summaryrefslogtreecommitdiffstats
path: root/video/out/opengl/video_shaders.c
blob: f56c687903aa3fb8d33ba4ffc75b9c118f5227ea (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
/*
 * This file is part of mpv.
 *
 * mpv 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.
 *
 * mpv 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 mpv.  If not, see <http://www.gnu.org/licenses/>.
 */

#include <math.h>

#include "video_shaders.h"
#include "video.h"

#define GLSL(x) gl_sc_add(sc, #x "\n");
#define GLSLF(...) gl_sc_addf(sc, __VA_ARGS__)
#define GLSLH(x) gl_sc_hadd(sc, #x "\n");
#define GLSLHF(...) gl_sc_haddf(sc, __VA_ARGS__)

// Set up shared/commonly used variables and macros
void sampler_prelude(struct gl_shader_cache *sc, int tex_num)
{
    GLSLF("#undef tex\n");
    GLSLF("#define tex texture%d\n", tex_num);
    GLSLF("vec2 pos = texcoord%d;\n", tex_num);
    GLSLF("vec2 size = texture_size%d;\n", tex_num);
    GLSLF("vec2 pt = pixel_size%d;\n", tex_num);
}

static void pass_sample_separated_get_weights(struct gl_shader_cache *sc,
                                              struct scaler *scaler)
{
    gl_sc_uniform_tex(sc, "lut", scaler->gl_target, scaler->gl_lut);
    // Define a new variable to cache the corrected fcoord.
    GLSLF("float fcoord_lut = LUT_POS(fcoord, %d.0);\n", scaler->lut_size);

    int N = scaler->kernel->size;
    if (N == 2) {
        GLSL(vec2 c1 = texture(lut, vec2(0.5, fcoord_lut)).rg;)
        GLSL(float weights[2] = float[](c1.r, c1.g);)
    } else if (N == 6) {
        GLSL(vec4 c1 = texture(lut, vec2(0.25, fcoord_lut));)
        GLSL(vec4 c2 = texture(lut, vec2(0.75, fcoord_lut));)
        GLSL(float weights[6] = float[](c1.r, c1.g, c1.b, c2.r, c2.g, c2.b);)
    } else {
        GLSLF("float weights[%d];\n", N);
        for (int n = 0; n < N / 4; n++) {
            GLSLF("c = texture(lut, vec2(1.0 / %d.0 + %d.0 / %d.0, fcoord_lut));\n",
                    N / 2, n, N / 4);
            GLSLF("weights[%d] = c.r;\n", n * 4 + 0);
            GLSLF("weights[%d] = c.g;\n", n * 4 + 1);
            GLSLF("weights[%d] = c.b;\n", n * 4 + 2);
            GLSLF("weights[%d] = c.a;\n", n * 4 + 3);
        }
    }
}

// Handle a single pass (either vertical or horizontal). The direction is given
// by the vector (d_x, d_y). If the vector is 0, then planar interpolation is
// used instead (samples from texture0 through textureN)
void pass_sample_separated_gen(struct gl_shader_cache *sc, struct scaler *scaler,
                               int d_x, int d_y)
{
    int N = scaler->kernel->size;
    bool use_ar = scaler->conf.antiring > 0;
    bool planar = d_x == 0 && d_y == 0;
    GLSL(color = vec4(0.0);)
    GLSLF("{\n");
    if (!planar) {
        GLSLF("vec2 dir = vec2(%d.0, %d.0);\n", d_x, d_y);
        GLSL(pt *= dir;)
        GLSL(float fcoord = dot(fract(pos * size - vec2(0.5)), dir);)
        GLSLF("vec2 base = pos - fcoord * pt - pt * vec2(%d.0);\n", N / 2 - 1);
    }
    GLSL(vec4 c;)
    if (use_ar) {
        GLSL(vec4 hi = vec4(0.0);)
        GLSL(vec4 lo = vec4(1.0);)
    }
    pass_sample_separated_get_weights(sc, scaler);
    GLSLF("// scaler samples\n");
    for (int n = 0; n < N; n++) {
        if (planar) {
            GLSLF("c = texture(texture%d, texcoord%d);\n", n, n);
        } else {
            GLSLF("c = texture(tex, base + pt * vec2(%d.0));\n", n);
        }
        GLSLF("color += vec4(weights[%d]) * c;\n", n);
        if (use_ar && (n == N/2-1 || n == N/2)) {
            GLSL(lo = min(lo, c);)
            GLSL(hi = max(hi, c);)
        }
    }
    if (use_ar)
        GLSLF("color = mix(color, clamp(color, lo, hi), %f);\n",
              scaler->conf.antiring);
    GLSLF("}\n");
}

