1// Copyright 2010 Google Inc. All Rights Reserved.
2//
3// Use of this source code is governed by a BSD-style license
4// that can be found in the COPYING file in the root of the source
5// tree. An additional intellectual property rights grant can be found
6// in the file PATENTS. All contributing project authors may
7// be found in the AUTHORS file in the root of the source tree.
8// -----------------------------------------------------------------------------
9//
10// Speed-critical decoding functions, default plain-C implementations.
11//
12// Author: Skal (pascal.massimino@gmail.com)
13
14#include "./dsp.h"
15#include "../dec/vp8i_dec.h"
16#include "../utils/utils.h"
17
18//------------------------------------------------------------------------------
19
20static WEBP_INLINE uint8_t clip_8b(int v) {
21 return (!(v & ~0xff)) ? v : (v < 0) ? 0 : 255;
22}
23
24//------------------------------------------------------------------------------
25// Transforms (Paragraph 14.4)
26
27#define STORE(x, y, v) \
28 dst[x + y * BPS] = clip_8b(dst[x + y * BPS] + ((v) >> 3))
29
30#define STORE2(y, dc, d, c) do { \
31 const int DC = (dc); \
32 STORE(0, y, DC + (d)); \
33 STORE(1, y, DC + (c)); \
34 STORE(2, y, DC - (c)); \
35 STORE(3, y, DC - (d)); \
36} while (0)
37
38#define MUL1(a) ((((a) * 20091) >> 16) + (a))
39#define MUL2(a) (((a) * 35468) >> 16)
40
41static void TransformOne(const int16_t* in, uint8_t* dst) {
42 int C[4 * 4], *tmp;
43 int i;
44 tmp = C;
45 for (i = 0; i < 4; ++i) { // vertical pass
46 const int a = in[0] + in[8]; // [-4096, 4094]
47 const int b = in[0] - in[8]; // [-4095, 4095]
48 const int c = MUL2(in[4]) - MUL1(in[12]); // [-3783, 3783]
49 const int d = MUL1(in[4]) + MUL2(in[12]); // [-3785, 3781]
50 tmp[0] = a + d; // [-7881, 7875]
51 tmp[1] = b + c; // [-7878, 7878]
52 tmp[2] = b - c; // [-7878, 7878]
53 tmp[3] = a - d; // [-7877, 7879]
54 tmp += 4;
55 in++;
56 }
57 // Each pass is expanding the dynamic range by ~3.85 (upper bound).
58 // The exact value is (2. + (20091 + 35468) / 65536).
59 // After the second pass, maximum interval is [-3794, 3794], assuming
60 // an input in [-2048, 2047] interval. We then need to add a dst value
61 // in the [0, 255] range.
62 // In the worst case scenario, the input to clip_8b() can be as large as
63 // [-60713, 60968].
64 tmp = C;
65 for (i = 0; i < 4; ++i) { // horizontal pass
66 const int dc = tmp[0] + 4;
67 const int a = dc + tmp[8];
68 const int b = dc - tmp[8];
69 const int c = MUL2(tmp[4]) - MUL1(tmp[12]);
70 const int d = MUL1(tmp[4]) + MUL2(tmp[12]);
71 STORE(0, 0, a + d);
72 STORE(1, 0, b + c);
73 STORE(2, 0, b - c);
74 STORE(3, 0, a - d);
75 tmp++;
76 dst += BPS;
77 }
78}
79
80// Simplified transform when only in[0], in[1] and in[4] are non-zero
81static void TransformAC3(const int16_t* in, uint8_t* dst) {
82 const int a = in[0] + 4;
83 const int c4 = MUL2(in[4]);
84 const int d4 = MUL1(in[4]);
85 const int c1 = MUL2(in[1]);
86 const int d1 = MUL1(in[1]);
87 STORE2(0, a + d4, d1, c1);
88 STORE2(1, a + c4, d1, c1);
