.git - libcamera pipeline handler for VIVID
summaryrefslogtreecommitdiff
path: root/src/ipa/raspberrypi/controller/rpi/alsc.cpp
blob: a4afaf841c4146d175fe23d99444f58852a62bd4 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
/* SPDX-License-Identifier: BSD-2-Clause */
/*
 * Copyright (C) 2019, Raspberry Pi Ltd
 *
 * alsc.cpp - ALSC (auto lens shading correction) control algorithm
 */

#include <functional>
#include <math.h>
#include <numeric>

#include <libcamera/base/log.h>
#include <libcamera/base/span.h>

#include "../awb_status.h"
#include "alsc.h"

/* Raspberry Pi ALSC (Auto Lens Shading Correction) algorithm. */

using namespace RPiController;
using namespace libcamera;

LOG_DEFINE_CATEGORY(RPiAlsc)

#define NAME "rpi.alsc"

static const int X = AlscCellsX;
static const int Y = AlscCellsY;
static const int XY = X * Y;
static const double InsufficientData = -1.0;

Alsc::Alsc(Controller *controller)
	: Algorithm(controller)
{
	asyncAbort_ = asyncStart_ = asyncStarted_ = asyncFinished_ = false;
	asyncThread_ = std::thread(std::bind(&Alsc::asyncFunc, this));
}

Alsc::~Alsc()
{
	{
		std::lock_guard<std::mutex> lock(mutex_);
		asyncAbort_ = true;
	}
	asyncSignal_.notify_one();
	asyncThread_.join();
}

char const *Alsc::name() const
{
	return NAME;
}

static int generateLut(double *lut, const libcamera::YamlObject &params)
{
	double cstrength = params["corner_strength"].get<double>(2.0);
	if (cstrength <= 1.0) {
		LOG(RPiAlsc, Error) << "corner_strength must be > 1.0";
		return -EINVAL;
	}

	double asymmetry = params["asymmetry"].get<double>(1.0);
	if (asymmetry < 0) {
		LOG(RPiAlsc, Error) << "asymmetry must be >= 0";
		return -EINVAL;
	}

	double f1 = cstrength - 1, f2 = 1 + sqrt(cstrength);
	double R2 = X * Y / 4 * (1 + asymmetry * asymmetry);
	int num = 0;
	for (int y = 0; y < Y; y++) {
		for (int x = 0; x < X; x++) {
			double dy = y - Y / 2 + 0.5,
			       dx = (x - X / 2 + 0.5) * asymmetry;
			double r2 = (dx * dx + dy * dy) / R2;
			lut[num++] =
				(f1 * r2 + f2) * (f1 * r2 + f2) /
				(f2 * f2); /* this reproduces the cos^4 rule */
		}
	}
	return 0;
}

static int readLut(double *lut, const libcamera::YamlObject &params)
{
	if (params.size() != XY) {
		LOG(RPiAlsc, Error) << "Invalid number of entries in LSC table";
		return -EINVAL;
	}

	int num = 0;
	for (const auto &p : params.asList()) {
		auto value = p.get<double>();
		if (!value)
			return -EINVAL;
		lut[num++] = *value;
	}

	return 0;
}

static int readCalibrations(std::vector<AlscCalibration> &calibrations,
			    const libcamera::YamlObject &params,
			    std::string const &name)
{
	if (params.contains(name)) {
		double lastCt = 0;
		for (const auto &p : params[name].asList()) {
			auto value = p["ct"].get<double>();
			if (!value)
				return -EINVAL;
			double ct = *value;
			if (ct <= lastCt) {
				LOG(RPiAlsc, Error)
					<< "Entries in " << name << " must be in increasing ct order";
				return -EINVAL;
			}
			AlscCalibration calibration;
			calibration.ct = lastCt = ct;

			const libcamera::YamlObject &table = p["table"];
			if (table.size() != XY) {
				LOG(RPiAlsc, Error)
					<< "Incorrect number of values for ct "
					<< ct << " in " << name;
				return -EINVAL;
			}

			int num = 0;
			for (const auto &elem : table.asList()) {
				value = elem.get<double>();
				if (!value)
					return -EINVAL;
				calibration.table[num++] = *value;
			}

			calibrations.push_back(calibration);
			LOG(RPiAlsc, Debug)
				<< "Read " << name << " calibration for ct " << ct;
		}
	}
	return 0;
}

