_CYCCNT   = (volatile unsigned int *)0xE0001004;
volatile unsigned int *DWT_CONTROL  = (volatile unsigned int *)0xE0001000;
volatile unsigned int *DWT_LAR      = (volatile unsigned int *)0xE0001FB0;
volatile unsigned int *SCB_DHCSR    = (volatile unsigned int *)0xE000EDF0;
volatile unsigned int *SCB_DEMCR    = (volatile unsigned int *)0xE000EDFC;
volatile unsigned int *ITM_TER      = (volatile unsigned int *)0xE0000E00;
volatile unsigned int *ITM_TCR      = (volatile unsigned int *)0xE0000E80;
static int Debug_ITMDebug = 0;

void EnableTiming(void)
{
  if ((*SCB_DHCSR & 1) && (*ITM_TER & 1)) // Enabled?
    Debug_ITMDebug = 1;

  *SCB_DEMCR |= 0x01000000;
  *DWT_LAR = 0xC5ACCE55; // enable access
  *DWT_CYCCNT = 0; // reset the counter
  *DWT_CONTROL |= 1 ; // enable the counter
}
#endif // PRINT_CPU_TIME

int tr_setup()
{
	printf("stringId: %s\n\r", kVerificationBlock.stringId);
#ifdef PRINT_CPU_TIME
	EnableTiming();
#endif // PRINT_CPU_TIME
#ifdef TRILL_BAR
	padsToOrderMap.resize(kNumPads);
	for(size_t n = 0; n < padsToOrderMap.size(); ++n)
		padsToOrderMap[n] = n;
#endif // TRILL_BAR
	// find unused pads and mark them as such,
	// then shift channel numbers accordingly
	uint32_t channelMask = 0;
	for(auto p : padsToOrderMap)
		channelMask |= 1 << p;
	assert(kNumPads == padsToOrderMap.size());
	assert(kNumPads == __builtin_popcount(channelMask));
	for(auto& p : padsToOrderMap)
	{
		assert(channelMask & (1 << p));
		unsigned int end = p;
		for(unsigned int n = 0; n < end; ++n)
			if(!(channelMask & (1 << n))) // decrement once for every skipped channel
				p--;
	}
	globalSlider.setup(padsToOrderMap, 5, 1);
#ifdef STM32_NEOPIXEL
	np.setSnp(&snp);
#endif // STM32_NEOPIXEL
	np.show();
#ifdef TRILL_BAR
	Trill::Device device = Trill::BAR;
	uint8_t startAddr = 0x20;
	const uint8_t kExpectedAddr = 0x20;
#else
	Trill::Device device = Trill::FLEX;
	uint8_t startAddr = 0x48;
	const uint8_t kExpectedAddr = 0x4c;
#endif
	// to speed up things, try the expected address and if it fails try the full range
	uint8_t foundAddress = 0;
	if(!trill.setup(1, device, kExpectedAddr))
	{
		foundAddress = kExpectedAddr;
	} else {
		for(uint8_t addr = startAddr; addr <= startAddr + 8; ++addr)
		{
			if(!trill.setup(1, device, addr))
			{
				foundAddress = addr;
				break;
			}
		}
	}
	if(!foundAddress)
		return false;
	if(trill.firmwareVersion() < 3)
	{
		printf("Incompatible Trill firwmware version: %d, should be >= %d\n\r", trill.firmwareVersion(), 3);
		return false;
	}
	trill.printDetails();
	if(trill.setMode(Trill::DIFF))
		return false;
	int prescaler;
#ifdef TRILL_BAR
	prescaler = 2;
#elif GLISS_HW_REV >= 3
	prescaler = 3;
#else
	prescaler = 5;
#endif

