-
Notifications
You must be signed in to change notification settings - Fork 9
/
effect_chain.cpp
2416 lines (2158 loc) · 82.2 KB
/
effect_chain.cpp
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
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
#include <epoxy/gl.h>
#include <assert.h>
#include <math.h>
#include <stddef.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <algorithm>
#include <set>
#include <stack>
#include <utility>
#include <vector>
#include <Eigen/Core>
#include "alpha_division_effect.h"
#include "alpha_multiplication_effect.h"
#include "colorspace_conversion_effect.h"
#include "dither_effect.h"
#include "effect.h"
#include "effect_chain.h"
#include "effect_util.h"
#include "gamma_compression_effect.h"
#include "gamma_expansion_effect.h"
#include "init.h"
#include "input.h"
#include "resource_pool.h"
#include "util.h"
#include "ycbcr_conversion_effect.h"
using namespace Eigen;
using namespace std;
namespace movit {
namespace {
// An effect whose only purpose is to sit in a phase on its own and take the
// texture output from a compute shader and display it to the normal backbuffer
// (or any FBO). That phase can be skipped when rendering using render_to_textures().
class ComputeShaderOutputDisplayEffect : public Effect {
public:
ComputeShaderOutputDisplayEffect() {}
string effect_type_id() const override { return "ComputeShaderOutputDisplayEffect"; }
string output_fragment_shader() override { return read_file("identity.frag"); }
bool needs_texture_bounce() const override { return true; }
};
} // namespace
EffectChain::EffectChain(float aspect_nom, float aspect_denom, ResourcePool *resource_pool)
: aspect_nom(aspect_nom),
aspect_denom(aspect_denom),
output_color_rgba(false),
num_output_color_ycbcr(0),
dither_effect(nullptr),
ycbcr_conversion_effect_node(nullptr),
intermediate_format(GL_RGBA16F),
intermediate_transformation(NO_FRAMEBUFFER_TRANSFORMATION),
num_dither_bits(0),
output_origin(OUTPUT_ORIGIN_BOTTOM_LEFT),
finalized(false),
resource_pool(resource_pool),
do_phase_timing(false) {
if (resource_pool == nullptr) {
this->resource_pool = new ResourcePool();
owns_resource_pool = true;
} else {
owns_resource_pool = false;
}
// Generate a VBO with some data in (shared position and texture coordinate data).
float vertices[] = {
0.0f, 2.0f,
0.0f, 0.0f,
2.0f, 0.0f
};
vbo = generate_vbo(2, GL_FLOAT, sizeof(vertices), vertices);
}
EffectChain::~EffectChain()
{
for (unsigned i = 0; i < nodes.size(); ++i) {
delete nodes[i]->effect;
delete nodes[i];
}
for (unsigned i = 0; i < phases.size(); ++i) {
resource_pool->release_glsl_program(phases[i]->glsl_program_num);
delete phases[i];
}
if (owns_resource_pool) {
delete resource_pool;
}
glDeleteBuffers(1, &vbo);
check_error();
}
Input *EffectChain::add_input(Input *input)
{
assert(!finalized);
inputs.push_back(input);
add_node(input);
return input;
}
void EffectChain::add_output(const ImageFormat &format, OutputAlphaFormat alpha_format)
{
assert(!finalized);
assert(!output_color_rgba);
output_format = format;
output_alpha_format = alpha_format;
output_color_rgba = true;
}
void EffectChain::add_ycbcr_output(const ImageFormat &format, OutputAlphaFormat alpha_format,
const YCbCrFormat &ycbcr_format, YCbCrOutputSplitting output_splitting,
GLenum output_type)
{
assert(!finalized);
assert(num_output_color_ycbcr < 2);
output_format = format;
output_alpha_format = alpha_format;
if (num_output_color_ycbcr == 1) {
// Check that the format is the same.
