-
Notifications
You must be signed in to change notification settings - Fork 30
/
rdo_bc_encoder.cpp
1270 lines (1024 loc) · 39.4 KB
/
rdo_bc_encoder.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
// rdo_bc_encoder.cpp
#include "rdo_bc_encoder.h"
#define RGBCX_IMPLEMENTATION
#include "rgbcx.h"
#define DECODE_BC4_TO_GRAYSCALE (0)
#ifdef _MSC_VER
#pragma warning (disable: 4127) // conditional expression is constant
#endif
using namespace utils;
namespace rdo_bc
{
static const char* get_dxgi_format_string(DXGI_FORMAT fmt)
{
switch (fmt)
{
case DXGI_FORMAT_BC1_UNORM: return "BC1_UNORM";
case DXGI_FORMAT_BC4_UNORM: return "BC4_UNORM";
case DXGI_FORMAT_BC3_UNORM: return "BC3_UNORM";
case DXGI_FORMAT_BC5_UNORM: return "BC5_UNORM";
case DXGI_FORMAT_BC7_UNORM: return "BC7_UNORM";
default: break;
}
return "?";
}
static std::vector<float> compute_block_mse_scales(const image_u8& source_image, uint32_t blocks_x, uint32_t blocks_y, uint32_t total_blocks, bool rdo_debug_output)
{
const float ULTRASMOOTH_BLOCK_STD_DEV_THRESHOLD = 2.9f;
const float DARK_THRESHOLD = 13.0f;
const float BRIGHT_THRESHOLD = 222.0f;
const float ULTRAMOOTH_BLOCK_MSE_SCALE = 120.0f;
const uint32_t ULTRASMOOTH_REGION_TOO_SMALL_THRESHOLD = 64;
image_u8 ultrasmooth_blocks_vis(blocks_x, blocks_y);
for (uint32_t by = 0; by < blocks_y; by++)
{
for (uint32_t bx = 0; bx < blocks_x; bx++)
{
color_quad_u8 block_pixels[16];
source_image.get_block(bx, by, 4, 4, block_pixels);
tracked_stat y_stats;
for (uint32_t y = 0; y < 4; y++)
for (uint32_t x = 0; x < 4; x++)
{
int l = block_pixels[x + y * 4].get_luma();
y_stats.update(l);
}
float max_std_dev = compute_block_max_std_dev((color_quad_u8*)block_pixels, 4, 4, 3);
float yl = max_std_dev / ULTRASMOOTH_BLOCK_STD_DEV_THRESHOLD;
yl = clamp(yl, 0.0f, 1.0f);
yl *= yl;
float y_avg = y_stats.get_mean();
if ((y_avg < DARK_THRESHOLD) || (y_avg >= BRIGHT_THRESHOLD))
yl = 1.0f;
int k = std::min<int>((int)(yl * 255.0f + .5f), 255);
ultrasmooth_blocks_vis.fill_box(bx, by, 1, 1, color_quad_u8((uint8_t)k, 255));
}
}
for (int pass = 0; pass < 1; pass++)
{
image_u8 next_vis(ultrasmooth_blocks_vis);
for (int y = 0; y < (int)blocks_y; y++)
{
for (int x = 0; x < (int)blocks_x; x++)
{
int m = 0;
for (int dy = -1; dy <= 1; dy++)
for (int dx = -1; dx <= 1; dx++)
{
if (ultrasmooth_blocks_vis.get_clamped(x + dx, y + dy).r == 255)
m = std::max<int>(m, ultrasmooth_blocks_vis.get_clamped(x + dx, y + dy).r);
}
next_vis(x, y).set((uint8_t)m, 255);
}
}
ultrasmooth_blocks_vis.swap(next_vis);
}
for (uint32_t pass = 0; pass < 32; pass++)
{
image_u8 next_vis(ultrasmooth_blocks_vis);
for (int y = 0; y < (int)blocks_y; y++)
{
for (int x = 0; x < (int)blocks_x; x++)
{
if (ultrasmooth_blocks_vis.get_clamped(x, y).r < 255)
{
int m = 0;
for (int dy = -1; dy <= 1; dy++)
for (int dx = -1; dx <= 1; dx++)
if (ultrasmooth_blocks_vis.get_clamped(x + dx, y + dy).r == 255)
m++;
if (m >= 5)
next_vis.set_pixel_clipped(x, y, color_quad_u8(255, 255, 255, 255));
}
}
}
ultrasmooth_blocks_vis.swap(next_vis);
}
image_u8 orig_ultrasmooth_blocks_vis(ultrasmooth_blocks_vis);
if (rdo_debug_output)
{
save_png("ultrasmooth_block_mask_pre_filter.png", ultrasmooth_blocks_vis, false);
}
for (uint32_t by = 0; by < blocks_y; by++)
{
for (uint32_t bx = 0; bx < blocks_x; bx++)
{
const bool is_ultrasmooth = ultrasmooth_blocks_vis(bx, by).r == 0;
if (!is_ultrasmooth)
continue;
std::vector<image_u8::pixel_coord> filled_pixels;
