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 ([email protected])
13
14 #include <assert.h>
15
16 #include "src/dsp/dsp.h"
17 #include "src/dec/vp8i_dec.h"
18 #include "src/utils/utils.h"
19
20 //------------------------------------------------------------------------------
21
clip_8b(int v)22 static WEBP_INLINE uint8_t clip_8b(int v) {
23 return (!(v & ~0xff)) ? v : (v < 0) ? 0 : 255;
24 }
25
26 //------------------------------------------------------------------------------
27 // Transforms (Paragraph 14.4)
28
29 #define STORE(x, y, v) \
30 dst[(x) + (y) * BPS] = clip_8b(dst[(x) + (y) * BPS] + ((v) >> 3))
31
32 #define STORE2(y, dc, d, c) do { \
33 const int DC = (dc); \
34 STORE(0, y, DC + (d)); \
35 STORE(1, y, DC + (c)); \
36 STORE(2, y, DC - (c)); \
37 STORE(3, y, DC - (d)); \
38 } while (0)
39
40 #if !WEBP_NEON_OMIT_C_CODE
TransformOne_C(const int16_t * in,uint8_t * dst)41 static void TransformOne_C(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 = WEBP_TRANSFORM_AC3_MUL2(in[4]) -
49 WEBP_TRANSFORM_AC3_MUL1(in[12]); // [-3783, 3783]
50 const int d = WEBP_TRANSFORM_AC3_MUL1(in[4]) +
51 WEBP_TRANSFORM_AC3_MUL2(in[12]); // [-3785, 3781]
52 tmp[0] = a + d; // [-7881, 7875]
53 tmp[1] = b + c; // [-7878, 7878]
54 tmp[2] = b - c; // [-7878, 7878]
55 tmp[3] = a - d; // [-7877, 7879]
56 tmp += 4;
57 in++;
58 }
59 // Each pass is expanding the dynamic range by ~3.85 (upper bound).
60 // The exact value is (2. + (20091 + 35468) / 65536).
61 // After the second pass, maximum interval is [-3794, 3794], assuming
62 // an input in [-2048, 2047] interval. We then need to add a dst value
63 // in the [0, 255] range.
64 // In the worst case scenario, the input to clip_8b() can be as large as
65 // [-60713, 60968].
66 tmp = C;
67 for (i = 0; i < 4; ++i) { // horizontal pass
68 const int dc = tmp[0] + 4;
69 const int a = dc + tmp[8];
70 const int b = dc - tmp[8];
71 const int c =
72 WEBP_TRANSFORM_AC3_MUL2(tmp[4]) - WEBP_TRANSFORM_AC3_MUL1(tmp[12]);
73 const int d =
74 WEBP_TRANSFORM_AC3_MUL1(tmp[4]) + WEBP_TRANSFORM_AC3_MUL2(tmp[12]);
75 STORE(0, 0, a + d);
76 STORE(1, 0, b + c);
77 STORE(2, 0, b - c);
78 STORE(3, 0, a - d);
79 tmp++;
80 dst += BPS;
81 }
82 }
83
84 // Simplified transform when only in[0], in[1] and in[4] are non-zero
TransformAC3_C(const int16_t * in,uint8_t * dst)85 static void TransformAC3_C(const int16_t* in, uint8_t* dst) {
86 const int a = in[0] + 4;
87 const int c4 = WEBP_TRANSFORM_AC3_MUL2(in[4]);
88 const int d4 = WEBP_TRANSFORM_AC3_MUL1(in[4]);
89 const int c1 = WEBP_TRANSFORM_AC3_MUL2(in[1]);
90 const int d1 = WEBP_TRANSFORM_AC3_MUL1(in[1]);
91 STORE2(0, a + d4, d1, c1);
92 STORE2(1, a + c4, d1, c1);
93 STORE2(2, a - c4, d1, c1);
94 STORE2(3, a - d4, d1, c1);
95 }
96 #undef STORE2
97
TransformTwo_C(const int16_t * in,uint8_t * dst,int do_two)98 static void TransformTwo_C(const int16_t* in, uint8_t* dst, int do_two) {
99 TransformOne_C(in, dst);
100 if (do_two) {
101 TransformOne_C(in + 16, dst + 4);
102 }
103 }
104 #endif // !WEBP_NEON_OMIT_C_CODE
105
TransformUV_C(const int16_t * in,uint8_t * dst)106 static void TransformUV_C(const int16_t* in, uint8_t* dst) {
107 VP8Transform(in + 0 * 16, dst, 1);
108 VP8Transform(in + 2 * 16, dst + 4 * BPS, 1);
109 }
110
111 #if !WEBP_NEON_OMIT_C_CODE
TransformDC_C(const int16_t * in,uint8_t * dst)112 static void TransformDC_C(const int16_t* in, uint8_t* dst) {
113 const int DC = in[0] + 4;
114 int i, j;
115 for (j = 0; j < 4; ++j) {
116 for (i = 0; i < 4; ++i) {
117 STORE(i, j, DC);
118 }
119 }
120 }
121 #endif // !WEBP_NEON_OMIT_C_CODE
122
TransformDCUV_C(const int16_t * in,uint8_t * dst)123 static void TransformDCUV_C(const int16_t* in, uint8_t* dst) {
124 if (in[0 * 16]) VP8TransformDC(in + 0 * 16, dst);