	if(trill.setChannelMask(channelMask))
		return false;;
	if(trill.setScanSettings(0, 12))
		return false;
	if(trill.setPrescaler(prescaler))
		return false;
	if(trill.setNoiseThreshold(0.06))
		return false;
	if(trill.updateBaseline())
		return false;
	if(trill.setScanTrigger(Trill::kScanTriggerI2c))
		return false;
	if(trill.setEventMode(Trill::kEventModeAlways)) // ... and set PSOC_EVENT pin when ready
		return false;
	if(trill.readStatusByte() < 0) // ensure that future reads via DMA have the correct offset
		return false;
	modeAlt_setup();
	PresetInitOptions_t presetType;
#ifdef TEST_MODE
	presetType = kPresetInit_LoadDefault;
#else // TEST_MODE
	presetType = kPresetInit_LoadLatest;
#endif // TEST_MODE
	int ret = presetInit(presetType, 2000, HAL_GetTick);
	printf("presetInit() loaded %d\n\r", ret);
	return foundAddress;
}

extern "C" void processMidiMessage(void);
#ifdef PRINT_CPU_TIME
static volatile unsigned int minTime = -1;
static volatile unsigned int maxTime = 0;
static volatile uint32_t cpuReportsCount = 0;
static size_t samplesPerBlock;
static float sampleRate;
float cpuPercentage(unsigned int val)
{
	return 100 * val / float(170000000) / (samplesPerBlock / sampleRate);
}
#endif // PRINT_CPU_TIME

void tr_mainLoop()
{
#ifndef TEST_MODE
#ifdef PRINT_CPU_TIME
	static uint32_t lastPrinted = 0;
	if(lastPrinted != cpuReportsCount)
	{
		lastPrinted = cpuReportsCount;
		printf("%7.4f%% %7.4f%%\n\r", cpuPercentage(minTime), cpuPercentage(maxTime));
		minTime = 0;
		maxTime = 0;
	}
#endif
	if(!gAlt)
	{
		int ret = presetCheckSave();
		if(ret >= 0)
			printf("presetCheckSave: %d\n\r", ret);
	}
	processMidiMessage();
	i2cProcessIncomingFromMainThread();
	gp_outgoing(kSysexUsb, sysexSend);
	gp_outgoing(kSysexI2c, sysexSend);
#endif // TEST_MODE
}

void tr_clearLeds()
{
#ifdef STM32_NEOPIXEL
	np.setSnp(&snp);
#endif // STM32_NEOPIXEL
	np.clear();
	np.show();
}

static int gShouldScan = 1;

void tr_requestScan(int val)
{
	gShouldScan = val;
}

int tr_scanRequested()
{
	return gShouldScan;
}

static int gShouldUpdateLeds = 1;

void tr_requestUpdateLeds(int val)
{
	gShouldUpdateLeds = val;
}

int tr_ledsUpdateRequested()
{
	return gShouldUpdateLeds;
}

constexpr uint32_t kInvalidFrameId = -1;
static std::atomic_uint32_t trillFrameId { kInvalidFrameId };

void tr_newData(const uint8_t* newData, size_t len)
{
	static uint32_t pastStatusByte = newData[0];
	// only process new data if they belong to a new frame
	if(pastStatusByte != newData[0])
	{
		pastStatusByte = newData[0];
		// mark data as invalid
		trillFrameId.store(kInvalidFrameId);
		// operate on it
		trill.newData(newData, len, true);
		// mark data as valid again
		uint32_t frameId = trill.getFrameIdUnwrapped();
		trillFrameId.store(frameId);
		if((gDebugFlags & 0x1) || ((gDebugFlags & 0x2) && (0 == (frameId % 5))))
		{
			const uint8_t* p = newData + 1; // skip status byte
			for(size_t n = 1; n < len; n += 2)
			{
				uint16_t val = (*p++) << 8;
				val |= *p++;
				printf("%d ", val);
			}
			printf("\n\r");
		}
	}
}

