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 <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 * WEBP_RESTRICT in,uint8_t * WEBP_RESTRICT dst)41 static void TransformOne_C(const int16_t* WEBP_RESTRICT in,
42 uint8_t* WEBP_RESTRICT dst) {
43 int C[4 * 4], *tmp;
44 int i;
45 tmp = C;
46 for (i = 0; i < 4; ++i) { // vertical pass
47 const int a = in[0] + in[8]; // [-4096, 4094]
48 const int b = in[0] - in[8]; // [-4095, 4095]
49 const int c = WEBP_TRANSFORM_AC3_MUL2(in[4]) -
50 WEBP_TRANSFORM_AC3_MUL1(in[12]); // [-3783, 3783]
51 const int d = WEBP_TRANSFORM_AC3_MUL1(in[4]) +
52 WEBP_TRANSFORM_AC3_MUL2(in[12]); // [-3785, 3781]
53 tmp[0] = a + d; // [-7881, 7875]
54 tmp[1] = b + c; // [-7878, 7878]
55 tmp[2] = b - c; // [-7878, 7878]
56 tmp[3] = a - d; // [-7877, 7879]
57 tmp += 4;
58 in++;
59 }
60 // Each pass is expanding the dynamic range by ~3.85 (upper bound).
61 // The exact value is (2. + (20091 + 35468) / 65536).
62 // After the second pass, maximum interval is [-3794, 3794], assuming
63 // an input in [-2048, 2047] interval. We then need to add a dst value
64 // in the [0, 255] range.
65 // In the worst case scenario, the input to clip_8b() can be as large as
66 // [-60713, 60968].
67 tmp = C;
68 for (i = 0; i < 4; ++i) { // horizontal pass
69 const int dc = tmp[0] + 4;
70 const int a = dc + tmp[8];
71 const int b = dc - tmp[8];
72 const int c =
73 WEBP_TRANSFORM_AC3_MUL2(tmp[4]) - WEBP_TRANSFORM_AC3_MUL1(tmp[12]);
74 const int d =
75 WEBP_TRANSFORM_AC3_MUL1(tmp[4]) + WEBP_TRANSFORM_AC3_MUL2(tmp[12]);
76 STORE(0, 0, a + d);
77 STORE(1, 0, b + c);
78 STORE(2, 0, b - c);
79 STORE(3, 0, a - d);
80 tmp++;
81 dst += BPS;
82 }
83 }
84
85 // Simplified transform when only in[0], in[1] and in[4] are non-zero
TransformAC3_C(const int16_t * WEBP_RESTRICT in,uint8_t * WEBP_RESTRICT dst)86 static void TransformAC3_C(const int16_t* WEBP_RESTRICT in,
87 uint8_t* WEBP_RESTRICT dst) {
88 const int a = in[0] + 4;
89 const int c4 = WEBP_TRANSFORM_AC3_MUL2(in[4]);
90 const int d4 = WEBP_TRANSFORM_AC3_MUL1(in[4]);
91 const int c1 = WEBP_TRANSFORM_AC3_MUL2(in[1]);
92 const int d1 = WEBP_TRANSFORM_AC3_MUL1(in[1]);
93 STORE2(0, a + d4, d1, c1);
94 STORE2(1, a + c4, d1, c1);
95 STORE2(2, a - c4, d1, c1);
96 STORE2(3, a - d4, d1, c1);
97 }
98 #undef STORE2
99
TransformTwo_C(const int16_t * WEBP_RESTRICT in,uint8_t * WEBP_RESTRICT dst,int do_two)100 static void TransformTwo_C(const int16_t* WEBP_RESTRICT in,
101 uint8_t* WEBP_RESTRICT dst, int do_two) {
102 TransformOne_C(in, dst);
103 if (do_two) {
104 TransformOne_C(in + 16, dst + 4);
105 }
106 }
107 #endif // !WEBP_NEON_OMIT_C_CODE
108
TransformUV_C(const int16_t * WEBP_RESTRICT in,uint8_t * WEBP_RESTRICT dst)109 static void TransformUV_C(const int16_t* WEBP_RESTRICT in,
110 uint8_t* WEBP_RESTRICT dst) {
111 VP8Transform(in + 0 * 16, dst, 1);
112 VP8Transform(in + 2 * 16, dst + 4 * BPS, 1);
113 }
114
115 #if !WEBP_NEON_OMIT_C_CODE
TransformDC_C(const int16_t * WEBP_RESTRICT in,uint8_t * WEBP_RESTRICT dst)116 static void TransformDC_C(const int16_t* WEBP_RESTRICT in,
117 uint8_t* WEBP_RESTRICT dst) {
118 const int DC = in[0] + 4;
119 int i, j;
120 for (j = 0; j < 4; ++j) {
121 for (i = 0; i < 4; ++i) {
122 STORE(i, j, DC);
123 }
124 }
125 }