filled_pixels.reserve(256);
uint32_t total_set_pixels = ultrasmooth_blocks_vis.flood_fill(bx, by, color_quad_u8(255, 255, 255, 255), color_quad_u8(0, 0, 0, 255), &filled_pixels);
if (total_set_pixels < ULTRASMOOTH_REGION_TOO_SMALL_THRESHOLD)
{
for (uint32_t i = 0; i < filled_pixels.size(); i++)
orig_ultrasmooth_blocks_vis(filled_pixels[i].m_x, filled_pixels[i].m_y) = color_quad_u8(255, 255, 255, 255);
}
} // bx
} // by
ultrasmooth_blocks_vis = orig_ultrasmooth_blocks_vis;
if (rdo_debug_output)
{
save_png("ultrasmooth_block_mask.png", ultrasmooth_blocks_vis, false);
}
std::vector<float> block_mse_scales(total_blocks);
uint32_t total_ultrasmooth_blocks = 0;
for (uint32_t by = 0; by < blocks_y; by++)
{
for (uint32_t bx = 0; bx < blocks_x; bx++)
{
const bool is_ultrasmooth = ultrasmooth_blocks_vis(bx, by).r == 0;
block_mse_scales[bx + by * blocks_x] = is_ultrasmooth ? ULTRAMOOTH_BLOCK_MSE_SCALE : -1.0f;
total_ultrasmooth_blocks += is_ultrasmooth;
}
}
if (rdo_debug_output)
printf("Total ultrasmooth blocks: %3.2f%%\n", total_ultrasmooth_blocks * 100.0f / total_blocks);
return block_mse_scales;
}
rdo_bc_encoder::rdo_bc_encoder() :
m_pOrig_source_image(nullptr),
m_orig_width(0),
m_orig_height(0),
m_blocks_x(0),
m_blocks_y(0),
m_total_blocks(0),
m_bytes_per_block(0),
m_pixel_format_bpp(0),
m_total_texels(0),
m_has_alpha(false)
{
}
void rdo_bc_encoder::clear()
{
m_pOrig_source_image = nullptr;
m_source_image.clear();
m_params.clear();
m_orig_width = 0;
m_orig_height = 0;
m_blocks_x = 0;
m_blocks_y = 0;
m_total_blocks = 0;
m_bytes_per_block = 0;
m_pixel_format_bpp = 0;
m_total_texels = 0;
m_has_alpha = false;
m_packed_image8.clear();
m_packed_image16.clear();
m_prerdo_packed_image8.clear();
m_prerdo_packed_image16.clear();
m_bc7enc_pack_params.clear();
#if SUPPORT_BC7E
memset(&m_bc7e_pack_params, 0, sizeof(m_bc7e_pack_params));
#endif
}
bool rdo_bc_encoder::init(const utils::image_u8& src_image, rdo_bc_params& params)
{
clear();
m_pOrig_source_image = &src_image;
m_params = params;
init_encoders();
if (!init_source_image())
return false;
return true;
}
bool rdo_bc_encoder::encode()
{
if (!m_packed_image8.size() && !m_packed_image16.size())
return false;
if (!init_encoder_params())
return false;
if (!encode_texture())
return false;
if (!postprocess_rdo())
return false;
return true;
}
void rdo_bc_encoder::init_encoders()
{
rgbcx::init(m_params.m_bc1_mode);
bc7enc_compress_block_init();
#if SUPPORT_BC7E
ispc::bc7e_compress_block_init();
#endif
}
bool rdo_bc_encoder::init_encoder_params()
{
bc7enc_compress_block_params_init(&m_bc7enc_pack_params);
if (!m_params.m_perceptual)
bc7enc_compress_block_params_init_linear_weights(&m_bc7enc_pack_params);
m_bc7enc_pack_params.m_max_partitions = m_params.m_bc7enc_max_partitions_to_scan;
m_bc7enc_pack_params.m_uber_level = std::min(BC7ENC_MAX_UBER_LEVEL, m_params.m_bc7_uber_level);
if (m_params.m_bc7enc_mode6_only)
m_bc7enc_pack_params.m_mode_mask = 1 << 6;
if ((m_params.m_dxgi_format == DXGI_FORMAT_BC7_UNORM) && (m_params.m_rdo_lambda > 0.0f))
{
// Slam off perceptual in RDO mode - we don't support it (too slow).
m_params.m_perceptual = false;
m_bc7enc_pack_params.m_perceptual = false;
bc7enc_compress_block_params_init_linear_weights(&m_bc7enc_pack_params);
}
if ((m_params.m_dxgi_format == DXGI_FORMAT_BC7_UNORM) && (m_params.m_bc7enc_reduce_entropy))
{
// Configure the BC7 encoder with some decent parameters for later RDO post-processing.