int Alsc::read(const libcamera::YamlObject &params)
{
	config_.framePeriod = params["frame_period"].get<uint16_t>(12);
	config_.startupFrames = params["startup_frames"].get<uint16_t>(10);
	config_.speed = params["speed"].get<double>(0.05);
	double sigma = params["sigma"].get<double>(0.01);
	config_.sigmaCr = params["sigma_Cr"].get<double>(sigma);
	config_.sigmaCb = params["sigma_Cb"].get<double>(sigma);
	config_.minCount = params["min_count"].get<double>(10.0);
	config_.minG = params["min_G"].get<uint16_t>(50);
	config_.omega = params["omega"].get<double>(1.3);
	config_.nIter = params["n_iter"].get<uint32_t>(X + Y);
	config_.luminanceStrength =
		params["luminance_strength"].get<double>(1.0);
	for (int i = 0; i < XY; i++)
		config_.luminanceLut[i] = 1.0;

	int ret = 0;

	if (params.contains("corner_strength"))
		ret = generateLut(config_.luminanceLut, params);
	else if (params.contains("luminance_lut"))
		ret = readLut(config_.luminanceLut, params["luminance_lut"]);
	else
		LOG(RPiAlsc, Warning)
			<< "no luminance table - assume unity everywhere";
	if (ret)
		return ret;

	ret = readCalibrations(config_.calibrationsCr, params, "calibrations_Cr");
	if (ret)
		return ret;
	ret = readCalibrations(config_.calibrationsCb, params, "calibrations_Cb");
	if (ret)
		return ret;

	config_.defaultCt = params["default_ct"].get<double>(4500.0);
	config_.threshold = params["threshold"].get<double>(1e-3);
	config_.lambdaBound = params["lambda_bound"].get<double>(0.05);

	return 0;
}

static double getCt(Metadata *metadata, double defaultCt);
static void getCalTable(double ct, std::vector<AlscCalibration> const &calibrations,
			double calTable[XY]);
static void resampleCalTable(double const calTableIn[XY], CameraMode const &cameraMode,
			     double calTableOut[XY]);
static void compensateLambdasForCal(double const calTable[XY], double const oldLambdas[XY],
				    double newLambdas[XY]);
static void addLuminanceToTables(double results[3][Y][X], double const lambdaR[XY], double lambdaG,
				 double const lambdaB[XY], double const luminanceLut[XY],
				 double luminanceStrength);

void Alsc::initialise()
{
	frameCount2_ = frameCount_ = framePhase_ = 0;
	firstTime_ = true;
	ct_ = config_.defaultCt;
	/* The lambdas are initialised in the SwitchMode. */
}

void Alsc::waitForAysncThread()
{
	if (asyncStarted_) {
		asyncStarted_ = false;
		std::unique_lock<std::mutex> lock(mutex_);
		syncSignal_.wait(lock, [&] {
			return asyncFinished_;
		});
		asyncFinished_ = false;
	}
}

static bool compareModes(CameraMode const &cm0, CameraMode const &cm1)
{
	/*
	 * Return true if the modes crop from the sensor significantly differently,
	 * or if the user transform has changed.
	 */
	if (cm0.transform != cm1.transform)
		return true;
	int leftDiff = abs(cm0.cropX - cm1.cropX);
	int topDiff = abs(cm0.cropY - cm1.cropY);
	int rightDiff = fabs(cm0.cropX + cm0.scaleX * cm0.width -
			     cm1.cropX - cm1.scaleX * cm1.width);
	int bottomDiff = fabs(cm0.cropY + cm0.scaleY * cm0.height -
			      cm1.cropY - cm1.scaleY * cm1.height);
	/*
	 * These thresholds are a rather arbitrary amount chosen to trigger
	 * when carrying on with the previously calculated tables might be
	 * worse than regenerating them (but without the adaptive algorithm).
	 */
	int thresholdX = cm0.sensorWidth >> 4;
	int thresholdY = cm0.sensorHeight >> 4;
	return leftDiff > thresholdX || rightDiff > thresholdX ||
	       topDiff > thresholdY || bottomDiff > thresholdY;
}

void Alsc::switchMode(CameraMode const &cameraMode,
		      [[maybe_unused]] Metadata *metadata)
{
	/*
	 * We're going to start over with the tables if there's any "significant"
	 * change.
	 */
	bool resetTables = firstTime_ || compareModes(cameraMode_, cameraMode);