89 STORE2(2, a - c4, d1, c1);
90 STORE2(3, a - d4, d1, c1);
91}
92#undef MUL1
93#undef MUL2
94#undef STORE2
95
96static void TransformTwo(const int16_t* in, uint8_t* dst, int do_two) {
97 TransformOne(in, dst);
98 if (do_two) {
99 TransformOne(in + 16, dst + 4);
100 }
101}
102
103static void TransformUV(const int16_t* in, uint8_t* dst) {
104 VP8Transform(in + 0 * 16, dst, 1);
105 VP8Transform(in + 2 * 16, dst + 4 * BPS, 1);
106}
107
108static void TransformDC(const int16_t* in, uint8_t* dst) {
109 const int DC = in[0] + 4;
110 int i, j;
111 for (j = 0; j < 4; ++j) {
112 for (i = 0; i < 4; ++i) {
113 STORE(i, j, DC);
114 }
115 }
116}
117
118static void TransformDCUV(const int16_t* in, uint8_t* dst) {
119 if (in[0 * 16]) VP8TransformDC(in + 0 * 16, dst);
120 if (in[1 * 16]) VP8TransformDC(in + 1 * 16, dst + 4);
121 if (in[2 * 16]) VP8TransformDC(in + 2 * 16, dst + 4 * BPS);
122 if (in[3 * 16]) VP8TransformDC(in + 3 * 16, dst + 4 * BPS + 4);
123}
124
125#undef STORE
126
127//------------------------------------------------------------------------------
128// Paragraph 14.3
129
130static void TransformWHT(const int16_t* in, int16_t* out) {
131 int tmp[16];
132 int i;
133 for (i = 0; i < 4; ++i) {
134 const int a0 = in[0 + i] + in[12 + i];
135 const int a1 = in[4 + i] + in[ 8 + i];
136 const int a2 = in[4 + i] - in[ 8 + i];
137 const int a3 = in[0 + i] - in[12 + i];
138 tmp[0 + i] = a0 + a1;
139 tmp[8 + i] = a0 - a1;
140 tmp[4 + i] = a3 + a2;
141 tmp[12 + i] = a3 - a2;
142 }
143 for (i = 0; i < 4; ++i) {
144 const int dc = tmp[0 + i * 4] + 3; // w/ rounder
145 const int a0 = dc + tmp[3 + i * 4];
146 const int a1 = tmp[1 + i * 4] + tmp[2 + i * 4];
147 const int a2 = tmp[1 + i * 4] - tmp[2 + i * 4];
148 const int a3 = dc - tmp[3 + i * 4];
149 out[ 0] = (a0 + a1) >> 3;
150 out[16] = (a3 + a2) >> 3;
151 out[32] = (a0 - a1) >> 3;
152 out[48] = (a3 - a2) >> 3;
153 out += 64;
154 }
155}
156
157void (*VP8TransformWHT)(const int16_t* in, int16_t* out);
158
159//------------------------------------------------------------------------------
160// Intra predictions
161
162#define DST(x, y) dst[(x) + (y) * BPS]
163
164static WEBP_INLINE void TrueMotion(uint8_t* dst, int size) {
165 const uint8_t* top = dst - BPS;
166 const uint8_t* const clip0 = VP8kclip1 - top[-1];
167 int y;
168 for (y = 0; y < size; ++y) {
169 const uint8_t* const clip = clip0 + dst[-1];
170 int x;
171 for (x = 0; x < size; ++x) {
172 dst[x] = clip[top[x]];
173 }
174 dst += BPS;
175 }
176}
177static void TM4(uint8_t* dst) { TrueMotion(dst, 4); }
178static void TM8uv(uint8_t* dst) { TrueMotion(dst, 8); }
179static void TM16(uint8_t* dst) { TrueMotion(dst, 16); }
180
181//------------------------------------------------------------------------------
182// 16x16
183
184static void VE16(uint8_t* dst) { // vertical
185 int j;
186 for (j = 0; j < 16; ++j) {
187 memcpy(dst + j * BPS, dst - BPS, 16);
188 }
189}
190
191static void HE16(uint8_t* dst) { // horizontal
192 int j;
193 for (j = 16; j > 0; --j) {
194 memset(dst, dst[-1], 16);
195 dst += BPS;