void tr_process(BelaContext* ptr)
{
	tri.process(ptr);
}

IoRange gInRange = {
	.min = 0,
	.max = 1,
	.range = kCvRangePositive10,
	.enabled = true,
};
IoRange gOutRangeTop = {
	.min = 0,
	.max = 1,
	.range = kCvRangePositive10,
	.enabled = true,
};
IoRange gOutRangeBottom = {
	.min = 0,
	.max = 1,
	.range = kCvRangePositive10,
	.enabled = true,
};

static inline void getRangeMinMax(bool input, size_t channel, float& min, float& max)
{
	const IoRange& ioRange = input ? gInRange : (0 == channel) ? gOutRangeTop : gOutRangeBottom;
	ioRange.getMinMax(min, max);
}

static float processRawThroughCalibration(const CalibrationData& cal, bool input, float raw)
{
	float value = 1;
	for(size_t n = 1; n < cal.points.size(); ++n)
	{
		// linear interpolation between the two nearest calibration points
		if(raw <= cal.points[n])
		{
			if(input)
				value = map(raw, cal.values[n - 1], cal.values[n], cal.points[n - 1], cal.points[n]);
			else
				value = map(raw, cal.points[n - 1], cal.points[n], cal.values[n - 1], cal.values[n]);
			break;
		}
	}
	return value;
}

static float reverseProcessRawThroughCalibration(const CalibrationData& cal, bool input, float cooked)
{
	return processRawThroughCalibration(cal, !input, cooked);
}

static float rescaleInput(const CalibrationData& inCal, float value)
{
	float min;
	float max;
	value = processRawThroughCalibration(inCal, true, value);
	getRangeMinMax(true, 0, min, max);
	return mapAndConstrain(value, min, max, 0, 1);
}

static float finalise(float value)
{
#if GLISS_HW_REV >= 2
	// inverting output
	return 1.f - value;
#else
	return value;
#endif
}

template <typename T>
class SmootherT
{
	T past = 0;
public:
	SmootherT() {};
	SmootherT(T val) : past(val) {}
	T process(T in, T alpha)
	{
		T out = past * alpha + in * (T(1) - alpha);
		past = out;
		return out;
	}
	void set(T val)
	{
		past = val;
	}
	T get()
	{
		return past;
	}
};

class Smoother
{
	SmootherT<double> db;
	SmootherT<float> fl;
	bool doub = false;
public:
	Smoother() {};
	void setDoublePrecision(bool doub)
	{
		this->doub = doub;
	}
	float process(float in, float alpha)
	{
		if(doub)
			return db.process(in, alpha);
		else
			return fl.process(in, alpha);
	}
	void set(float val)
	{
		db.set(val);
		fl.set(val);
	}
	float get()
	{
		if(doub)
			return db.get();
		else
			return fl.get();
	}
};

static float rescaleOutput(bool ignoreRange, size_t channel, const CalibrationData& cal, float value)
{
	float gnd = cal.values[1];
	if(kNoOutput == value)
		return gnd;
	float min = 0;
	float top = 1;
	if(!ignoreRange)
		getRangeMinMax(false, channel, min, top);

	// the input can be negative: out of the specified range but possibly still within the capabilities of the module
	value = map(value, 0, 1, min, top);
	value = processRawThroughCalibration(cal, false, value);
	// we only constrain at the end: can't write outside full scale
	value = constrain(value, 0, 1);
	return value;
}

static float reverseRescaleOutput(bool ignoreRange, size_t channel, const CalibrationData& cal, float value)
{
	float min = 0;
	float top = 1;
	if(!ignoreRange)
		getRangeMinMax(false, channel, min, top);
	// in the reverse order as rescaleOutput()
	value = reverseProcessRawThroughCalibration(cal, false, value);
	value = mapAndConstrain(value, min, top, 0, 1); // swapped args 1,2 and 3,4
	return value;
}