	/* Believe the colour temperature from the AWB, if there is one. */
	ct_ = getCt(metadata, ct_);

	/* Ensure the other thread isn't running while we do this. */
	waitForAysncThread();

	cameraMode_ = cameraMode;

	/*
	 * We must resample the luminance table like we do the others, but it's
	 * fixed so we can simply do it up front here.
	 */
	resampleCalTable(config_.luminanceLut, cameraMode_, luminanceTable_);

	if (resetTables) {
		/*
		 * Upon every "table reset", arrange for something sensible to be
		 * generated. Construct the tables for the previous recorded colour
		 * temperature. In order to start over from scratch we initialise
		 * the lambdas, but the rest of this code then echoes the code in
		 * doAlsc, without the adaptive algorithm.
		 */
		for (int i = 0; i < XY; i++)
			lambdaR_[i] = lambdaB_[i] = 1.0;
		double calTableR[XY], calTableB[XY], calTableTmp[XY];
		getCalTable(ct_, config_.calibrationsCr, calTableTmp);
		resampleCalTable(calTableTmp, cameraMode_, calTableR);
		getCalTable(ct_, config_.calibrationsCb, calTableTmp);
		resampleCalTable(calTableTmp, cameraMode_, calTableB);
		compensateLambdasForCal(calTableR, lambdaR_, asyncLambdaR_);
		compensateLambdasForCal(calTableB, lambdaB_, asyncLambdaB_);
		addLuminanceToTables(syncResults_, asyncLambdaR_, 1.0, asyncLambdaB_,
				     luminanceTable_, config_.luminanceStrength);
		memcpy(prevSyncResults_, syncResults_, sizeof(prevSyncResults_));
		framePhase_ = config_.framePeriod; /* run the algo again asap */
		firstTime_ = false;
	}
}

void Alsc::fetchAsyncResults()
{
	LOG(RPiAlsc, Debug) << "Fetch ALSC results";
	asyncFinished_ = false;
	asyncStarted_ = false;
	memcpy(syncResults_, asyncResults_, sizeof(syncResults_));
}

double getCt(Metadata *metadata, double defaultCt)
{
	AwbStatus awbStatus;
	awbStatus.temperatureK = defaultCt; /* in case nothing found */
	if (metadata->get("awb.status", awbStatus) != 0)
		LOG(RPiAlsc, Debug) << "no AWB results found, using "
				    << awbStatus.temperatureK;
	else
		LOG(RPiAlsc, Debug) << "AWB results found, using "
				    << awbStatus.temperatureK;
	return awbStatus.temperatureK;
}

static void copyStats(bcm2835_isp_stats_region regions[XY], StatisticsPtr &stats,
		      AlscStatus const &status)
{
	bcm2835_isp_stats_region *inputRegions = stats->awb_stats;
	double *rTable = (double *)status.r;
	double *gTable = (double *)status.g;
	double *bTable = (double *)status.b;
	for (int i = 0; i < XY; i++) {
		regions[i].r_sum = inputRegions[i].r_sum / rTable[i];
		regions[i].g_sum = inputRegions[i].g_sum / gTable[i];
		regions[i].b_sum = inputRegions[i].b_sum / bTable[i];
		regions[i].counted = inputRegions[i].counted;
		/* (don't care about the uncounted value) */
	}
}

void Alsc::restartAsync(StatisticsPtr &stats, Metadata *imageMetadata)
{
	LOG(RPiAlsc, Debug) << "Starting ALSC calculation";
	/*
	 * Get the current colour temperature. It's all we need from the
	 * metadata. Default to the last CT value (which could be the default).
	 */
	ct_ = getCt(imageMetadata, ct_);
	/*
	 * We have to copy the statistics here, dividing out our best guess of
	 * the LSC table that the pipeline applied to them.
	 */
	AlscStatus alscStatus;
	if (imageMetadata->get("alsc.status", alscStatus) != 0) {
		LOG(RPiAlsc, Warning)
			<< "No ALSC status found for applied gains!";
		for (int y = 0; y < Y; y++)
			for (int x = 0; x < X; x++) {
				alscStatus.r[y][x] = 1.0;
				alscStatus.g[y][x] = 1.0;
				alscStatus.b[y][x] = 1.0;
			}
	}
	copyStats(statistics_, stats, alscStatus);
	framePhase_ = 0;
	asyncStarted_ = true;
	{
		std::lock_guard<std::mutex> lock(mutex_);
		asyncStart_ = true;
	}
	asyncSignal_.notify_one();
}