125 if (in[1 * 16]) VP8TransformDC(in + 1 * 16, dst + 4);
126 if (in[2 * 16]) VP8TransformDC(in + 2 * 16, dst + 4 * BPS);
127 if (in[3 * 16]) VP8TransformDC(in + 3 * 16, dst + 4 * BPS + 4);
128 }
129
130 #undef STORE
131
132 //------------------------------------------------------------------------------
133 // Paragraph 14.3
134
135 #if !WEBP_NEON_OMIT_C_CODE
TransformWHT_C(const int16_t * in,int16_t * out)136 static void TransformWHT_C(const int16_t* in, int16_t* out) {
137 int tmp[16];
138 int i;
139 for (i = 0; i < 4; ++i) {
140 const int a0 = in[0 + i] + in[12 + i];
141 const int a1 = in[4 + i] + in[ 8 + i];
142 const int a2 = in[4 + i] - in[ 8 + i];
143 const int a3 = in[0 + i] - in[12 + i];
144 tmp[0 + i] = a0 + a1;
145 tmp[8 + i] = a0 - a1;
146 tmp[4 + i] = a3 + a2;
147 tmp[12 + i] = a3 - a2;
148 }
149 for (i = 0; i < 4; ++i) {
150 const int dc = tmp[0 + i * 4] + 3; // w/ rounder
151 const int a0 = dc + tmp[3 + i * 4];
152 const int a1 = tmp[1 + i * 4] + tmp[2 + i * 4];
153 const int a2 = tmp[1 + i * 4] - tmp[2 + i * 4];
154 const int a3 = dc - tmp[3 + i * 4];
155 out[ 0] = (a0 + a1) >> 3;
156 out[16] = (a3 + a2) >> 3;
157 out[32] = (a0 - a1) >> 3;
158 out[48] = (a3 - a2) >> 3;
159 out += 64;
160 }
161 }
162 #endif // !WEBP_NEON_OMIT_C_CODE
163
164 void (*VP8TransformWHT)(const int16_t* in, int16_t* out);
165
166 //------------------------------------------------------------------------------
167 // Intra predictions
168
169 #define DST(x, y) dst[(x) + (y) * BPS]
170
171 #if !WEBP_NEON_OMIT_C_CODE
TrueMotion(uint8_t * dst,int size)172 static WEBP_INLINE void TrueMotion(uint8_t* dst, int size) {
173 const uint8_t* top = dst - BPS;
174 const uint8_t* const clip0 = VP8kclip1 - top[-1];
175 int y;
176 for (y = 0; y < size; ++y) {
177 const uint8_t* const clip = clip0 + dst[-1];
178 int x;
179 for (x = 0; x < size; ++x) {
180 dst[x] = clip[top[x]];
181 }
182 dst += BPS;
183 }
184 }
TM4_C(uint8_t * dst)185 static void TM4_C(uint8_t* dst) { TrueMotion(dst, 4); }
TM8uv_C(uint8_t * dst)186 static void TM8uv_C(uint8_t* dst) { TrueMotion(dst, 8); }
TM16_C(uint8_t * dst)187 static void TM16_C(uint8_t* dst) { TrueMotion(dst, 16); }
188
189 //------------------------------------------------------------------------------
190 // 16x16
191
VE16_C(uint8_t * dst)192 static void VE16_C(uint8_t* dst) { // vertical
193 int j;
194 for (j = 0; j < 16; ++j) {
195 memcpy(dst + j * BPS, dst - BPS, 16);
196 }
197 }
198
HE16_C(uint8_t * dst)199 static void HE16_C(uint8_t* dst) { // horizontal
200 int j;
201 for (j = 16; j > 0; --j) {
202 memset(dst, dst[-1], 16);
203 dst += BPS;
204 }
205 }
206
Put16(int v,uint8_t * dst)207 static WEBP_INLINE void Put16(int v, uint8_t* dst) {
208 int j;
209 for (j = 0; j < 16; ++j) {
210 memset(dst + j * BPS, v, 16);
211 }
212 }
213
DC16_C(uint8_t * dst)214 static void DC16_C(uint8_t* dst) { // DC
215 int DC = 16;
216 int j;
217 for (j = 0; j < 16; ++j) {
218 DC += dst[-1 + j * BPS] + dst[j - BPS];
219 }
220 Put16(DC >> 5, dst);
221 }
222
DC16NoTop_C(uint8_t * dst)223 static void DC16NoTop_C(uint8_t* dst) { // DC with top samples not available
224 int DC = 8;
225 int j;
226 for (j = 0; j < 16; ++j) {
227 DC += dst[-1 + j * BPS];
228 }
229 Put16(DC >> 4, dst);
230 }
231
DC16NoLeft_C(uint8_t * dst)232 static void DC16NoLeft_C(uint8_t* dst) { // DC with left samples not available
233 int DC = 8;
234 int i;
235 for (i = 0; i < 16; ++i) {
236 DC += dst[i - BPS];
237 }
238 Put16(DC >> 4, dst);
239 }
240
DC16NoTopLeft_C(uint8_t * dst)241 static void DC16NoTopLeft_C(uint8_t* dst) { // DC with no top and left samples
242 Put16(0x80, dst);
243 }
244 #endif // !WEBP_NEON_OMIT_C_CODE
245
246 VP8PredFunc VP8PredLuma16[NUM_B_DC_MODES];
247