196 }
197}
198
199static WEBP_INLINE void Put16(int v, uint8_t* dst) {
200 int j;
201 for (j = 0; j < 16; ++j) {
202 memset(dst + j * BPS, v, 16);
203 }
204}
205
206static void DC16(uint8_t* dst) { // DC
207 int DC = 16;
208 int j;
209 for (j = 0; j < 16; ++j) {
210 DC += dst[-1 + j * BPS] + dst[j - BPS];
211 }
212 Put16(DC >> 5, dst);
213}
214
215static void DC16NoTop(uint8_t* dst) { // DC with top samples not available
216 int DC = 8;
217 int j;
218 for (j = 0; j < 16; ++j) {
219 DC += dst[-1 + j * BPS];
220 }
221 Put16(DC >> 4, dst);
222}
223
224static void DC16NoLeft(uint8_t* dst) { // DC with left samples not available
225 int DC = 8;
226 int i;
227 for (i = 0; i < 16; ++i) {
228 DC += dst[i - BPS];
229 }
230 Put16(DC >> 4, dst);
231}
232
233static void DC16NoTopLeft(uint8_t* dst) { // DC with no top and left samples
234 Put16(0x80, dst);
235}
236
237VP8PredFunc VP8PredLuma16[NUM_B_DC_MODES];
238
239//------------------------------------------------------------------------------
240// 4x4
241
242#define AVG3(a, b, c) ((uint8_t)(((a) + 2 * (b) + (c) + 2) >> 2))
243#define AVG2(a, b) (((a) + (b) + 1) >> 1)
244
245static void VE4(uint8_t* dst) { // vertical
246 const uint8_t* top = dst - BPS;
247 const uint8_t vals[4] = {
248 AVG3(top[-1], top[0], top[1]),
249 AVG3(top[ 0], top[1], top[2]),
250 AVG3(top[ 1], top[2], top[3]),
251 AVG3(top[ 2], top[3], top[4])
252 };
253 int i;
254 for (i = 0; i < 4; ++i) {
255 memcpy(dst + i * BPS, vals, sizeof(vals));
256 }
257}
258
259static void HE4(uint8_t* dst) { // horizontal
260 const int A = dst[-1 - BPS];
261 const int B = dst[-1];
262 const int C = dst[-1 + BPS];
263 const int D = dst[-1 + 2 * BPS];
264 const int E = dst[-1 + 3 * BPS];
265 WebPUint32ToMem(dst + 0 * BPS, 0x01010101U * AVG3(A, B, C));
266 WebPUint32ToMem(dst + 1 * BPS, 0x01010101U * AVG3(B, C, D));
267 WebPUint32ToMem(dst + 2 * BPS, 0x01010101U * AVG3(C, D, E));
268 WebPUint32ToMem(dst + 3 * BPS, 0x01010101U * AVG3(D, E, E));
269}
270
271static void DC4(uint8_t* dst) { // DC
272 uint32_t dc = 4;
273 int i;
274 for (i = 0; i < 4; ++i) dc += dst[i - BPS] + dst[-1 + i * BPS];
275 dc >>= 3;
276 for (i = 0; i < 4; ++i) memset(dst + i * BPS, dc, 4);
277}
278
279static void RD4(uint8_t* dst) { // Down-right
280 const int I = dst[-1 + 0 * BPS];
281 const int J = dst[-1 + 1 * BPS];
282 const int K = dst[-1 + 2 * BPS];
283 const int L = dst[-1 + 3 * BPS];
284 const int X = dst[-1 - BPS];
285 const int A = dst[0 - BPS];
286 const int B = dst[1 - BPS];
287 const int C = dst[2 - BPS];
288 const int D = dst[3 - BPS];
289 DST(0, 3) = AVG3(J, K, L);
290 DST(1, 3) = DST(0, 2) = AVG3(I, J, K);
291 DST(2, 3) = DST(1, 2) = DST(0, 1) = AVG3(X, I, J);
292 DST(3, 3) = DST(2, 2) = DST(1, 1) = DST(0, 0) = AVG3(A, X, I);
293 DST(3, 2) = DST(2, 1) = DST(1, 0) = AVG3(B, A, X);
294 DST(3, 1) = DST(2, 0) = AVG3(C, B, A);
295 DST(3, 0) = AVG3(D, C, B);
296}
297
298static void LD4(uint8_t* dst) { // Down-Left
299 const int A = dst[0 - BPS];
300 const int B = dst[1 - BPS];
301 const int C = dst[2 - BPS];