126 #endif // !WEBP_NEON_OMIT_C_CODE
127
TransformDCUV_C(const int16_t * WEBP_RESTRICT in,uint8_t * WEBP_RESTRICT dst)128 static void TransformDCUV_C(const int16_t* WEBP_RESTRICT in,
129 uint8_t* WEBP_RESTRICT dst) {
130 if (in[0 * 16]) VP8TransformDC(in + 0 * 16, dst);
131 if (in[1 * 16]) VP8TransformDC(in + 1 * 16, dst + 4);
132 if (in[2 * 16]) VP8TransformDC(in + 2 * 16, dst + 4 * BPS);
133 if (in[3 * 16]) VP8TransformDC(in + 3 * 16, dst + 4 * BPS + 4);
134 }
135
136 #undef STORE
137
138 //------------------------------------------------------------------------------
139 // Paragraph 14.3
140
141 #if !WEBP_NEON_OMIT_C_CODE
TransformWHT_C(const int16_t * WEBP_RESTRICT in,int16_t * WEBP_RESTRICT out)142 static void TransformWHT_C(const int16_t* WEBP_RESTRICT in,
143 int16_t* WEBP_RESTRICT out) {
144 int tmp[16];
145 int i;
146 for (i = 0; i < 4; ++i) {
147 const int a0 = in[0 + i] + in[12 + i];
148 const int a1 = in[4 + i] + in[ 8 + i];
149 const int a2 = in[4 + i] - in[ 8 + i];
150 const int a3 = in[0 + i] - in[12 + i];
151 tmp[0 + i] = a0 + a1;
152 tmp[8 + i] = a0 - a1;
153 tmp[4 + i] = a3 + a2;
154 tmp[12 + i] = a3 - a2;
155 }
156 for (i = 0; i < 4; ++i) {
157 const int dc = tmp[0 + i * 4] + 3; // w/ rounder
158 const int a0 = dc + tmp[3 + i * 4];
159 const int a1 = tmp[1 + i * 4] + tmp[2 + i * 4];
160 const int a2 = tmp[1 + i * 4] - tmp[2 + i * 4];
161 const int a3 = dc - tmp[3 + i * 4];
162 out[ 0] = (a0 + a1) >> 3;
163 out[16] = (a3 + a2) >> 3;
164 out[32] = (a0 - a1) >> 3;
165 out[48] = (a3 - a2) >> 3;
166 out += 64;
167 }
168 }
169 #endif // !WEBP_NEON_OMIT_C_CODE
170
171 VP8WHT VP8TransformWHT;
172
173 //------------------------------------------------------------------------------
174 // Intra predictions
175
176 #define DST(x, y) dst[(x) + (y) * BPS]
177
178 #if !WEBP_NEON_OMIT_C_CODE
TrueMotion(uint8_t * dst,int size)179 static WEBP_INLINE void TrueMotion(uint8_t* dst, int size) {
180 const uint8_t* top = dst - BPS;
181 const uint8_t* const clip0 = VP8kclip1 - top[-1];
182 int y;
183 for (y = 0; y < size; ++y) {
184 const uint8_t* const clip = clip0 + dst[-1];
185 int x;
186 for (x = 0; x < size; ++x) {
187 dst[x] = clip[top[x]];
188 }
189 dst += BPS;
190 }
191 }
TM4_C(uint8_t * dst)192 static void TM4_C(uint8_t* dst) { TrueMotion(dst, 4); }
TM8uv_C(uint8_t * dst)193 static void TM8uv_C(uint8_t* dst) { TrueMotion(dst, 8); }
TM16_C(uint8_t * dst)194 static void TM16_C(uint8_t* dst) { TrueMotion(dst, 16); }
195
196 //------------------------------------------------------------------------------
197 // 16x16
198
VE16_C(uint8_t * dst)199 static void VE16_C(uint8_t* dst) { // vertical
200 int j;
201 for (j = 0; j < 16; ++j) {
202 memcpy(dst + j * BPS, dst - BPS, 16);
203 }
204 }
205
HE16_C(uint8_t * dst)206 static void HE16_C(uint8_t* dst) { // horizontal
207 int j;
208 for (j = 16; j > 0; --j) {
209 memset(dst, dst[-1], 16);
210 dst += BPS;
211 }
212 }
213
Put16(int v,uint8_t * dst)214 static WEBP_INLINE void Put16(int v, uint8_t* dst) {
215 int j;
216 for (j = 0; j < 16; ++j) {
217 memset(dst + j * BPS, v, 16);
218 }
219 }
220
DC16_C(uint8_t * dst)221 static void DC16_C(uint8_t* dst) { // DC
222 int DC = 16;
223 int j;
224 for (j = 0; j < 16; ++j) {
225 DC += dst[-1 + j * BPS] + dst[j - BPS];
226 }