// Textures with alpha are harder for BC7 to handle, so we use more conservative defaults.
m_bc7enc_pack_params.m_mode17_partition_estimation_filterbank = false;
if (m_params.m_bc7enc_rdo_bc7_weight_modes)
{
// Weight modes 5 and especially 6 more highly than the other modes.
if (m_has_alpha)
{
m_bc7enc_pack_params.m_mode5_error_weight = .7f;
m_bc7enc_pack_params.m_mode6_error_weight = .6f;
}
else
{
m_bc7enc_pack_params.m_mode6_error_weight = .4f;
}
}
if (m_params.m_bc7enc_rdo_bc7_weight_low_frequency_partitions)
{
// Slightly prefer the lower frequency partition patterns.
m_bc7enc_pack_params.m_low_frequency_partition_weight = .9999f;
}
if (m_params.m_bc7enc_rdo_bc7_quant_mode6_endpoints)
{
// As a good default, don't quantize mode 6 endpoints if the texture has alpha. This isn't required, but helps mask textures.
//if (!has_alpha)
m_bc7enc_pack_params.m_quant_mode6_endpoints = true;
}
if (m_params.m_bc7enc_rdo_bc7_pbit1_weighting)
{
// Favor p-bit 0 vs. 1, to slightly lower the entropy of output blocks with p-bits
m_bc7enc_pack_params.m_pbit1_weight = 1.3f;
}
}
#if SUPPORT_BC7E
// Now initialize the BC7 compressor's parameters.
memset(&m_bc7e_pack_params, 0, sizeof(m_bc7e_pack_params));
switch (m_params.m_bc7_uber_level)
{
case 0:
ispc::bc7e_compress_block_params_init_ultrafast(&m_bc7e_pack_params, m_params.m_perceptual);
break;
case 1:
ispc::bc7e_compress_block_params_init_veryfast(&m_bc7e_pack_params, m_params.m_perceptual);
break;
case 2:
ispc::bc7e_compress_block_params_init_fast(&m_bc7e_pack_params, m_params.m_perceptual);
break;
case 3:
ispc::bc7e_compress_block_params_init_basic(&m_bc7e_pack_params, m_params.m_perceptual);
break;
case 4:
ispc::bc7e_compress_block_params_init_slow(&m_bc7e_pack_params, m_params.m_perceptual);
break;
case 5:
ispc::bc7e_compress_block_params_init_veryslow(&m_bc7e_pack_params, m_params.m_perceptual);
break;
case 6:
default:
ispc::bc7e_compress_block_params_init_slowest(&m_bc7e_pack_params, m_params.m_perceptual);
break;
}
#endif
if (m_params.m_status_output)
{
if (m_params.m_dxgi_format == DXGI_FORMAT_BC7_UNORM)
{
if ((SUPPORT_BC7E) && (m_params.m_use_bc7e))
printf("bc7e.ispc uber level: %u, perceptual: %u\n", m_params.m_bc7_uber_level, m_params.m_perceptual);
else
{
printf("\nbc7enc parameters:\n");
m_bc7enc_pack_params.print();
}
}
else
{
printf("BC1 level: %u, use 3-color mode: %u, use 3-color mode for black: %u, bc1_mode: %u\n",
m_params.m_bc1_quality_level, m_params.m_use_bc1_3color_mode, m_params.m_use_bc1_3color_mode_for_black, (int)m_params.m_bc1_mode);
}
if ((m_params.m_dxgi_format == DXGI_FORMAT_BC3_UNORM) || (m_params.m_dxgi_format == DXGI_FORMAT_BC4_UNORM) || (m_params.m_dxgi_format == DXGI_FORMAT_BC5_UNORM))
{
printf("Use high quality BC4 block encoder: %u, BC4 block radius: %u, use 6 value mode: %u, use 8 value mode: %u\n",
m_params.m_use_hq_bc345, m_params.m_bc345_search_rad, (m_params.m_bc345_mode_mask & 2) != 0, (m_params.m_bc345_mode_mask & 1) != 0);
}
printf("\nrdo_bc_params:\n");
printf(" Perceptual: %u\n", m_params.m_perceptual);
printf(" Y Flip: %u\n", m_params.m_y_flip);
printf(" DXGI format: 0x%X %s\n", m_params.m_dxgi_format, get_dxgi_format_string(m_params.m_dxgi_format));
printf("BC1-5 parameters:\n");
printf(" BC45 channels: %u %u\n", m_params.m_bc45_channel0, m_params.m_bc45_channel1);
printf(" BC1 approximation mode: %u\n", (int)m_params.m_bc1_mode);
printf(" Use BC1 3-color mode: %u\n", m_params.m_use_bc1_3color_mode);
printf(" Use BC1 3-color mode for black: %u\n", m_params.m_use_bc1_3color_mode_for_black);
printf(" BC1 quality level: %u\n", m_params.m_bc1_quality_level);
printf(" Use HQ BC345: %u\n", m_params.m_use_hq_bc345);