302 const int D = dst[3 - BPS];
303 const int E = dst[4 - BPS];
304 const int F = dst[5 - BPS];
305 const int G = dst[6 - BPS];
306 const int H = dst[7 - BPS];
307 DST(0, 0) = AVG3(A, B, C);
308 DST(1, 0) = DST(0, 1) = AVG3(B, C, D);
309 DST(2, 0) = DST(1, 1) = DST(0, 2) = AVG3(C, D, E);
310 DST(3, 0) = DST(2, 1) = DST(1, 2) = DST(0, 3) = AVG3(D, E, F);
311 DST(3, 1) = DST(2, 2) = DST(1, 3) = AVG3(E, F, G);
312 DST(3, 2) = DST(2, 3) = AVG3(F, G, H);
313 DST(3, 3) = AVG3(G, H, H);
314}
315
316static void VR4(uint8_t* dst) { // Vertical-Right
317 const int I = dst[-1 + 0 * BPS];
318 const int J = dst[-1 + 1 * BPS];
319 const int K = dst[-1 + 2 * BPS];
320 const int X = dst[-1 - BPS];
321 const int A = dst[0 - BPS];
322 const int B = dst[1 - BPS];
323 const int C = dst[2 - BPS];
324 const int D = dst[3 - BPS];
325 DST(0, 0) = DST(1, 2) = AVG2(X, A);
326 DST(1, 0) = DST(2, 2) = AVG2(A, B);
327 DST(2, 0) = DST(3, 2) = AVG2(B, C);
328 DST(3, 0) = AVG2(C, D);
329
330 DST(0, 3) = AVG3(K, J, I);
331 DST(0, 2) = AVG3(J, I, X);
332 DST(0, 1) = DST(1, 3) = AVG3(I, X, A);
333 DST(1, 1) = DST(2, 3) = AVG3(X, A, B);
334 DST(2, 1) = DST(3, 3) = AVG3(A, B, C);
335 DST(3, 1) = AVG3(B, C, D);
336}
337
338static void VL4(uint8_t* dst) { // Vertical-Left
339 const int A = dst[0 - BPS];
340 const int B = dst[1 - BPS];
341 const int C = dst[2 - BPS];
342 const int D = dst[3 - BPS];
343 const int E = dst[4 - BPS];
344 const int F = dst[5 - BPS];
345 const int G = dst[6 - BPS];
346 const int H = dst[7 - BPS];
347 DST(0, 0) = AVG2(A, B);
348 DST(1, 0) = DST(0, 2) = AVG2(B, C);
349 DST(2, 0) = DST(1, 2) = AVG2(C, D);
350 DST(3, 0) = DST(2, 2) = AVG2(D, E);
351
352 DST(0, 1) = AVG3(A, B, C);
353 DST(1, 1) = DST(0, 3) = AVG3(B, C, D);
354 DST(2, 1) = DST(1, 3) = AVG3(C, D, E);
355 DST(3, 1) = DST(2, 3) = AVG3(D, E, F);
356 DST(3, 2) = AVG3(E, F, G);
357 DST(3, 3) = AVG3(F, G, H);
358}
359
360static void HU4(uint8_t* dst) { // Horizontal-Up
361 const int I = dst[-1 + 0 * BPS];
362 const int J = dst[-1 + 1 * BPS];
363 const int K = dst[-1 + 2 * BPS];
364 const int L = dst[-1 + 3 * BPS];
365 DST(0, 0) = AVG2(I, J);
366 DST(2, 0) = DST(0, 1) = AVG2(J, K);
367 DST(2, 1) = DST(0, 2) = AVG2(K, L);
368 DST(1, 0) = AVG3(I, J, K);
369 DST(3, 0) = DST(1, 1) = AVG3(J, K, L);
370 DST(3, 1) = DST(1, 2) = AVG3(K, L, L);
371 DST(3, 2) = DST(2, 2) =
372 DST(0, 3) = DST(1, 3) = DST(2, 3) = DST(3, 3) = L;
373}
374
375static void HD4(uint8_t* dst) { // Horizontal-Down
376 const int I = dst[-1 + 0 * BPS];
377 const int J = dst[-1 + 1 * BPS];
378 const int K = dst[-1 + 2 * BPS];
379 const int L = dst[-1 + 3 * BPS];
380 const int X = dst[-1 - BPS];
381 const int A = dst[0 - BPS];
382 const int B = dst[1 - BPS];
383 const int C = dst[2 - BPS];
384
385 DST(0, 0) = DST(2, 1) = AVG2(I, X);
386 DST(0, 1) = DST(2, 2) = AVG2(J, I);
387 DST(0, 2) = DST(2, 3) = AVG2(K, J);
388 DST(0, 3) = AVG2(L, K);
389
390 DST(3, 0) = AVG3(A, B, C);
391 DST(2, 0) = AVG3(X, A, B);
392 DST(1, 0) = DST(3, 1) = AVG3(I, X, A);
393 DST(1, 1) = DST(3, 2) = AVG3(J, I, X);