assert(output_ycbcr_format.luma_coefficients == ycbcr_format.luma_coefficients);
assert(output_ycbcr_format.full_range == ycbcr_format.full_range);
assert(output_ycbcr_format.num_levels == ycbcr_format.num_levels);
assert(output_ycbcr_format.chroma_subsampling_x == 1);
assert(output_ycbcr_format.chroma_subsampling_y == 1);
assert(output_ycbcr_type == output_type);
} else {
output_ycbcr_format = ycbcr_format;
output_ycbcr_type = output_type;
}
output_ycbcr_splitting[num_output_color_ycbcr++] = output_splitting;
assert(ycbcr_format.chroma_subsampling_x == 1);
assert(ycbcr_format.chroma_subsampling_y == 1);
}
void EffectChain::change_ycbcr_output_format(const YCbCrFormat &ycbcr_format)
{
assert(num_output_color_ycbcr > 0);
assert(output_ycbcr_format.chroma_subsampling_x == 1);
assert(output_ycbcr_format.chroma_subsampling_y == 1);
output_ycbcr_format = ycbcr_format;
if (finalized) {
YCbCrConversionEffect *effect = (YCbCrConversionEffect *)(ycbcr_conversion_effect_node->effect);
effect->change_output_format(ycbcr_format);
}
}
Node *EffectChain::add_node(Effect *effect)
{
for (unsigned i = 0; i < nodes.size(); ++i) {
assert(nodes[i]->effect != effect);
}
Node *node = new Node;
node->effect = effect;
node->disabled = false;
node->output_color_space = COLORSPACE_INVALID;
node->output_gamma_curve = GAMMA_INVALID;
node->output_alpha_type = ALPHA_INVALID;
node->needs_mipmaps = Effect::DOES_NOT_NEED_MIPMAPS;
node->one_to_one_sampling = false;
node->strong_one_to_one_sampling = false;
nodes.push_back(node);
node_map[effect] = node;
effect->inform_added(this);
return node;
}
void EffectChain::connect_nodes(Node *sender, Node *receiver)
{
sender->outgoing_links.push_back(receiver);
receiver->incoming_links.push_back(sender);
}
void EffectChain::replace_receiver(Node *old_receiver, Node *new_receiver)
{
new_receiver->incoming_links = old_receiver->incoming_links;
old_receiver->incoming_links.clear();
for (unsigned i = 0; i < new_receiver->incoming_links.size(); ++i) {
Node *sender = new_receiver->incoming_links[i];
for (unsigned j = 0; j < sender->outgoing_links.size(); ++j) {
if (sender->outgoing_links[j] == old_receiver) {
sender->outgoing_links[j] = new_receiver;
}
}
}
}
void EffectChain::replace_sender(Node *old_sender, Node *new_sender)
{
new_sender->outgoing_links = old_sender->outgoing_links;
old_sender->outgoing_links.clear();
for (unsigned i = 0; i < new_sender->outgoing_links.size(); ++i) {
Node *receiver = new_sender->outgoing_links[i];
for (unsigned j = 0; j < receiver->incoming_links.size(); ++j) {
if (receiver->incoming_links[j] == old_sender) {
receiver->incoming_links[j] = new_sender;
}
}
}
}
void EffectChain::insert_node_between(Node *sender, Node *middle, Node *receiver)
{
for (unsigned i = 0; i < sender->outgoing_links.size(); ++i) {
if (sender->outgoing_links[i] == receiver) {
sender->outgoing_links[i] = middle;
middle->incoming_links.push_back(sender);
}
}
for (unsigned i = 0; i < receiver->incoming_links.size(); ++i) {
if (receiver->incoming_links[i] == sender) {
receiver->incoming_links[i] = middle;
middle->outgoing_links.push_back(receiver);
}
}
assert(middle->incoming_links.size() == middle->effect->num_inputs());
}
GLenum EffectChain::get_input_sampler(Node *node, unsigned input_num) const
{
assert(node->effect->needs_texture_bounce());
assert(input_num < node->incoming_links.size());
assert(node->incoming_links[input_num]->bound_sampler_num >= 0);
assert(node->incoming_links[input_num]->bound_sampler_num < 8);
return GL_TEXTURE0 + node->incoming_links[input_num]->bound_sampler_num;
}
GLenum EffectChain::has_input_sampler(Node *node, unsigned input_num) const
{
assert(input_num < node->incoming_links.size());
return node->incoming_links[input_num]->bound_sampler_num >= 0 &&
node->incoming_links[input_num]->bound_sampler_num < 8;
}
void EffectChain::find_all_nonlinear_inputs(Node *node, vector<Node *> *nonlinear_inputs)
{
if (node->output_gamma_curve == GAMMA_LINEAR &&
node->effect->effect_type_id() != "GammaCompressionEffect") {
return;
}
if (node->effect->num_inputs() == 0) {
nonlinear_inputs->push_back(node);
} else {
assert(node->effect->num_inputs() == node->incoming_links.size());
for (unsigned i = 0; i < node->incoming_links.size(); ++i) {
find_all_nonlinear_inputs(node->incoming_links[i], nonlinear_inputs);
}
}
}
Effect *EffectChain::add_effect(Effect *effect, const vector<Effect *> &inputs)
{
assert(!finalized);
assert(inputs.size() == effect->num_inputs());
Node *node = add_node(effect);
for (unsigned i = 0; i < inputs.size(); ++i) {
assert(node_map.count(inputs[i]) != 0);
connect_nodes(node_map[inputs[i]], node);
}
return effect;
}
// ESSL doesn't support token pasting. Replace PREFIX(x) with <effect_id>_x.
string replace_prefix(const string &text, const string &prefix)
{
string output;
size_t start = 0;
while (start < text.size()) {
size_t pos = text.find("PREFIX(", start);
if (pos == string::npos) {
output.append(text.substr(start, string::npos));
break;
}
output.append(text.substr(start, pos - start));
output.append(prefix);
output.append("_");
pos += strlen("PREFIX(");
// Output stuff until we find the matching ), which we then eat.
