1 | // SPDX-License-Identifier: Apache-2.0 |
2 | // ---------------------------------------------------------------------------- |
3 | // Copyright 2011-2021 Arm Limited |
4 | // |
5 | // Licensed under the Apache License, Version 2.0 (the "License"); you may not |
6 | // use this file except in compliance with the License. You may obtain a copy |
7 | // of the License at: |
8 | // |
9 | // http://www.apache.org/licenses/LICENSE-2.0 |
10 | // |
11 | // Unless required by applicable law or agreed to in writing, software |
12 | // distributed under the License is distributed on an "AS IS" BASIS, WITHOUT |
13 | // WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the |
14 | // License for the specific language governing permissions and limitations |
15 | // under the License. |
16 | // ---------------------------------------------------------------------------- |
17 | |
18 | #include <utility> |
19 | |
20 | /** |
21 | * @brief Functions for color unquantization. |
22 | */ |
23 | |
24 | #include "astcenc_internal.h" |
25 | |
26 | /** |
27 | * @brief Un-blue-contract a color. |
28 | * |
29 | * This function reverses any applied blue contraction. |
30 | * |
31 | * @param input The input color that has been blue-contracted. |
32 | * |
33 | * @return The uncontracted color. |
34 | */ |
35 | static ASTCENC_SIMD_INLINE vint4 uncontract_color( |
36 | vint4 input |
37 | ) { |
38 | vmask4 mask(true, true, false, false); |
39 | vint4 bc0 = asr<1>(input + input.lane<2>()); |
40 | return select(input, bc0, mask); |
41 | } |
42 | |
43 | /** |
44 | * @brief Unpack an LDR RGBA color that uses delta encoding. |
45 | * |
46 | * @param input0 The packed endpoint 0 color. |
47 | * @param input1 The packed endpoint 1 color deltas. |
48 | * @param[out] output0 The unpacked endpoint 0 color. |
49 | * @param[out] output1 The unpacked endpoint 1 color. |
50 | */ |
51 | static void rgba_delta_unpack( |
52 | vint4 input0, |
53 | vint4 input1, |
54 | vint4& output0, |
55 | vint4& output1 |
56 | ) { |
57 | // Apply bit transfer |
58 | bit_transfer_signed(input1, input0); |
59 | |
60 | // Apply blue-uncontraction if needed |
61 | int rgb_sum = hadd_rgb_s(input1); |
62 | input1 = input1 + input0; |
63 | if (rgb_sum < 0) |
64 | { |
65 | input0 = uncontract_color(input0); |
66 | input1 = uncontract_color(input1); |
67 | std::swap(input0, input1); |
68 | } |
69 | |
70 | output0 = clamp(0, 255, input0); |
71 | output1 = clamp(0, 255, input1); |
72 | } |
73 | |
74 | /** |
75 | * @brief Unpack an LDR RGB color that uses delta encoding. |
76 | * |
77 | * Output alpha set to 255. |
78 | * |
79 | * @param input0 The packed endpoint 0 color. |
80 | * @param input1 The packed endpoint 1 color deltas. |
81 | * @param[out] output0 The unpacked endpoint 0 color. |
82 | * @param[out] output1 The unpacked endpoint 1 color. |
83 | */ |
84 | static void rgb_delta_unpack( |
85 | vint4 input0, |
86 | vint4 input1, |
87 | vint4& output0, |
88 | vint4& output1 |
89 | ) { |
90 | rgba_delta_unpack(input0, input1, output0, output1); |
91 | output0.set_lane<3>(255); |
92 | output1.set_lane<3>(255); |
93 | } |
94 | |
95 | /** |
96 | * @brief Unpack an LDR RGBA color that uses direct encoding. |
97 | * |
98 | * @param input0 The packed endpoint 0 color. |
99 | * @param input1 The packed endpoint 1 color. |