void Alsc::prepare(Metadata *imageMetadata)
{
	/*
	 * Count frames since we started, and since we last poked the async
	 * thread.
	 */
	if (frameCount_ < (int)config_.startupFrames)
		frameCount_++;
	double speed = frameCount_ < (int)config_.startupFrames
			       ? 1.0
			       : config_.speed;
	LOG(RPiAlsc, Debug)
		<< "frame count " << frameCount_ << " speed " << speed;
	{
		std::unique_lock<std::mutex> lock(mutex_);
		if (asyncStarted_ && asyncFinished_)
			fetchAsyncResults();
	}
	/* Apply IIR filter to results and program into the pipeline. */
	double *ptr = (double *)syncResults_,
	       *pptr = (double *)prevSyncResults_;
	for (unsigned int i = 0; i < sizeof(syncResults_) / sizeof(double); i++)
		pptr[i] = speed * ptr[i] + (1.0 - speed) * pptr[i];
	/* Put output values into status metadata. */
	AlscStatus status;
	memcpy(status.r, prevSyncResults_[0], sizeof(status.r));
	memcpy(status.g, prevSyncResults_[1], sizeof(status.g));
	memcpy(status.b, prevSyncResults_[2], sizeof(status.b));
	imageMetadata->set("alsc.status", status);
}

void Alsc::process(StatisticsPtr &stats, Metadata *imageMetadata)
{
	/*
	 * Count frames since we started, and since we last poked the async
	 * thread.
	 */
	if (framePhase_ < (int)config_.framePeriod)
		framePhase_++;
	if (frameCount2_ < (int)config_.startupFrames)
		frameCount2_++;
	LOG(RPiAlsc, Debug) << "frame_phase " << framePhase_;
	if (framePhase_ >= (int)config_.framePeriod ||
	    frameCount2_ < (int)config_.startupFrames) {
		if (asyncStarted_ == false)
			restartAsync(stats, imageMetadata);
	}
}

void Alsc::asyncFunc()
{
	while (true) {
		{
			std::unique_lock<std::mutex> lock(mutex_);
			asyncSignal_.wait(lock, [&] {
				return asyncStart_ || asyncAbort_;
			});
			asyncStart_ = false;
			if (asyncAbort_)
				break;
		}
		doAlsc();
		{
			std::lock_guard<std::mutex> lock(mutex_);
			asyncFinished_ = true;
		}
		syncSignal_.notify_one();
	}
}

void getCalTable(double ct, std::vector<AlscCalibration> const &calibrations,
		 double calTable[XY])
{
	if (calibrations.empty()) {
		for (int i = 0; i < XY; i++)
			calTable[i] = 1.0;
		LOG(RPiAlsc, Debug) << "no calibrations found";
	} else if (ct <= calibrations.front().ct) {
		memcpy(calTable, calibrations.front().table, XY * sizeof(double));
		LOG(RPiAlsc, Debug) << "using calibration for "
				    << calibrations.front().ct;
	} else if (ct >= calibrations.back().ct) {
		memcpy(calTable, calibrations.back().table, XY * sizeof(double));
		LOG(RPiAlsc, Debug) << "using calibration for "
				    << calibrations.back().ct;
	} else {
		int idx = 0;
		while (ct > calibrations[idx + 1].ct)
			idx++;
		double ct0 = calibrations[idx].ct, ct1 = calibrations[idx + 1].ct;
		LOG(RPiAlsc, Debug)
			<< "ct is " << ct << ", interpolating between "
			<< ct0 << " and " << ct1;
		for (int i = 0; i < XY; i++)
			calTable[i] =
				(calibrations[idx].table[i] * (ct1 - ct) +
				 calibrations[idx + 1].table[i] * (ct - ct0)) /
				(ct1 - ct0);
	}
}