227 Put16(DC >> 5, dst);
228 }
229
DC16NoTop_C(uint8_t * dst)230 static void DC16NoTop_C(uint8_t* dst) { // DC with top samples not available
231 int DC = 8;
232 int j;
233 for (j = 0; j < 16; ++j) {
234 DC += dst[-1 + j * BPS];
235 }
236 Put16(DC >> 4, dst);
237 }
238
DC16NoLeft_C(uint8_t * dst)239 static void DC16NoLeft_C(uint8_t* dst) { // DC with left samples not available
240 int DC = 8;
241 int i;
242 for (i = 0; i < 16; ++i) {
243 DC += dst[i - BPS];
244 }
245 Put16(DC >> 4, dst);
246 }
247
DC16NoTopLeft_C(uint8_t * dst)248 static void DC16NoTopLeft_C(uint8_t* dst) { // DC with no top and left samples
249 Put16(0x80, dst);
250 }
251 #endif // !WEBP_NEON_OMIT_C_CODE
252
253 VP8PredFunc VP8PredLuma16[NUM_B_DC_MODES];
254
255 //------------------------------------------------------------------------------
256 // 4x4
257
258 #define AVG3(a, b, c) ((uint8_t)(((a) + 2 * (b) + (c) + 2) >> 2))
259 #define AVG2(a, b) (((a) + (b) + 1) >> 1)
260
261 #if !WEBP_NEON_OMIT_C_CODE
VE4_C(uint8_t * dst)262 static void VE4_C(uint8_t* dst) { // vertical
263 const uint8_t* top = dst - BPS;
264 const uint8_t vals[4] = {
265 AVG3(top[-1], top[0], top[1]),
266 AVG3(top[ 0], top[1], top[2]),
267 AVG3(top[ 1], top[2], top[3]),
268 AVG3(top[ 2], top[3], top[4])
269 };
270 int i;
271 for (i = 0; i < 4; ++i) {
272 memcpy(dst + i * BPS, vals, sizeof(vals));
273 }
274 }
275 #endif // !WEBP_NEON_OMIT_C_CODE
276
HE4_C(uint8_t * dst)277 static void HE4_C(uint8_t* dst) { // horizontal
278 const int A = dst[-1 - BPS];
279 const int B = dst[-1];
280 const int C = dst[-1 + BPS];
281 const int D = dst[-1 + 2 * BPS];
282 const int E = dst[-1 + 3 * BPS];
283 WebPUint32ToMem(dst + 0 * BPS, 0x01010101U * AVG3(A, B, C));
284 WebPUint32ToMem(dst + 1 * BPS, 0x01010101U * AVG3(B, C, D));
285 WebPUint32ToMem(dst + 2 * BPS, 0x01010101U * AVG3(C, D, E));
286 WebPUint32ToMem(dst + 3 * BPS, 0x01010101U * AVG3(D, E, E));
287 }
288
289 #if !WEBP_NEON_OMIT_C_CODE
DC4_C(uint8_t * dst)290 static void DC4_C(uint8_t* dst) { // DC
291 uint32_t dc = 4;
292 int i;
293 for (i = 0; i < 4; ++i) dc += dst[i - BPS] + dst[-1 + i * BPS];
294 dc >>= 3;
295 for (i = 0; i < 4; ++i) memset(dst + i * BPS, dc, 4);
296 }
297
RD4_C(uint8_t * dst)298 static void RD4_C(uint8_t* dst) { // Down-right
299 const int I = dst[-1 + 0 * BPS];
300 const int J = dst[-1 + 1 * BPS];
301 const int K = dst[-1 + 2 * BPS];
302 const int L = dst[-1 + 3 * BPS];
303 const int X = dst[-1 - BPS];
304 const int A = dst[0 - BPS];
305 const int B = dst[1 - BPS];
306 const int C = dst[2 - BPS];
307 const int D = dst[3 - BPS];
308 DST(0, 3) = AVG3(J, K, L);
309 DST(1, 3) = DST(0, 2) = AVG3(I, J, K);
310 DST(2, 3) = DST(1, 2) = DST(0, 1) = AVG3(X, I, J);
311 DST(3, 3) = DST(2, 2) = DST(1, 1) = DST(0, 0) = AVG3(A, X, I);
312 DST(3, 2) = DST(2, 1) = DST(1, 0) = AVG3(B, A, X);
313 DST(3, 1) = DST(2, 0) = AVG3(C, B, A);
314 DST(3, 0) = AVG3(D, C, B);
315 }
316
LD4_C(uint8_t * dst)317 static void LD4_C(uint8_t* dst) { // Down-Left
318 const int A = dst[0 - BPS];
319 const int B = dst[1 - BPS];
320 const int C = dst[2 - BPS];
321 const int D = dst[3 - BPS];
322 const int E = dst[4 - BPS];
323 const int F = dst[5 - BPS];
324 const int G = dst[6 - BPS];
325 const int H = dst[7 - BPS];
326 DST(0, 0) = AVG3(A, B, C);