100 | * @param[out] output0 The unpacked endpoint 0 color. |
101 | * @param[out] output1 The unpacked endpoint 1 color. |
102 | */ |
103 | static void rgba_unpack( |
104 | vint4 input0, |
105 | vint4 input1, |
106 | vint4& output0, |
107 | vint4& output1 |
108 | ) { |
109 | // Apply blue-uncontraction if needed |
110 | if (hadd_rgb_s(input0) > hadd_rgb_s(input1)) |
111 | { |
112 | input0 = uncontract_color(input0); |
113 | input1 = uncontract_color(input1); |
114 | std::swap(input0, input1); |
115 | } |
116 | |
117 | output0 = input0; |
118 | output1 = input1; |
119 | } |
120 | |
121 | /** |
122 | * @brief Unpack an LDR RGB color that uses direct encoding. |
123 | * |
124 | * Output alpha set to 255. |
125 | * |
126 | * @param input0 The packed endpoint 0 color. |
127 | * @param input1 The packed endpoint 1 color. |
128 | * @param[out] output0 The unpacked endpoint 0 color. |
129 | * @param[out] output1 The unpacked endpoint 1 color. |
130 | */ |
131 | static void rgb_unpack( |
132 | vint4 input0, |
133 | vint4 input1, |
134 | vint4& output0, |
135 | vint4& output1 |
136 | ) { |
137 | rgba_unpack(input0, input1, output0, output1); |
138 | output0.set_lane<3>(255); |
139 | output1.set_lane<3>(255); |
140 | } |
141 | |
142 | /** |
143 | * @brief Unpack an LDR RGBA color that uses scaled encoding. |
144 | * |
145 | * Note only the RGB channels use the scaled encoding, alpha uses direct. |
146 | * |
147 | * @param input0 The packed endpoint 0 color. |
148 | * @param alpha1 The packed endpoint 1 alpha value. |
149 | * @param scale The packed quantized scale. |
150 | * @param[out] output0 The unpacked endpoint 0 color. |
151 | * @param[out] output1 The unpacked endpoint 1 color. |
152 | */ |
153 | static void rgb_scale_alpha_unpack( |
154 | vint4 input0, |
155 | uint8_t alpha1, |
156 | uint8_t scale, |
157 | vint4& output0, |
158 | vint4& output1 |
159 | ) { |
160 | output1 = input0; |
161 | output1.set_lane<3>(alpha1); |
162 | |
163 | output0 = asr<8>(input0 * scale); |
164 | output0.set_lane<3>(input0.lane<3>()); |
165 | } |
166 | |
167 | /** |
168 | * @brief Unpack an LDR RGB color that uses scaled encoding. |
169 | * |
170 | * Output alpha is 255. |
171 | * |
172 | * @param input0 The packed endpoint 0 color. |
173 | * @param scale The packed scale. |
174 | * @param[out] output0 The unpacked endpoint 0 color. |
175 | * @param[out] output1 The unpacked endpoint 1 color. |
176 | */ |
177 | static void rgb_scale_unpack( |
178 | vint4 input0, |
179 | int scale, |
180 | vint4& output0, |
181 | vint4& output1 |
182 | ) { |
183 | output1 = input0; |
184 | output1.set_lane<3>(255); |
185 | |
186 | output0 = asr<8>(input0 * scale); |
187 | output0.set_lane<3>(255); |
188 | } |
189 | |
190 | /** |
191 | * @brief Unpack an LDR L color that uses direct encoding. |
192 | * |
193 | * Output alpha is 255. |
194 | * |
195 | * @param input The packed endpoints. |
196 | * @param[out] output0 The unpacked endpoint 0 color. |
197 | * @param[out] output1 The unpacked endpoint 1 color. |
198 | */ |
199 | static void luminance_unpack( |
200 | const uint8_t input[2], |
201 | vint4& output0, |
202 | vint4& output1 |
203 | ) { |
204 | int lum0 = input[0]; |
205 | int lum1 = input[1]; |
206 | output0 = vint4(lum0, lum0, lum0, 255); |
207 | output1 = vint4(lum1, lum1, lum1, 255); |