static void analogWriteJacks(BelaContext* context, unsigned int frame, unsigned int channel, float value)
{
	// swap output channels if needed
	unsigned int c = uio.outputsSwapped() ? !channel : channel;
	analogWriteOnce(context, frame, c, value);
}

static void renderMenuEnter(unsigned int n)
{
	menu_setup(n);
	gAlt = 1;
	tri.buttonLedSet(TrillRackInterface::kSolid, TrillRackInterface::kG, 1, 100);
}

static std::array<float,kNumOutChannels> outDiffs {};
static std::array<float,kNumOutChannels> pastOutReverseMapped {};

void tr_render(BelaContext* context)
{
#ifdef PRINT_CPU_TIME
	auto cpuCyclesStart = *DWT_CYCCNT;
	samplesPerBlock = context->analogFrames;
	sampleRate = context->analogSampleRate;
#endif // PRINT_CPU_TIME
	gp_processIncoming();
	static uint32_t pastFrameId = kInvalidFrameId;
	uint32_t frameId = trillFrameId.load();
	bool newFrame = false;
	if(frameId != kInvalidFrameId && frameId != pastFrameId)
	{
		newFrame = true;
		pastFrameId = frameId;
	}
#if 0 //  count average frames of capacitive data
	static uint32_t firstFrameId = frameId;
	static uint32_t count = 0;
	if(newFrame)
	{
		static uint32_t lastCount = 0;
		if(lastCount && count - lastCount != 5)
			printf("WRONG COUNT: %lu\n", count - lastCount);
		lastCount = count;
		if(count > 1000) {
			float mean = (frameId - firstFrameId) / double(count);
			printf(" %.4f\n\r", mean);
			count = 0;
			lastCount = 0;
			firstFrameId = frameId;
		}
	}
	count++;
#endif
	FrameData frameData = {
			.id = pastFrameId,
			.isNew = newFrame,
	};
#ifdef STM32
	static std::array<uint32_t, 2> pastTicks;
	static size_t pastTicksIdx = 0;
	static size_t backoff = 0;
	// ensure that we are not taking up all CPU
	// i.e.: check that systick had a chance to run recently
	if(backoff)
		backoff--;
	if(backoff)
		return;

	uint32_t tick = HAL_GetTick();
	bool different = false;
	for(auto& t : pastTicks) {
		if(tick != t) {
			different = true;
			break;
		}
	}
	pastTicks[pastTicksIdx] = tick;
	static_assert(!((pastTicks.size() - 1) & pastTicks.size())); // ensure it's a power of 2 so that the next line works
	pastTicksIdx = (pastTicksIdx + 1) & (pastTicks.size() - 1);
	if(pastTicks.size() == pastTicksIdx)
		pastTicksIdx &= pastTicks.size() - 1;

	if(!different){
		if(!(0x3 & gDebugFlags))
			printf("C");
		backoff = 2;
		return;
	}
#endif // STM32
	np.clear(); // clear display before we start writing to it
	triggerInToClock(context);

	float min = kSliderBottomMargin;
	float max = 1.f - kSliderTopMargin;
	if(gAlt && 5 == ledSlidersAlt.sliders.size())
	{
		// tweak to get more accurate mapping of the menu selectors to the LEDs
		max = uio.touchStripSwapped() ? 1 : max;
		min = uio.menuSwapped() ? 0.1 : min;
	}
	if(uio.touchStripSwapped())
		std::swap(min, max);
	globalSlider.setUsableRange(min, max);

	const CalibrationData& inCal = getCalibrationInput();
	// rescale analog inputs according to range
	// TODO: don't do it if we are using this input for trig instead.
	// TODO: don't do it if we are only using one or 0 frames
	for(size_t idx = 0; idx < context->analogFrames * context->analogInChannels; ++idx)
		context->analogIn[idx] = rescaleInput(inCal, context->analogIn[idx]);