248 //------------------------------------------------------------------------------
249 // 4x4
250
251 #define AVG3(a, b, c) ((uint8_t)(((a) + 2 * (b) + (c) + 2) >> 2))
252 #define AVG2(a, b) (((a) + (b) + 1) >> 1)
253
254 #if !WEBP_NEON_OMIT_C_CODE
VE4_C(uint8_t * dst)255 static void VE4_C(uint8_t* dst) { // vertical
256 const uint8_t* top = dst - BPS;
257 const uint8_t vals[4] = {
258 AVG3(top[-1], top[0], top[1]),
259 AVG3(top[ 0], top[1], top[2]),
260 AVG3(top[ 1], top[2], top[3]),
261 AVG3(top[ 2], top[3], top[4])
262 };
263 int i;
264 for (i = 0; i < 4; ++i) {
265 memcpy(dst + i * BPS, vals, sizeof(vals));
266 }
267 }
268 #endif // !WEBP_NEON_OMIT_C_CODE
269
HE4_C(uint8_t * dst)270 static void HE4_C(uint8_t* dst) { // horizontal
271 const int A = dst[-1 - BPS];
272 const int B = dst[-1];
273 const int C = dst[-1 + BPS];
274 const int D = dst[-1 + 2 * BPS];
275 const int E = dst[-1 + 3 * BPS];
276 WebPUint32ToMem(dst + 0 * BPS, 0x01010101U * AVG3(A, B, C));
277 WebPUint32ToMem(dst + 1 * BPS, 0x01010101U * AVG3(B, C, D));
278 WebPUint32ToMem(dst + 2 * BPS, 0x01010101U * AVG3(C, D, E));
279 WebPUint32ToMem(dst + 3 * BPS, 0x01010101U * AVG3(D, E, E));
280 }
281
282 #if !WEBP_NEON_OMIT_C_CODE
DC4_C(uint8_t * dst)283 static void DC4_C(uint8_t* dst) { // DC
284 uint32_t dc = 4;
285 int i;
286 for (i = 0; i < 4; ++i) dc += dst[i - BPS] + dst[-1 + i * BPS];
287 dc >>= 3;
288 for (i = 0; i < 4; ++i) memset(dst + i * BPS, dc, 4);
289 }
290
RD4_C(uint8_t * dst)291 static void RD4_C(uint8_t* dst) { // Down-right
292 const int I = dst[-1 + 0 * BPS];
293 const int J = dst[-1 + 1 * BPS];
294 const int K = dst[-1 + 2 * BPS];
295 const int L = dst[-1 + 3 * BPS];
296 const int X = dst[-1 - BPS];
297 const int A = dst[0 - BPS];
298 const int B = dst[1 - BPS];
299 const int C = dst[2 - BPS];
300 const int D = dst[3 - BPS];
301 DST(0, 3) = AVG3(J, K, L);
302 DST(1, 3) = DST(0, 2) = AVG3(I, J, K);
303 DST(2, 3) = DST(1, 2) = DST(0, 1) = AVG3(X, I, J);
304 DST(3, 3) = DST(2, 2) = DST(1, 1) = DST(0, 0) = AVG3(A, X, I);
305 DST(3, 2) = DST(2, 1) = DST(1, 0) = AVG3(B, A, X);
306 DST(3, 1) = DST(2, 0) = AVG3(C, B, A);
307 DST(3, 0) = AVG3(D, C, B);
308 }
309
LD4_C(uint8_t * dst)310 static void LD4_C(uint8_t* dst) { // Down-Left
311 const int A = dst[0 - BPS];
312 const int B = dst[1 - BPS];
313 const int C = dst[2 - BPS];
314 const int D = dst[3 - BPS];
315 const int E = dst[4 - BPS];
316 const int F = dst[5 - BPS];
317 const int G = dst[6 - BPS];
318 const int H = dst[7 - BPS];
319 DST(0, 0) = AVG3(A, B, C);
320 DST(1, 0) = DST(0, 1) = AVG3(B, C, D);
321 DST(2, 0) = DST(1, 1) = DST(0, 2) = AVG3(C, D, E);
322 DST(3, 0) = DST(2, 1) = DST(1, 2) = DST(0, 3) = AVG3(D, E, F);
323 DST(3, 1) = DST(2, 2) = DST(1, 3) = AVG3(E, F, G);
324 DST(3, 2) = DST(2, 3) = AVG3(F, G, H);
325 DST(3, 3) = AVG3(G, H, H);
326 }
327 #endif // !WEBP_NEON_OMIT_C_CODE
328
VR4_C(uint8_t * dst)329 static void VR4_C(uint8_t* dst) { // Vertical-Right
330 const int I = dst[-1 + 0 * BPS];
331 const int J = dst[-1 + 1 * BPS];
332 const int K = dst[-1 + 2 * BPS];
333 const int X = dst[-1 - BPS];
334 const int A = dst[0 - BPS];
335 const int B = dst[1 - BPS];
336 const int C = dst[2 - BPS];
337 const int D = dst[3 - BPS];
338 DST(0, 0) = DST(1, 2) = AVG2(X, A);
339 DST(1, 0) = DST(2, 2) = AVG2(A, B);
340 DST(2, 0) = DST(3, 2) = AVG2(B, C);
341 DST(3, 0) = AVG2(C, D);
342
343 DST(0, 3) = AVG3(K, J, I);
344 DST(0, 2) = AVG3(J, I, X);
345 DST(0, 1) = DST(1, 3) = AVG3(I, X, A);
346 DST(1, 1) = DST(2, 3) = AVG3(X, A, B);
347 DST(2, 1) = DST(3, 3) = AVG3(A, B, C);
348 DST(3, 1) = AVG3(B, C, D);
349 }
350
VL4_C(uint8_t * dst)351 static void VL4_C(uint8_t* dst) { // Vertical-Left
352 const int A = dst[0 - BPS];
353 const int B = dst[1 - BPS];