printf(" BC345 search radius: %u\n", m_params.m_bc345_search_rad);
printf(" BC345 mode mask: 0x%X\n", m_params.m_bc345_mode_mask);
printf("BC7 parameters:\n");
printf(" Use bc7e: %u\n", m_params.m_use_bc7e);
printf(" BC7 uber level: %u\n", m_params.m_bc7_uber_level);
printf("RDO parameters:\n");
printf(" Lambda: %f\n", m_params.m_rdo_lambda);
printf(" Lookback window size: %u\n", m_params.m_lookback_window_size);
printf(" Custom lookback window size: %u\n", m_params.m_custom_lookback_window_size);
printf(" Try 2 matches: %u\n", m_params.m_rdo_try_2_matches);
printf(" Smooth block error scale: %f\n", m_params.m_rdo_smooth_block_error_scale);
printf(" Custom RDO smooth block error scale: %u\n", m_params.m_custom_rdo_smooth_block_error_scale);
printf(" Max smooth block std dev: %f\n", m_params.m_rdo_max_smooth_block_std_dev);
printf(" Allow relative movement: %u\n", m_params.m_rdo_allow_relative_movement);
printf(" Ultrasmooth block handling: %u\n", m_params.m_rdo_ultrasmooth_block_handling);
printf(" Multithreading: %u, max threads: %u\n", m_params.m_rdo_multithreading, m_params.m_rdo_max_threads);
printf("bc7enc parameters:\n");
printf(" Mode 6 only: %u\n", m_params.m_bc7enc_mode6_only);
printf(" Max partitions to scan: %u\n", m_params.m_bc7enc_max_partitions_to_scan);
printf(" Quant mode 6 endpoints: %u\n", m_params.m_bc7enc_rdo_bc7_quant_mode6_endpoints);
printf(" Weight modes: %u\n", m_params.m_bc7enc_rdo_bc7_weight_modes);
printf(" Weight low freq partitions: %u\n", m_params.m_bc7enc_rdo_bc7_weight_low_frequency_partitions);
printf(" P-bit1 weighting: %u\n", m_params.m_bc7enc_rdo_bc7_pbit1_weighting);
printf(" Reduce entropy mode: %u\n", m_params.m_bc7enc_reduce_entropy);
printf("\n");
}
return true;
}
bool rdo_bc_encoder::init_source_image()
{
switch (m_params.m_dxgi_format)
{
case DXGI_FORMAT_BC1_UNORM:
case DXGI_FORMAT_BC4_UNORM:
m_pixel_format_bpp = 4;
break;
case DXGI_FORMAT_BC3_UNORM:
case DXGI_FORMAT_BC5_UNORM:
case DXGI_FORMAT_BC7_UNORM:
m_pixel_format_bpp = 8;
break;
default:
return false;
}
m_bytes_per_block = (16 * m_pixel_format_bpp) / 8;
assert((m_bytes_per_block == 8) || (m_bytes_per_block == 16));
m_source_image = *m_pOrig_source_image;
m_orig_width = m_source_image.width();
m_orig_height = m_source_image.height();
if (m_params.m_y_flip)
{
utils::image_u8 temp;
temp.init(m_orig_width, m_orig_height);
for (uint32_t y = 0; y < m_orig_height; y++)
for (uint32_t x = 0; x < m_orig_width; x++)
temp(x, (m_orig_height - 1) - y) = m_source_image(x, y);
temp.swap(m_source_image);
}
m_source_image.crop_dup_borders((m_source_image.width() + 3) & ~3, (m_source_image.height() + 3) & ~3);
m_blocks_x = m_source_image.width() / 4;
m_blocks_y = m_source_image.height() / 4;
m_total_blocks = m_blocks_x * m_blocks_y;
m_total_texels = m_total_blocks * 16;
bool has_alpha = false;
for (int by = 0; by < ((int)m_blocks_y) && !has_alpha; by++)
{
for (uint32_t bx = 0; bx < m_blocks_x; bx++)
{
color_quad_u8 pixels[16];
m_source_image.get_block(bx, by, 4, 4, pixels);
for (uint32_t i = 0; i < 16; i++)
{
if (pixels[i].m_c[3] < 255)
{
has_alpha = true;
break;
}
}
}
}
if (m_pixel_format_bpp == 8)
m_packed_image16.resize(m_total_blocks);
else
m_packed_image8.resize(m_total_blocks);
return true;
}
bool rdo_bc_encoder::encode_texture()
{
clock_t start_t = clock();
uint32_t bc7_mode_hist[8];
memset(bc7_mode_hist, 0, sizeof(bc7_mode_hist));
#if SUPPORT_BC7E
if ((m_params.m_dxgi_format == DXGI_FORMAT_BC7_UNORM) && (m_params.m_use_bc7e))
{
if (m_params.m_status_output)
printf("Using bc7e: ");
#pragma omp parallel for
for (int32_t by = 0; by < static_cast<int32_t>(m_blocks_y); by++)
{
// Process 64 blocks at a time, for efficient SIMD processing.