int depth = 1;
size_t end_arg_pos = pos;
while (end_arg_pos < text.size()) {
if (text[end_arg_pos] == '(') {
++depth;
} else if (text[end_arg_pos] == ')') {
--depth;
if (depth == 0) {
break;
}
}
++end_arg_pos;
}
output.append(text.substr(pos, end_arg_pos - pos));
++end_arg_pos;
assert(depth == 0);
start = end_arg_pos;
}
return output;
}
namespace {
template<class T>
void extract_uniform_declarations(const vector<Uniform<T>> &effect_uniforms,
const string &type_specifier,
const string &effect_id,
vector<Uniform<T>> *phase_uniforms,
string *glsl_string)
{
for (unsigned i = 0; i < effect_uniforms.size(); ++i) {
phase_uniforms->push_back(effect_uniforms[i]);
phase_uniforms->back().prefix = effect_id;
*glsl_string += string("uniform ") + type_specifier + " " + effect_id
+ "_" + effect_uniforms[i].name + ";\n";
}
}
template<class T>
void extract_uniform_array_declarations(const vector<Uniform<T>> &effect_uniforms,
const string &type_specifier,
const string &effect_id,
vector<Uniform<T>> *phase_uniforms,
string *glsl_string)
{
for (unsigned i = 0; i < effect_uniforms.size(); ++i) {
phase_uniforms->push_back(effect_uniforms[i]);
phase_uniforms->back().prefix = effect_id;
char buf[256];
snprintf(buf, sizeof(buf), "uniform %s %s_%s[%d];\n",
type_specifier.c_str(), effect_id.c_str(),
effect_uniforms[i].name.c_str(),
int(effect_uniforms[i].num_values));
*glsl_string += buf;
}
}
template<class T>
void collect_uniform_locations(GLuint glsl_program_num, vector<Uniform<T>> *phase_uniforms)
{
for (unsigned i = 0; i < phase_uniforms->size(); ++i) {
Uniform<T> &uniform = (*phase_uniforms)[i];
uniform.location = get_uniform_location(glsl_program_num, uniform.prefix, uniform.name);
}
}
} // namespace
void EffectChain::compile_glsl_program(Phase *phase)
{
string frag_shader_header;
if (phase->is_compute_shader) {
frag_shader_header = read_file("header.comp");
} else {
frag_shader_header = read_version_dependent_file("header", "frag");
}
string frag_shader = "";
// Create functions and uniforms for all the texture inputs that we need.
for (unsigned i = 0; i < phase->inputs.size(); ++i) {
Node *input = phase->inputs[i]->output_node;
char effect_id[256];
sprintf(effect_id, "in%u", i);
phase->effect_ids.insert(make_pair(make_pair(input, IN_ANOTHER_PHASE), effect_id));
frag_shader += string("uniform sampler2D tex_") + effect_id + ";\n";
frag_shader += string("vec4 ") + effect_id + "(vec2 tc) {\n";
frag_shader += "\tvec4 tmp = tex2D(tex_" + string(effect_id) + ", tc);\n";
if (intermediate_transformation == SQUARE_ROOT_FRAMEBUFFER_TRANSFORMATION &&
phase->inputs[i]->output_node->output_gamma_curve == GAMMA_LINEAR) {
frag_shader += "\ttmp.rgb *= tmp.rgb;\n";
}
frag_shader += "\treturn tmp;\n";
frag_shader += "}\n";
frag_shader += "\n";
Uniform<int> uniform;
uniform.name = effect_id;
uniform.value = &phase->input_samplers[i];
uniform.prefix = "tex";
uniform.num_values = 1;
uniform.location = -1;
phase->uniforms_sampler2d.push_back(uniform);
}
// Give each effect in the phase its own ID.
for (unsigned i = 0; i < phase->effects.size(); ++i) {
Node *node = phase->effects[i];
char effect_id[256];
sprintf(effect_id, "eff%u", i);
bool inserted = phase->effect_ids.insert(make_pair(make_pair(node, IN_SAME_PHASE), effect_id)).second;
assert(inserted);
}
for (unsigned i = 0; i < phase->effects.size(); ++i) {
Node *node = phase->effects[i];
const string effect_id = phase->effect_ids[make_pair(node, IN_SAME_PHASE)];
for (unsigned j = 0; j < node->incoming_links.size(); ++j) {
if (node->incoming_links.size() == 1) {
frag_shader += "#define INPUT";
} else {
char buf[256];
sprintf(buf, "#define INPUT%d", j + 1);
frag_shader += buf;
}
Node *input = node->incoming_links[j];
NodeLinkType link_type = node->incoming_link_type[j];
if (i != 0 &&
input->effect->is_compute_shader() &&
node->incoming_link_type[j] == IN_SAME_PHASE) {
// First effect after the compute shader reads the value
// that cs_output() wrote to a global variable,
// ignoring the tc (since all such effects have to be
// strong one-to-one).