354 const int C = dst[2 - BPS];
355 const int D = dst[3 - BPS];
356 const int E = dst[4 - BPS];
357 const int F = dst[5 - BPS];
358 const int G = dst[6 - BPS];
359 const int H = dst[7 - BPS];
360 DST(0, 0) = AVG2(A, B);
361 DST(1, 0) = DST(0, 2) = AVG2(B, C);
362 DST(2, 0) = DST(1, 2) = AVG2(C, D);
363 DST(3, 0) = DST(2, 2) = AVG2(D, E);
364
365 DST(0, 1) = AVG3(A, B, C);
366 DST(1, 1) = DST(0, 3) = AVG3(B, C, D);
367 DST(2, 1) = DST(1, 3) = AVG3(C, D, E);
368 DST(3, 1) = DST(2, 3) = AVG3(D, E, F);
369 DST(3, 2) = AVG3(E, F, G);
370 DST(3, 3) = AVG3(F, G, H);
371 }
372
HU4_C(uint8_t * dst)373 static void HU4_C(uint8_t* dst) { // Horizontal-Up
374 const int I = dst[-1 + 0 * BPS];
375 const int J = dst[-1 + 1 * BPS];
376 const int K = dst[-1 + 2 * BPS];
377 const int L = dst[-1 + 3 * BPS];
378 DST(0, 0) = AVG2(I, J);
379 DST(2, 0) = DST(0, 1) = AVG2(J, K);
380 DST(2, 1) = DST(0, 2) = AVG2(K, L);
381 DST(1, 0) = AVG3(I, J, K);
382 DST(3, 0) = DST(1, 1) = AVG3(J, K, L);
383 DST(3, 1) = DST(1, 2) = AVG3(K, L, L);
384 DST(3, 2) = DST(2, 2) =
385 DST(0, 3) = DST(1, 3) = DST(2, 3) = DST(3, 3) = L;
386 }
387
HD4_C(uint8_t * dst)388 static void HD4_C(uint8_t* dst) { // Horizontal-Down
389 const int I = dst[-1 + 0 * BPS];
390 const int J = dst[-1 + 1 * BPS];
391 const int K = dst[-1 + 2 * BPS];
392 const int L = dst[-1 + 3 * BPS];
393 const int X = dst[-1 - BPS];
394 const int A = dst[0 - BPS];
395 const int B = dst[1 - BPS];
396 const int C = dst[2 - BPS];
397
398 DST(0, 0) = DST(2, 1) = AVG2(I, X);
399 DST(0, 1) = DST(2, 2) = AVG2(J, I);
400 DST(0, 2) = DST(2, 3) = AVG2(K, J);
401 DST(0, 3) = AVG2(L, K);
402
403 DST(3, 0) = AVG3(A, B, C);
404 DST(2, 0) = AVG3(X, A, B);
405 DST(1, 0) = DST(3, 1) = AVG3(I, X, A);
406 DST(1, 1) = DST(3, 2) = AVG3(J, I, X);
407 DST(1, 2) = DST(3, 3) = AVG3(K, J, I);
408 DST(1, 3) = AVG3(L, K, J);
409 }
410
411 #undef DST
412 #undef AVG3
413 #undef AVG2
414
415 VP8PredFunc VP8PredLuma4[NUM_BMODES];
416
417 //------------------------------------------------------------------------------
418 // Chroma
419
420 #if !WEBP_NEON_OMIT_C_CODE
VE8uv_C(uint8_t * dst)421 static void VE8uv_C(uint8_t* dst) { // vertical
422 int j;
423 for (j = 0; j < 8; ++j) {
424 memcpy(dst + j * BPS, dst - BPS, 8);
425 }
426 }
427
HE8uv_C(uint8_t * dst)428 static void HE8uv_C(uint8_t* dst) { // horizontal
429 int j;
430 for (j = 0; j < 8; ++j) {
431 memset(dst, dst[-1], 8);
432 dst += BPS;
433 }
434 }
435
436 // helper for chroma-DC predictions
Put8x8uv(uint8_t value,uint8_t * dst)437 static WEBP_INLINE void Put8x8uv(uint8_t value, uint8_t* dst) {
438 int j;
439 for (j = 0; j < 8; ++j) {
440 memset(dst + j * BPS, value, 8);
441 }
442 }
443
DC8uv_C(uint8_t * dst)444 static void DC8uv_C(uint8_t* dst) { // DC
445 int dc0 = 8;
446 int i;
447 for (i = 0; i < 8; ++i) {
448 dc0 += dst[i - BPS] + dst[-1 + i * BPS];
449 }
450 Put8x8uv(dc0 >> 4, dst);
451 }
452
DC8uvNoLeft_C(uint8_t * dst)453 static void DC8uvNoLeft_C(uint8_t* dst) { // DC with no left samples
454 int dc0 = 4;
455 int i;
456 for (i = 0; i < 8; ++i) {
457 dc0 += dst[i - BPS];
458 }
459 Put8x8uv(dc0 >> 3, dst);
460 }
461
DC8uvNoTop_C(uint8_t * dst)462 static void DC8uvNoTop_C(uint8_t* dst) { // DC with no top samples
463 int dc0 = 4;
464 int i;
465 for (i = 0; i < 8; ++i) {
466 dc0 += dst[-1 + i * BPS];
467 }
468 Put8x8uv(dc0 >> 3, dst);
469 }
470
DC8uvNoTopLeft_C(uint8_t * dst)471 static void DC8uvNoTopLeft_C(uint8_t* dst) { // DC with nothing
472 Put8x8uv(0x80, dst);
473 }
474 #endif // !WEBP_NEON_OMIT_C_CODE
475
476 VP8PredFunc VP8PredChroma8[NUM_B_DC_MODES];
477
478 //------------------------------------------------------------------------------
479 // Edge filtering functions
480
481 #if !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