327 DST(1, 0) = DST(0, 1) = AVG3(B, C, D);
328 DST(2, 0) = DST(1, 1) = DST(0, 2) = AVG3(C, D, E);
329 DST(3, 0) = DST(2, 1) = DST(1, 2) = DST(0, 3) = AVG3(D, E, F);
330 DST(3, 1) = DST(2, 2) = DST(1, 3) = AVG3(E, F, G);
331 DST(3, 2) = DST(2, 3) = AVG3(F, G, H);
332 DST(3, 3) = AVG3(G, H, H);
333 }
334 #endif // !WEBP_NEON_OMIT_C_CODE
335
VR4_C(uint8_t * dst)336 static void VR4_C(uint8_t* dst) { // Vertical-Right
337 const int I = dst[-1 + 0 * BPS];
338 const int J = dst[-1 + 1 * BPS];
339 const int K = dst[-1 + 2 * BPS];
340 const int X = dst[-1 - BPS];
341 const int A = dst[0 - BPS];
342 const int B = dst[1 - BPS];
343 const int C = dst[2 - BPS];
344 const int D = dst[3 - BPS];
345 DST(0, 0) = DST(1, 2) = AVG2(X, A);
346 DST(1, 0) = DST(2, 2) = AVG2(A, B);
347 DST(2, 0) = DST(3, 2) = AVG2(B, C);
348 DST(3, 0) = AVG2(C, D);
349
350 DST(0, 3) = AVG3(K, J, I);
351 DST(0, 2) = AVG3(J, I, X);
352 DST(0, 1) = DST(1, 3) = AVG3(I, X, A);
353 DST(1, 1) = DST(2, 3) = AVG3(X, A, B);
354 DST(2, 1) = DST(3, 3) = AVG3(A, B, C);
355 DST(3, 1) = AVG3(B, C, D);
356 }
357
VL4_C(uint8_t * dst)358 static void VL4_C(uint8_t* dst) { // Vertical-Left
359 const int A = dst[0 - BPS];
360 const int B = dst[1 - BPS];
361 const int C = dst[2 - BPS];
362 const int D = dst[3 - BPS];
363 const int E = dst[4 - BPS];
364 const int F = dst[5 - BPS];
365 const int G = dst[6 - BPS];
366 const int H = dst[7 - BPS];
367 DST(0, 0) = AVG2(A, B);
368 DST(1, 0) = DST(0, 2) = AVG2(B, C);
369 DST(2, 0) = DST(1, 2) = AVG2(C, D);
370 DST(3, 0) = DST(2, 2) = AVG2(D, E);
371
372 DST(0, 1) = AVG3(A, B, C);
373 DST(1, 1) = DST(0, 3) = AVG3(B, C, D);
374 DST(2, 1) = DST(1, 3) = AVG3(C, D, E);
375 DST(3, 1) = DST(2, 3) = AVG3(D, E, F);
376 DST(3, 2) = AVG3(E, F, G);
377 DST(3, 3) = AVG3(F, G, H);
378 }
379
HU4_C(uint8_t * dst)380 static void HU4_C(uint8_t* dst) { // Horizontal-Up
381 const int I = dst[-1 + 0 * BPS];
382 const int J = dst[-1 + 1 * BPS];
383 const int K = dst[-1 + 2 * BPS];
384 const int L = dst[-1 + 3 * BPS];
385 DST(0, 0) = AVG2(I, J);
386 DST(2, 0) = DST(0, 1) = AVG2(J, K);
387 DST(2, 1) = DST(0, 2) = AVG2(K, L);
388 DST(1, 0) = AVG3(I, J, K);
389 DST(3, 0) = DST(1, 1) = AVG3(J, K, L);
390 DST(3, 1) = DST(1, 2) = AVG3(K, L, L);
391 DST(3, 2) = DST(2, 2) =
392 DST(0, 3) = DST(1, 3) = DST(2, 3) = DST(3, 3) = L;
393 }
394
HD4_C(uint8_t * dst)395 static void HD4_C(uint8_t* dst) { // Horizontal-Down
396 const int I = dst[-1 + 0 * BPS];
397 const int J = dst[-1 + 1 * BPS];
398 const int K = dst[-1 + 2 * BPS];
399 const int L = dst[-1 + 3 * BPS];
400 const int X = dst[-1 - BPS];
401 const int A = dst[0 - BPS];
402 const int B = dst[1 - BPS];
403 const int C = dst[2 - BPS];
404
405 DST(0, 0) = DST(2, 1) = AVG2(I, X);
406 DST(0, 1) = DST(2, 2) = AVG2(J, I);
407 DST(0, 2) = DST(2, 3) = AVG2(K, J);
408 DST(0, 3) = AVG2(L, K);
409
410 DST(3, 0) = AVG3(A, B, C);
411 DST(2, 0) = AVG3(X, A, B);
412 DST(1, 0) = DST(3, 1) = AVG3(I, X, A);
413 DST(1, 1) = DST(3, 2) = AVG3(J, I, X);
414 DST(1, 2) = DST(3, 3) = AVG3(K, J, I);
415 DST(1, 3) = AVG3(L, K, J);
416 }
417
418 #undef DST
419 #undef AVG3
420 #undef AVG2
421
422 VP8PredFunc VP8PredLuma4[NUM_BMODES];
423
424 //------------------------------------------------------------------------------
425 // Chroma