frag_shader += "(tc) CS_OUTPUT_VAL\n";
} else {
assert(phase->effect_ids.count(make_pair(input, link_type)));
frag_shader += string(" ") + phase->effect_ids[make_pair(input, link_type)] + "\n";
}
}
frag_shader += "\n";
frag_shader += string("#define FUNCNAME ") + effect_id + "\n";
if (node->effect->is_compute_shader()) {
frag_shader += string("#define NORMALIZE_TEXTURE_COORDS(tc) ((tc) * ") + effect_id + "_inv_output_size + " + effect_id + "_output_texcoord_adjust)\n";
}
frag_shader += replace_prefix(node->effect->output_fragment_shader(), effect_id);
frag_shader += "#undef FUNCNAME\n";
if (node->incoming_links.size() == 1) {
frag_shader += "#undef INPUT\n";
} else {
for (unsigned j = 0; j < node->incoming_links.size(); ++j) {
char buf[256];
sprintf(buf, "#undef INPUT%d\n", j + 1);
frag_shader += buf;
}
}
frag_shader += "\n";
}
if (phase->is_compute_shader) {
assert(phase->effect_ids.count(make_pair(phase->compute_shader_node, IN_SAME_PHASE)));
frag_shader += string("#define INPUT ") + phase->effect_ids[make_pair(phase->compute_shader_node, IN_SAME_PHASE)] + "\n";
if (phase->compute_shader_node == phase->effects.back()) {
// No postprocessing.
frag_shader += "#define CS_POSTPROC(tc) CS_OUTPUT_VAL\n";
} else {
frag_shader += string("#define CS_POSTPROC ") + phase->effect_ids[make_pair(phase->effects.back(), IN_SAME_PHASE)] + "\n";
}
} else {
assert(phase->effect_ids.count(make_pair(phase->effects.back(), IN_SAME_PHASE)));
frag_shader += string("#define INPUT ") + phase->effect_ids[make_pair(phase->effects.back(), IN_SAME_PHASE)] + "\n";
}
// If we're the last phase, add the right #defines for Y'CbCr multi-output as needed.
vector<string> frag_shader_outputs; // In order.
if (phase->output_node->outgoing_links.empty() && num_output_color_ycbcr > 0) {
switch (output_ycbcr_splitting[0]) {
case YCBCR_OUTPUT_INTERLEAVED:
// No #defines set.
frag_shader_outputs.push_back("FragColor");
break;
case YCBCR_OUTPUT_SPLIT_Y_AND_CBCR:
frag_shader += "#define YCBCR_OUTPUT_SPLIT_Y_AND_CBCR 1\n";
frag_shader_outputs.push_back("Y");
frag_shader_outputs.push_back("Chroma");
break;
case YCBCR_OUTPUT_PLANAR:
frag_shader += "#define YCBCR_OUTPUT_PLANAR 1\n";
frag_shader_outputs.push_back("Y");
frag_shader_outputs.push_back("Cb");
frag_shader_outputs.push_back("Cr");
break;
default:
assert(false);
}
if (num_output_color_ycbcr > 1) {
switch (output_ycbcr_splitting[1]) {
case YCBCR_OUTPUT_INTERLEAVED:
frag_shader += "#define SECOND_YCBCR_OUTPUT_INTERLEAVED 1\n";
frag_shader_outputs.push_back("YCbCr2");
break;
case YCBCR_OUTPUT_SPLIT_Y_AND_CBCR:
frag_shader += "#define SECOND_YCBCR_OUTPUT_SPLIT_Y_AND_CBCR 1\n";
frag_shader_outputs.push_back("Y2");
frag_shader_outputs.push_back("Chroma2");
break;
case YCBCR_OUTPUT_PLANAR:
frag_shader += "#define SECOND_YCBCR_OUTPUT_PLANAR 1\n";
frag_shader_outputs.push_back("Y2");
frag_shader_outputs.push_back("Cb2");
frag_shader_outputs.push_back("Cr2");
break;
default:
assert(false);
}
}
if (output_color_rgba) {
// Note: Needs to come in the header, because not only the
// output needs to see it (YCbCrConversionEffect and DitherEffect
// do, too).
frag_shader_header += "#define YCBCR_ALSO_OUTPUT_RGBA 1\n";
frag_shader_outputs.push_back("RGBA");
}
}
// If we're bouncing to a temporary texture, signal transformation if desired.
if (!phase->output_node->outgoing_links.empty()) {
if (intermediate_transformation == SQUARE_ROOT_FRAMEBUFFER_TRANSFORMATION &&
phase->output_node->output_gamma_curve == GAMMA_LINEAR) {
frag_shader += "#define SQUARE_ROOT_TRANSFORMATION 1\n";
}
}
if (phase->is_compute_shader) {
frag_shader.append(read_file("footer.comp"));
phase->compute_shader_node->effect->register_uniform_ivec2("output_size", phase->uniform_output_size);
phase->compute_shader_node->effect->register_uniform_vec2("inv_output_size", (float *)&phase->inv_output_size);
phase->compute_shader_node->effect->register_uniform_vec2("output_texcoord_adjust", (float *)&phase->output_texcoord_adjust);
} else {
frag_shader.append(read_file("footer.frag"));
}
// Collect uniforms from all effects and output them. Note that this needs
// to happen after output_fragment_shader(), even though the uniforms come
// before in the output source, since output_fragment_shader() is allowed
// to register new uniforms (e.g. arrays that are of unknown length until
// finalization time).