482 // 4 pixels in, 2 pixels out
DoFilter2_C(uint8_t * p,int step)483 static WEBP_INLINE void DoFilter2_C(uint8_t* p, int step) {
484 const int p1 = p[-2*step], p0 = p[-step], q0 = p[0], q1 = p[step];
485 const int a = 3 * (q0 - p0) + VP8ksclip1[p1 - q1]; // in [-893,892]
486 const int a1 = VP8ksclip2[(a + 4) >> 3]; // in [-16,15]
487 const int a2 = VP8ksclip2[(a + 3) >> 3];
488 p[-step] = VP8kclip1[p0 + a2];
489 p[ 0] = VP8kclip1[q0 - a1];
490 }
491
492 // 4 pixels in, 4 pixels out
DoFilter4_C(uint8_t * p,int step)493 static WEBP_INLINE void DoFilter4_C(uint8_t* p, int step) {
494 const int p1 = p[-2*step], p0 = p[-step], q0 = p[0], q1 = p[step];
495 const int a = 3 * (q0 - p0);
496 const int a1 = VP8ksclip2[(a + 4) >> 3];
497 const int a2 = VP8ksclip2[(a + 3) >> 3];
498 const int a3 = (a1 + 1) >> 1;
499 p[-2*step] = VP8kclip1[p1 + a3];
500 p[- step] = VP8kclip1[p0 + a2];
501 p[ 0] = VP8kclip1[q0 - a1];
502 p[ step] = VP8kclip1[q1 - a3];
503 }
504
505 // 6 pixels in, 6 pixels out
DoFilter6_C(uint8_t * p,int step)506 static WEBP_INLINE void DoFilter6_C(uint8_t* p, int step) {
507 const int p2 = p[-3*step], p1 = p[-2*step], p0 = p[-step];
508 const int q0 = p[0], q1 = p[step], q2 = p[2*step];
509 const int a = VP8ksclip1[3 * (q0 - p0) + VP8ksclip1[p1 - q1]];
510 // a is in [-128,127], a1 in [-27,27], a2 in [-18,18] and a3 in [-9,9]
511 const int a1 = (27 * a + 63) >> 7; // eq. to ((3 * a + 7) * 9) >> 7
512 const int a2 = (18 * a + 63) >> 7; // eq. to ((2 * a + 7) * 9) >> 7
513 const int a3 = (9 * a + 63) >> 7; // eq. to ((1 * a + 7) * 9) >> 7
514 p[-3*step] = VP8kclip1[p2 + a3];
515 p[-2*step] = VP8kclip1[p1 + a2];
516 p[- step] = VP8kclip1[p0 + a1];
517 p[ 0] = VP8kclip1[q0 - a1];
518 p[ step] = VP8kclip1[q1 - a2];
519 p[ 2*step] = VP8kclip1[q2 - a3];
520 }
521
Hev(const uint8_t * p,int step,int thresh)522 static WEBP_INLINE int Hev(const uint8_t* p, int step, int thresh) {
523 const int p1 = p[-2*step], p0 = p[-step], q0 = p[0], q1 = p[step];
524 return (VP8kabs0[p1 - p0] > thresh) || (VP8kabs0[q1 - q0] > thresh);
525 }
526 #endif // !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
527
528 #if !WEBP_NEON_OMIT_C_CODE
NeedsFilter_C(const uint8_t * p,int step,int t)529 static WEBP_INLINE int NeedsFilter_C(const uint8_t* p, int step, int t) {
530 const int p1 = p[-2 * step], p0 = p[-step], q0 = p[0], q1 = p[step];
531 return ((4 * VP8kabs0[p0 - q0] + VP8kabs0[p1 - q1]) <= t);
532 }
533 #endif // !WEBP_NEON_OMIT_C_CODE
534
535 #if !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
NeedsFilter2_C(const uint8_t * p,int step,int t,int it)536 static WEBP_INLINE int NeedsFilter2_C(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];
539 const int p0 = p[-step], q0 = p[0];
540 const int q1 = p[step], q2 = p[2 * step], q3 = p[3 * step];
541 if ((4 * VP8kabs0[p0 - q0] + VP8kabs0[p1 - q1]) > t) return 0;
542 return VP8kabs0[p3 - p2] <= it && VP8kabs0[p2 - p1] <= it &&
543 VP8kabs0[p1 - p0] <= it && VP8kabs0[q3 - q2] <= it &&
544 VP8kabs0[q2 - q1] <= it && VP8kabs0[q1 - q0] <= it;
545 }
546 #endif // !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
547
548 //------------------------------------------------------------------------------
549 // Simple In-loop filtering (Paragraph 15.2)
550
551 #if !WEBP_NEON_OMIT_C_CODE
SimpleVFilter16_C(uint8_t * p,int stride,int thresh)552 static void SimpleVFilter16_C(uint8_t* p, int stride, int thresh) {
553 int i;
554 const int thresh2 = 2 * thresh + 1;
555 for (i = 0; i < 16; ++i) {
556 if (NeedsFilter_C(p + i, stride, thresh2)) {
557 DoFilter2_C(p + i, stride);
558 }
559 }
560 }
561
SimpleHFilter16_C(uint8_t * p,int stride,int thresh)562 static void SimpleHFilter16_C(uint8_t* p, int stride, int thresh) {