394 DST(1, 2) = DST(3, 3) = AVG3(K, J, I);
395 DST(1, 3) = AVG3(L, K, J);
396}
397
398#undef DST
399#undef AVG3
400#undef AVG2
401
402VP8PredFunc VP8PredLuma4[NUM_BMODES];
403
404//------------------------------------------------------------------------------
405// Chroma
406
407static void VE8uv(uint8_t* dst) { // vertical
408 int j;
409 for (j = 0; j < 8; ++j) {
410 memcpy(dst + j * BPS, dst - BPS, 8);
411 }
412}
413
414static void HE8uv(uint8_t* dst) { // horizontal
415 int j;
416 for (j = 0; j < 8; ++j) {
417 memset(dst, dst[-1], 8);
418 dst += BPS;
419 }
420}
421
422// helper for chroma-DC predictions
423static WEBP_INLINE void Put8x8uv(uint8_t value, uint8_t* dst) {
424 int j;
425 for (j = 0; j < 8; ++j) {
426 memset(dst + j * BPS, value, 8);
427 }
428}
429
430static void DC8uv(uint8_t* dst) { // DC
431 int dc0 = 8;
432 int i;
433 for (i = 0; i < 8; ++i) {
434 dc0 += dst[i - BPS] + dst[-1 + i * BPS];
435 }
436 Put8x8uv(dc0 >> 4, dst);
437}
438
439static void DC8uvNoLeft(uint8_t* dst) { // DC with no left samples
440 int dc0 = 4;
441 int i;
442 for (i = 0; i < 8; ++i) {
443 dc0 += dst[i - BPS];
444 }
445 Put8x8uv(dc0 >> 3, dst);
446}
447
448static void DC8uvNoTop(uint8_t* dst) { // DC with no top samples
449 int dc0 = 4;
450 int i;
451 for (i = 0; i < 8; ++i) {
452 dc0 += dst[-1 + i * BPS];
453 }
454 Put8x8uv(dc0 >> 3, dst);
455}
456
457static void DC8uvNoTopLeft(uint8_t* dst) { // DC with nothing
458 Put8x8uv(0x80, dst);
459}
460
461VP8PredFunc VP8PredChroma8[NUM_B_DC_MODES];
462
463//------------------------------------------------------------------------------
464// Edge filtering functions
465
466// 4 pixels in, 2 pixels out
467static WEBP_INLINE void do_filter2(uint8_t* p, int step) {
468 const int p1 = p[-2*step], p0 = p[-step], q0 = p[0], q1 = p[step];
469 const int a = 3 * (q0 - p0) + VP8ksclip1[p1 - q1]; // in [-893,892]
470 const int a1 = VP8ksclip2[(a + 4) >> 3]; // in [-16,15]
471 const int a2 = VP8ksclip2[(a + 3) >> 3];
472 p[-step] = VP8kclip1[p0 + a2];
473 p[ 0] = VP8kclip1[q0 - a1];
474}
475
476// 4 pixels in, 4 pixels out
477static WEBP_INLINE void do_filter4(uint8_t* p, int step) {
478 const int p1 = p[-2*step], p0 = p[-step], q0 = p[0], q1 = p[step];
479 const int a = 3 * (q0 - p0);
480 const int a1 = VP8ksclip2[(a + 4) >> 3];
481 const int a2 = VP8ksclip2[(a + 3) >> 3];
482 const int a3 = (a1 + 1) >> 1;
483 p[-2*step] = VP8kclip1[p1 + a3];
484 p[- step] = VP8kclip1[p0 + a2];
485 p[ 0] = VP8kclip1[q0 - a1];
486 p[ step] = VP8kclip1[q1 - a3];
487}
488
489// 6 pixels in, 6 pixels out
490static WEBP_INLINE void do_filter6(uint8_t* p, int step) {
491 const int p2 = p[-3*step], p1 = p[-2*step], p0 = p[-step];
492 const int q0 = p[0], q1 = p[step], q2 = p[2*step];
493 const int a = VP8ksclip1[3 * (q0 - p0) + VP8ksclip1[p1 - q1]];
494 // a is in [-128,127], a1 in [-27,27], a2 in [-18,18] and a3 in [-9,9]
495 const int a1 = (27 * a + 63) >> 7; // eq. to ((3 * a + 7) * 9) >> 7
496 const int a2 = (18 * a + 63) >> 7; // eq. to ((2 * a + 7) * 9) >> 7