// Ideally, N >= 8 (or more) and (N % 8) == 0.
const int N = 64;
for (uint32_t bx = 0; bx < m_blocks_x; bx += N)
{
const uint32_t num_blocks_to_process = std::min<uint32_t>(m_blocks_x - bx, N);
color_quad_u8 pixels[16 * N];
// Extract num_blocks_to_process 4x4 pixel blocks from the source image and put them into the pixels[] array.
for (uint32_t b = 0; b < num_blocks_to_process; b++)
m_source_image.get_block(bx + b, by, 4, 4, pixels + b * 16);
// Compress the blocks to BC7.
// Note: If you've used Intel's ispc_texcomp, the input pixels are different. BC7E requires a pointer to an array of 16 pixels for each block.
block16* pBlock = &m_packed_image16[bx + by * m_blocks_x];
ispc::bc7e_compress_blocks(num_blocks_to_process, reinterpret_cast<uint64_t*>(pBlock), reinterpret_cast<const uint32_t*>(pixels), &m_bc7e_pack_params);
}
if (m_params.m_status_output)
{
if ((by & 63) == 0)
printf(".");
}
}
for (int by = 0; by < (int)m_blocks_y; by++)
{
for (uint32_t bx = 0; bx < m_blocks_x; bx++)
{
block16* pBlock = &m_packed_image16[bx + by * m_blocks_x];
uint32_t mode = ((uint8_t*)pBlock)[0];
for (uint32_t m = 0; m <= 7; m++)
{
if (mode & (1 << m))
{
bc7_mode_hist[m]++;
break;
}
}
}
}
}
else
#endif
{
#pragma omp parallel for
for (int by = 0; by < (int)m_blocks_y; by++)
{
for (uint32_t bx = 0; bx < m_blocks_x; bx++)
{
color_quad_u8 pixels[16];
m_source_image.get_block(bx, by, 4, 4, pixels);
switch (m_params.m_dxgi_format)
{
case DXGI_FORMAT_BC1_UNORM:
{
block8* pBlock = &m_packed_image8[bx + by * m_blocks_x];
rgbcx::encode_bc1(m_params.m_bc1_quality_level, pBlock, &pixels[0].m_c[0], m_params.m_use_bc1_3color_mode, m_params.m_use_bc1_3color_mode_for_black);
break;
}
case DXGI_FORMAT_BC3_UNORM:
{
block16* pBlock = &m_packed_image16[bx + by * m_blocks_x];
if (m_params.m_use_hq_bc345)
rgbcx::encode_bc3_hq(m_params.m_bc1_quality_level, pBlock, &pixels[0].m_c[0], m_params.m_bc345_search_rad, m_params.m_bc345_mode_mask);
else
rgbcx::encode_bc3(m_params.m_bc1_quality_level, pBlock, &pixels[0].m_c[0]);
break;
}
case DXGI_FORMAT_BC4_UNORM:
{
block8* pBlock = &m_packed_image8[bx + by * m_blocks_x];
if (m_params.m_use_hq_bc345)
rgbcx::encode_bc4_hq(pBlock, &pixels[0].m_c[m_params.m_bc45_channel0], 4, m_params.m_bc345_search_rad, m_params.m_bc345_mode_mask);
else
rgbcx::encode_bc4(pBlock, &pixels[0].m_c[m_params.m_bc45_channel0], 4);
break;
}
case DXGI_FORMAT_BC5_UNORM:
{
block16* pBlock = &m_packed_image16[bx + by * m_blocks_x];
if (m_params.m_use_hq_bc345)
rgbcx::encode_bc5_hq(pBlock, &pixels[0].m_c[0], m_params.m_bc45_channel0, m_params.m_bc45_channel1, 4, m_params.m_bc345_search_rad, m_params.m_bc345_mode_mask);
else
rgbcx::encode_bc5(pBlock, &pixels[0].m_c[0], m_params.m_bc45_channel0, m_params.m_bc45_channel1, 4);
break;
}
case DXGI_FORMAT_BC7_UNORM:
{
block16* pBlock = &m_packed_image16[bx + by * m_blocks_x];
bc7enc_compress_block(pBlock, pixels, &m_bc7enc_pack_params);
#pragma omp critical
{
uint32_t mode = ((uint8_t*)pBlock)[0];
for (uint32_t m = 0; m <= 7; m++)
{
if (mode & (1 << m))
{
bc7_mode_hist[m]++;
break;
}
}
}
break;
}
default:
{