void resampleCalTable(double const calTableIn[XY],
		      CameraMode const &cameraMode, double calTableOut[XY])
{
	/*
	 * Precalculate and cache the x sampling locations and phases to save
	 * recomputing them on every row.
	 */
	int xLo[X], xHi[X];
	double xf[X];
	double scaleX = cameraMode.sensorWidth /
			(cameraMode.width * cameraMode.scaleX);
	double xOff = cameraMode.cropX / (double)cameraMode.sensorWidth;
	double x = .5 / scaleX + xOff * X - .5;
	double xInc = 1 / scaleX;
	for (int i = 0; i < X; i++, x += xInc) {
		xLo[i] = floor(x);
		xf[i] = x - xLo[i];
		xHi[i] = std::min(xLo[i] + 1, X - 1);
		xLo[i] = std::max(xLo[i], 0);
		if (!!(cameraMode.transform & libcamera::Transform::HFlip)) {
			xLo[i] = X - 1 - xLo[i];
			xHi[i] = X - 1 - xHi[i];
		}
	}
	/* Now march over the output table generating the new values. */
	double scaleY = cameraMode.sensorHeight /
			(cameraMode.height * cameraMode.scaleY);
	double yOff = cameraMode.cropY / (double)cameraMode.sensorHeight;
	double y = .5 / scaleY + yOff * Y - .5;
	double yInc = 1 / scaleY;
	for (int j = 0; j < Y; j++, y += yInc) {
		int yLo = floor(y);
		double yf = y - yLo;
		int yHi = std::min(yLo + 1, Y - 1);
		yLo = std::max(yLo, 0);
		if (!!(cameraMode.transform & libcamera::Transform::VFlip)) {
			yLo = Y - 1 - yLo;
			yHi = Y - 1 - yHi;
		}
		double const *rowAbove = calTableIn + X * yLo;
		double const *rowBelow = calTableIn + X * yHi;
		for (int i = 0; i < X; i++) {
			double above = rowAbove[xLo[i]] * (1 - xf[i]) +
				       rowAbove[xHi[i]] * xf[i];
			double below = rowBelow[xLo[i]] * (1 - xf[i]) +
				       rowBelow[xHi[i]] * xf[i];
			*(calTableOut++) = above * (1 - yf) + below * yf;
		}
	}
}

/* Calculate chrominance statistics (R/G and B/G) for each region. */
static_assert(XY == AWB_REGIONS, "ALSC/AWB statistics region mismatch");
static void calculateCrCb(bcm2835_isp_stats_region *awbRegion, double cr[XY],
			  double cb[XY], uint32_t minCount, uint16_t minG)
{
	for (int i = 0; i < XY; i++) {
		bcm2835_isp_stats_region &zone = awbRegion[i];
		if (zone.counted <= minCount ||
		    zone.g_sum / zone.counted <= minG) {
			cr[i] = cb[i] = InsufficientData;
			continue;
		}
		cr[i] = zone.r_sum / (double)zone.g_sum;
		cb[i] = zone.b_sum / (double)zone.g_sum;
	}
}

static void applyCalTable(double const calTable[XY], double C[XY])
{
	for (int i = 0; i < XY; i++)
		if (C[i] != InsufficientData)
			C[i] *= calTable[i];
}

void compensateLambdasForCal(double const calTable[XY],
			     double const oldLambdas[XY],
			     double newLambdas[XY])
{
	double minNewLambda = std::numeric_limits<double>::max();
	for (int i = 0; i < XY; i++) {
		newLambdas[i] = oldLambdas[i] * calTable[i];
		minNewLambda = std::min(minNewLambda, newLambdas[i]);
	}
	for (int i = 0; i < XY; i++)
		newLambdas[i] /= minNewLambda;
}

[[maybe_unused]] static void printCalTable(double const C[XY])
{
	printf("table: [\n");
	for (int j = 0; j < Y; j++) {
		for (int i = 0; i < X; i++) {
			printf("%5.3f", 1.0 / C[j * X + i]);
			if (i != X - 1 || j != Y - 1)
				printf(",");
		}
		printf("\n");
	}
	printf("]\n");
}

/*
 * Compute weight out of 1.0 which reflects how similar we wish to make the
 * colours of these two regions.
 */
static double computeWeight(double Ci, double Cj, double sigma)
{
	if (Ci == InsufficientData || Cj == InsufficientData)
		return 0;
	double diff = (Ci - Cj) / sigma;
	return exp(-diff * diff / 2);
}

