1 // Copyright 2010 Google Inc. All Rights Reserved.
2 //
3 // This code is licensed under the same terms as WebM:
4 // Software License Agreement: http://www.webmproject.org/license/software/
5 // Additional IP Rights Grant: http://www.webmproject.org/license/additional/
6 // -----------------------------------------------------------------------------
7 //
8 // Speed-critical decoding functions.
9 //
10 // Author: Skal (pascal.massimino@gmail.com)
11
12 #include "./dsp.h"
13 #include "../dec/vp8i.h"
14
15 #if defined(__cplusplus) || defined(c_plusplus)
16 extern "C" {
17 #endif
18
19 //------------------------------------------------------------------------------
20 // run-time tables (~4k)
21
22 static uint8_t abs0[255 + 255 + 1]; // abs(i)
23 static uint8_t abs1[255 + 255 + 1]; // abs(i)>>1
24 static int8_t sclip1[1020 + 1020 + 1]; // clips [-1020, 1020] to [-128, 127]
25 static int8_t sclip2[112 + 112 + 1]; // clips [-112, 112] to [-16, 15]
26 static uint8_t clip1[255 + 510 + 1]; // clips [-255,510] to [0,255]
27
28 // We declare this variable 'volatile' to prevent instruction reordering
29 // and make sure it's set to true _last_ (so as to be thread-safe)
30 static volatile int tables_ok = 0;
31
DspInitTables(void)32 static void DspInitTables(void) {
33 if (!tables_ok) {
34 int i;
35 for (i = -255; i <= 255; ++i) {
36 abs0[255 + i] = (i < 0) ? -i : i;
37 abs1[255 + i] = abs0[255 + i] >> 1;
38 }
39 for (i = -1020; i <= 1020; ++i) {
40 sclip1[1020 + i] = (i < -128) ? -128 : (i > 127) ? 127 : i;
41 }
42 for (i = -112; i <= 112; ++i) {
43 sclip2[112 + i] = (i < -16) ? -16 : (i > 15) ? 15 : i;
44 }
45 for (i = -255; i <= 255 + 255; ++i) {
46 clip1[255 + i] = (i < 0) ? 0 : (i > 255) ? 255 : i;
47 }
48 tables_ok = 1;
49 }
50 }
51
clip_8b(int v)52 static WEBP_INLINE uint8_t clip_8b(int v) {
53 return (!(v & ~0xff)) ? v : (v < 0) ? 0 : 255;
54 }
55
56 //------------------------------------------------------------------------------
57 // Transforms (Paragraph 14.4)
58
59 #define STORE(x, y, v) \
60 dst[x + y * BPS] = clip_8b(dst[x + y * BPS] + ((v) >> 3))
61
62 static const int kC1 = 20091 + (1 << 16);
63 static const int kC2 = 35468;
64 #define MUL(a, b) (((a) * (b)) >> 16)
65
TransformOne(const int16_t * in,uint8_t * dst)66 static void TransformOne(const int16_t* in, uint8_t* dst) {
67 int C[4 * 4], *tmp;
68 int i;
69 tmp = C;
70 for (i = 0; i < 4; ++i) { // vertical pass
71 const int a = in[0] + in[8]; // [-4096, 4094]
72 const int b = in[0] - in[8]; // [-4095, 4095]
73 const int c = MUL(in[4], kC2) - MUL(in[12], kC1); // [-3783, 3783]
74 const int d = MUL(in[4], kC1) + MUL(in[12], kC2); // [-3785, 3781]
75 tmp[0] = a + d; // [-7881, 7875]
76 tmp[1] = b + c; // [-7878, 7878]
77 tmp[2] = b - c; // [-7878, 7878]
78 tmp[3] = a - d; // [-7877, 7879]
79 tmp += 4;
80 in++;
81 }
82 // Each pass is expanding the dynamic range by ~3.85 (upper bound).
83 // The exact value is (2. + (kC1 + kC2) / 65536).
84 // After the second pass, maximum interval is [-3794, 3794], assuming
85 // an input in [-2048, 2047] interval. We then need to add a dst value
86 // in the [0, 255] range.
87 // In the worst case scenario, the input to clip_8b() can be as large as
88 // [-60713, 60968].