void pass_sample_polar(struct gl_shader_cache *sc, struct scaler *scaler)
{
    double radius = scaler->kernel->f.radius * scaler->kernel->filter_scale;
    int bound = ceil(radius);
    bool use_ar = scaler->conf.antiring > 0;
    GLSL(color = vec4(0.0);)
    GLSLF("{\n");
    GLSL(vec2 fcoord = fract(pos * size - vec2(0.5));)
    GLSL(vec2 base = pos - fcoord * pt;)
    GLSL(vec4 c;)
    GLSLF("float w, d, wsum = 0.0;\n");
    if (use_ar) {
        GLSL(vec4 lo = vec4(1.0);)
        GLSL(vec4 hi = vec4(0.0);)
    }
    gl_sc_uniform_tex(sc, "lut", scaler->gl_target, scaler->gl_lut);
    GLSLF("// scaler samples\n");
    for (int y = 1-bound; y <= bound; y++) {
        for (int x = 1-bound; x <= bound; x++) {
            // Since we can't know the subpixel position in advance, assume a
            // worst case scenario
            int yy = y > 0 ? y-1 : y;
            int xx = x > 0 ? x-1 : x;
            double dmax = sqrt(xx*xx + yy*yy);
            // Skip samples definitely outside the radius
            if (dmax >= radius)
                continue;
            GLSLF("d = length(vec2(%d.0, %d.0) - fcoord)/%f;\n", x, y, radius);
            // Check for samples that might be skippable
            if (dmax >= radius - M_SQRT2)
                GLSLF("if (d < 1.0) {\n");
            if (scaler->gl_target == GL_TEXTURE_1D) {
                GLSLF("w = texture1D(lut, LUT_POS(d, %d.0)).r;\n",
                      scaler->lut_size);
            } else {
                GLSLF("w = texture(lut, vec2(0.5, LUT_POS(d, %d.0))).r;\n",
                      scaler->lut_size);
            }
            GLSL(wsum += w;)
            GLSLF("c = texture(tex, base + pt * vec2(%d.0, %d.0));\n", x, y);
            GLSL(color += vec4(w) * c;)
            if (use_ar && x >= 0 && y >= 0 && x <= 1 && y <= 1) {
                GLSL(lo = min(lo, c);)
                GLSL(hi = max(hi, c);)
            }
            if (dmax >= radius - M_SQRT2)
                GLSLF("}\n");
        }
    }
    GLSL(color = color / vec4(wsum);)
    if (use_ar)
        GLSLF("color = mix(color, clamp(color, lo, hi), %f);\n",
              scaler->conf.antiring);
    GLSLF("}\n");
}

static void bicubic_calcweights(struct gl_shader_cache *sc, const char *t, const char *s)
{
    // Explanation of how bicubic scaling with only 4 texel fetches is done:
    //   http://www.mate.tue.nl/mate/pdfs/10318.pdf
    //   'Efficient GPU-Based Texture Interpolation using Uniform B-Splines'
    // Explanation why this algorithm normally always blurs, even with unit
    // scaling:
    //   http://bigwww.epfl.ch/preprints/ruijters1001p.pdf
    //   'GPU Prefilter for Accurate Cubic B-spline Interpolation'
    GLSLF("vec4 %s = vec4(-0.5, 0.1666, 0.3333, -0.3333) * %s"
                " + vec4(1, 0, -0.5, 0.5);\n", t, s);
    GLSLF("%s = %s * %s + vec4(0, 0, -0.5, 0.5);\n", t, t, s);
    GLSLF("%s = %s * %s + vec4(-0.6666, 0, 0.8333, 0.1666);\n", t, t, s);
    GLSLF("%s.xy *= vec2(1, 1) / vec2(%s.z, %s.w);\n", t, t, t);
    GLSLF("%s.xy += vec2(1.0 + %s, 1.0 - %s);\n", t, s, s);
}