208 | } |
209 | |
210 | /** |
211 | * @brief Unpack an LDR L color that uses delta encoding. |
212 | * |
213 | * Output alpha is 255. |
214 | * |
215 | * @param input The packed endpoints (L0, L1). |
216 | * @param[out] output0 The unpacked endpoint 0 color. |
217 | * @param[out] output1 The unpacked endpoint 1 color. |
218 | */ |
219 | static void luminance_delta_unpack( |
220 | const uint8_t input[2], |
221 | vint4& output0, |
222 | vint4& output1 |
223 | ) { |
224 | int v0 = input[0]; |
225 | int v1 = input[1]; |
226 | int l0 = (v0 >> 2) | (v1 & 0xC0); |
227 | int l1 = l0 + (v1 & 0x3F); |
228 | |
229 | l1 = astc::min(l1, 255); |
230 | |
231 | output0 = vint4(l0, l0, l0, 255); |
232 | output1 = vint4(l1, l1, l1, 255); |
233 | } |
234 | |
235 | /** |
236 | * @brief Unpack an LDR LA color that uses direct encoding. |
237 | * |
238 | * @param input The packed endpoints (L0, L1, A0, A1). |
239 | * @param[out] output0 The unpacked endpoint 0 color. |
240 | * @param[out] output1 The unpacked endpoint 1 color. |
241 | */ |
242 | static void luminance_alpha_unpack( |
243 | const uint8_t input[4], |
244 | vint4& output0, |
245 | vint4& output1 |
246 | ) { |
247 | int lum0 = input[0]; |
248 | int lum1 = input[1]; |
249 | int alpha0 = input[2]; |
250 | int alpha1 = input[3]; |
251 | output0 = vint4(lum0, lum0, lum0, alpha0); |
252 | output1 = vint4(lum1, lum1, lum1, alpha1); |
253 | } |
254 | |
255 | /** |
256 | * @brief Unpack an LDR LA color that uses delta encoding. |
257 | * |
258 | * @param input The packed endpoints (L0, L1, A0, A1). |
259 | * @param[out] output0 The unpacked endpoint 0 color. |
260 | * @param[out] output1 The unpacked endpoint 1 color. |
261 | */ |
262 | static void luminance_alpha_delta_unpack( |
263 | const uint8_t input[4], |
264 | vint4& output0, |
265 | vint4& output1 |
266 | ) { |
267 | int lum0 = input[0]; |
268 | int lum1 = input[1]; |
269 | int alpha0 = input[2]; |
270 | int alpha1 = input[3]; |
271 | |
272 | lum0 |= (lum1 & 0x80) << 1; |
273 | alpha0 |= (alpha1 & 0x80) << 1; |
274 | lum1 &= 0x7F; |
275 | alpha1 &= 0x7F; |
276 | |
277 | if (lum1 & 0x40) |
278 | { |
279 | lum1 -= 0x80; |
280 | } |
281 | |
282 | if (alpha1 & 0x40) |
283 | { |
284 | alpha1 -= 0x80; |
285 | } |
286 | |
287 | lum0 >>= 1; |
288 | lum1 >>= 1; |
289 | alpha0 >>= 1; |
290 | alpha1 >>= 1; |
291 | lum1 += lum0; |
292 | alpha1 += alpha0; |
293 | |
294 | lum1 = astc::clamp(lum1, 0, 255); |
295 | alpha1 = astc::clamp(alpha1, 0, 255); |
296 | |
297 | output0 = vint4(lum0, lum0, lum0, alpha0); |
298 | output1 = vint4(lum1, lum1, lum1, alpha1); |
299 | } |
300 | |
301 | /** |
302 | * @brief Unpack an HDR RGB + offset encoding. |
303 | * |
304 | * @param input The packed endpoints (packed and modal). |
305 | * @param[out] output0 The unpacked endpoint 0 color. |
306 | * @param[out] output1 The unpacked endpoint 1 color. |
307 | */ |
308 | static void hdr_rgbo_unpack( |
309 | const uint8_t input[4], |
310 | vint4& output0, |
311 | vint4& output1 |
312 | ) { |
313 | int v0 = input[0]; |
314 | int v1 = input[1]; |
315 | int v2 = input[2]; |
316 | int v3 = input[3]; |
317 | |
318 | int modeval = ((v0 & 0xC0) >> 6) | (((v1 & 0x80) >> 7) << 2) | (((v2 & 0x80) >> 7) << 3); |
319 | |
320 | int majcomp; |
321 | int mode; |
322 | if ((modeval & 0xC) != 0xC) |
323 | { |