89 tmp = C;
90 for (i = 0; i < 4; ++i) { // horizontal pass
91 const int dc = tmp[0] + 4;
92 const int a = dc + tmp[8];
93 const int b = dc - tmp[8];
94 const int c = MUL(tmp[4], kC2) - MUL(tmp[12], kC1);
95 const int d = MUL(tmp[4], kC1) + MUL(tmp[12], kC2);
96 STORE(0, 0, a + d);
97 STORE(1, 0, b + c);
98 STORE(2, 0, b - c);
99 STORE(3, 0, a - d);
100 tmp++;
101 dst += BPS;
102 }
103 }
104 #undef MUL
105
TransformTwo(const int16_t * in,uint8_t * dst,int do_two)106 static void TransformTwo(const int16_t* in, uint8_t* dst, int do_two) {
107 TransformOne(in, dst);
108 if (do_two) {
109 TransformOne(in + 16, dst + 4);
110 }
111 }
112
TransformUV(const int16_t * in,uint8_t * dst)113 static void TransformUV(const int16_t* in, uint8_t* dst) {
114 VP8Transform(in + 0 * 16, dst, 1);
115 VP8Transform(in + 2 * 16, dst + 4 * BPS, 1);
116 }
117
TransformDC(const int16_t * in,uint8_t * dst)118 static void TransformDC(const int16_t *in, uint8_t* dst) {
119 const int DC = in[0] + 4;
120 int i, j;
121 for (j = 0; j < 4; ++j) {
122 for (i = 0; i < 4; ++i) {
123 STORE(i, j, DC);
124 }
125 }
126 }
127
TransformDCUV(const int16_t * in,uint8_t * dst)128 static void TransformDCUV(const int16_t* in, uint8_t* dst) {
129 if (in[0 * 16]) TransformDC(in + 0 * 16, dst);
130 if (in[1 * 16]) TransformDC(in + 1 * 16, dst + 4);
131 if (in[2 * 16]) TransformDC(in + 2 * 16, dst + 4 * BPS);
132 if (in[3 * 16]) TransformDC(in + 3 * 16, dst + 4 * BPS + 4);
133 }
134
135 #undef STORE
136
137 //------------------------------------------------------------------------------
138 // Paragraph 14.3
139
TransformWHT(const int16_t * in,int16_t * out)140 static void TransformWHT(const int16_t* in, int16_t* out) {
141 int tmp[16];
142 int i;
143 for (i = 0; i < 4; ++i) {
144 const int a0 = in[0 + i] + in[12 + i];
145 const int a1 = in[4 + i] + in[ 8 + i];
146 const int a2 = in[4 + i] - in[ 8 + i];
147 const int a3 = in[0 + i] - in[12 + i];
148 tmp[0 + i] = a0 + a1;
149 tmp[8 + i] = a0 - a1;
150 tmp[4 + i] = a3 + a2;
151 tmp[12 + i] = a3 - a2;
152 }
153 for (i = 0; i < 4; ++i) {
154 const int dc = tmp[0 + i * 4] + 3; // w/ rounder
155 const int a0 = dc + tmp[3 + i * 4];
156 const int a1 = tmp[1 + i * 4] + tmp[2 + i * 4];
157 const int a2 = tmp[1 + i * 4] - tmp[2 + i * 4];
158 const int a3 = dc - tmp[3 + i * 4];
159 out[ 0] = (a0 + a1) >> 3;
160 out[16] = (a3 + a2) >> 3;
161 out[32] = (a0 - a1) >> 3;
162 out[48] = (a3 - a2) >> 3;
163 out += 64;
164 }
165 }
166
167 void (*VP8TransformWHT)(const int16_t* in, int16_t* out) = TransformWHT;
168
169 //------------------------------------------------------------------------------
170 // Intra predictions
171
172 #define DST(x, y) dst[(x) + (y) * BPS]
173
TrueMotion(uint8_t * dst,int size)174 static WEBP_INLINE void TrueMotion(uint8_t *dst, int size) {
175 const uint8_t* top = dst - BPS;
176 const uint8_t* const clip0 = clip1 + 255 - top[-1];
177 int y;
178 for (y = 0; y < size; ++y) {
179 const uint8_t* const clip = clip0 + dst[-1];
180 int x;
181 for (x = 0; x < size; ++x) {
182 dst[x] = clip[top[x]];
183 }
184 dst += BPS;
185 }
186 }
TM4(uint8_t * dst)187 static void TM4(uint8_t *dst) { TrueMotion(dst, 4); }