426
427 #if !WEBP_NEON_OMIT_C_CODE
VE8uv_C(uint8_t * dst)428 static void VE8uv_C(uint8_t* dst) { // vertical
429 int j;
430 for (j = 0; j < 8; ++j) {
431 memcpy(dst + j * BPS, dst - BPS, 8);
432 }
433 }
434
HE8uv_C(uint8_t * dst)435 static void HE8uv_C(uint8_t* dst) { // horizontal
436 int j;
437 for (j = 0; j < 8; ++j) {
438 memset(dst, dst[-1], 8);
439 dst += BPS;
440 }
441 }
442
443 // helper for chroma-DC predictions
Put8x8uv(uint8_t value,uint8_t * dst)444 static WEBP_INLINE void Put8x8uv(uint8_t value, uint8_t* dst) {
445 int j;
446 for (j = 0; j < 8; ++j) {
447 memset(dst + j * BPS, value, 8);
448 }
449 }
450
DC8uv_C(uint8_t * dst)451 static void DC8uv_C(uint8_t* dst) { // DC
452 int dc0 = 8;
453 int i;
454 for (i = 0; i < 8; ++i) {
455 dc0 += dst[i - BPS] + dst[-1 + i * BPS];
456 }
457 Put8x8uv(dc0 >> 4, dst);
458 }
459
DC8uvNoLeft_C(uint8_t * dst)460 static void DC8uvNoLeft_C(uint8_t* dst) { // DC with no left samples
461 int dc0 = 4;
462 int i;
463 for (i = 0; i < 8; ++i) {
464 dc0 += dst[i - BPS];
465 }
466 Put8x8uv(dc0 >> 3, dst);
467 }
468
DC8uvNoTop_C(uint8_t * dst)469 static void DC8uvNoTop_C(uint8_t* dst) { // DC with no top samples
470 int dc0 = 4;
471 int i;
472 for (i = 0; i < 8; ++i) {
473 dc0 += dst[-1 + i * BPS];
474 }
475 Put8x8uv(dc0 >> 3, dst);
476 }
477
DC8uvNoTopLeft_C(uint8_t * dst)478 static void DC8uvNoTopLeft_C(uint8_t* dst) { // DC with nothing
479 Put8x8uv(0x80, dst);
480 }
481 #endif // !WEBP_NEON_OMIT_C_CODE
482
483 VP8PredFunc VP8PredChroma8[NUM_B_DC_MODES];
484
485 //------------------------------------------------------------------------------
486 // Edge filtering functions
487
488 #if !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
489 // 4 pixels in, 2 pixels out
DoFilter2_C(uint8_t * p,int step)490 static WEBP_INLINE void DoFilter2_C(uint8_t* p, int step) {
491 const int p1 = p[-2*step], p0 = p[-step], q0 = p[0], q1 = p[step];
492 const int a = 3 * (q0 - p0) + VP8ksclip1[p1 - q1]; // in [-893,892]
493 const int a1 = VP8ksclip2[(a + 4) >> 3]; // in [-16,15]
494 const int a2 = VP8ksclip2[(a + 3) >> 3];
495 p[-step] = VP8kclip1[p0 + a2];
496 p[ 0] = VP8kclip1[q0 - a1];
497 }
498
499 // 4 pixels in, 4 pixels out
DoFilter4_C(uint8_t * p,int step)500 static WEBP_INLINE void DoFilter4_C(uint8_t* p, int step) {
501 const int p1 = p[-2*step], p0 = p[-step], q0 = p[0], q1 = p[step];
502 const int a = 3 * (q0 - p0);
503 const int a1 = VP8ksclip2[(a + 4) >> 3];
504 const int a2 = VP8ksclip2[(a + 3) >> 3];
505 const int a3 = (a1 + 1) >> 1;
506 p[-2*step] = VP8kclip1[p1 + a3];
507 p[- step] = VP8kclip1[p0 + a2];
508 p[ 0] = VP8kclip1[q0 - a1];
509 p[ step] = VP8kclip1[q1 - a3];
510 }
511
512 // 6 pixels in, 6 pixels out
DoFilter6_C(uint8_t * p,int step)513 static WEBP_INLINE void DoFilter6_C(uint8_t* p, int step) {
514 const int p2 = p[-3*step], p1 = p[-2*step], p0 = p[-step];
515 const int q0 = p[0], q1 = p[step], q2 = p[2*step];
516 const int a = VP8ksclip1[3 * (q0 - p0) + VP8ksclip1[p1 - q1]];
517 // a is in [-128,127], a1 in [-27,27], a2 in [-18,18] and a3 in [-9,9]
518 const int a1 = (27 * a + 63) >> 7; // eq. to ((3 * a + 7) * 9) >> 7
519 const int a2 = (18 * a + 63) >> 7; // eq. to ((2 * a + 7) * 9) >> 7
520 const int a3 = (9 * a + 63) >> 7; // eq. to ((1 * a + 7) * 9) >> 7