324 | majcomp = modeval >> 2; |
325 | mode = modeval & 3; |
326 | } |
327 | else if (modeval != 0xF) |
328 | { |
329 | majcomp = modeval & 3; |
330 | mode = 4; |
331 | } |
332 | else |
333 | { |
334 | majcomp = 0; |
335 | mode = 5; |
336 | } |
337 | |
338 | int red = v0 & 0x3F; |
339 | int green = v1 & 0x1F; |
340 | int blue = v2 & 0x1F; |
341 | int scale = v3 & 0x1F; |
342 | |
343 | int bit0 = (v1 >> 6) & 1; |
344 | int bit1 = (v1 >> 5) & 1; |
345 | int bit2 = (v2 >> 6) & 1; |
346 | int bit3 = (v2 >> 5) & 1; |
347 | int bit4 = (v3 >> 7) & 1; |
348 | int bit5 = (v3 >> 6) & 1; |
349 | int bit6 = (v3 >> 5) & 1; |
350 | |
351 | int ohcomp = 1 << mode; |
352 | |
353 | if (ohcomp & 0x30) |
354 | green |= bit0 << 6; |
355 | if (ohcomp & 0x3A) |
356 | green |= bit1 << 5; |
357 | if (ohcomp & 0x30) |
358 | blue |= bit2 << 6; |
359 | if (ohcomp & 0x3A) |
360 | blue |= bit3 << 5; |
361 | |
362 | if (ohcomp & 0x3D) |
363 | scale |= bit6 << 5; |
364 | if (ohcomp & 0x2D) |
365 | scale |= bit5 << 6; |
366 | if (ohcomp & 0x04) |
367 | scale |= bit4 << 7; |
368 | |
369 | if (ohcomp & 0x3B) |
370 | red |= bit4 << 6; |
371 | if (ohcomp & 0x04) |
372 | red |= bit3 << 6; |
373 | |
374 | if (ohcomp & 0x10) |
375 | red |= bit5 << 7; |
376 | if (ohcomp & 0x0F) |
377 | red |= bit2 << 7; |
378 | |
379 | if (ohcomp & 0x05) |
380 | red |= bit1 << 8; |
381 | if (ohcomp & 0x0A) |
382 | red |= bit0 << 8; |
383 | |
384 | if (ohcomp & 0x05) |
385 | red |= bit0 << 9; |
386 | if (ohcomp & 0x02) |
387 | red |= bit6 << 9; |
388 | |
389 | if (ohcomp & 0x01) |
390 | red |= bit3 << 10; |
391 | if (ohcomp & 0x02) |
392 | red |= bit5 << 10; |
393 | |
394 | // expand to 12 bits. |
395 | static const int shamts[6] { 1, 1, 2, 3, 4, 5 }; |
396 | int shamt = shamts[mode]; |
397 | red <<= shamt; |
398 | green <<= shamt; |
399 | blue <<= shamt; |
400 | scale <<= shamt; |
401 | |
402 | // on modes 0 to 4, the values stored for "green" and "blue" are differentials, |
403 | // not absolute values. |
404 | if (mode != 5) |
405 | { |
406 | green = red - green; |
407 | blue = red - blue; |
408 | } |
409 | |
410 | // switch around components. |
411 | int temp; |
412 | switch (majcomp) |
413 | { |
414 | case 1: |
415 | temp = red; |
416 | red = green; |
417 | green = temp; |
418 | break; |
419 | case 2: |
420 | temp = red; |
421 | red = blue; |
422 | blue = temp; |
423 | break; |
424 | default: |
425 | break; |
426 | } |
427 | |
428 | int red0 = red - scale; |
429 | int green0 = green - scale; |
430 | int blue0 = blue - scale; |
431 | |
432 | // clamp to [0,0xFFF]. |
433 | if (red < 0) |
434 | red = 0; |
435 | if (green < 0) |
436 | green = 0; |
437 | if (blue < 0) |
438 | blue = 0; |
439 | |
440 | if (red0 < 0) |
441 | red0 = 0; |
442 | if (green0 < 0) |
443 | green0 = 0; |
444 | if (blue0 < 0) |
445 | blue0 = 0; |
446 | |
447 | output0 = vint4(red0 << 4, green0 << 4, blue0 << 4, 0x7800); |
448 | output1 = vint4(red << 4, green << 4, blue << 4, 0x7800); |
449 | } |
450 | |
451 | /** |
452 | * @brief Unpack an HDR RGB direct encoding. |
453 | * |
454 | * @param input The packed endpoints (packed and modal). |
455 | * @param[out] output0 The unpacked endpoint 0 color. |
456 | * @param[out] output1 The unpacked endpoint 1 color. |
457 | */ |
458 | static void hdr_rgb_unpack( |
459 | const uint8_t input[6], |