	static_assert(inCal.points[1] == CalibrationData::kGnd); // we assume points[1] represents gnd
	extern size_t msToNumBlocks(BelaContext* context, float ms);
	static const uint32_t kDoubleClickTime = msToNumBlocks(context, 300);
	static const uint32_t kTripleClickTime = msToNumBlocks(context, 700);
	static uint32_t timeNow = 0;
	static uint32_t lastOnsetTime = 0;
	static uint32_t lastLastOnsetTime = 0;
	static ButtonView btn; // reflects reality
	static bool wasPressed = !tri.digitalRead(0);
	static uint32_t doubleClickPressId = -1;
	static uint32_t tripleClickPressId = -1;
	btn = {false, false, false, false, false, false, false, false, btn.pressId, btn.pressDuration};
	bool isPressed = !tri.digitalRead(0);
	btn.enabled = true;
	btn.offset = wasPressed && !isPressed;
	btn.onset = isPressed && !wasPressed;
	if(btn.onset)
	{
		btn.pressId++;
		if(ButtonView::kPressIdInvalid == btn.pressId) // reserved value
			btn.pressId = 0;
		btn.pressDuration = 0;
		if(timeNow - lastLastOnsetTime < kTripleClickTime)
		{
			btn.tripleClick = true;
			tripleClickPressId = btn.pressId;
		}
		else if(timeNow - lastOnsetTime < kDoubleClickTime)
		{
			btn.doubleClick = true;
			doubleClickPressId = btn.pressId;
		}
		lastLastOnsetTime = lastOnsetTime;
		lastOnsetTime = timeNow;
	}
	if(btn.offset)
	{
		 if(btn.pressId == tripleClickPressId)
			 btn.tripleClickOffset = true;
		 else if(btn.pressId == doubleClickPressId)
			btn.doubleClickOffset = true;
	}
	btn.pressed = isPressed;
	if(btn.pressed)
		btn.pressDuration += isPressed;
	else
		btn.pressDuration = 0;
	wasPressed = isPressed;
	timeNow++;

	static bool hadTouch = false;
	if(newFrame)
		globalSlider.process(trill.rawData.data());
	size_t numTouches = globalSlider.getNumTouches();
	static enum {
		kMenuChangeDisabled = 0,
		kMenuPre,
	} menuState = kMenuChangeDisabled;
	static constexpr size_t kTouchesForMenuLock = 4;
	static size_t maxTouchesThisMenuPre = 0;
	if(btn.pressed)
	{
		static constexpr size_t kTouchesForFactoryTest = 5;
		static constexpr std::array<size_t, 3> touchesForMenu = {
			2, // kTouchesForLocalSettings
			3, // kTouchesForGlobalSettings
			kTouchesForFactoryTest,
		};

		// keep tracking of max touches to avoid entering the wrong mode when releasing
		if(numTouches && !hadTouch) {// we start touching
			menuState = kMenuPre;
			maxTouchesThisMenuPre = 0;
		}
		if(!numTouches) // we no longer touch
			menuState = kMenuChangeDisabled;
		if(kMenuPre == menuState && (!gModeWantsMenuDelay || timeNow - lastOnsetTime > msToNumBlocks(context, 500)))
		{
			if(numTouches > maxTouchesThisMenuPre && !menu_isLocked())
			{
				// we have a new touch
				for(ssize_t n = touchesForMenu.size() - 1; n >= 0; --n)
				{
					if(touchesForMenu[n] == numTouches)
					{
						renderMenuEnter(n);
						break;
					}
				}
			}
			static size_t pastTouches = 0;
			static uint64_t touchesLastChanged = 0;
			if(numTouches != pastTouches)
			{
				touchesLastChanged = context->audioFramesElapsed;
			}
			pastTouches = numTouches;
			// some additional special cases are handled here:
			if(menu_isLocked())
			{
				if(kTouchesForMenuLock == numTouches && context->audioFramesElapsed > touchesLastChanged + 4 * context->analogSampleRate)
				{
					menu_setLocked(false);
					renderMenuEnter(0);
				}
			} else if(kTouchesForFactoryTest == numTouches)
			{
				if(context->audioFramesElapsed > touchesLastChanged + 10 * context->analogSampleRate)
				{
					menu_setup(touchesForMenu.size());
					menuState = kMenuChangeDisabled;
					gAlt = 0;
					menu_exit();
				}
			}