521 p[-3*step] = VP8kclip1[p2 + a3];
522 p[-2*step] = VP8kclip1[p1 + a2];
523 p[- step] = VP8kclip1[p0 + a1];
524 p[ 0] = VP8kclip1[q0 - a1];
525 p[ step] = VP8kclip1[q1 - a2];
526 p[ 2*step] = VP8kclip1[q2 - a3];
527 }
528
Hev(const uint8_t * p,int step,int thresh)529 static WEBP_INLINE int Hev(const uint8_t* p, int step, int thresh) {
530 const int p1 = p[-2*step], p0 = p[-step], q0 = p[0], q1 = p[step];
531 return (VP8kabs0[p1 - p0] > thresh) || (VP8kabs0[q1 - q0] > thresh);
532 }
533 #endif // !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
534
535 #if !WEBP_NEON_OMIT_C_CODE
NeedsFilter_C(const uint8_t * p,int step,int t)536 static WEBP_INLINE int NeedsFilter_C(const uint8_t* p, int step, int t) {
537 const int p1 = p[-2 * step], p0 = p[-step], q0 = p[0], q1 = p[step];
538 return ((4 * VP8kabs0[p0 - q0] + VP8kabs0[p1 - q1]) <= t);
539 }
540 #endif // !WEBP_NEON_OMIT_C_CODE
541
542 #if !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
NeedsFilter2_C(const uint8_t * p,int step,int t,int it)543 static WEBP_INLINE int NeedsFilter2_C(const uint8_t* p,
544 int step, int t, int it) {
545 const int p3 = p[-4 * step], p2 = p[-3 * step], p1 = p[-2 * step];
546 const int p0 = p[-step], q0 = p[0];
547 const int q1 = p[step], q2 = p[2 * step], q3 = p[3 * step];
548 if ((4 * VP8kabs0[p0 - q0] + VP8kabs0[p1 - q1]) > t) return 0;
549 return VP8kabs0[p3 - p2] <= it && VP8kabs0[p2 - p1] <= it &&
550 VP8kabs0[p1 - p0] <= it && VP8kabs0[q3 - q2] <= it &&
551 VP8kabs0[q2 - q1] <= it && VP8kabs0[q1 - q0] <= it;
552 }
553 #endif // !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
554
555 //------------------------------------------------------------------------------
556 // Simple In-loop filtering (Paragraph 15.2)
557
558 #if !WEBP_NEON_OMIT_C_CODE
SimpleVFilter16_C(uint8_t * p,int stride,int thresh)559 static void SimpleVFilter16_C(uint8_t* p, int stride, int thresh) {
560 int i;
561 const int thresh2 = 2 * thresh + 1;
562 for (i = 0; i < 16; ++i) {
563 if (NeedsFilter_C(p + i, stride, thresh2)) {
564 DoFilter2_C(p + i, stride);
565 }
566 }
567 }
568
SimpleHFilter16_C(uint8_t * p,int stride,int thresh)569 static void SimpleHFilter16_C(uint8_t* p, int stride, int thresh) {
570 int i;
571 const int thresh2 = 2 * thresh + 1;
572 for (i = 0; i < 16; ++i) {
573 if (NeedsFilter_C(p + i * stride, 1, thresh2)) {
574 DoFilter2_C(p + i * stride, 1);
575 }
576 }
577 }
578
SimpleVFilter16i_C(uint8_t * p,int stride,int thresh)579 static void SimpleVFilter16i_C(uint8_t* p, int stride, int thresh) {
580 int k;
581 for (k = 3; k > 0; --k) {
582 p += 4 * stride;
583 SimpleVFilter16_C(p, stride, thresh);
584 }
585 }
586
SimpleHFilter16i_C(uint8_t * p,int stride,int thresh)587 static void SimpleHFilter16i_C(uint8_t* p, int stride, int thresh) {
588 int k;
589 for (k = 3; k > 0; --k) {
590 p += 4;
591 SimpleHFilter16_C(p, stride, thresh);
592 }
593 }
594 #endif // !WEBP_NEON_OMIT_C_CODE
595
596 //------------------------------------------------------------------------------
597 // Complex In-loop filtering (Paragraph 15.3)
598
599 #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)600 static WEBP_INLINE void FilterLoop26_C(uint8_t* p,
601 int hstride, int vstride, int size,
602 int thresh, int ithresh,