assert(0);
break;
}
}
}
if (m_params.m_status_output)
{
if ((by & 127) == 0)
printf(".");
}
}
}
clock_t end_t = clock();
if (m_params.m_status_output)
{
printf("\nTotal encoding time: %f secs\n", (double)(end_t - start_t) / CLOCKS_PER_SEC);
if (m_params.m_dxgi_format == DXGI_FORMAT_BC7_UNORM)
{
printf("BC7 mode histogram:\n");
for (uint32_t i = 0; i < 8; i++)
printf("%u: %u\n", i, bc7_mode_hist[i]);
}
}
return true;
}
bool rdo_bc_encoder::postprocess_rdo()
{
m_prerdo_packed_image8 = m_packed_image8;
m_prerdo_packed_image16 = m_packed_image16;
// Post-process the data with Rate Distortion Optimization
if (m_params.m_rdo_lambda <= 0.0f)
return true;
const uint32_t MIN_RDO_MULTITHREADING_BLOCKS = 4096;
const int rdo_total_threads = (m_params.m_rdo_multithreading && (m_params.m_rdo_max_threads > 1) && (m_total_blocks >= MIN_RDO_MULTITHREADING_BLOCKS)) ? m_params.m_rdo_max_threads : 1;
if (m_params.m_status_output)
printf("rdo_total_threads: %u\n", rdo_total_threads);
int blocks_remaining = m_total_blocks, cur_block_index = 0;
std::vector<int> blocks_to_do(rdo_total_threads), first_block_index(rdo_total_threads);
for (int p = 0; p < rdo_total_threads; p++)
{
const int num_blocks = (p == (rdo_total_threads - 1)) ? blocks_remaining : (m_total_blocks / rdo_total_threads);
blocks_to_do[p] = num_blocks;
first_block_index[p] = cur_block_index;
cur_block_index += num_blocks;
blocks_remaining -= num_blocks;
}
assert(!blocks_remaining && cur_block_index == (int)m_total_blocks);
ert::reduce_entropy_params ert_p;
ert_p.m_lambda = m_params.m_rdo_lambda;
ert_p.m_lookback_window_size = m_params.m_lookback_window_size;
ert_p.m_smooth_block_max_mse_scale = m_params.m_rdo_smooth_block_error_scale;
ert_p.m_max_smooth_block_std_dev = m_params.m_rdo_max_smooth_block_std_dev;
ert_p.m_debug_output = m_params.m_rdo_debug_output;
ert_p.m_try_two_matches = m_params.m_rdo_try_2_matches;
ert_p.m_allow_relative_movement = m_params.m_rdo_allow_relative_movement;
ert_p.m_skip_zero_mse_blocks = false;
std::vector<float> block_rgb_mse_scales(compute_block_mse_scales(m_source_image, m_blocks_x, m_blocks_y, m_total_blocks, m_params.m_rdo_debug_output));
std::vector<rgbcx::color32> block_pixels(m_total_blocks * 16);
for (uint32_t by = 0; by < m_blocks_y; by++)
for (uint32_t bx = 0; bx < m_blocks_x; bx++)
m_source_image.get_block(bx, by, 4, 4, (color_quad_u8*)&block_pixels[(bx + by * m_blocks_x) * 16]);
unpacker_funcs block_unpackers;
block_unpackers.m_allow_3color_mode = m_params.m_use_bc1_3color_mode;
block_unpackers.m_use_bc1_3color_mode_for_black = m_params.m_use_bc1_3color_mode_for_black;
block_unpackers.m_mode = m_params.m_bc1_mode;
if (m_params.m_dxgi_format == DXGI_FORMAT_BC7_UNORM)
{
ert_p.m_lookback_window_size = std::max(16U, m_params.m_lookback_window_size);
// BC7 RDO
const uint32_t NUM_COMPONENTS = 4;
if (!m_params.m_custom_rdo_smooth_block_error_scale)
{
// Attempt to compute a decent conservative smooth block MSE max scaling factor.
// No single smooth block scale setting can work for all textures (unless it's ridiuclously large, killing efficiency).