497 const int a3 = (9 * a + 63) >> 7; // eq. to ((1 * a + 7) * 9) >> 7
498 p[-3*step] = VP8kclip1[p2 + a3];
499 p[-2*step] = VP8kclip1[p1 + a2];
500 p[- step] = VP8kclip1[p0 + a1];
501 p[ 0] = VP8kclip1[q0 - a1];
502 p[ step] = VP8kclip1[q1 - a2];
503 p[ 2*step] = VP8kclip1[q2 - a3];
504}
505
506static WEBP_INLINE int hev(const uint8_t* p, int step, int thresh) {
507 const int p1 = p[-2*step], p0 = p[-step], q0 = p[0], q1 = p[step];
508 return (VP8kabs0[p1 - p0] > thresh) || (VP8kabs0[q1 - q0] > thresh);
509}
510
511static WEBP_INLINE int needs_filter(const uint8_t* p, int step, int t) {
512 const int p1 = p[-2 * step], p0 = p[-step], q0 = p[0], q1 = p[step];
513 return ((4 * VP8kabs0[p0 - q0] + VP8kabs0[p1 - q1]) <= t);
514}
515
516static WEBP_INLINE int needs_filter2(const uint8_t* p,
517 int step, int t, int it) {
518 const int p3 = p[-4 * step], p2 = p[-3 * step], p1 = p[-2 * step];
519 const int p0 = p[-step], q0 = p[0];
520 const int q1 = p[step], q2 = p[2 * step], q3 = p[3 * step];
521 if ((4 * VP8kabs0[p0 - q0] + VP8kabs0[p1 - q1]) > t) return 0;
522 return VP8kabs0[p3 - p2] <= it && VP8kabs0[p2 - p1] <= it &&
523 VP8kabs0[p1 - p0] <= it && VP8kabs0[q3 - q2] <= it &&
524 VP8kabs0[q2 - q1] <= it && VP8kabs0[q1 - q0] <= it;
525}
526
527//------------------------------------------------------------------------------
528// Simple In-loop filtering (Paragraph 15.2)
529
530static void SimpleVFilter16(uint8_t* p, int stride, int thresh) {
531 int i;
532 const int thresh2 = 2 * thresh + 1;
533 for (i = 0; i < 16; ++i) {
534 if (needs_filter(p + i, stride, thresh2)) {
535 do_filter2(p + i, stride);
536 }
537 }
538}
539
540static void SimpleHFilter16(uint8_t* p, int stride, int thresh) {
541 int i;
542 const int thresh2 = 2 * thresh + 1;
543 for (i = 0; i < 16; ++i) {
544 if (needs_filter(p + i * stride, 1, thresh2)) {
545 do_filter2(p + i * stride, 1);
546 }
547 }
548}
549
550static void SimpleVFilter16i(uint8_t* p, int stride, int thresh) {
551 int k;
552 for (k = 3; k > 0; --k) {
553 p += 4 * stride;
554 SimpleVFilter16(p, stride, thresh);
555 }
556}
557
558static void SimpleHFilter16i(uint8_t* p, int stride, int thresh) {
559 int k;
560 for (k = 3; k > 0; --k) {
561 p += 4;
562 SimpleHFilter16(p, stride, thresh);
563 }
564}
565
566//------------------------------------------------------------------------------
567// Complex In-loop filtering (Paragraph 15.3)
568
569static WEBP_INLINE void FilterLoop26(uint8_t* p,
570 int hstride, int vstride, int size,
571 int thresh, int ithresh, int hev_thresh) {
572 const int thresh2 = 2 * thresh + 1;
573 while (size-- > 0) {
574 if (needs_filter2(p, hstride, thresh2, ithresh)) {
575 if (hev(p, hstride, hev_thresh)) {
576 do_filter2(p, hstride);
577 } else {
578 do_filter6(p, hstride);
579 }
580 }
581 p += vstride;
582 }
583}
584
585static WEBP_INLINE void FilterLoop24(uint8_t* p,
586 int hstride, int vstride, int size,
587 int thresh, int ithresh, int hev_thresh) {
588 const int thresh2 = 2 * thresh + 1;
589 while (size-- > 0) {
590 if (needs_filter2(p, hstride, thresh2, ithresh)) {