void pass_sample_bicubic_fast(struct gl_shader_cache *sc)
{
    GLSLF("{\n");
    GLSL(vec2 fcoord = fract(pos * size + vec2(0.5, 0.5));)
    bicubic_calcweights(sc, "parmx", "fcoord.x");
    bicubic_calcweights(sc, "parmy", "fcoord.y");
    GLSL(vec4 cdelta;)
    GLSL(cdelta.xz = parmx.rg * vec2(-pt.x, pt.x);)
    GLSL(cdelta.yw = parmy.rg * vec2(-pt.y, pt.y);)
    // first y-interpolation
    GLSL(vec4 ar = texture(tex, pos + cdelta.xy);)
    GLSL(vec4 ag = texture(tex, pos + cdelta.xw);)
    GLSL(vec4 ab = mix(ag, ar, parmy.b);)
    // second y-interpolation
    GLSL(vec4 br = texture(tex, pos + cdelta.zy);)
    GLSL(vec4 bg = texture(tex, pos + cdelta.zw);)
    GLSL(vec4 aa = mix(bg, br, parmy.b);)
    // x-interpolation
    GLSL(color = mix(aa, ab, parmx.b);)
    GLSLF("}\n");
}

void pass_sample_oversample(struct gl_shader_cache *sc, struct scaler *scaler,
                                   int w, int h)
{
    GLSLF("{\n");
    GLSL(vec2 pos = pos + vec2(0.5) * pt;) // round to nearest
    GLSL(vec2 fcoord = fract(pos * size - vec2(0.5));)
    // Determine the mixing coefficient vector
    gl_sc_uniform_vec2(sc, "output_size", (float[2]){w, h});
    GLSL(vec2 coeff = fcoord * output_size/size;)
    float threshold = scaler->conf.kernel.params[0];
    threshold = isnan(threshold) ? 0.0 : threshold;
    GLSLF("coeff = (coeff - %f) / %f;\n", threshold, 1.0 - 2 * threshold);
    GLSL(coeff = clamp(coeff, 0.0, 1.0);)
    // Compute the right blend of colors
    GLSL(color = texture(tex, pos + pt * (coeff - fcoord));)
    GLSLF("}\n");
}

// Common constants for SMPTE ST.2084 (HDR)
static const float PQ_M1 = 2610./4096 * 1./4,
                   PQ_M2 = 2523./4096 * 128,
                   PQ_C1 = 3424./4096,
                   PQ_C2 = 2413./4096 * 32,
                   PQ_C3 = 2392./4096 * 32;

// Common constants for ARIB STD-B67 (HLG)
static const float HLG_A = 0.17883277,
                   HLG_B = 0.28466892,
                   HLG_C = 0.55991073;

// Common constants for Panasonic V-Log
static const float VLOG_B = 0.00873,
                   VLOG_C = 0.241514,
                   VLOG_D = 0.598206;

// Linearize (expand), given a TRC as input. In essence, this is the ITU-R
// EOTF, calculated on an idealized (reference) monitor with a white point of
// MP_REF_WHITE and infinite contrast.
void pass_linearize(struct gl_shader_cache *sc, enum mp_csp_trc trc)
{
    if (trc == MP_CSP_TRC_LINEAR)
        return;

    GLSLF("// linearize\n");