460 | vint4& output0, |
461 | vint4& output1 |
462 | ) { |
463 | |
464 | int v0 = input[0]; |
465 | int v1 = input[1]; |
466 | int v2 = input[2]; |
467 | int v3 = input[3]; |
468 | int v4 = input[4]; |
469 | int v5 = input[5]; |
470 | |
471 | // extract all the fixed-placement bitfields |
472 | int modeval = ((v1 & 0x80) >> 7) | (((v2 & 0x80) >> 7) << 1) | (((v3 & 0x80) >> 7) << 2); |
473 | |
474 | int majcomp = ((v4 & 0x80) >> 7) | (((v5 & 0x80) >> 7) << 1); |
475 | |
476 | if (majcomp == 3) |
477 | { |
478 | output0 = vint4(v0 << 8, v2 << 8, (v4 & 0x7F) << 9, 0x7800); |
479 | output1 = vint4(v1 << 8, v3 << 8, (v5 & 0x7F) << 9, 0x7800); |
480 | return; |
481 | } |
482 | |
483 | int a = v0 | ((v1 & 0x40) << 2); |
484 | int b0 = v2 & 0x3f; |
485 | int b1 = v3 & 0x3f; |
486 | int c = v1 & 0x3f; |
487 | int d0 = v4 & 0x7f; |
488 | int d1 = v5 & 0x7f; |
489 | |
490 | // get hold of the number of bits in 'd0' and 'd1' |
491 | static const int dbits_tab[8] { 7, 6, 7, 6, 5, 6, 5, 6 }; |
492 | int dbits = dbits_tab[modeval]; |
493 | |
494 | // extract six variable-placement bits |
495 | int bit0 = (v2 >> 6) & 1; |
496 | int bit1 = (v3 >> 6) & 1; |
497 | int bit2 = (v4 >> 6) & 1; |
498 | int bit3 = (v5 >> 6) & 1; |
499 | int bit4 = (v4 >> 5) & 1; |
500 | int bit5 = (v5 >> 5) & 1; |
501 | |
502 | // and prepend the variable-placement bits depending on mode. |
503 | int ohmod = 1 << modeval; // one-hot-mode |
504 | if (ohmod & 0xA4) |
505 | a |= bit0 << 9; |
506 | if (ohmod & 0x8) |
507 | a |= bit2 << 9; |
508 | if (ohmod & 0x50) |
509 | a |= bit4 << 9; |
510 | |
511 | if (ohmod & 0x50) |
512 | a |= bit5 << 10; |
513 | if (ohmod & 0xA0) |
514 | a |= bit1 << 10; |
515 | |
516 | if (ohmod & 0xC0) |
517 | a |= bit2 << 11; |
518 | |
519 | if (ohmod & 0x4) |
520 | c |= bit1 << 6; |
521 | if (ohmod & 0xE8) |
522 | c |= bit3 << 6; |
523 | |
524 | if (ohmod & 0x20) |
525 | c |= bit2 << 7; |
526 | |
527 | if (ohmod & 0x5B) |
528 | { |
529 | b0 |= bit0 << 6; |
530 | b1 |= bit1 << 6; |
531 | } |
532 | |
533 | if (ohmod & 0x12) |
534 | { |
535 | b0 |= bit2 << 7; |
536 | b1 |= bit3 << 7; |
537 | } |
538 | |
539 | if (ohmod & 0xAF) |
540 | { |
541 | d0 |= bit4 << 5; |
542 | d1 |= bit5 << 5; |
543 | } |
544 | |
545 | if (ohmod & 0x5) |
546 | { |
547 | d0 |= bit2 << 6; |
548 | d1 |= bit3 << 6; |
549 | } |
550 | |
551 | // sign-extend 'd0' and 'd1' |
552 | // note: this code assumes that signed right-shift actually sign-fills, not zero-fills. |
553 | int32_t d0x = d0; |
554 | int32_t d1x = d1; |
555 | int sx_shamt = 32 - dbits; |
556 | d0x <<= sx_shamt; |
557 | d0x >>= sx_shamt; |
558 | d1x <<= sx_shamt; |
559 | d1x >>= sx_shamt; |
560 | d0 = d0x; |
561 | d1 = d1x; |
562 | |
563 | // expand all values to 12 bits, with left-shift as needed. |
564 | int val_shamt = (modeval >> 1) ^ 3; |
565 | a <<= val_shamt; |
566 | b0 <<= val_shamt; |
567 | b1 <<= val_shamt; |
568 | c <<= val_shamt; |
569 | d0 <<= val_shamt; |
570 | d1 <<= val_shamt; |
571 | |
572 | // then compute the actual color values. |
573 | int red1 = a; |
574 | int green1 = a - b0; |
575 | int blue1 = a - b1; |
576 | int red0 = a - c; |
577 | int green0 = a - b0 - c - d0; |
578 | int blue0 = a - b1 - c - d1; |
579 | |
580 | // clamp the color components to [0,2^12 - 1] |
581 | red0 = astc::clamp(red0, 0, 4095); |