TM8uv(uint8_t * dst)188 static void TM8uv(uint8_t *dst) { TrueMotion(dst, 8); }
TM16(uint8_t * dst)189 static void TM16(uint8_t *dst) { TrueMotion(dst, 16); }
190
191 //------------------------------------------------------------------------------
192 // 16x16
193
VE16(uint8_t * dst)194 static void VE16(uint8_t *dst) { // vertical
195 int j;
196 for (j = 0; j < 16; ++j) {
197 memcpy(dst + j * BPS, dst - BPS, 16);
198 }
199 }
200
HE16(uint8_t * dst)201 static void HE16(uint8_t *dst) { // horizontal
202 int j;
203 for (j = 16; j > 0; --j) {
204 memset(dst, dst[-1], 16);
205 dst += BPS;
206 }
207 }
208
Put16(int v,uint8_t * dst)209 static WEBP_INLINE void Put16(int v, uint8_t* dst) {
210 int j;
211 for (j = 0; j < 16; ++j) {
212 memset(dst + j * BPS, v, 16);
213 }
214 }
215
DC16(uint8_t * dst)216 static void DC16(uint8_t *dst) { // DC
217 int DC = 16;
218 int j;
219 for (j = 0; j < 16; ++j) {
220 DC += dst[-1 + j * BPS] + dst[j - BPS];
221 }
222 Put16(DC >> 5, dst);
223 }
224
DC16NoTop(uint8_t * dst)225 static void DC16NoTop(uint8_t *dst) { // DC with top samples not available
226 int DC = 8;
227 int j;
228 for (j = 0; j < 16; ++j) {
229 DC += dst[-1 + j * BPS];
230 }
231 Put16(DC >> 4, dst);
232 }
233
DC16NoLeft(uint8_t * dst)234 static void DC16NoLeft(uint8_t *dst) { // DC with left samples not available
235 int DC = 8;
236 int i;
237 for (i = 0; i < 16; ++i) {
238 DC += dst[i - BPS];
239 }
240 Put16(DC >> 4, dst);
241 }
242
DC16NoTopLeft(uint8_t * dst)243 static void DC16NoTopLeft(uint8_t *dst) { // DC with no top and left samples
244 Put16(0x80, dst);
245 }
246
247 //------------------------------------------------------------------------------
248 // 4x4
249
250 #define AVG3(a, b, c) (((a) + 2 * (b) + (c) + 2) >> 2)
251 #define AVG2(a, b) (((a) + (b) + 1) >> 1)
252
VE4(uint8_t * dst)253 static void VE4(uint8_t *dst) { // vertical
254 const uint8_t* top = dst - BPS;
255 const uint8_t vals[4] = {
256 AVG3(top[-1], top[0], top[1]),
257 AVG3(top[ 0], top[1], top[2]),
258 AVG3(top[ 1], top[2], top[3]),
259 AVG3(top[ 2], top[3], top[4])
260 };
261 int i;
262 for (i = 0; i < 4; ++i) {
263 memcpy(dst + i * BPS, vals, sizeof(vals));
264 }
265 }
266
HE4(uint8_t * dst)267 static void HE4(uint8_t *dst) { // horizontal
268 const int A = dst[-1 - BPS];
269 const int B = dst[-1];
270 const int C = dst[-1 + BPS];
271 const int D = dst[-1 + 2 * BPS];
272 const int E = dst[-1 + 3 * BPS];
273 *(uint32_t*)(dst + 0 * BPS) = 0x01010101U * AVG3(A, B, C);
274 *(uint32_t*)(dst + 1 * BPS) = 0x01010101U * AVG3(B, C, D);
275 *(uint32_t*)(dst + 2 * BPS) = 0x01010101U * AVG3(C, D, E);
276 *(uint32_t*)(dst + 3 * BPS) = 0x01010101U * AVG3(D, E, E);
277 }
278
DC4(uint8_t * dst)279 static void DC4(uint8_t *dst) { // DC
280 uint32_t dc = 4;
281 int i;
282 for (i = 0; i < 4; ++i) dc += dst[i - BPS] + dst[-1 + i * BPS];
283 dc >>= 3;
284 for (i = 0; i < 4; ++i) memset(dst + i * BPS, dc, 4);
285 }
286
RD4(uint8_t * dst)287 static void RD4(uint8_t *dst) { // Down-right
288 const int I = dst[-1 + 0 * BPS];
289 const int J = dst[-1 + 1 * BPS];
290 const int K = dst[-1 + 2 * BPS];