    // Note that this clamp may technically violate the definition of
    // ITU-R BT.2100, which allows for sub-blacks and super-whites to be
    // displayed on the display where such would be possible. That said, the
    // problem is that not all gamma curves are well-defined on the values
    // outside this range, so we ignore it and just clip anyway for sanity.
    GLSL(color.rgb = clamp(color.rgb, 0.0, 1.0);)

    switch (trc) {
    case MP_CSP_TRC_SRGB:
        GLSL(color.rgb = mix(color.rgb / vec3(12.92),
                             pow((color.rgb + vec3(0.055))/vec3(1.055), vec3(2.4)),
                             lessThan(vec3(0.04045), color.rgb));)
        break;
    case MP_CSP_TRC_BT_1886:
        GLSL(color.rgb = pow(color.rgb, vec3(2.4));)
        break;
    case MP_CSP_TRC_GAMMA18:
        GLSL(color.rgb = pow(color.rgb, vec3(1.8));)
        break;
    case MP_CSP_TRC_GAMMA22:
        GLSL(color.rgb = pow(color.rgb, vec3(2.2));)
        break;
    case MP_CSP_TRC_GAMMA28:
        GLSL(color.rgb = pow(color.rgb, vec3(2.8));)
        break;
    case MP_CSP_TRC_PRO_PHOTO:
        GLSL(color.rgb = mix(color.rgb / vec3(16.0),
                             pow(color.rgb, vec3(1.8)),
                             lessThan(vec3(0.03125), color.rgb));)
        break;
    case MP_CSP_TRC_PQ:
        GLSLF("color.rgb = pow(color.rgb, vec3(1.0/%f));\n", PQ_M2);
        GLSLF("color.rgb = max(color.rgb - vec3(%f), vec3(0.0)) \n"
              "             / (vec3(%f) - vec3(%f) * color.rgb);\n",
              PQ_C1, PQ_C2, PQ_C3);
        GLSLF("color.rgb = pow(color.rgb, vec3(1.0/%f));\n", PQ_M1);
        // PQ's output range is 0-10000, but we need it to be relative to to
        // MP_REF_WHITE instead, so rescale
        GLSLF("color.rgb *= vec3(%f);\n", 10000 / MP_REF_WHITE);
        break;
    case MP_CSP_TRC_HLG:
        GLSLF("color.rgb = mix(vec3(4.0) * color.rgb * color.rgb,\n"
              "                exp((color.rgb - vec3(%f)) / vec3(%f)) + vec3(%f),\n"
              "                lessThan(vec3(0.5), color.rgb));\n",
              HLG_C, HLG_A, HLG_B);
        break;
    case MP_CSP_TRC_V_LOG:
        GLSLF("color.rgb = mix((color.rgb - vec3(0.125)) / vec3(5.6), \n"
              "    pow(vec3(10.0), (color.rgb - vec3(%f)) / vec3(%f)) \n"
              "              - vec3(%f),                              \n"
              "    lessThanEqual(vec3(0.181), color.rgb));            \n",
              VLOG_D, VLOG_C, VLOG_B);
        break;
    default:
        abort();
    }

    // Rescale to prevent clipping on non-float textures
    GLSLF("color.rgb /= vec3(%f);\n", mp_trc_nom_peak(trc));
}

// Delinearize (compress), given a TRC as output. This corresponds to the
// inverse EOTF (not the OETF) in ITU-R terminology, again assuming a
// reference monitor.
void pass_delinearize(struct gl_shader_cache *sc, enum mp_csp_trc trc)
{
    if (trc == MP_CSP_TRC_LINEAR)
        return;

    GLSLF("// delinearize\n");
    GLSL(color.rgb = clamp(color.rgb, 0.0, 1.0);)
    GLSLF("color.rgb *= vec3(%f);\n", mp_trc_nom_peak(trc));