ert_p.m_smooth_block_max_mse_scale = lerp(15.0f, 50.0f, std::min(1.0f, ert_p.m_lambda / 4.0f));
if (m_params.m_status_output)
printf("Using an automatically computed smooth block error scale of %f (use -zb# to override)\n", ert_p.m_smooth_block_max_mse_scale);
}
for (uint32_t by = 0; by < m_blocks_y; by++)
for (uint32_t bx = 0; bx < m_blocks_x; bx++)
{
float& s = block_rgb_mse_scales[bx + by * m_blocks_x];
if (s > 0.0f)
s = std::max(ert_p.m_smooth_block_max_mse_scale, s * std::min(ert_p.m_lambda, 3.0f));
}
if (m_params.m_status_output)
{
printf("\nERT parameters:\n");
ert_p.print();
printf("\n");
}
uint32_t total_modified = 0;
clock_t rdo_start_t = clock();
#pragma omp parallel for
for (int p = 0; p < rdo_total_threads; p++)
{
const int first_block_to_encode = first_block_index[p];
const int num_blocks_to_encode = blocks_to_do[p];
if (!num_blocks_to_encode)
continue;
uint32_t total_modified_local = 0;
std::vector<float> local_block_rgb_mse_scales(num_blocks_to_encode);
for (int i = 0; i < num_blocks_to_encode; i++)
local_block_rgb_mse_scales[i] = block_rgb_mse_scales[first_block_to_encode + i];
ert::reduce_entropy(&m_packed_image16[first_block_to_encode], num_blocks_to_encode,
16, 16, 4, 4, NUM_COMPONENTS,
(ert::color_rgba*)&block_pixels[16 * first_block_to_encode], ert_p, total_modified_local,
unpacker_funcs::unpack_bc7_block, &block_unpackers,
m_params.m_rdo_ultrasmooth_block_handling ? &local_block_rgb_mse_scales : nullptr);
#pragma omp critical
{
total_modified += total_modified_local;
}
} // p
clock_t rdo_end_t = clock();
if (m_params.m_status_output)
{
printf("Total RDO time: %f secs\n", (double)(rdo_end_t - rdo_start_t) / CLOCKS_PER_SEC);
printf("Total blocks modified: %u %3.2f%%\n", total_modified, total_modified * 100.0f / m_total_blocks);
uint32_t bc7_mode_hist[8];
memset(bc7_mode_hist, 0, sizeof(bc7_mode_hist));
for (int by = 0; by < (int)m_blocks_y; by++)
{
for (uint32_t bx = 0; bx < m_blocks_x; bx++)
{
block16* pBlock = &m_packed_image16[bx + by * m_blocks_x];
const uint32_t mode_byte = ((uint8_t*)pBlock)[0];
uint32_t m;
for (m = 0; m <= 7; m++)
{
if (mode_byte & (1 << m))
{
bc7_mode_hist[m]++;
break;
}
}
assert(m != 8);
}
}
printf("BC7 mode histogram:\n");
for (uint32_t i = 0; i < 8; i++)
printf("%u: %u\n", i, bc7_mode_hist[i]);
}
}
else if (m_params.m_dxgi_format == DXGI_FORMAT_BC5_UNORM)
{
// BC5 RDO - One BC4 block for R followed by one BC4 block for G
ert_p.m_lookback_window_size = std::max(16U, m_params.m_lookback_window_size);
std::vector<rgbcx::color32> block_pixels_r(m_total_blocks * 16), block_pixels_g(m_total_blocks * 16);
for (uint32_t by = 0; by < m_blocks_y; by++)
{
for (uint32_t bx = 0; bx < m_blocks_x; bx++)
{
color_quad_u8 orig_block[16];
m_source_image.get_block(bx, by, 4, 4, orig_block);
color_quad_u8* pDst_block_r = (color_quad_u8*)&block_pixels_r[(bx + by * m_blocks_x) * 16];
color_quad_u8* pDst_block_g = (color_quad_u8*)&block_pixels_g[(bx + by * m_blocks_x) * 16];
for (uint32_t i = 0; i < 16; i++)
{
pDst_block_r[i].set(orig_block[i].r, 0, 0, 0);
pDst_block_g[i].set(orig_block[i].g, 0, 0, 0);
}
}
}
const uint32_t NUM_COMPONENTS = 1;
ert_p.m_color_weights[1] = 0;
ert_p.m_color_weights[2] = 0;
ert_p.m_color_weights[3] = 0;
if (!m_params.m_custom_rdo_smooth_block_error_scale)
{
// Attempt to compute a decent conservative smooth block MSE max scaling factor.
// No single smooth block scale setting can work for all textures (unless it's ridiuclously large, killing efficiency).