// TODO: Make a uniform block for platforms that support it.
string frag_shader_uniforms = "";
for (unsigned i = 0; i < phase->effects.size(); ++i) {
Node *node = phase->effects[i];
Effect *effect = node->effect;
const string effect_id = phase->effect_ids[make_pair(node, IN_SAME_PHASE)];
extract_uniform_declarations(effect->uniforms_image2d, "image2D", effect_id, &phase->uniforms_image2d, &frag_shader_uniforms);
extract_uniform_declarations(effect->uniforms_sampler2d, "sampler2D", effect_id, &phase->uniforms_sampler2d, &frag_shader_uniforms);
extract_uniform_declarations(effect->uniforms_bool, "bool", effect_id, &phase->uniforms_bool, &frag_shader_uniforms);
extract_uniform_declarations(effect->uniforms_int, "int", effect_id, &phase->uniforms_int, &frag_shader_uniforms);
extract_uniform_declarations(effect->uniforms_ivec2, "ivec2", effect_id, &phase->uniforms_ivec2, &frag_shader_uniforms);
extract_uniform_declarations(effect->uniforms_float, "float", effect_id, &phase->uniforms_float, &frag_shader_uniforms);
extract_uniform_declarations(effect->uniforms_vec2, "vec2", effect_id, &phase->uniforms_vec2, &frag_shader_uniforms);
extract_uniform_declarations(effect->uniforms_vec3, "vec3", effect_id, &phase->uniforms_vec3, &frag_shader_uniforms);
extract_uniform_declarations(effect->uniforms_vec4, "vec4", effect_id, &phase->uniforms_vec4, &frag_shader_uniforms);
extract_uniform_array_declarations(effect->uniforms_float_array, "float", effect_id, &phase->uniforms_float, &frag_shader_uniforms);
extract_uniform_array_declarations(effect->uniforms_vec2_array, "vec2", effect_id, &phase->uniforms_vec2, &frag_shader_uniforms);
extract_uniform_array_declarations(effect->uniforms_vec3_array, "vec3", effect_id, &phase->uniforms_vec3, &frag_shader_uniforms);
extract_uniform_array_declarations(effect->uniforms_vec4_array, "vec4", effect_id, &phase->uniforms_vec4, &frag_shader_uniforms);
extract_uniform_declarations(effect->uniforms_mat3, "mat3", effect_id, &phase->uniforms_mat3, &frag_shader_uniforms);
}
string vert_shader = read_version_dependent_file("vs", "vert");
// If we're the last phase and need to flip the picture to compensate for
// the origin, tell the vertex or compute shader so.
bool is_last_phase;
if (has_dummy_effect) {
is_last_phase = (phase->output_node->outgoing_links.size() == 1 &&
phase->output_node->outgoing_links[0]->effect->effect_type_id() == "ComputeShaderOutputDisplayEffect");
} else {
is_last_phase = phase->output_node->outgoing_links.empty();
}
if (is_last_phase && output_origin == OUTPUT_ORIGIN_TOP_LEFT) {
if (phase->is_compute_shader) {
frag_shader_header += "#define FLIP_ORIGIN 1\n";
} else {
const string needle = "#define FLIP_ORIGIN 0";
size_t pos = vert_shader.find(needle);
assert(pos != string::npos);
vert_shader[pos + needle.size() - 1] = '1';
}
}
frag_shader = frag_shader_header + frag_shader_uniforms + frag_shader;
if (phase->is_compute_shader) {
phase->glsl_program_num = resource_pool->compile_glsl_compute_program(frag_shader);
Uniform<int> uniform;
uniform.name = "outbuf";
uniform.value = &phase->outbuf_image_unit;
uniform.prefix = "tex";
uniform.num_values = 1;
uniform.location = -1;
phase->uniforms_image2d.push_back(uniform);
} else {
phase->glsl_program_num = resource_pool->compile_glsl_program(vert_shader, frag_shader, frag_shader_outputs);
}
GLint position_attribute_index = glGetAttribLocation(phase->glsl_program_num, "position");
GLint texcoord_attribute_index = glGetAttribLocation(phase->glsl_program_num, "texcoord");
if (position_attribute_index != -1) {
phase->attribute_indexes.insert(position_attribute_index);
}
if (texcoord_attribute_index != -1) {
phase->attribute_indexes.insert(texcoord_attribute_index);
}
// Collect the resulting location numbers for each uniform.