291 const int L = dst[-1 + 3 * BPS];
292 const int X = dst[-1 - BPS];
293 const int A = dst[0 - BPS];
294 const int B = dst[1 - BPS];
295 const int C = dst[2 - BPS];
296 const int D = dst[3 - BPS];
297 DST(0, 3) = AVG3(J, K, L);
298 DST(0, 2) = DST(1, 3) = AVG3(I, J, K);
299 DST(0, 1) = DST(1, 2) = DST(2, 3) = AVG3(X, I, J);
300 DST(0, 0) = DST(1, 1) = DST(2, 2) = DST(3, 3) = AVG3(A, X, I);
301 DST(1, 0) = DST(2, 1) = DST(3, 2) = AVG3(B, A, X);
302 DST(2, 0) = DST(3, 1) = AVG3(C, B, A);
303 DST(3, 0) = AVG3(D, C, B);
304 }
305
LD4(uint8_t * dst)306 static void LD4(uint8_t *dst) { // Down-Left
307 const int A = dst[0 - BPS];
308 const int B = dst[1 - BPS];
309 const int C = dst[2 - BPS];
310 const int D = dst[3 - BPS];
311 const int E = dst[4 - BPS];
312 const int F = dst[5 - BPS];
313 const int G = dst[6 - BPS];
314 const int H = dst[7 - BPS];
315 DST(0, 0) = AVG3(A, B, C);
316 DST(1, 0) = DST(0, 1) = AVG3(B, C, D);
317 DST(2, 0) = DST(1, 1) = DST(0, 2) = AVG3(C, D, E);
318 DST(3, 0) = DST(2, 1) = DST(1, 2) = DST(0, 3) = AVG3(D, E, F);
319 DST(3, 1) = DST(2, 2) = DST(1, 3) = AVG3(E, F, G);
320 DST(3, 2) = DST(2, 3) = AVG3(F, G, H);
321 DST(3, 3) = AVG3(G, H, H);
322 }
323
VR4(uint8_t * dst)324 static void VR4(uint8_t *dst) { // Vertical-Right
325 const int I = dst[-1 + 0 * BPS];
326 const int J = dst[-1 + 1 * BPS];
327 const int K = dst[-1 + 2 * BPS];
328 const int X = dst[-1 - BPS];
329 const int A = dst[0 - BPS];
330 const int B = dst[1 - BPS];
331 const int C = dst[2 - BPS];
332 const int D = dst[3 - BPS];
333 DST(0, 0) = DST(1, 2) = AVG2(X, A);
334 DST(1, 0) = DST(2, 2) = AVG2(A, B);
335 DST(2, 0) = DST(3, 2) = AVG2(B, C);
336 DST(3, 0) = AVG2(C, D);
337
338 DST(0, 3) = AVG3(K, J, I);
339 DST(0, 2) = AVG3(J, I, X);
340 DST(0, 1) = DST(1, 3) = AVG3(I, X, A);
341 DST(1, 1) = DST(2, 3) = AVG3(X, A, B);
342 DST(2, 1) = DST(3, 3) = AVG3(A, B, C);
343 DST(3, 1) = AVG3(B, C, D);
344 }
345
VL4(uint8_t * dst)346 static void VL4(uint8_t *dst) { // Vertical-Left
347 const int A = dst[0 - BPS];
348 const int B = dst[1 - BPS];
349 const int C = dst[2 - BPS];
350 const int D = dst[3 - BPS];
351 const int E = dst[4 - BPS];
352 const int F = dst[5 - BPS];
353 const int G = dst[6 - BPS];
354 const int H = dst[7 - BPS];
355 DST(0, 0) = AVG2(A, B);
356 DST(1, 0) = DST(0, 2) = AVG2(B, C);
357 DST(2, 0) = DST(1, 2) = AVG2(C, D);
358 DST(3, 0) = DST(2, 2) = AVG2(D, E);
359
360 DST(0, 1) = AVG3(A, B, C);
361 DST(1, 1) = DST(0, 3) = AVG3(B, C, D);
362 DST(2, 1) = DST(1, 3) = AVG3(C, D, E);
363 DST(3, 1) = DST(2, 3) = AVG3(D, E, F);
364 DST(3, 2) = AVG3(E, F, G);
365 DST(3, 3) = AVG3(F, G, H);
366 }
367
HU4(uint8_t * dst)368 static void HU4(uint8_t *dst) { // Horizontal-Up
369 const int I = dst[-1 + 0 * BPS];
370 const int J = dst[-1 + 1 * BPS];
371 const int K = dst[-1 + 2 * BPS];
372 const int L = dst[-1 + 3 * BPS];
373 DST(0, 0) = AVG2(I, J);
374 DST(2, 0) = DST(0, 1) = AVG2(J, K);
375 DST(2, 1) = DST(0, 2) = AVG2(K, L);
376 DST(1, 0) = AVG3(I, J, K);
377 DST(3, 0) = DST(1, 1) = AVG3(J, K, L);
378 DST(3, 1) = DST(1, 2) = AVG3(K, L, L);
379 DST(3, 2) = DST(2, 2) =