591 if (hev(p, hstride, hev_thresh)) {
592 do_filter2(p, hstride);
593 } else {
594 do_filter4(p, hstride);
595 }
596 }
597 p += vstride;
598 }
599}
600
601// on macroblock edges
602static void VFilter16(uint8_t* p, int stride,
603 int thresh, int ithresh, int hev_thresh) {
604 FilterLoop26(p, stride, 1, 16, thresh, ithresh, hev_thresh);
605}
606
607static void HFilter16(uint8_t* p, int stride,
608 int thresh, int ithresh, int hev_thresh) {
609 FilterLoop26(p, 1, stride, 16, thresh, ithresh, hev_thresh);
610}
611
612// on three inner edges
613static void VFilter16i(uint8_t* p, int stride,
614 int thresh, int ithresh, int hev_thresh) {
615 int k;
616 for (k = 3; k > 0; --k) {
617 p += 4 * stride;
618 FilterLoop24(p, stride, 1, 16, thresh, ithresh, hev_thresh);
619 }
620}
621
622static void HFilter16i(uint8_t* p, int stride,
623 int thresh, int ithresh, int hev_thresh) {
624 int k;
625 for (k = 3; k > 0; --k) {
626 p += 4;
627 FilterLoop24(p, 1, stride, 16, thresh, ithresh, hev_thresh);
628 }
629}
630
631// 8-pixels wide variant, for chroma filtering
632static void VFilter8(uint8_t* u, uint8_t* v, int stride,
633 int thresh, int ithresh, int hev_thresh) {
634 FilterLoop26(u, stride, 1, 8, thresh, ithresh, hev_thresh);
635 FilterLoop26(v, stride, 1, 8, thresh, ithresh, hev_thresh);
636}
637
638static void HFilter8(uint8_t* u, uint8_t* v, int stride,
639 int thresh, int ithresh, int hev_thresh) {
640 FilterLoop26(u, 1, stride, 8, thresh, ithresh, hev_thresh);
641 FilterLoop26(v, 1, stride, 8, thresh, ithresh, hev_thresh);
642}
643
644static void VFilter8i(uint8_t* u, uint8_t* v, int stride,
645 int thresh, int ithresh, int hev_thresh) {
646 FilterLoop24(u + 4 * stride, stride, 1, 8, thresh, ithresh, hev_thresh);
647 FilterLoop24(v + 4 * stride, stride, 1, 8, thresh, ithresh, hev_thresh);
648}
649
650static void HFilter8i(uint8_t* u, uint8_t* v, int stride,
651 int thresh, int ithresh, int hev_thresh) {
652 FilterLoop24(u + 4, 1, stride, 8, thresh, ithresh, hev_thresh);
653 FilterLoop24(v + 4, 1, stride, 8, thresh, ithresh, hev_thresh);
654}
655
656//------------------------------------------------------------------------------
657
658static void DitherCombine8x8(const uint8_t* dither, uint8_t* dst,
659 int dst_stride) {
660 int i, j;
661 for (j = 0; j < 8; ++j) {
662 for (i = 0; i < 8; ++i) {
663 const int delta0 = dither[i] - VP8_DITHER_AMP_CENTER;
664 const int delta1 =
665 (delta0 + VP8_DITHER_DESCALE_ROUNDER) >> VP8_DITHER_DESCALE;
666 dst[i] = clip_8b((int)dst[i] + delta1);
667 }
668 dst += dst_stride;
669 dither += 8;
670 }
671}
672
673//------------------------------------------------------------------------------
674
675VP8DecIdct2 VP8Transform;
676VP8DecIdct VP8TransformAC3;
677VP8DecIdct VP8TransformUV;
678VP8DecIdct VP8TransformDC;
679VP8DecIdct VP8TransformDCUV;
680
681VP8LumaFilterFunc VP8VFilter16;
682VP8LumaFilterFunc VP8HFilter16;
683VP8ChromaFilterFunc VP8VFilter8;
684VP8ChromaFilterFunc VP8HFilter8;
685VP8LumaFilterFunc VP8VFilter16i;
686VP8LumaFilterFunc VP8HFilter16i;