			maxTouchesThisMenuPre = std::max(maxTouchesThisMenuPre, numTouches);
		}
	} else
		menuState = kMenuChangeDisabled; // shouldn't be needed, but, you know ...
	hadTouch = numTouches > 0;
	bool menuActive = (1 == gAlt);

	// multiplexer part 1
	const static ButtonView disBtn = {0};
	menuBtn = menuActive ? btn : disBtn;
	ledSlidersAlt.enableTouch(menuActive);
	ledSlidersAlt.enableLeds(menuActive);

	if(menuActive)
	{
		// has to run before multiplexer part 2 so
		// that any slider that gets recreated because of menu action
		// doesn't lose its enables. TODO: fix this better
		menu_render(context, &frameData); // this will set gAlt back to 0 when exiting menu
	}

	static unsigned int showLock = 0;
	static bool menuExitWaitingButtonRelease = false;
	static bool menuExitWaitingTouchRelease = false;
	static unsigned int menuExitTouchHoldButtonPresses = 0;
	static int oldAlt = gAlt;
	if(1 == oldAlt && 0 == gAlt)
	{
		// we just got out of menu mode. We may have done so by pressing the button (with or without a touch)
		// or by releasing a touch (in which case we won't have any touches here)
		// If button was pressed and/or touch was active when exiting from menu,
		// we should wait for each to be released before they get re-enabled for performance
		if(btn.pressed)
			menuExitWaitingButtonRelease = true;
		if(globalSlider.getNumTouches())
			menuExitWaitingTouchRelease = true;
		menuExitTouchHoldButtonPresses = 0;
		// also reset the button's state machine
		lastOnsetTime = 0;
		lastLastOnsetTime = 0;
	} else { // else ensures we don't run this uselessly in the same block where they were set
		if(menuExitWaitingButtonRelease) {
			if(!btn.pressed && !btn.offset)
				menuExitWaitingButtonRelease = false;
		}
		if(menuExitWaitingTouchRelease) {
			if(btn.onset)
			{
				menuExitTouchHoldButtonPresses++;
				if(kTouchesForMenuLock == maxTouchesThisMenuPre)
				{
					if(3 == menuExitTouchHoldButtonPresses)
					{
						menu_setLocked(true);
						showLock = 400;
					}
				}
			}
			if(!globalSlider.getNumTouches())
				menuExitWaitingTouchRelease = false;
		}
	}
	oldAlt = gAlt;
	if(showLock)
	{
		showLock--;
		for(size_t n = np.getNumPixels() * 0.25f; n < np.getNumPixels() * 0.75f; ++n)
			np.setPixelColor(n, {100, 0, 0});
	}


	// multiplexer part 2
	bool performanceActive = (0 == gAlt) && !menuActive && kMenuChangeDisabled == menuState && !showLock;
	performanceBtn = (performanceActive && !menuExitWaitingButtonRelease) ? btn : disBtn;
	bool forceAllowPerformanceInteraction = !menuActive && gModeWantsInteractionPreMenu;
	gInPreMenu = (!menuActive && !performanceActive);
	ledSliders.enableTouch((performanceActive && !menuExitWaitingTouchRelease) || forceAllowPerformanceInteraction);
	ledSliders.enableLeds(performanceActive || forceAllowPerformanceInteraction);

	static double setupMs;
	static bool setupDone = false;
	static int lastMode = -1;
	if(lastMode != gNewMode) {
		setupMs = tri.getTimeMs();
		printf("new mode: %d to %d\n\r", lastMode, gNewMode);
		lastMode = gNewMode;
		setupDone = false;
	}