collect_uniform_locations(phase->glsl_program_num, &phase->uniforms_image2d);
collect_uniform_locations(phase->glsl_program_num, &phase->uniforms_sampler2d);
collect_uniform_locations(phase->glsl_program_num, &phase->uniforms_bool);
collect_uniform_locations(phase->glsl_program_num, &phase->uniforms_int);
collect_uniform_locations(phase->glsl_program_num, &phase->uniforms_ivec2);
collect_uniform_locations(phase->glsl_program_num, &phase->uniforms_float);
collect_uniform_locations(phase->glsl_program_num, &phase->uniforms_vec2);
collect_uniform_locations(phase->glsl_program_num, &phase->uniforms_vec3);
collect_uniform_locations(phase->glsl_program_num, &phase->uniforms_vec4);
collect_uniform_locations(phase->glsl_program_num, &phase->uniforms_mat3);
}
// Construct GLSL programs, starting at the given effect and following
// the chain from there. We end a program every time we come to an effect
// marked as "needs texture bounce", one that is used by multiple other
// effects, every time we need to bounce due to output size change
// (not all size changes require ending), and of course at the end.
//
// We follow a quite simple depth-first search from the output, although
// without recursing explicitly within each phase.
Phase *EffectChain::construct_phase(Node *output, map<Node *, Phase *> *completed_effects)
{
if (completed_effects->count(output)) {
return (*completed_effects)[output];
}
Phase *phase = new Phase;
phase->output_node = output;
phase->is_compute_shader = false;
phase->compute_shader_node = nullptr;
// If the output effect has one-to-one sampling, we try to trace this
// status down through the dependency chain. This is important in case
// we hit an effect that changes output size (and not sets a virtual
// output size); if we have one-to-one sampling, we don't have to break
// the phase.
output->one_to_one_sampling = output->effect->one_to_one_sampling();
output->strong_one_to_one_sampling = output->effect->strong_one_to_one_sampling();
// Effects that we have yet to calculate, but that we know should
// be in the current phase.
stack<Node *> effects_todo_this_phase;
effects_todo_this_phase.push(output);
while (!effects_todo_this_phase.empty()) {
Node *node = effects_todo_this_phase.top();
effects_todo_this_phase.pop();
assert(node->effect->one_to_one_sampling() >= node->effect->strong_one_to_one_sampling());
if (node->effect->needs_mipmaps() != Effect::DOES_NOT_NEED_MIPMAPS) {
// Can't have incompatible requirements imposed on us from a dependent effect;
// if so, it should have started a new phase instead.
assert(node->needs_mipmaps == Effect::DOES_NOT_NEED_MIPMAPS ||
node->needs_mipmaps == node->effect->needs_mipmaps());
node->needs_mipmaps = node->effect->needs_mipmaps();
}
// This should currently only happen for effects that are inputs
// (either true inputs or phase outputs). We special-case inputs,
// and then deduplicate phase outputs below.
if (node->effect->num_inputs() == 0) {
if (find(phase->effects.begin(), phase->effects.end(), node) != phase->effects.end()) {
continue;
}
} else {
assert(completed_effects->count(node) == 0);
}
phase->effects.push_back(node);
if (node->effect->is_compute_shader()) {
assert(phase->compute_shader_node == nullptr ||
phase->compute_shader_node == node);
phase->is_compute_shader = true;
phase->compute_shader_node = node;
}
// Find all the dependencies of this effect, and add them to the stack.
assert(node->effect->num_inputs() == node->incoming_links.size());
for (Node *dep : node->incoming_links) {
bool start_new_phase = false;
Effect::MipmapRequirements save_needs_mipmaps = dep->needs_mipmaps;
if (node->effect->needs_texture_bounce() &&
!dep->effect->is_single_texture() &&
!dep->effect->override_disable_bounce()) {
start_new_phase = true;
}
// Propagate information about needing mipmaps down the chain,
// breaking the phase if we notice an incompatibility.
//
// Note that we cannot do this propagation as a normal pass,
// because it needs information about where the phases end
// (we should not propagate the flag across phases).
if (node->needs_mipmaps != Effect::DOES_NOT_NEED_MIPMAPS) {
// The node can have a value set (ie. not DOES_NOT_NEED_MIPMAPS)
// if we have diamonds in the graph; if so, choose that.
// If not, the effect on the node can also decide (this is the
// more common case).