582 | green0 = astc::clamp(green0, 0, 4095); |
583 | blue0 = astc::clamp(blue0, 0, 4095); |
584 | |
585 | red1 = astc::clamp(red1, 0, 4095); |
586 | green1 = astc::clamp(green1, 0, 4095); |
587 | blue1 = astc::clamp(blue1, 0, 4095); |
588 | |
589 | // switch around the color components |
590 | int temp0, temp1; |
591 | switch (majcomp) |
592 | { |
593 | case 1: // switch around red and green |
594 | temp0 = red0; |
595 | temp1 = red1; |
596 | red0 = green0; |
597 | red1 = green1; |
598 | green0 = temp0; |
599 | green1 = temp1; |
600 | break; |
601 | case 2: // switch around red and blue |
602 | temp0 = red0; |
603 | temp1 = red1; |
604 | red0 = blue0; |
605 | red1 = blue1; |
606 | blue0 = temp0; |
607 | blue1 = temp1; |
608 | break; |
609 | case 0: // no switch |
610 | break; |
611 | } |
612 | |
613 | output0 = vint4(red0 << 4, green0 << 4, blue0 << 4, 0x7800); |
614 | output1 = vint4(red1 << 4, green1 << 4, blue1 << 4, 0x7800); |
615 | } |
616 | |
617 | /** |
618 | * @brief Unpack an HDR RGB + LDR A direct encoding. |
619 | * |
620 | * @param input The packed endpoints (packed and modal). |
621 | * @param[out] output0 The unpacked endpoint 0 color. |
622 | * @param[out] output1 The unpacked endpoint 1 color. |
623 | */ |
624 | static void hdr_rgb_ldr_alpha_unpack( |
625 | const uint8_t input[8], |
626 | vint4& output0, |
627 | vint4& output1 |
628 | ) { |
629 | hdr_rgb_unpack(input, output0, output1); |
630 | |
631 | int v6 = input[6]; |
632 | int v7 = input[7]; |
633 | output0.set_lane<3>(v6); |
634 | output1.set_lane<3>(v7); |
635 | } |
636 | |
637 | /** |
638 | * @brief Unpack an HDR L (small range) direct encoding. |
639 | * |
640 | * @param input The packed endpoints (packed and modal). |
641 | * @param[out] output0 The unpacked endpoint 0 color. |
642 | * @param[out] output1 The unpacked endpoint 1 color. |
643 | */ |
644 | static void hdr_luminance_small_range_unpack( |
645 | const uint8_t input[2], |
646 | vint4& output0, |
647 | vint4& output1 |
648 | ) { |
649 | int v0 = input[0]; |
650 | int v1 = input[1]; |
651 | |
652 | int y0, y1; |
653 | if (v0 & 0x80) |
654 | { |
655 | y0 = ((v1 & 0xE0) << 4) | ((v0 & 0x7F) << 2); |
656 | y1 = (v1 & 0x1F) << 2; |
657 | } |
658 | else |
659 | { |
660 | y0 = ((v1 & 0xF0) << 4) | ((v0 & 0x7F) << 1); |
661 | y1 = (v1 & 0xF) << 1; |
662 | } |
663 | |
664 | y1 += y0; |
665 | if (y1 > 0xFFF) |
666 | { |
667 | y1 = 0xFFF; |
668 | } |
669 | |
670 | output0 = vint4(y0 << 4, y0 << 4, y0 << 4, 0x7800); |
671 | output1 = vint4(y1 << 4, y1 << 4, y1 << 4, 0x7800); |
672 | } |
673 | |
674 | /** |
675 | * @brief Unpack an HDR L (large range) direct encoding. |
676 | * |
677 | * @param input The packed endpoints (packed and modal). |
678 | * @param[out] output0 The unpacked endpoint 0 color. |
679 | * @param[out] output1 The unpacked endpoint 1 color. |
680 | */ |
681 | static void hdr_luminance_large_range_unpack( |
682 | const uint8_t input[2], |
683 | vint4& output0, |
684 | vint4& output1 |
685 | ) { |
686 | int v0 = input[0]; |
687 | int v1 = input[1]; |
688 | |
689 | int y0, y1; |
690 | if (v1 >= v0) |
691 | { |
692 | y0 = v0 << 4; |
693 | y1 = v1 << 4; |
694 | } |
695 | else |
696 | { |
697 | y0 = (v1 << 4) + 8; |
698 | y1 = (v0 << 4) - 8; |
699 | } |
700 | |
701 | output0 = vint4(y0 << 4, y0 << 4, y0 << 4, 0x7800); |