380 DST(0, 3) = DST(1, 3) = DST(2, 3) = DST(3, 3) = L;
381 }
382
HD4(uint8_t * dst)383 static void HD4(uint8_t *dst) { // Horizontal-Down
384 const int I = dst[-1 + 0 * BPS];
385 const int J = dst[-1 + 1 * BPS];
386 const int K = dst[-1 + 2 * BPS];
387 const int L = dst[-1 + 3 * BPS];
388 const int X = dst[-1 - BPS];
389 const int A = dst[0 - BPS];
390 const int B = dst[1 - BPS];
391 const int C = dst[2 - BPS];
392
393 DST(0, 0) = DST(2, 1) = AVG2(I, X);
394 DST(0, 1) = DST(2, 2) = AVG2(J, I);
395 DST(0, 2) = DST(2, 3) = AVG2(K, J);
396 DST(0, 3) = AVG2(L, K);
397
398 DST(3, 0) = AVG3(A, B, C);
399 DST(2, 0) = AVG3(X, A, B);
400 DST(1, 0) = DST(3, 1) = AVG3(I, X, A);
401 DST(1, 1) = DST(3, 2) = AVG3(J, I, X);
402 DST(1, 2) = DST(3, 3) = AVG3(K, J, I);
403 DST(1, 3) = AVG3(L, K, J);
404 }
405
406 #undef DST
407 #undef AVG3
408 #undef AVG2
409
410 //------------------------------------------------------------------------------
411 // Chroma
412
VE8uv(uint8_t * dst)413 static void VE8uv(uint8_t *dst) { // vertical
414 int j;
415 for (j = 0; j < 8; ++j) {
416 memcpy(dst + j * BPS, dst - BPS, 8);
417 }
418 }
419
HE8uv(uint8_t * dst)420 static void HE8uv(uint8_t *dst) { // horizontal
421 int j;
422 for (j = 0; j < 8; ++j) {
423 memset(dst, dst[-1], 8);
424 dst += BPS;
425 }
426 }
427
428 // helper for chroma-DC predictions
Put8x8uv(uint64_t v,uint8_t * dst)429 static WEBP_INLINE void Put8x8uv(uint64_t v, uint8_t* dst) {
430 int j;
431 for (j = 0; j < 8; ++j) {
432 *(uint64_t*)(dst + j * BPS) = v;
433 }
434 }
435
DC8uv(uint8_t * dst)436 static void DC8uv(uint8_t *dst) { // DC
437 int dc0 = 8;
438 int i;
439 for (i = 0; i < 8; ++i) {
440 dc0 += dst[i - BPS] + dst[-1 + i * BPS];
441 }
442 Put8x8uv((uint64_t)((dc0 >> 4) * 0x0101010101010101ULL), dst);
443 }
444
DC8uvNoLeft(uint8_t * dst)445 static void DC8uvNoLeft(uint8_t *dst) { // DC with no left samples
446 int dc0 = 4;
447 int i;
448 for (i = 0; i < 8; ++i) {
449 dc0 += dst[i - BPS];
450 }
451 Put8x8uv((uint64_t)((dc0 >> 3) * 0x0101010101010101ULL), dst);
452 }
453
DC8uvNoTop(uint8_t * dst)454 static void DC8uvNoTop(uint8_t *dst) { // DC with no top samples
455 int dc0 = 4;
456 int i;
457 for (i = 0; i < 8; ++i) {
458 dc0 += dst[-1 + i * BPS];
459 }
460 Put8x8uv((uint64_t)((dc0 >> 3) * 0x0101010101010101ULL), dst);
461 }
462
DC8uvNoTopLeft(uint8_t * dst)463 static void DC8uvNoTopLeft(uint8_t *dst) { // DC with nothing
464 Put8x8uv(0x8080808080808080ULL, dst);
465 }
466
467 //------------------------------------------------------------------------------
468 // default C implementations
469
470 const VP8PredFunc VP8PredLuma4[NUM_BMODES] = {
471 DC4, TM4, VE4, HE4, RD4, VR4, LD4, VL4, HD4, HU4
472 };
473
474 const VP8PredFunc VP8PredLuma16[NUM_B_DC_MODES] = {
475 DC16, TM16, VE16, HE16,
476 DC16NoTop, DC16NoLeft, DC16NoTopLeft
477 };
478
479 const VP8PredFunc VP8PredChroma8[NUM_B_DC_MODES] = {
480 DC8uv, TM8uv, VE8uv, HE8uv,
481 DC8uvNoTop, DC8uvNoLeft, DC8uvNoTopLeft
482 };
483
484 //------------------------------------------------------------------------------
485 // Edge filtering functions
486
487 // 4 pixels in, 2 pixels out