	if(!setupDone) {
		setupDone = performanceMode_setup(tri.getTimeMs() - setupMs);
	}
	if(setupDone)
	{
		if(newFrame)
			ledSliders.process(globalSlider);
		performanceMode_render(context, &frameData);
	} else {
		// zero the outputs
		gManualAnOut[0] = gManualAnOut[1] = 0;
		for(size_t n = 0; n < context->analogFrames; ++n)
		{
			for(size_t c = 0; c < context->analogOutChannels; ++c)
				analogWriteJacks(context, n, c, 0);
		}
	}
	if(uio.touchStripSwapped())
		np.reverse();
	// actually display the updated LEDs
	// this may have been written by alt, mode_setups or mode_renders, whatever last wrote it is whatever we display
	// TODO: clear separation of concerns: at any time make it clear who can write to each pixel.
	if(tr_ledsUpdateRequested())
	{
		static float brightness = 0.2f;
		static bool justStarted = true;
		if(justStarted)
		{
			// at startup, fade in brightness if it is not the default
			float alpha = 0.0005;
			brightness = brightness * (1.f - alpha) + gBrightness * alpha;
			if(std::abs(brightness - gBrightness) < 0.01) // close enough, stop smoothing
				justStarted = false;
		} else {
			brightness = gBrightness;
		}
		np.scaleBy(brightness);
		np.show();
	}
//	tri.buttonLedWrite(gMtrClkTriggerLED);
	
	// write analog outputs
	const CalibrationData& outCal = getCalibrationOutput();

	assert(kNumOutChannels == context->analogOutChannels);

	std::array<bool,kNumOutChannels> smoothed {};
	std::array<bool,kNumOutChannels> block {};
	for(size_t c = 0; c < kNumOutChannels; ++c)
	{
		smoothed[c] = (gOutMode[c] == kOutModeManualBlock) || (gOutMode[c] == kOutModeManualSampleSmoothed || (gOutMode[c] == kOutModeManualBlockCustomSmoothed));
		block[c] = (gOutMode[c] == kOutModeManualBlock) || (gOutMode[c] == kOutModeManualBlockCustomSmoothed);
	}
	std::array<float,kNumOutChannels> rangeGnd;
	for(size_t n = 0; n < kNumOutChannels; ++n)
	{
		float gnd = (0 == n ? gOutRangeTop : gOutRangeBottom).getGnd();
		rangeGnd[n] = gnd; // we do not constrain, so this could be negative if gnd is out of the range
	}
	static std::array<Smoother,kNumOutChannels> smoothers {};
	for(size_t c = 0; c < kNumOutChannels; ++c)
	{
		// double precision is expensive so we use it only if we are expecting
		// alphas very close to 1
		smoothers[c].setDoublePrecision(kOutModeManualBlockCustomSmoothed == gOutMode[c]);
	}
	for(unsigned int n = 0; n < context->analogFrames; ++n)
	{
		for(unsigned int channel = 0; channel < kNumOutChannels; ++channel)
		{
			float start;
			if(block[channel])
				start = gManualAnOut[channel];
			else // analogOut has already been written. Rescale in-place
				start = context->analogOut[n * kNumOutChannels + channel];
			if(kOutModeManualBlockCustomSmoothed == gOutMode[channel])
			{
				if(kNoOutput == start) // pretend we have some actual data: go to 0V
					start = rangeGnd[channel];
			}
			static auto pastSmoothed = smoothed;
			static std::array<bool,kNumOutChannels> pastStartWasNoOutput{true, true};
			float rescaled = rescaleOutput(false, channel, outCal, start);
			if(smoothed[channel] && !pastSmoothed[channel])
			{
				// if we haven't processed the channel for some time, reset the filter
				smoothers[channel].set(rescaled);
				pastStartWasNoOutput[channel] = true;
			}
			pastSmoothed[channel] = smoothed[channel];
			float alpha;
			bool startIsNoOutput = (start == kNoOutput);
			if(smoothed[channel])
			{
				if(startIsNoOutput || pastStartWasNoOutput[channel])
					alpha = 0;
				else
					alpha = gOutMode[channel] == kOutModeManualBlockCustomSmoothed ? gCustomSmoothedAlpha[channel] : kAlphaDefault;
			} else
				alpha = 0;
			pastStartWasNoOutput[channel] = startIsNoOutput;
			float out = smoothers[channel].process(rescaled, alpha);
			analogWriteOnce(context, n, channel, out);
			outDiffs[channel] = out - rescaled;
		}
	}
	// Store the past smoothed output after reverse-applying the voltage range.
	// This may be used by some modes that need it for visualisation purposes.
	for(size_t c = 0; c < kNumOutChannels; ++c)
	{
		pastOutReverseMapped[c] = reverseRescaleOutput(false, c, outCal, smoothers[c].get());
	}