    switch (trc) {
    case MP_CSP_TRC_SRGB:
        GLSL(color.rgb = mix(color.rgb * vec3(12.92),
                             vec3(1.055) * pow(color.rgb, vec3(1.0/2.4))
                                 - vec3(0.055),
                             lessThanEqual(vec3(0.0031308), color.rgb));)
        break;
    case MP_CSP_TRC_BT_1886:
        GLSL(color.rgb = pow(color.rgb, vec3(1.0/2.4));)
        break;
    case MP_CSP_TRC_GAMMA18:
        GLSL(color.rgb = pow(color.rgb, vec3(1.0/1.8));)
        break;
    case MP_CSP_TRC_GAMMA22:
        GLSL(color.rgb = pow(color.rgb, vec3(1.0/2.2));)
        break;
    case MP_CSP_TRC_GAMMA28:
        GLSL(color.rgb = pow(color.rgb, vec3(1.0/2.8));)
        break;
    case MP_CSP_TRC_PRO_PHOTO:
        GLSL(color.rgb = mix(color.rgb * vec3(16.0),
                             pow(color.rgb, vec3(1.0/1.8)),
                             lessThanEqual(vec3(0.001953), color.rgb));)
        break;
    case MP_CSP_TRC_PQ:
        GLSLF("color.rgb /= vec3(%f);\n", 10000 / MP_REF_WHITE);
        GLSLF("color.rgb = pow(color.rgb, vec3(%f));\n", PQ_M1);
        GLSLF("color.rgb = (vec3(%f) + vec3(%f) * color.rgb) \n"
              "             / (vec3(1.0) + vec3(%f) * color.rgb);\n",
              PQ_C1, PQ_C2, PQ_C3);
        GLSLF("color.rgb = pow(color.rgb, vec3(%f));\n", PQ_M2);
        break;
    case MP_CSP_TRC_HLG:
        GLSLF("color.rgb = mix(vec3(0.5) * sqrt(color.rgb),\n"
              "                vec3(%f) * log(color.rgb - vec3(%f)) + vec3(%f),\n"
              "                lessThan(vec3(1.0), color.rgb));\n",
              HLG_A, HLG_B, HLG_C);
        break;
    case MP_CSP_TRC_V_LOG:
        GLSLF("color.rgb = mix(vec3(5.6) * color.rgb + vec3(0.125),   \n"
              "                vec3(%f) * log(color.rgb + vec3(%f))   \n"
              "                    + vec3(%f),                        \n"
              "                lessThanEqual(vec3(0.01), color.rgb)); \n",
              VLOG_C / M_LN10, VLOG_B, VLOG_D);
        break;
    default:
        abort();
    }
}

// Apply the OOTF mapping from a given light type to display-referred light.
// The extra peak parameter is used to scale the values before and after
// the OOTF, and can be inferred using mp_trc_nom_peak
void pass_ootf(struct gl_shader_cache *sc, enum mp_csp_light light, float peak)
{
    if (light == MP_CSP_LIGHT_DISPLAY)
        return;

    GLSLF("// apply ootf\n", sc);
    GLSLF("color.rgb *= vec3(%f);\n", peak);

    switch (light)
    {
    case MP_CSP_LIGHT_SCENE_HLG:
        // HLG OOTF from BT.2100, assuming a reference display with a
        // peak of 1000 cd/m² -> gamma = 1.2
        GLSL(float luma = dot(color.rgb, vec3(0.2627, 0.6780, 0.0593));)
        GLSLF("color.rgb *= vec3(%f * pow(luma, 0.2));\n",
              (1000 / MP_REF_WHITE) / pow(12, 1.2));
        break;
    case MP_CSP_LIGHT_SCENE_709_1886:
        // This OOTF is defined by encoding the result as 709 and then decoding
        // it as 1886; although this is called 709_1886 we actually use the
        // more precise (by one decimal) values from BT.2020 instead
        GLSL(color.rgb = mix(color.rgb * vec3(4.5),
                             vec3(1.0993) * pow(color.rgb, vec3(0.45)) - vec3(0.0993),
                             lessThan(vec3(0.0181), color.rgb));)
        GLSL(color.rgb = pow(color.rgb, vec3(2.4));)
        break;
    case MP_CSP_LIGHT_SCENE_1_2:
        GLSL(color.rgb = pow(color.rgb, vec3(1.2));)
        break;
    default:
        abort();
    }