do_filter2(uint8_t * p,int step)488 static WEBP_INLINE void do_filter2(uint8_t* p, int step) {
489 const int p1 = p[-2*step], p0 = p[-step], q0 = p[0], q1 = p[step];
490 const int a = 3 * (q0 - p0) + sclip1[1020 + p1 - q1];
491 const int a1 = sclip2[112 + ((a + 4) >> 3)];
492 const int a2 = sclip2[112 + ((a + 3) >> 3)];
493 p[-step] = clip1[255 + p0 + a2];
494 p[ 0] = clip1[255 + q0 - a1];
495 }
496
497 // 4 pixels in, 4 pixels out
do_filter4(uint8_t * p,int step)498 static WEBP_INLINE void do_filter4(uint8_t* p, int step) {
499 const int p1 = p[-2*step], p0 = p[-step], q0 = p[0], q1 = p[step];
500 const int a = 3 * (q0 - p0);
501 const int a1 = sclip2[112 + ((a + 4) >> 3)];
502 const int a2 = sclip2[112 + ((a + 3) >> 3)];
503 const int a3 = (a1 + 1) >> 1;
504 p[-2*step] = clip1[255 + p1 + a3];
505 p[- step] = clip1[255 + p0 + a2];
506 p[ 0] = clip1[255 + q0 - a1];
507 p[ step] = clip1[255 + q1 - a3];
508 }
509
510 // 6 pixels in, 6 pixels out
do_filter6(uint8_t * p,int step)511 static WEBP_INLINE void do_filter6(uint8_t* p, int step) {
512 const int p2 = p[-3*step], p1 = p[-2*step], p0 = p[-step];
513 const int q0 = p[0], q1 = p[step], q2 = p[2*step];
514 const int a = sclip1[1020 + 3 * (q0 - p0) + sclip1[1020 + p1 - q1]];
515 const int a1 = (27 * a + 63) >> 7; // eq. to ((3 * a + 7) * 9) >> 7
516 const int a2 = (18 * a + 63) >> 7; // eq. to ((2 * a + 7) * 9) >> 7
517 const int a3 = (9 * a + 63) >> 7; // eq. to ((1 * a + 7) * 9) >> 7
518 p[-3*step] = clip1[255 + p2 + a3];
519 p[-2*step] = clip1[255 + p1 + a2];
520 p[- step] = clip1[255 + p0 + a1];
521 p[ 0] = clip1[255 + q0 - a1];
522 p[ step] = clip1[255 + q1 - a2];
523 p[ 2*step] = clip1[255 + q2 - a3];
524 }
525
hev(const uint8_t * p,int step,int thresh)526 static WEBP_INLINE int hev(const uint8_t* p, int step, int thresh) {
527 const int p1 = p[-2*step], p0 = p[-step], q0 = p[0], q1 = p[step];
528 return (abs0[255 + p1 - p0] > thresh) || (abs0[255 + q1 - q0] > thresh);
529 }
530
needs_filter(const uint8_t * p,int step,int thresh)531 static WEBP_INLINE int needs_filter(const uint8_t* p, int step, int thresh) {
532 const int p1 = p[-2*step], p0 = p[-step], q0 = p[0], q1 = p[step];
533 return (2 * abs0[255 + p0 - q0] + abs1[255 + p1 - q1]) <= thresh;
534 }
535
needs_filter2(const uint8_t * p,int step,int t,int it)536 static WEBP_INLINE int needs_filter2(const uint8_t* p,
537 int step, int t, int it) {
538 const int p3 = p[-4*step], p2 = p[-3*step], p1 = p[-2*step], p0 = p[-step];
539 const int q0 = p[0], q1 = p[step], q2 = p[2*step], q3 = p[3*step];
540 if ((2 * abs0[255 + p0 - q0] + abs1[255 + p1 - q1]) > t)
541 return 0;
542 return abs0[255 + p3 - p2] <= it && abs0[255 + p2 - p1] <= it &&
543 abs0[255 + p1 - p0] <= it && abs0[255 + q3 - q2] <= it &&
544 abs0[255 + q2 - q1] <= it && abs0[255 + q1 - q0] <= it;
545 }
546
547 //------------------------------------------------------------------------------
548 // Simple In-loop filtering (Paragraph 15.2)
549
SimpleVFilter16(uint8_t * p,int stride,int thresh)550 static void SimpleVFilter16(uint8_t* p, int stride, int thresh) {
551 int i;
552 for (i = 0; i < 16; ++i) {
553 if (needs_filter(p + i, stride, thresh)) {