	// we do the loop again to swap channels if needed
	// this can be incorporated in the above loop
	// but it needs a lot of care not to overwrite the other channel
	// note: if the two channels have a different gOutMode, this may swap some stale data
	// that will be overwritten below
	// TODO: consolidate these three loops. Try to incorporate this above, or just use it on
	// all channels after the next loop.
		if(uio.outputsSwapped())
		{
			for(unsigned int n = 0; n < context->analogFrames; ++n)
			{
				size_t idx0 = n * kNumOutChannels;
				std::swap(context->analogOut[idx0], context->analogOut[idx0 + 1]);
			}
		}

	// finalise
	for(size_t n = 0; n < context->analogFrames; ++n)
	{
		for(size_t c = 0; c < context->analogOutChannels; ++c)
		{
			size_t idx = n * context->analogOutChannels + c;
			context->analogOut[idx] = finalise(context->analogOut[idx]);
		}
	}

	bool overrideOutput = tick - gOverride.started < 10;
	if(overrideOutput)
	{
		unsigned int c = gOverride.ch;
		float value = rescaleOutput(gOverride.bypassOutRange, c, outCal, gOverride.out);
		for(unsigned int n = 0; n < context->analogFrames; ++n)
			analogWriteJacks(context, n, c, finalise(value));
	}
#if 0 // send out a quiet tone on one channel and loop back the input on the other
	for(unsigned int n = 0; n < context->analogFrames; ++n)
	{
		analogWriteJacks(context, n, 0, analogRead(context, n, 0));
		float osc = (n % 32 > 16);
		analogWriteOnce(context, n , 1, (osc + 2048) / 4096.f);
	}
#endif
#ifdef PRINT_CPU_TIME
	{
		auto time = *DWT_CYCCNT - cpuCyclesStart;
		static decltype(time) minTime = -1;
		static decltype(time) maxTime = 0;
		static unsigned int count = 0;

		if(time < minTime)
			minTime = time;
		if(time > maxTime)
			maxTime = time;
		if(++count == 500)
		{
			count = 0;
			::maxTime = maxTime;
			::minTime = minTime;
			cpuReportsCount++;
			minTime = -1;
			maxTime = 0;
		}
	}
#endif // PRINT_CPU_TIME
}

float getOutputSmoothDiff(size_t idx)
{
	if(idx < kNumOutChannels)
		return outDiffs[idx];
	return 0;
}

float getOutputSmoothDiffNormalised(size_t idx)
{
	if(idx < kNumOutChannels)
	{
		float min;
		float max;
		getRangeMinMax(false, idx, min, max);
		return getOutputSmoothDiff(idx) * 15.f / std::abs(max - min);
	}
	return 0;
}

float getOutputReverseMap(size_t idx)
{
	if(idx < kNumOutChannels)
	{
		return pastOutReverseMapped[idx];
	}
	return 0;
}