603 int hev_thresh) {
604 const int thresh2 = 2 * thresh + 1;
605 while (size-- > 0) {
606 if (NeedsFilter2_C(p, hstride, thresh2, ithresh)) {
607 if (Hev(p, hstride, hev_thresh)) {
608 DoFilter2_C(p, hstride);
609 } else {
610 DoFilter6_C(p, hstride);
611 }
612 }
613 p += vstride;
614 }
615 }
616
FilterLoop24_C(uint8_t * p,int hstride,int vstride,int size,int thresh,int ithresh,int hev_thresh)617 static WEBP_INLINE void FilterLoop24_C(uint8_t* p,
618 int hstride, int vstride, int size,
619 int thresh, int ithresh,
620 int hev_thresh) {
621 const int thresh2 = 2 * thresh + 1;
622 while (size-- > 0) {
623 if (NeedsFilter2_C(p, hstride, thresh2, ithresh)) {
624 if (Hev(p, hstride, hev_thresh)) {
625 DoFilter2_C(p, hstride);
626 } else {
627 DoFilter4_C(p, hstride);
628 }
629 }
630 p += vstride;
631 }
632 }
633 #endif // !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
634
635 #if !WEBP_NEON_OMIT_C_CODE
636 // on macroblock edges
VFilter16_C(uint8_t * p,int stride,int thresh,int ithresh,int hev_thresh)637 static void VFilter16_C(uint8_t* p, int stride,
638 int thresh, int ithresh, int hev_thresh) {
639 FilterLoop26_C(p, stride, 1, 16, thresh, ithresh, hev_thresh);
640 }
641
HFilter16_C(uint8_t * p,int stride,int thresh,int ithresh,int hev_thresh)642 static void HFilter16_C(uint8_t* p, int stride,
643 int thresh, int ithresh, int hev_thresh) {
644 FilterLoop26_C(p, 1, stride, 16, thresh, ithresh, hev_thresh);
645 }
646
647 // on three inner edges
VFilter16i_C(uint8_t * p,int stride,int thresh,int ithresh,int hev_thresh)648 static void VFilter16i_C(uint8_t* p, int stride,
649 int thresh, int ithresh, int hev_thresh) {
650 int k;
651 for (k = 3; k > 0; --k) {
652 p += 4 * stride;
653 FilterLoop24_C(p, stride, 1, 16, thresh, ithresh, hev_thresh);
654 }
655 }
656 #endif // !WEBP_NEON_OMIT_C_CODE
657
658 #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)659 static void HFilter16i_C(uint8_t* p, int stride,
660 int thresh, int ithresh, int hev_thresh) {
661 int k;
662 for (k = 3; k > 0; --k) {
663 p += 4;
664 FilterLoop24_C(p, 1, stride, 16, thresh, ithresh, hev_thresh);
665 }
666 }
667 #endif // !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
668
669 #if !WEBP_NEON_OMIT_C_CODE
670 // 8-pixels wide variant, for chroma filtering
VFilter8_C(uint8_t * WEBP_RESTRICT u,uint8_t * WEBP_RESTRICT v,int stride,int thresh,int ithresh,int hev_thresh)671 static void VFilter8_C(uint8_t* WEBP_RESTRICT u, uint8_t* WEBP_RESTRICT v,
672 int stride, int thresh, int ithresh, int hev_thresh) {
673 FilterLoop26_C(u, stride, 1, 8, thresh, ithresh, hev_thresh);
674 FilterLoop26_C(v, stride, 1, 8, thresh, ithresh, hev_thresh);
675 }
676 #endif // !WEBP_NEON_OMIT_C_CODE
677
678 #if !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
HFilter8_C(uint8_t * WEBP_RESTRICT u,uint8_t * WEBP_RESTRICT v,int stride,int thresh,int ithresh,int hev_thresh)679 static void HFilter8_C(uint8_t* WEBP_RESTRICT u, uint8_t* WEBP_RESTRICT v,
680 int stride, int thresh, int ithresh, int hev_thresh) {
681 FilterLoop26_C(u, 1, stride, 8, thresh, ithresh, hev_thresh);
682 FilterLoop26_C(v, 1, stride, 8, thresh, ithresh, hev_thresh);
683 }
684 #endif // !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
685
686 #if !WEBP_NEON_OMIT_C_CODE