Effect::MipmapRequirements dep_mipmaps = dep->needs_mipmaps;
if (dep_mipmaps == Effect::DOES_NOT_NEED_MIPMAPS) {
if (dep->effect->num_inputs() == 0) {
Input *input = static_cast<Input *>(dep->effect);
dep_mipmaps = input->can_supply_mipmaps() ? Effect::DOES_NOT_NEED_MIPMAPS : Effect::CANNOT_ACCEPT_MIPMAPS;
} else {
dep_mipmaps = dep->effect->needs_mipmaps();
}
}
if (dep_mipmaps == Effect::DOES_NOT_NEED_MIPMAPS) {
dep->needs_mipmaps = node->needs_mipmaps;
} else if (dep_mipmaps != node->needs_mipmaps) {
// The dependency cannot supply our mipmap demands
// (either because it's an input that can't do mipmaps,
// or because there's a conflict between mipmap-needing
// and mipmap-refusing effects somewhere in the graph),
// so they cannot be in the same phase.
start_new_phase = true;
}
}
if (dep->outgoing_links.size() > 1) {
if (!dep->effect->is_single_texture()) {
// More than one effect uses this as the input,
// and it is not a texture itself.
// The easiest thing to do (and probably also the safest
// performance-wise in most cases) is to bounce it to a texture
// and then let the next passes read from that.
start_new_phase = true;
} else {
assert(dep->effect->num_inputs() == 0);
// For textures, we try to be slightly more clever;
// if none of our outputs need a bounce, we don't bounce
// but instead simply use the effect many times.
//
// Strictly speaking, we could bounce it for some outputs
// and use it directly for others, but the processing becomes
// somewhat simpler if the effect is only used in one such way.
for (unsigned j = 0; j < dep->outgoing_links.size(); ++j) {
Node *rdep = dep->outgoing_links[j];
start_new_phase |= rdep->effect->needs_texture_bounce();
}
}
}
if (dep->effect->is_compute_shader()) {
if (phase->is_compute_shader) {
// Only one compute shader per phase.
start_new_phase = true;
} else if (!node->strong_one_to_one_sampling) {
// If all nodes so far are strong one-to-one, we can put them after
// the compute shader (ie., process them on the output).
start_new_phase = true;
} else if (!start_new_phase) {
phase->is_compute_shader = true;
phase->compute_shader_node = dep;
}
} else if (dep->effect->sets_virtual_output_size()) {
assert(dep->effect->changes_output_size());
// If the next effect sets a virtual size to rely on OpenGL's
// bilinear sampling, we'll really need to break the phase here.
start_new_phase = true;
} else if (dep->effect->changes_output_size() && !node->one_to_one_sampling) {
// If the next effect changes size and we don't have one-to-one sampling,
// we also need to break here.
start_new_phase = true;
}
if (start_new_phase) {
// Since we're starting a new phase here, we don't need to impose any
// new demands on this effect. Restore the status we had before we
// started looking at it.
dep->needs_mipmaps = save_needs_mipmaps;
phase->inputs.push_back(construct_phase(dep, completed_effects));
} else {
effects_todo_this_phase.push(dep);
// Propagate the one-to-one status down through the dependency.
dep->one_to_one_sampling = node->one_to_one_sampling &&
dep->effect->one_to_one_sampling();
dep->strong_one_to_one_sampling = node->strong_one_to_one_sampling &&
dep->effect->strong_one_to_one_sampling();
}
node->incoming_link_type.push_back(start_new_phase ? IN_ANOTHER_PHASE : IN_SAME_PHASE);
}
}
// No more effects to do this phase. Take all the ones we have,
// and create a GLSL program for it.
assert(!phase->effects.empty());
// Deduplicate the inputs, but don't change the ordering e.g. by sorting;
// that would be nondeterministic and thus reduce cacheability.
// TODO: Make this even more deterministic.
vector<Phase *> dedup_inputs;
set<Phase *> seen_inputs;
for (size_t i = 0; i < phase->inputs.size(); ++i) {
if (seen_inputs.insert(phase->inputs[i]).second) {
dedup_inputs.push_back(phase->inputs[i]);
}
}
swap(phase->inputs, dedup_inputs);
// Allocate samplers for each input.
phase->input_samplers.resize(phase->inputs.size());
// We added the effects from the output and back, but we need to output
// them in topological sort order in the shader.
phase->effects = topological_sort(phase->effects);
// Figure out if we need mipmaps or not, and if so, tell the inputs that.
// (RTT inputs have different logic, which is checked in execute_phase().)
for (unsigned i = 0; i < phase->effects.size(); ++i) {
Node *node = phase->effects[i];
if (node->effect->num_inputs() == 0) {
Input *input = static_cast<Input *>(node->effect);
assert(node->needs_mipmaps != Effect::NEEDS_MIPMAPS || input->can_supply_mipmaps());
CHECK(input->set_int("needs_mipmaps", node->needs_mipmaps == Effect::NEEDS_MIPMAPS));
}
}
// Tell each node which phase it ended up in, so that the unit test
// can check that the phases were split in the right place.