ert_p.m_smooth_block_max_mse_scale = lerp(10.0f, 30.0f, std::min(1.0f, ert_p.m_lambda / 4.0f));
if (m_params.m_status_output)
printf("Using an automatically computed smooth block error scale of %f (use -zb# to override)\n", ert_p.m_smooth_block_max_mse_scale);
}
if (m_params.m_status_output)
{
printf("\nERT parameters:\n");
ert_p.print();
printf("\n");
}
uint32_t total_modified_r = 0, total_modified_g = 0;
clock_t rdo_start_t = clock();
#pragma omp parallel for
for (int p = 0; p < rdo_total_threads; p++)
{
const int first_block_to_encode = first_block_index[p];
const int num_blocks_to_encode = blocks_to_do[p];
if (!num_blocks_to_encode)
continue;
uint32_t total_modified_local_r = 0, total_modified_local_g = 0;
ert::reduce_entropy(&m_packed_image16[first_block_to_encode], num_blocks_to_encode,
2 * sizeof(rgbcx::bc4_block), sizeof(rgbcx::bc4_block), 4, 4, NUM_COMPONENTS,
(ert::color_rgba*)&block_pixels_r[16 * first_block_to_encode], ert_p, total_modified_local_r,
unpacker_funcs::unpack_bc4_block, &block_unpackers);
ert::reduce_entropy((uint8_t*)&m_packed_image16[first_block_to_encode] + sizeof(rgbcx::bc4_block), num_blocks_to_encode,
2 * sizeof(rgbcx::bc4_block), sizeof(rgbcx::bc4_block), 4, 4, NUM_COMPONENTS,
(ert::color_rgba*)&block_pixels_g[16 * first_block_to_encode], ert_p, total_modified_local_g,
unpacker_funcs::unpack_bc4_block, &block_unpackers);
#pragma omp critical
{
total_modified_r += total_modified_local_r;
total_modified_g += total_modified_local_g;
}
} // p
clock_t rdo_end_t = clock();
if (m_params.m_status_output)
{
printf("Total RDO time: %f secs\n", (double)(rdo_end_t - rdo_start_t) / CLOCKS_PER_SEC);
printf("Total blocks modified R: %u %3.2f%%\n", total_modified_r, total_modified_r * 100.0f / m_total_blocks);
printf("Total blocks modified G: %u %3.2f%%\n", total_modified_g, total_modified_g * 100.0f / m_total_blocks);
}
}
else if (m_params.m_dxgi_format == DXGI_FORMAT_BC4_UNORM)
{
// BC4 RDO - One BC4 block for R
const uint32_t NUM_COMPONENTS = 1;
ert_p.m_color_weights[1] = 0;
ert_p.m_color_weights[2] = 0;
ert_p.m_color_weights[3] = 0;
if (!m_params.m_custom_rdo_smooth_block_error_scale)
{
// Attempt to compute a decent conservative smooth block MSE max scaling factor.
// No single smooth block scale setting can work for all textures (unless it's ridiuclously large, killing efficiency).
ert_p.m_smooth_block_max_mse_scale = lerp(10.0f, 30.0f, std::min(1.0f, ert_p.m_lambda / 4.0f));
if (m_params.m_status_output)
printf("Using an automatically computed smooth block error scale of %f (use -zb# to override)\n", ert_p.m_smooth_block_max_mse_scale);
}
if (m_params.m_status_output)
{
printf("\nERT parameters:\n");
ert_p.print();
printf("\n");
}
uint32_t total_modified = 0;
clock_t rdo_start_t = clock();
#pragma omp parallel for
for (int p = 0; p < rdo_total_threads; p++)
{
const int first_block_to_encode = first_block_index[p];
const int num_blocks_to_encode = blocks_to_do[p];
if (!num_blocks_to_encode)
continue;
uint32_t total_modified_local = 0;
ert::reduce_entropy(&m_packed_image8[first_block_to_encode], num_blocks_to_encode,
sizeof(rgbcx::bc4_block), sizeof(rgbcx::bc4_block), 4, 4, NUM_COMPONENTS,
(ert::color_rgba*)&block_pixels[16 * first_block_to_encode], ert_p, total_modified_local,
unpacker_funcs::unpack_bc4_block, &block_unpackers);
#pragma omp critical
{
total_modified += total_modified_local;
}
} // p
clock_t rdo_end_t = clock();
if (m_params.m_status_output)
{
printf("Total RDO time: %f secs\n", (double)(rdo_end_t - rdo_start_t) / CLOCKS_PER_SEC);
printf("Total blocks modified: %u %3.2f%%\n", total_modified, total_modified * 100.0f / m_total_blocks);
}
}
else if (m_params.m_dxgi_format == DXGI_FORMAT_BC1_UNORM)
{
// BC1 RDO - One BC1 block
const uint32_t NUM_COMPONENTS = 3;
ert_p.m_color_weights[3] = 0;
if (!m_params.m_custom_rdo_smooth_block_error_scale)
{
// This is just a hack - no single setting can work for all textures.
ert_p.m_smooth_block_max_mse_scale = lerp(15.0f, 50.0f, std::min(1.0f, ert_p.m_lambda / 8.0f));
if (m_params.m_status_output)
printf("Using an automatically computed smooth block error scale of %f (use -zb# to override)\n", ert_p.m_smooth_block_max_mse_scale);
}
for (uint32_t by = 0; by < m_blocks_y; by++)
for (uint32_t bx = 0; bx < m_blocks_x; bx++)
{
float& s = block_rgb_mse_scales[bx + by * m_blocks_x];
if (s > 0.0f)
s = std::max(ert_p.m_smooth_block_max_mse_scale, s * std::min(ert_p.m_lambda, 3.0f));
}
printf("\nERT parameters:\n");
ert_p.print();
printf("\n");
uint32_t total_modified = 0;
clock_t rdo_start_t = clock();
#pragma omp parallel for
for (int p = 0; p < rdo_total_threads; p++)
{
const int first_block_to_encode = first_block_index[p];