    GLSLF("color.rgb /= vec3(%f);\n", peak);
}

// Inverse of the function pass_ootf, for completeness' sake. Note that the
// inverse OOTF for MP_CSP_LIGHT_SCENE_HLG has no analytical solution and is
// therefore unimplemented. Care must be used to never call this function
// in that way.(In principle, a iterative algorithm can approach
// the solution numerically, but this is tricky and we don't really need it
// since mpv currently only supports outputting display-referred light)
void pass_inverse_ootf(struct gl_shader_cache *sc, enum mp_csp_light light, float peak)
{
    if (light == MP_CSP_LIGHT_DISPLAY)
        return;

    GLSLF("// apply inverse ootf\n");
    GLSLF("color.rgb *= vec3(%f);\n", peak);

    switch (light)
    {
    case MP_CSP_LIGHT_SCENE_HLG:
        // Has no analytical solution
        abort();
        break;
    case MP_CSP_LIGHT_SCENE_709_1886:
        GLSL(color.rgb = pow(color.rgb, vec3(1/2.4));)
        GLSL(color.rgb = mix(color.rgb / vec3(4.5),
                             pow((color.rgb + vec3(0.0993)) / vec3(1.0993), vec3(1/0.45)),
                             lessThan(vec3(0.08145), color.rgb));)
        break;
    case MP_CSP_LIGHT_SCENE_1_2:
        GLSL(color.rgb = pow(color.rgb, vec3(1/1.2));)
        break;
    default:
        abort();
    }

    GLSLF("color.rgb /= vec3(%f);\n", peak);
}

// Tone map from a known peak brightness to the range [0,1]
static void pass_tone_map(struct gl_shader_cache *sc, float ref_peak,
                          enum tone_mapping algo, float param)
{
    GLSLF("// HDR tone mapping\n");

    switch (algo) {
    case TONE_MAPPING_CLIP:
        GLSL(color.rgb = clamp(color.rgb, 0.0, 1.0);)
        break;

    case TONE_MAPPING_MOBIUS: {
        float j = isnan(param) ? 0.3 : param;
        // solve for M(j) = j; M(ref_peak) = 1.0; M'(j) = 1.0
        // where M(x) = scale * (x+a)/(x+b)
        float a = -j*j * (ref_peak - 1) / (j*j - 2*j + ref_peak),
              b = (j*j - 2*j*ref_peak + ref_peak) / (ref_peak - 1);

        GLSLF("color.rgb = mix(vec3(%f) * (color.rgb + vec3(%f))\n"
              "                         / (color.rgb + vec3(%f)),\n"
              "                color.rgb,\n"
              "                lessThanEqual(color.rgb, vec3(%f)));\n",
              (b*b + 2*b*j + j*j) / (b - a), a, b, j);
        break;
    }

    case TONE_MAPPING_REINHARD: {
        float contrast = isnan(param) ? 0.5 : param,
              offset = (1.0 - contrast) / contrast;
        GLSLF("color.rgb = color.rgb / (color.rgb + vec3(%f));\n", offset);
        GLSLF("color.rgb *= vec3(%f);\n", (ref_peak + offset) / ref_peak);
        break;
    }

    case TONE_MAPPING_HABLE: {
        float A = 0.15, B = 0.50, C = 0.10, D = 0.20, E = 0.02, F = 0.30;
        GLSLHF("vec3 hable(vec3 x) {\n");
        GLSLHF("return ((x * (%f*x + %f)+%f)/(x * (%f*x + %f) + %f)) - %f;\n",
               A, C*B, D*E, A, B, D*F, E/F);
        GLSLHF("}\n");