702 | output1 = vint4(y1 << 4, y1 << 4, y1 << 4, 0x7800); |
703 | } |
704 | |
705 | /** |
706 | * @brief Unpack an HDR A direct encoding. |
707 | * |
708 | * @param input The packed endpoints (packed and modal). |
709 | * @param[out] output0 The unpacked endpoint 0 color. |
710 | * @param[out] output1 The unpacked endpoint 1 color. |
711 | */ |
712 | static void hdr_alpha_unpack( |
713 | const uint8_t input[2], |
714 | int& output0, |
715 | int& output1 |
716 | ) { |
717 | |
718 | int v6 = input[0]; |
719 | int v7 = input[1]; |
720 | |
721 | int selector = ((v6 >> 7) & 1) | ((v7 >> 6) & 2); |
722 | v6 &= 0x7F; |
723 | v7 &= 0x7F; |
724 | if (selector == 3) |
725 | { |
726 | output0 = v6 << 5; |
727 | output1 = v7 << 5; |
728 | } |
729 | else |
730 | { |
731 | v6 |= (v7 << (selector + 1)) & 0x780; |
732 | v7 &= (0x3f >> selector); |
733 | v7 ^= 32 >> selector; |
734 | v7 -= 32 >> selector; |
735 | v6 <<= (4 - selector); |
736 | v7 <<= (4 - selector); |
737 | v7 += v6; |
738 | |
739 | if (v7 < 0) |
740 | { |
741 | v7 = 0; |
742 | } |
743 | else if (v7 > 0xFFF) |
744 | { |
745 | v7 = 0xFFF; |
746 | } |
747 | |
748 | output0 = v6; |
749 | output1 = v7; |
750 | } |
751 | |
752 | output0 <<= 4; |
753 | output1 <<= 4; |
754 | } |
755 | |
756 | /** |
757 | * @brief Unpack an HDR RGBA direct encoding. |
758 | * |
759 | * @param input The packed endpoints (packed and modal). |
760 | * @param[out] output0 The unpacked endpoint 0 color. |
761 | * @param[out] output1 The unpacked endpoint 1 color. |
762 | */ |
763 | static void hdr_rgb_hdr_alpha_unpack( |
764 | const uint8_t input[8], |
765 | vint4& output0, |
766 | vint4& output1 |
767 | ) { |
768 | hdr_rgb_unpack(input, output0, output1); |
769 | |
770 | int alpha0, alpha1; |
771 | hdr_alpha_unpack(input + 6, alpha0, alpha1); |
772 | |
773 | output0.set_lane<3>(alpha0); |
774 | output1.set_lane<3>(alpha1); |
775 | } |
776 | |
777 | /* See header for documentation. */ |
778 | void unpack_color_endpoints( |
779 | astcenc_profile decode_mode, |
780 | int format, |
781 | const uint8_t* input, |
782 | bool& rgb_hdr, |
783 | bool& alpha_hdr, |
784 | vint4& output0, |
785 | vint4& output1 |
786 | ) { |
787 | // Assume no NaNs and LDR endpoints unless set later |
788 | rgb_hdr = false; |
789 | alpha_hdr = false; |
790 | |
791 | bool alpha_hdr_default = false; |
792 | |
793 | switch (format) |
794 | { |
795 | case FMT_LUMINANCE: |
796 | luminance_unpack(input, output0, output1); |
797 | break; |
798 | |
799 | case FMT_LUMINANCE_DELTA: |
800 | luminance_delta_unpack(input, output0, output1); |
801 | break; |
802 | |
803 | case FMT_HDR_LUMINANCE_SMALL_RANGE: |
804 | rgb_hdr = true; |
805 | alpha_hdr_default = true; |
806 | hdr_luminance_small_range_unpack(input, output0, output1); |
807 | break; |
808 | |
809 | case FMT_HDR_LUMINANCE_LARGE_RANGE: |
810 | rgb_hdr = true; |
811 | alpha_hdr_default = true; |
812 | hdr_luminance_large_range_unpack(input, output0, output1); |
813 | break; |
814 | |
815 | case FMT_LUMINANCE_ALPHA: |
816 | luminance_alpha_unpack(input, output0, output1); |
817 | break; |
818 | |
819 | case FMT_LUMINANCE_ALPHA_DELTA: |
820 | luminance_alpha_delta_unpack(input, output0, output1); |
821 | break; |
822 | |
823 | case FMT_RGB_SCALE: |
824 | { |
825 | vint4 input0q(input[0], input[1], input[2], 0); |