554 do_filter2(p + i, stride);
555 }
556 }
557 }
558
SimpleHFilter16(uint8_t * p,int stride,int thresh)559 static void SimpleHFilter16(uint8_t* p, int stride, int thresh) {
560 int i;
561 for (i = 0; i < 16; ++i) {
562 if (needs_filter(p + i * stride, 1, thresh)) {
563 do_filter2(p + i * stride, 1);
564 }
565 }
566 }
567
SimpleVFilter16i(uint8_t * p,int stride,int thresh)568 static void SimpleVFilter16i(uint8_t* p, int stride, int thresh) {
569 int k;
570 for (k = 3; k > 0; --k) {
571 p += 4 * stride;
572 SimpleVFilter16(p, stride, thresh);
573 }
574 }
575
SimpleHFilter16i(uint8_t * p,int stride,int thresh)576 static void SimpleHFilter16i(uint8_t* p, int stride, int thresh) {
577 int k;
578 for (k = 3; k > 0; --k) {
579 p += 4;
580 SimpleHFilter16(p, stride, thresh);
581 }
582 }
583
584 //------------------------------------------------------------------------------
585 // Complex In-loop filtering (Paragraph 15.3)
586
FilterLoop26(uint8_t * p,int hstride,int vstride,int size,int thresh,int ithresh,int hev_thresh)587 static WEBP_INLINE void FilterLoop26(uint8_t* p,
588 int hstride, int vstride, int size,
589 int thresh, int ithresh, int hev_thresh) {
590 while (size-- > 0) {
591 if (needs_filter2(p, hstride, thresh, ithresh)) {
592 if (hev(p, hstride, hev_thresh)) {
593 do_filter2(p, hstride);
594 } else {
595 do_filter6(p, hstride);
596 }
597 }
598 p += vstride;
599 }
600 }
601
FilterLoop24(uint8_t * p,int hstride,int vstride,int size,int thresh,int ithresh,int hev_thresh)602 static WEBP_INLINE void FilterLoop24(uint8_t* p,
603 int hstride, int vstride, int size,
604 int thresh, int ithresh, int hev_thresh) {
605 while (size-- > 0) {
606 if (needs_filter2(p, hstride, thresh, ithresh)) {
607 if (hev(p, hstride, hev_thresh)) {
608 do_filter2(p, hstride);
609 } else {
610 do_filter4(p, hstride);
611 }
612 }
613 p += vstride;
614 }
615 }
616
617 // on macroblock edges
VFilter16(uint8_t * p,int stride,int thresh,int ithresh,int hev_thresh)618 static void VFilter16(uint8_t* p, int stride,
619 int thresh, int ithresh, int hev_thresh) {
620 FilterLoop26(p, stride, 1, 16, thresh, ithresh, hev_thresh);
621 }
622
HFilter16(uint8_t * p,int stride,int thresh,int ithresh,int hev_thresh)623 static void HFilter16(uint8_t* p, int stride,
624 int thresh, int ithresh, int hev_thresh) {
625 FilterLoop26(p, 1, stride, 16, thresh, ithresh, hev_thresh);
626 }
627
628 // on three inner edges
VFilter16i(uint8_t * p,int stride,int thresh,int ithresh,int hev_thresh)629 static void VFilter16i(uint8_t* p, int stride,
630 int thresh, int ithresh, int hev_thresh) {
631 int k;
632 for (k = 3; k > 0; --k) {
633 p += 4 * stride;
634 FilterLoop24(p, stride, 1, 16, thresh, ithresh, hev_thresh);
635 }
636 }
637
HFilter16i(uint8_t * p,int stride,int thresh,int ithresh,int hev_thresh)638 static void HFilter16i(uint8_t* p, int stride,
639 int thresh, int ithresh, int hev_thresh) {
640 int k;
641 for (k = 3; k > 0; --k) {
642 p += 4;
643 FilterLoop24(p, 1, stride, 16, thresh, ithresh, hev_thresh);
644 }
645 }
646
647 // 8-pixels wide variant, for chroma filtering