// Note that this ignores that effects may be part of multiple phases;
// if the unit tests need to test such cases, we'll reconsider.
for (unsigned i = 0; i < phase->effects.size(); ++i) {
phase->effects[i]->containing_phase = phase;
}
// Actually make the shader for this phase.
compile_glsl_program(phase);
// Initialize timers.
if (movit_timer_queries_supported) {
phase->time_elapsed_ns = 0;
phase->num_measured_iterations = 0;
}
assert(completed_effects->count(output) == 0);
completed_effects->insert(make_pair(output, phase));
phases.push_back(phase);
return phase;
}
void EffectChain::output_dot(const char *filename)
{
if (movit_debug_level != MOVIT_DEBUG_ON) {
return;
}
FILE *fp = fopen(filename, "w");
if (fp == nullptr) {
perror(filename);
exit(1);
}
fprintf(fp, "digraph G {\n");
fprintf(fp, " output [shape=box label=\"(output)\"];\n");
for (unsigned i = 0; i < nodes.size(); ++i) {
// Find out which phase this event belongs to.
vector<int> in_phases;
for (unsigned j = 0; j < phases.size(); ++j) {
const Phase* p = phases[j];
if (find(p->effects.begin(), p->effects.end(), nodes[i]) != p->effects.end()) {
in_phases.push_back(j);
}
}
if (in_phases.empty()) {
fprintf(fp, " n%ld [label=\"%s\"];\n", (long)nodes[i], nodes[i]->effect->effect_type_id().c_str());
} else if (in_phases.size() == 1) {
fprintf(fp, " n%ld [label=\"%s\" style=\"filled\" fillcolor=\"/accent8/%d\"];\n",
(long)nodes[i], nodes[i]->effect->effect_type_id().c_str(),
(in_phases[0] % 8) + 1);
} else {
// If we had new enough Graphviz, style="wedged" would probably be ideal here.
// But alas.
fprintf(fp, " n%ld [label=\"%s [in multiple phases]\" style=\"filled\" fillcolor=\"/accent8/%d\"];\n",
(long)nodes[i], nodes[i]->effect->effect_type_id().c_str(),
(in_phases[0] % 8) + 1);
}
char from_node_id[256];
snprintf(from_node_id, 256, "n%ld", (long)nodes[i]);
for (unsigned j = 0; j < nodes[i]->outgoing_links.size(); ++j) {
char to_node_id[256];
snprintf(to_node_id, 256, "n%ld", (long)nodes[i]->outgoing_links[j]);
vector<string> labels = get_labels_for_edge(nodes[i], nodes[i]->outgoing_links[j]);
output_dot_edge(fp, from_node_id, to_node_id, labels);
}
if (nodes[i]->outgoing_links.empty() && !nodes[i]->disabled) {
// Output node.
vector<string> labels = get_labels_for_edge(nodes[i], nullptr);
output_dot_edge(fp, from_node_id, "output", labels);
}
}
fprintf(fp, "}\n");
fclose(fp);
}
vector<string> EffectChain::get_labels_for_edge(const Node *from, const Node *to)
{
vector<string> labels;
if (to != nullptr && to->effect->needs_texture_bounce()) {
labels.push_back("needs_bounce");
}
if (from->effect->changes_output_size()) {
labels.push_back("resize");
}
switch (from->output_color_space) {
case COLORSPACE_INVALID:
labels.push_back("spc[invalid]");
break;
case COLORSPACE_REC_601_525:
labels.push_back("spc[rec601-525]");
break;
case COLORSPACE_REC_601_625:
labels.push_back("spc[rec601-625]");
break;
default:
break;
}
switch (from->output_gamma_curve) {
case GAMMA_INVALID:
labels.push_back("gamma[invalid]");
break;
case GAMMA_sRGB:
labels.push_back("gamma[sRGB]");
break;
case GAMMA_REC_601: // and GAMMA_REC_709
labels.push_back("gamma[rec601/709]");
break;
default:
break;
}
switch (from->output_alpha_type) {
case ALPHA_INVALID:
labels.push_back("alpha[invalid]");
break;
case ALPHA_BLANK:
labels.push_back("alpha[blank]");
break;
case ALPHA_POSTMULTIPLIED:
labels.push_back("alpha[postmult]");
break;
default:
break;
}
return labels;
}
void EffectChain::output_dot_edge(FILE *fp,
const string &from_node_id,
const string &to_node_id,
const vector<string> &labels)
{
if (labels.empty()) {
fprintf(fp, " %s -> %s;\n", from_node_id.c_str(), to_node_id.c_str());
} else {
string label = labels[0];
for (unsigned k = 1; k < labels.size(); ++k) {
label += ", " + labels[k];
}
fprintf(fp, " %s -> %s [label=\"%s\"];\n", from_node_id.c_str(), to_node_id.c_str(), label.c_str());
}
}
void EffectChain::size_rectangle_to_fit(unsigned width, unsigned height, unsigned *output_width, unsigned *output_height)
{
unsigned scaled_width, scaled_height;