826 | uint8_t scale = input[3]; |
827 | rgb_scale_unpack(input0q, scale, output0, output1); |
828 | } |
829 | break; |
830 | |
831 | case FMT_RGB_SCALE_ALPHA: |
832 | { |
833 | vint4 input0q(input[0], input[1], input[2], input[4]); |
834 | uint8_t alpha1q = input[5]; |
835 | uint8_t scaleq = input[3]; |
836 | rgb_scale_alpha_unpack(input0q, alpha1q, scaleq, output0, output1); |
837 | } |
838 | break; |
839 | |
840 | case FMT_HDR_RGB_SCALE: |
841 | rgb_hdr = true; |
842 | alpha_hdr_default = true; |
843 | hdr_rgbo_unpack(input, output0, output1); |
844 | break; |
845 | |
846 | case FMT_RGB: |
847 | { |
848 | vint4 input0q(input[0], input[2], input[4], 0); |
849 | vint4 input1q(input[1], input[3], input[5], 0); |
850 | rgb_unpack(input0q, input1q, output0, output1); |
851 | } |
852 | break; |
853 | |
854 | case FMT_RGB_DELTA: |
855 | { |
856 | vint4 input0q(input[0], input[2], input[4], 0); |
857 | vint4 input1q(input[1], input[3], input[5], 0); |
858 | rgb_delta_unpack(input0q, input1q, output0, output1); |
859 | } |
860 | break; |
861 | |
862 | case FMT_HDR_RGB: |
863 | rgb_hdr = true; |
864 | alpha_hdr_default = true; |
865 | hdr_rgb_unpack(input, output0, output1); |
866 | break; |
867 | |
868 | case FMT_RGBA: |
869 | { |
870 | vint4 input0q(input[0], input[2], input[4], input[6]); |
871 | vint4 input1q(input[1], input[3], input[5], input[7]); |
872 | rgba_unpack(input0q, input1q, output0, output1); |
873 | } |
874 | break; |
875 | |
876 | case FMT_RGBA_DELTA: |
877 | { |
878 | vint4 input0q(input[0], input[2], input[4], input[6]); |
879 | vint4 input1q(input[1], input[3], input[5], input[7]); |
880 | rgba_delta_unpack(input0q, input1q, output0, output1); |
881 | } |
882 | break; |
883 | |
884 | case FMT_HDR_RGB_LDR_ALPHA: |
885 | rgb_hdr = true; |
886 | hdr_rgb_ldr_alpha_unpack(input, output0, output1); |
887 | break; |
888 | |
889 | case FMT_HDR_RGBA: |
890 | rgb_hdr = true; |
891 | alpha_hdr = true; |
892 | hdr_rgb_hdr_alpha_unpack(input, output0, output1); |
893 | break; |
894 | } |
895 | |
896 | // Assign a correct default alpha |
897 | if (alpha_hdr_default) |
898 | { |
899 | if (decode_mode == ASTCENC_PRF_HDR) |
900 | { |
901 | output0.set_lane<3>(0x7800); |
902 | output1.set_lane<3>(0x7800); |
903 | alpha_hdr = true; |
904 | } |
905 | else |
906 | { |
907 | output0.set_lane<3>(0x00FF); |
908 | output1.set_lane<3>(0x00FF); |
909 | alpha_hdr = false; |
910 | } |
911 | } |
912 | |
913 | vint4 ldr_scale(257); |
914 | vint4 hdr_scale(1); |
915 | vint4 output_scale = ldr_scale; |
916 | |
917 | // An LDR profile image |
918 | if ((decode_mode == ASTCENC_PRF_LDR) || |
919 | (decode_mode == ASTCENC_PRF_LDR_SRGB)) |
920 | { |
921 | // Also matches HDR alpha, as cannot have HDR alpha without HDR RGB |
922 | if (rgb_hdr == true) |
923 | { |
924 | output0 = vint4(0xFF00, 0x0000, 0xFF00, 0xFF00); |
925 | output1 = vint4(0xFF00, 0x0000, 0xFF00, 0xFF00); |
926 | output_scale = hdr_scale; |
927 | |
928 | rgb_hdr = false; |
929 | alpha_hdr = false; |
930 | } |
931 | } |
932 | // An HDR profile image |
933 | else |
934 | { |
935 | vmask4 hdr_lanes(rgb_hdr, rgb_hdr, rgb_hdr, alpha_hdr); |
936 | output_scale = select(ldr_scale, hdr_scale, hdr_lanes); |
937 | } |
938 | |
939 | output0 = output0 * output_scale; |
940 | output1 = output1 * output_scale; |
941 | } |
942 | |