        GLSLF("color.rgb = hable(color.rgb) / hable(vec3(%f));\n", ref_peak);
        break;
    }

    case TONE_MAPPING_GAMMA: {
        float gamma = isnan(param) ? 1.8 : param;
        GLSLF("color.rgb = pow(color.rgb / vec3(%f), vec3(%f));\n",
              ref_peak, 1.0/gamma);
        break;
    }

    case TONE_MAPPING_LINEAR: {
        float coeff = isnan(param) ? 1.0 : param;
        GLSLF("color.rgb = vec3(%f) * color.rgb;\n", coeff / ref_peak);
        break;
    }

    default:
        abort();
    }
}

// Map colors from one source space to another. These source spaces must be
// known (i.e. not MP_CSP_*_AUTO), as this function won't perform any
// auto-guessing. If is_linear is true, we assume the input has already been
// linearized (e.g. for linear-scaling)
void pass_color_map(struct gl_shader_cache *sc,
                    struct mp_colorspace src, struct mp_colorspace dst,
                    enum tone_mapping algo, float tone_mapping_param,
                    bool is_linear)
{
    GLSLF("// color mapping\n");

    // Compute the highest encodable level
    float src_range = mp_trc_nom_peak(src.gamma),
          dst_range = mp_trc_nom_peak(dst.gamma);

    // All operations from here on require linear light as a starting point,
    // so we linearize even if src.gamma == dst.gamma when one of the other
    // operations needs it
    bool need_gamma = src.gamma != dst.gamma ||
                      src.primaries != dst.primaries ||
                      src_range != dst_range ||
                      src.sig_peak > dst_range ||
                      src.light != dst.light;

    if (need_gamma && !is_linear) {
        pass_linearize(sc, src.gamma);
        is_linear= true;
    }

    if (src.light != dst.light)
        pass_ootf(sc, src.light, mp_trc_nom_peak(src.gamma));

    // Rescale the signal to compensate for differences in the encoding range
    // and reference white level. This is necessary because of how mpv encodes
    // brightness in textures.
    if (src_range != dst_range) {
        GLSLF("// rescale value range;\n");
        GLSLF("color.rgb *= vec3(%f);\n", src_range / dst_range);
    }

    // Tone map to prevent clipping when the source signal peak exceeds the
    // encodable range
    if (src.sig_peak > dst_range)
        pass_tone_map(sc, src.sig_peak / dst_range, algo, tone_mapping_param);

    // Adapt to the right colorspace if necessary
    if (src.primaries != dst.primaries) {
        struct mp_csp_primaries csp_src = mp_get_csp_primaries(src.primaries),
                                csp_dst = mp_get_csp_primaries(dst.primaries);
        float m[3][3] = {{0}};
        mp_get_cms_matrix(csp_src, csp_dst, MP_INTENT_RELATIVE_COLORIMETRIC, m);
        gl_sc_uniform_mat3(sc, "cms_matrix", true, &m[0][0]);
        GLSL(color.rgb = cms_matrix * color.rgb;)
    }

    if (src.light != dst.light)
        pass_inverse_ootf(sc, dst.light, mp_trc_nom_peak(dst.gamma));

    if (is_linear)
        pass_delinearize(sc, dst.gamma);
}

// Wide usage friendly PRNG, shamelessly stolen from a GLSL tricks forum post.
// Obtain random numbers by calling rand(h), followed by h = permute(h) to
// update the state. Assumes the texture was hooked.
static void prng_init(struct gl_shader_cache *sc, AVLFG *lfg)
{
    GLSLH(float mod289(float x)  { return x - floor(x / 289.0) * 289.0; })
    GLSLH(float permute(float x) { return mod289((34.0*x + 1.0) * x); })
    GLSLH(float rand(float x)    { return fract(x / 41.0); })

    // Initialize the PRNG by hashing the position + a random uniform
    GLSL(vec3 _m = vec3(HOOKED_pos, random) + vec3(1.0);)
    GLSL(float h = permute(permute(permute(_m.x)+_m.y)+_m.z);)
    gl_sc_uniform_f(sc, "random", (double)av_lfg_get(lfg) / UINT32_MAX);
}

struct deband_opts {
    int enabled;
    int iterations;
    float threshold;
    float range;
    float grain;
};

const struct deband_opts deband_opts_def = {
    .iterations = 1,
    .threshold = 64.0,
    .range = 16.0,
    .grain = 48.0,
};