563 int i;
564 const int thresh2 = 2 * thresh + 1;
565 for (i = 0; i < 16; ++i) {
566 if (NeedsFilter_C(p + i * stride, 1, thresh2)) {
567 DoFilter2_C(p + i * stride, 1);
568 }
569 }
570 }
571
SimpleVFilter16i_C(uint8_t * p,int stride,int thresh)572 static void SimpleVFilter16i_C(uint8_t* p, int stride, int thresh) {
573 int k;
574 for (k = 3; k > 0; --k) {
575 p += 4 * stride;
576 SimpleVFilter16_C(p, stride, thresh);
577 }
578 }
579
SimpleHFilter16i_C(uint8_t * p,int stride,int thresh)580 static void SimpleHFilter16i_C(uint8_t* p, int stride, int thresh) {
581 int k;
582 for (k = 3; k > 0; --k) {
583 p += 4;
584 SimpleHFilter16_C(p, stride, thresh);
585 }
586 }
587 #endif // !WEBP_NEON_OMIT_C_CODE
588
589 //------------------------------------------------------------------------------
590 // Complex In-loop filtering (Paragraph 15.3)
591
592 #if !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
FilterLoop26_C(uint8_t * p,int hstride,int vstride,int size,int thresh,int ithresh,int hev_thresh)593 static WEBP_INLINE void FilterLoop26_C(uint8_t* p,
594 int hstride, int vstride, int size,
595 int thresh, int ithresh,
596 int hev_thresh) {
597 const int thresh2 = 2 * thresh + 1;
598 while (size-- > 0) {
599 if (NeedsFilter2_C(p, hstride, thresh2, ithresh)) {
600 if (Hev(p, hstride, hev_thresh)) {
601 DoFilter2_C(p, hstride);
602 } else {
603 DoFilter6_C(p, hstride);
604 }
605 }
606 p += vstride;
607 }
608 }
609
FilterLoop24_C(uint8_t * p,int hstride,int vstride,int size,int thresh,int ithresh,int hev_thresh)610 static WEBP_INLINE void FilterLoop24_C(uint8_t* p,
611 int hstride, int vstride, int size,
612 int thresh, int ithresh,
613 int hev_thresh) {
614 const int thresh2 = 2 * thresh + 1;
615 while (size-- > 0) {
616 if (NeedsFilter2_C(p, hstride, thresh2, ithresh)) {
617 if (Hev(p, hstride, hev_thresh)) {
618 DoFilter2_C(p, hstride);
619 } else {
620 DoFilter4_C(p, hstride);
621 }
622 }
623 p += vstride;
624 }
625 }
626 #endif // !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
627
628 #if !WEBP_NEON_OMIT_C_CODE
629 // on macroblock edges
VFilter16_C(uint8_t * p,int stride,int thresh,int ithresh,int hev_thresh)630 static void VFilter16_C(uint8_t* p, int stride,
631 int thresh, int ithresh, int hev_thresh) {
632 FilterLoop26_C(p, stride, 1, 16, thresh, ithresh, hev_thresh);
633 }
634
HFilter16_C(uint8_t * p,int stride,int thresh,int ithresh,int hev_thresh)635 static void HFilter16_C(uint8_t* p, int stride,
636 int thresh, int ithresh, int hev_thresh) {
637 FilterLoop26_C(p, 1, stride, 16, thresh, ithresh, hev_thresh);
638 }
639
640 // on three inner edges
VFilter16i_C(uint8_t * p,int stride,int thresh,int ithresh,int hev_thresh)641 static void VFilter16i_C(uint8_t* p, int stride,
642 int thresh, int ithresh, int hev_thresh) {
643 int k;
644 for (k = 3; k > 0; --k) {
645 p += 4 * stride;
646 FilterLoop24_C(p, stride, 1, 16, thresh, ithresh, hev_thresh);
647 }
648 }
649 #endif // !WEBP_NEON_OMIT_C_CODE
650
651 #if !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
HFilter16i_C(uint8_t * p,int stride,int thresh,int ithresh,int hev_thresh)652 static void HFilter16i_C(uint8_t* p, int stride,
653 int thresh, int ithresh, int hev_thresh) {
654 int k;
655 for (k = 3; k > 0; --k) {
656 p += 4;
657 FilterLoop24_C(p, 1, stride, 16, thresh, ithresh, hev_thresh);
658 }
659 }
660 #endif // !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
661
662 #if !WEBP_NEON_OMIT_C_CODE
663 // 8-pixels wide variant, for chroma filtering
VFilter8_C(uint8_t * u,uint8_t * v,int stride,int thresh,int ithresh,int hev_thresh)664 static void VFilter8_C(uint8_t* u, uint8_t* v, int stride,
665 int thresh, int ithresh, int hev_thresh) {
666 FilterLoop26_C(u, stride, 1, 8, thresh, ithresh, hev_thresh);