687VP8ChromaFilterFunc VP8VFilter8i;
688VP8ChromaFilterFunc VP8HFilter8i;
689VP8SimpleFilterFunc VP8SimpleVFilter16;
690VP8SimpleFilterFunc VP8SimpleHFilter16;
691VP8SimpleFilterFunc VP8SimpleVFilter16i;
692VP8SimpleFilterFunc VP8SimpleHFilter16i;
693
694void (*VP8DitherCombine8x8)(const uint8_t* dither, uint8_t* dst,
695 int dst_stride);
696
697extern void VP8DspInitSSE2(void);
698extern void VP8DspInitSSE41(void);
699extern void VP8DspInitNEON(void);
700extern void VP8DspInitMIPS32(void);
701extern void VP8DspInitMIPSdspR2(void);
702extern void VP8DspInitMSA(void);
703
704static volatile VP8CPUInfo dec_last_cpuinfo_used =
705 (VP8CPUInfo)&dec_last_cpuinfo_used;
706
707WEBP_TSAN_IGNORE_FUNCTION void VP8DspInit(void) {
708 if (dec_last_cpuinfo_used == VP8GetCPUInfo) return;
709
710 VP8InitClipTables();
711
712 VP8TransformWHT = TransformWHT;
713 VP8Transform = TransformTwo;
714 VP8TransformUV = TransformUV;
715 VP8TransformDC = TransformDC;
716 VP8TransformDCUV = TransformDCUV;
717 VP8TransformAC3 = TransformAC3;
718
719 VP8VFilter16 = VFilter16;
720 VP8HFilter16 = HFilter16;
721 VP8VFilter8 = VFilter8;
722 VP8HFilter8 = HFilter8;
723 VP8VFilter16i = VFilter16i;
724 VP8HFilter16i = HFilter16i;
725 VP8VFilter8i = VFilter8i;
726 VP8HFilter8i = HFilter8i;
727 VP8SimpleVFilter16 = SimpleVFilter16;
728 VP8SimpleHFilter16 = SimpleHFilter16;
729 VP8SimpleVFilter16i = SimpleVFilter16i;
730 VP8SimpleHFilter16i = SimpleHFilter16i;
731
732 VP8PredLuma4[0] = DC4;
733 VP8PredLuma4[1] = TM4;
734 VP8PredLuma4[2] = VE4;
735 VP8PredLuma4[3] = HE4;
736 VP8PredLuma4[4] = RD4;
737 VP8PredLuma4[5] = VR4;
738 VP8PredLuma4[6] = LD4;
739 VP8PredLuma4[7] = VL4;
740 VP8PredLuma4[8] = HD4;
741 VP8PredLuma4[9] = HU4;
742
743 VP8PredLuma16[0] = DC16;
744 VP8PredLuma16[1] = TM16;
745 VP8PredLuma16[2] = VE16;
746 VP8PredLuma16[3] = HE16;
747 VP8PredLuma16[4] = DC16NoTop;
748 VP8PredLuma16[5] = DC16NoLeft;
749 VP8PredLuma16[6] = DC16NoTopLeft;
750
751 VP8PredChroma8[0] = DC8uv;
752 VP8PredChroma8[1] = TM8uv;
753 VP8PredChroma8[2] = VE8uv;
754 VP8PredChroma8[3] = HE8uv;
755 VP8PredChroma8[4] = DC8uvNoTop;
756 VP8PredChroma8[5] = DC8uvNoLeft;
757 VP8PredChroma8[6] = DC8uvNoTopLeft;
758
759 VP8DitherCombine8x8 = DitherCombine8x8;
760
761 // If defined, use CPUInfo() to overwrite some pointers with faster versions.
762 if (VP8GetCPUInfo != NULL) {
763#if defined(WEBP_USE_SSE2)
764 if (VP8GetCPUInfo(kSSE2)) {
765 VP8DspInitSSE2();
766#if defined(WEBP_USE_SSE41)
767 if (VP8GetCPUInfo(kSSE4_1)) {
768 VP8DspInitSSE41();
769 }
770#endif
771 }
772#endif
773#if defined(WEBP_USE_NEON)
774 if (VP8GetCPUInfo(kNEON)) {
775 VP8DspInitNEON();
776 }
777#endif
778#if defined(WEBP_USE_MIPS32)
779 if (VP8GetCPUInfo(kMIPS32)) {
780 VP8DspInitMIPS32();
781 }
782#endif
783#if defined(WEBP_USE_MIPS_DSP_R2)
784 if (VP8GetCPUInfo(kMIPSdspR2)) {
785 VP8DspInitMIPSdspR2();
786 }
787#endif
788#if defined(WEBP_USE_MSA)
789 if (VP8GetCPUInfo(kMSA)) {
790 VP8DspInitMSA();
791 }
792#endif
793 }
794 dec_last_cpuinfo_used = VP8GetCPUInfo;
795}
796