#define OPT_BASE_STRUCT struct deband_opts
const struct m_sub_options deband_conf = {
    .opts = (const m_option_t[]) {
        OPT_INTRANGE("iterations", iterations, 0, 1, 16),
        OPT_FLOATRANGE("threshold", threshold, 0, 0.0, 4096.0),
        OPT_FLOATRANGE("range", range, 0, 1.0, 64.0),
        OPT_FLOATRANGE("grain", grain, 0, 0.0, 4096.0),
        {0}
    },
    .size = sizeof(struct deband_opts),
    .defaults = &deband_opts_def,
};

// Stochastically sample a debanded result from a hooked texture.
void pass_sample_deband(struct gl_shader_cache *sc, struct deband_opts *opts,
                        AVLFG *lfg)
{
    // Initialize the PRNG
    GLSLF("{\n");
    prng_init(sc, lfg);

    // Helper: Compute a stochastic approximation of the avg color around a
    // pixel
    GLSLHF("vec4 average(float range, inout float h) {\n");
        // Compute a random rangle and distance
        GLSLH(float dist = rand(h) * range;     h = permute(h);)
        GLSLH(float dir  = rand(h) * 6.2831853; h = permute(h);)
        GLSLH(vec2 o = dist * vec2(cos(dir), sin(dir));)

        // Sample at quarter-turn intervals around the source pixel
        GLSLH(vec4 ref[4];)
        GLSLH(ref[0] = HOOKED_texOff(vec2( o.x,  o.y));)
        GLSLH(ref[1] = HOOKED_texOff(vec2(-o.y,  o.x));)
        GLSLH(ref[2] = HOOKED_texOff(vec2(-o.x, -o.y));)
        GLSLH(ref[3] = HOOKED_texOff(vec2( o.y, -o.x));)

        // Return the (normalized) average
        GLSLH(return (ref[0] + ref[1] + ref[2] + ref[3])/4.0;)
    GLSLHF("}\n");

    // Sample the source pixel
    GLSL(color = HOOKED_tex(HOOKED_pos);)
    GLSLF("vec4 avg, diff;\n");
    for (int i = 1; i <= opts->iterations; i++) {
        // Sample the average pixel and use it instead of the original if
        // the difference is below the given threshold
        GLSLF("avg = average(%f, h);\n", i * opts->range);
        GLSL(diff = abs(color - avg);)
        GLSLF("color = mix(avg, color, greaterThan(diff, vec4(%f)));\n",
              opts->threshold / (i * 16384.0));
    }

    // Add some random noise to smooth out residual differences
    GLSL(vec3 noise;)
    GLSL(noise.x = rand(h); h = permute(h);)
    GLSL(noise.y = rand(h); h = permute(h);)
    GLSL(noise.z = rand(h); h = permute(h);)
    GLSLF("color.xyz += %f * (noise - vec3(0.5));\n", opts->grain/8192.0);
    GLSLF("}\n");
}

// Assumes the texture was hooked
void pass_sample_unsharp(struct gl_shader_cache *sc, float param) {
    GLSLF("// unsharp\n");
    GLSLF("{\n");
    GLSL(float st1 = 1.2;)
    GLSL(vec4 p = HOOKED_tex(HOOKED_pos);)
    GLSL(vec4 sum1 = HOOKED_texOff(st1 * vec2(+1, +1))
                   + HOOKED_texOff(st1 * vec2(+1, -1))
                   + HOOKED_texOff(st1 * vec2(-1, +1))
                   + HOOKED_texOff(st1 * vec2(-1, -1));)
    GLSL(float st2 = 1.5;)
    GLSL(vec4 sum2 = HOOKED_texOff(st2 * vec2(+1,  0))
                   + HOOKED_texOff(st2 * vec2( 0, +1))
                   + HOOKED_texOff(st2 * vec2(-1,  0))
                   + HOOKED_texOff(st2 * vec2( 0, -1));)
    GLSL(vec4 t = p * 0.859375 + sum2 * -0.1171875 + sum1 * -0.09765625;)
    GLSLF("color = p + t * %f;\n", param);
    GLSLF("}\n");
}