/* Compute all weights. */
static void computeW(double const C[XY], double sigma, double W[XY][4])
{
	for (int i = 0; i < XY; i++) {
		/* Start with neighbour above and go clockwise. */
		W[i][0] = i >= X ? computeWeight(C[i], C[i - X], sigma) : 0;
		W[i][1] = i % X < X - 1 ? computeWeight(C[i], C[i + 1], sigma) : 0;
		W[i][2] = i < XY - X ? computeWeight(C[i], C[i + X], sigma) : 0;
		W[i][3] = i % X ? computeWeight(C[i], C[i - 1], sigma) : 0;
	}
}

/* Compute M, the large but sparse matrix such that M * lambdas = 0. */
static void constructM(double const C[XY], double const W[XY][4],
		       double M[XY][4])
{
	double epsilon = 0.001;
	for (int i = 0; i < XY; i++) {
		/*
		 * Note how, if C[i] == INSUFFICIENT_DATA, the weights will all
		 * be zero so the equation is still set up correctly.
		 */
		int m = !!(i >= X) + !!(i % X < X - 1) + !!(i < XY - X) +
			!!(i % X); /* total number of neighbours */
		/* we'll divide the diagonal out straight away */
		double diagonal = (epsilon + W[i][0] + W[i][1] + W[i][2] + W[i][3]) * C[i];
		M[i][0] = i >= X ? (W[i][0] * C[i - X] + epsilon / m * C[i]) / diagonal : 0;
		M[i][1] = i % X < X - 1 ? (W[i][1] * C[i + 1] + epsilon / m * C[i]) / diagonal : 0;
		M[i][2] = i < XY - X ? (W[i][2] * C[i + X] + epsilon / m * C[i]) / diagonal : 0;
		M[i][3] = i % X ? (W[i][3] * C[i - 1] + epsilon / m * C[i]) / diagonal : 0;
	}
}

/*
 * In the compute_lambda_ functions, note that the matrix coefficients for the
 * left/right neighbours are zero down the left/right edges, so we don't need
 * need to test the i value to exclude them.
 */
static double computeLambdaBottom(int i, double const M[XY][4],
				  double lambda[XY])
{
	return M[i][1] * lambda[i + 1] + M[i][2] * lambda[i + X] +
	       M[i][3] * lambda[i - 1];
}
static double computeLambdaBottomStart(int i, double const M[XY][4],
				       double lambda[XY])
{
	return M[i][1] * lambda[i + 1] + M[i][2] * lambda[i + X];
}
static double computeLambdaInterior(int i, double const M[XY][4],
				    double lambda[XY])
{
	return M[i][0] * lambda[i - X] + M[i][1] * lambda[i + 1] +
	       M[i][2] * lambda[i + X] + M[i][3] * lambda[i - 1];
}
static double computeLambdaTop(int i, double const M[XY][4],
			       double lambda[XY])
{
	return M[i][0] * lambda[i - X] + M[i][1] * lambda[i + 1] +
	       M[i][3] * lambda[i - 1];
}
static double computeLambdaTopEnd(int i, double const M[XY][4],
				  double lambda[XY])
{
	return M[i][0] * lambda[i - X] + M[i][3] * lambda[i - 1];
}

/* Gauss-Seidel iteration with over-relaxation. */
static double gaussSeidel2Sor(double const M[XY][4], double omega,
			      double lambda[XY], double lambdaBound)
{
	const double min = 1 - lambdaBound, max = 1 + lambdaBound;
	double oldLambda[XY];
	int i;
	for (i = 0; i < XY; i++)
		oldLambda[i] = lambda[i];
	lambda[0] = computeLambdaBottomStart(0, M, lambda);
	lambda[0] = std::clamp(lambda[0], min, max);
	for (i = 1; i < X; i++) {
		lambda[i] = computeLambdaBottom(i, M, lambda);
		lambda[i] = std::clamp(lambda[i], min, max);
	}
	for (; i < XY - X; i++) {
		lambda[i] = computeLambdaInterior(i, M, lambda);
		lambda[i] = std::clamp(lambda[i], min, max);
	}
	for (; i < XY - 1; i++) {
		lambda[i] = computeLambdaTop(i, M, lambda);
		lambda[i] = std::clamp(lambda[i], min, max);
	}
	lambda[i] = computeLambdaTopEnd(i, M, lambda);
	lambda[i] = std::clamp(lambda[i], min, max);
	/*
	 * Also solve the system from bottom to top, to help spread the updates
	 * better.
	 */
	lambda[i] = computeLambdaTopEnd(i, M, lambda);