667 FilterLoop26_C(v, stride, 1, 8, thresh, ithresh, hev_thresh);
668 }
669 #endif // !WEBP_NEON_OMIT_C_CODE
670
671 #if !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
HFilter8_C(uint8_t * u,uint8_t * v,int stride,int thresh,int ithresh,int hev_thresh)672 static void HFilter8_C(uint8_t* u, uint8_t* v, int stride,
673 int thresh, int ithresh, int hev_thresh) {
674 FilterLoop26_C(u, 1, stride, 8, thresh, ithresh, hev_thresh);
675 FilterLoop26_C(v, 1, stride, 8, thresh, ithresh, hev_thresh);
676 }
677 #endif // !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
678
679 #if !WEBP_NEON_OMIT_C_CODE
VFilter8i_C(uint8_t * u,uint8_t * v,int stride,int thresh,int ithresh,int hev_thresh)680 static void VFilter8i_C(uint8_t* u, uint8_t* v, int stride,
681 int thresh, int ithresh, int hev_thresh) {
682 FilterLoop24_C(u + 4 * stride, stride, 1, 8, thresh, ithresh, hev_thresh);
683 FilterLoop24_C(v + 4 * stride, stride, 1, 8, thresh, ithresh, hev_thresh);
684 }
685 #endif // !WEBP_NEON_OMIT_C_CODE
686
687 #if !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
HFilter8i_C(uint8_t * u,uint8_t * v,int stride,int thresh,int ithresh,int hev_thresh)688 static void HFilter8i_C(uint8_t* u, uint8_t* v, int stride,
689 int thresh, int ithresh, int hev_thresh) {
690 FilterLoop24_C(u + 4, 1, stride, 8, thresh, ithresh, hev_thresh);
691 FilterLoop24_C(v + 4, 1, stride, 8, thresh, ithresh, hev_thresh);
692 }
693 #endif // !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
694
695 //------------------------------------------------------------------------------
696
DitherCombine8x8_C(const uint8_t * dither,uint8_t * dst,int dst_stride)697 static void DitherCombine8x8_C(const uint8_t* dither, uint8_t* dst,
698 int dst_stride) {
699 int i, j;
700 for (j = 0; j < 8; ++j) {
701 for (i = 0; i < 8; ++i) {
702 const int delta0 = dither[i] - VP8_DITHER_AMP_CENTER;
703 const int delta1 =
704 (delta0 + VP8_DITHER_DESCALE_ROUNDER) >> VP8_DITHER_DESCALE;
705 dst[i] = clip_8b((int)dst[i] + delta1);
706 }
707 dst += dst_stride;
708 dither += 8;
709 }
710 }
711
712 //------------------------------------------------------------------------------
713
714 VP8DecIdct2 VP8Transform;
715 VP8DecIdct VP8TransformAC3;
716 VP8DecIdct VP8TransformUV;
717 VP8DecIdct VP8TransformDC;
718 VP8DecIdct VP8TransformDCUV;
719
720 VP8LumaFilterFunc VP8VFilter16;
721 VP8LumaFilterFunc VP8HFilter16;
722 VP8ChromaFilterFunc VP8VFilter8;
723 VP8ChromaFilterFunc VP8HFilter8;
724 VP8LumaFilterFunc VP8VFilter16i;
725 VP8LumaFilterFunc VP8HFilter16i;
726 VP8ChromaFilterFunc VP8VFilter8i;
727 VP8ChromaFilterFunc VP8HFilter8i;
728 VP8SimpleFilterFunc VP8SimpleVFilter16;
729 VP8SimpleFilterFunc VP8SimpleHFilter16;
730 VP8SimpleFilterFunc VP8SimpleVFilter16i;
731 VP8SimpleFilterFunc VP8SimpleHFilter16i;
732
733 void (*VP8DitherCombine8x8)(const uint8_t* dither, uint8_t* dst,
734 int dst_stride);
735
736 extern VP8CPUInfo VP8GetCPUInfo;
737 extern void VP8DspInitSSE2(void);
738 extern void VP8DspInitSSE41(void);
739 extern void VP8DspInitNEON(void);
740 extern void VP8DspInitMIPS32(void);
741 extern void VP8DspInitMIPSdspR2(void);
742 extern void VP8DspInitMSA(void);
743
WEBP_DSP_INIT_FUNC(VP8DspInit)744 WEBP_DSP_INIT_FUNC(VP8DspInit) {
745 VP8InitClipTables();
746
747 #if !WEBP_NEON_OMIT_C_CODE
748 VP8TransformWHT = TransformWHT_C;
749 VP8Transform = TransformTwo_C;
750 VP8TransformDC = TransformDC_C;
751 VP8TransformAC3 = TransformAC3_C;
752 #endif
753 VP8TransformUV = TransformUV_C;
754 VP8TransformDCUV = TransformDCUV_C;
755
756 #if !WEBP_NEON_OMIT_C_CODE
757 VP8VFilter16 = VFilter16_C;
758 VP8VFilter16i = VFilter16i_C;
759 VP8HFilter16 = HFilter16_C;
760 VP8VFilter8 = VFilter8_C;
761 VP8VFilter8i = VFilter8i_C;