VFilter8(uint8_t * u,uint8_t * v,int stride,int thresh,int ithresh,int hev_thresh)648 static void VFilter8(uint8_t* u, uint8_t* v, int stride,
649 int thresh, int ithresh, int hev_thresh) {
650 FilterLoop26(u, stride, 1, 8, thresh, ithresh, hev_thresh);
651 FilterLoop26(v, stride, 1, 8, thresh, ithresh, hev_thresh);
652 }
653
HFilter8(uint8_t * u,uint8_t * v,int stride,int thresh,int ithresh,int hev_thresh)654 static void HFilter8(uint8_t* u, uint8_t* v, int stride,
655 int thresh, int ithresh, int hev_thresh) {
656 FilterLoop26(u, 1, stride, 8, thresh, ithresh, hev_thresh);
657 FilterLoop26(v, 1, stride, 8, thresh, ithresh, hev_thresh);
658 }
659
VFilter8i(uint8_t * u,uint8_t * v,int stride,int thresh,int ithresh,int hev_thresh)660 static void VFilter8i(uint8_t* u, uint8_t* v, int stride,
661 int thresh, int ithresh, int hev_thresh) {
662 FilterLoop24(u + 4 * stride, stride, 1, 8, thresh, ithresh, hev_thresh);
663 FilterLoop24(v + 4 * stride, stride, 1, 8, thresh, ithresh, hev_thresh);
664 }
665
HFilter8i(uint8_t * u,uint8_t * v,int stride,int thresh,int ithresh,int hev_thresh)666 static void HFilter8i(uint8_t* u, uint8_t* v, int stride,
667 int thresh, int ithresh, int hev_thresh) {
668 FilterLoop24(u + 4, 1, stride, 8, thresh, ithresh, hev_thresh);
669 FilterLoop24(v + 4, 1, stride, 8, thresh, ithresh, hev_thresh);
670 }
671
672 //------------------------------------------------------------------------------
673
674 VP8DecIdct2 VP8Transform;
675 VP8DecIdct VP8TransformUV;
676 VP8DecIdct VP8TransformDC;
677 VP8DecIdct VP8TransformDCUV;
678
679 VP8LumaFilterFunc VP8VFilter16;
680 VP8LumaFilterFunc VP8HFilter16;
681 VP8ChromaFilterFunc VP8VFilter8;
682 VP8ChromaFilterFunc VP8HFilter8;
683 VP8LumaFilterFunc VP8VFilter16i;
684 VP8LumaFilterFunc VP8HFilter16i;
685 VP8ChromaFilterFunc VP8VFilter8i;
686 VP8ChromaFilterFunc VP8HFilter8i;
687 VP8SimpleFilterFunc VP8SimpleVFilter16;
688 VP8SimpleFilterFunc VP8SimpleHFilter16;
689 VP8SimpleFilterFunc VP8SimpleVFilter16i;
690 VP8SimpleFilterFunc VP8SimpleHFilter16i;
691
692 extern void VP8DspInitSSE2(void);
693 extern void VP8DspInitNEON(void);
694
VP8DspInit(void)695 void VP8DspInit(void) {
696 DspInitTables();
697
698 VP8Transform = TransformTwo;
699 VP8TransformUV = TransformUV;
700 VP8TransformDC = TransformDC;
701 VP8TransformDCUV = TransformDCUV;
702
703 VP8VFilter16 = VFilter16;
704 VP8HFilter16 = HFilter16;
705 VP8VFilter8 = VFilter8;
706 VP8HFilter8 = HFilter8;
707 VP8VFilter16i = VFilter16i;
708 VP8HFilter16i = HFilter16i;
709 VP8VFilter8i = VFilter8i;
710 VP8HFilter8i = HFilter8i;
711 VP8SimpleVFilter16 = SimpleVFilter16;
712 VP8SimpleHFilter16 = SimpleHFilter16;
713 VP8SimpleVFilter16i = SimpleVFilter16i;
714 VP8SimpleHFilter16i = SimpleHFilter16i;
715
716 // If defined, use CPUInfo() to overwrite some pointers with faster versions.
717 if (VP8GetCPUInfo) {
718 #if defined(WEBP_USE_SSE2)
719 if (VP8GetCPUInfo(kSSE2)) {
720 VP8DspInitSSE2();
721 }
722 #elif defined(WEBP_USE_NEON)
723 if (VP8GetCPUInfo(kNEON)) {
724 VP8DspInitNEON();
725 }
726 #endif
727 }
728 }
729
730 #if defined(__cplusplus) || defined(c_plusplus)
731 } // extern "C"
732 #endif
733