762 VP8SimpleVFilter16 = SimpleVFilter16_C;
763 VP8SimpleHFilter16 = SimpleHFilter16_C;
764 VP8SimpleVFilter16i = SimpleVFilter16i_C;
765 VP8SimpleHFilter16i = SimpleHFilter16i_C;
766 #endif
767
768 #if !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
769 VP8HFilter16i = HFilter16i_C;
770 VP8HFilter8 = HFilter8_C;
771 VP8HFilter8i = HFilter8i_C;
772 #endif
773
774 #if !WEBP_NEON_OMIT_C_CODE
775 VP8PredLuma4[0] = DC4_C;
776 VP8PredLuma4[1] = TM4_C;
777 VP8PredLuma4[2] = VE4_C;
778 VP8PredLuma4[4] = RD4_C;
779 VP8PredLuma4[6] = LD4_C;
780 #endif
781
782 VP8PredLuma4[3] = HE4_C;
783 VP8PredLuma4[5] = VR4_C;
784 VP8PredLuma4[7] = VL4_C;
785 VP8PredLuma4[8] = HD4_C;
786 VP8PredLuma4[9] = HU4_C;
787
788 #if !WEBP_NEON_OMIT_C_CODE
789 VP8PredLuma16[0] = DC16_C;
790 VP8PredLuma16[1] = TM16_C;
791 VP8PredLuma16[2] = VE16_C;
792 VP8PredLuma16[3] = HE16_C;
793 VP8PredLuma16[4] = DC16NoTop_C;
794 VP8PredLuma16[5] = DC16NoLeft_C;
795 VP8PredLuma16[6] = DC16NoTopLeft_C;
796
797 VP8PredChroma8[0] = DC8uv_C;
798 VP8PredChroma8[1] = TM8uv_C;
799 VP8PredChroma8[2] = VE8uv_C;
800 VP8PredChroma8[3] = HE8uv_C;
801 VP8PredChroma8[4] = DC8uvNoTop_C;
802 VP8PredChroma8[5] = DC8uvNoLeft_C;
803 VP8PredChroma8[6] = DC8uvNoTopLeft_C;
804 #endif
805
806 VP8DitherCombine8x8 = DitherCombine8x8_C;
807
808 // If defined, use CPUInfo() to overwrite some pointers with faster versions.
809 if (VP8GetCPUInfo != NULL) {
810 #if defined(WEBP_HAVE_SSE2)
811 if (VP8GetCPUInfo(kSSE2)) {
812 VP8DspInitSSE2();
813 #if defined(WEBP_HAVE_SSE41)
814 if (VP8GetCPUInfo(kSSE4_1)) {
815 VP8DspInitSSE41();
816 }
817 #endif
818 }
819 #endif
820 #if defined(WEBP_USE_MIPS32)
821 if (VP8GetCPUInfo(kMIPS32)) {
822 VP8DspInitMIPS32();
823 }
824 #endif
825 #if defined(WEBP_USE_MIPS_DSP_R2)
826 if (VP8GetCPUInfo(kMIPSdspR2)) {
827 VP8DspInitMIPSdspR2();
828 }
829 #endif
830 #if defined(WEBP_USE_MSA)
831 if (VP8GetCPUInfo(kMSA)) {
832 VP8DspInitMSA();
833 }
834 #endif
835 }
836
837 #if defined(WEBP_HAVE_NEON)
838 if (WEBP_NEON_OMIT_C_CODE ||
839 (VP8GetCPUInfo != NULL && VP8GetCPUInfo(kNEON))) {
840 VP8DspInitNEON();
841 }
842 #endif
843
844 assert(VP8TransformWHT != NULL);
845 assert(VP8Transform != NULL);
846 assert(VP8TransformDC != NULL);
847 assert(VP8TransformAC3 != NULL);
848 assert(VP8TransformUV != NULL);
849 assert(VP8TransformDCUV != NULL);
850 assert(VP8VFilter16 != NULL);
851 assert(VP8HFilter16 != NULL);
852 assert(VP8VFilter8 != NULL);
853 assert(VP8HFilter8 != NULL);
854 assert(VP8VFilter16i != NULL);
855 assert(VP8HFilter16i != NULL);
856 assert(VP8VFilter8i != NULL);
857 assert(VP8HFilter8i != NULL);
858 assert(VP8SimpleVFilter16 != NULL);
859 assert(VP8SimpleHFilter16 != NULL);
860 assert(VP8SimpleVFilter16i != NULL);
861 assert(VP8SimpleHFilter16i != NULL);
862 assert(VP8PredLuma4[0] != NULL);
863 assert(VP8PredLuma4[1] != NULL);
864 assert(VP8PredLuma4[2] != NULL);
865 assert(VP8PredLuma4[3] != NULL);
866 assert(VP8PredLuma4[4] != NULL);
867 assert(VP8PredLuma4[5] != NULL);
868 assert(VP8PredLuma4[6] != NULL);
869 assert(VP8PredLuma4[7] != NULL);
870 assert(VP8PredLuma4[8] != NULL);
871 assert(VP8PredLuma4[9] != NULL);
872 assert(VP8PredLuma16[0] != NULL);
873 assert(VP8PredLuma16[1] != NULL);
874 assert(VP8PredLuma16[2] != NULL);
875 assert(VP8PredLuma16[3] != NULL);
876 assert(VP8PredLuma16[4] != NULL);
877 assert(VP8PredLuma16[5] != NULL);
878 assert(VP8PredLuma16[6] != NULL);
879 assert(VP8PredChroma8[0] != NULL);
880 assert(VP8PredChroma8[1] != NULL);
881 assert(VP8PredChroma8[2] != NULL);
882 assert(VP8PredChroma8[3] != NULL);
883 assert(VP8PredChroma8[4] != NULL);
884 assert(VP8PredChroma8[5] != NULL);
885 assert(VP8PredChroma8[6] != NULL);
886 assert(VP8DitherCombine8x8 != NULL);
887 }
888