VFilter8i_C(uint8_t * WEBP_RESTRICT u,uint8_t * WEBP_RESTRICT v,int stride,int thresh,int ithresh,int hev_thresh)687 static void VFilter8i_C(uint8_t* WEBP_RESTRICT u, uint8_t* WEBP_RESTRICT v,
688 int stride, int thresh, int ithresh, int hev_thresh) {
689 FilterLoop24_C(u + 4 * stride, stride, 1, 8, thresh, ithresh, hev_thresh);
690 FilterLoop24_C(v + 4 * stride, stride, 1, 8, thresh, ithresh, hev_thresh);
691 }
692 #endif // !WEBP_NEON_OMIT_C_CODE
693
694 #if !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
HFilter8i_C(uint8_t * WEBP_RESTRICT u,uint8_t * WEBP_RESTRICT v,int stride,int thresh,int ithresh,int hev_thresh)695 static void HFilter8i_C(uint8_t* WEBP_RESTRICT u, uint8_t* WEBP_RESTRICT v,
696 int stride, int thresh, int ithresh, int hev_thresh) {
697 FilterLoop24_C(u + 4, 1, stride, 8, thresh, ithresh, hev_thresh);
698 FilterLoop24_C(v + 4, 1, stride, 8, thresh, ithresh, hev_thresh);
699 }
700 #endif // !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
701
702 //------------------------------------------------------------------------------
703
DitherCombine8x8_C(const uint8_t * WEBP_RESTRICT dither,uint8_t * WEBP_RESTRICT dst,int dst_stride)704 static void DitherCombine8x8_C(const uint8_t* WEBP_RESTRICT dither,
705 uint8_t* WEBP_RESTRICT dst, int dst_stride) {
706 int i, j;
707 for (j = 0; j < 8; ++j) {
708 for (i = 0; i < 8; ++i) {
709 const int delta0 = dither[i] - VP8_DITHER_AMP_CENTER;
710 const int delta1 =
711 (delta0 + VP8_DITHER_DESCALE_ROUNDER) >> VP8_DITHER_DESCALE;
712 dst[i] = clip_8b((int)dst[i] + delta1);
713 }
714 dst += dst_stride;
715 dither += 8;
716 }
717 }
718
719 //------------------------------------------------------------------------------
720
721 VP8DecIdct2 VP8Transform;
722 VP8DecIdct VP8TransformAC3;
723 VP8DecIdct VP8TransformUV;
724 VP8DecIdct VP8TransformDC;
725 VP8DecIdct VP8TransformDCUV;
726
727 VP8LumaFilterFunc VP8VFilter16;
728 VP8LumaFilterFunc VP8HFilter16;
729 VP8ChromaFilterFunc VP8VFilter8;
730 VP8ChromaFilterFunc VP8HFilter8;
731 VP8LumaFilterFunc VP8VFilter16i;
732 VP8LumaFilterFunc VP8HFilter16i;
733 VP8ChromaFilterFunc VP8VFilter8i;
734 VP8ChromaFilterFunc VP8HFilter8i;
735 VP8SimpleFilterFunc VP8SimpleVFilter16;
736 VP8SimpleFilterFunc VP8SimpleHFilter16;
737 VP8SimpleFilterFunc VP8SimpleVFilter16i;
738 VP8SimpleFilterFunc VP8SimpleHFilter16i;
739
740 void (*VP8DitherCombine8x8)(const uint8_t* WEBP_RESTRICT dither,
741 uint8_t* WEBP_RESTRICT dst, int dst_stride);
742
743 extern VP8CPUInfo VP8GetCPUInfo;
744 extern void VP8DspInitSSE2(void);
745 extern void VP8DspInitSSE41(void);
746 extern void VP8DspInitNEON(void);
747 extern void VP8DspInitMIPS32(void);
748 extern void VP8DspInitMIPSdspR2(void);
749 extern void VP8DspInitMSA(void);
750
WEBP_DSP_INIT_FUNC(VP8DspInit)751 WEBP_DSP_INIT_FUNC(VP8DspInit) {
752 VP8InitClipTables();
753
754 #if !WEBP_NEON_OMIT_C_CODE
755 VP8TransformWHT = TransformWHT_C;
756 VP8Transform = TransformTwo_C;
757 VP8TransformDC = TransformDC_C;
758 VP8TransformAC3 = TransformAC3_C;
759 #endif
760 VP8TransformUV = TransformUV_C;
761 VP8TransformDCUV = TransformDCUV_C;
762
763 #if !WEBP_NEON_OMIT_C_CODE
764 VP8VFilter16 = VFilter16_C;
765 VP8VFilter16i = VFilter16i_C;
766 VP8HFilter16 = HFilter16_C;
767 VP8VFilter8 = VFilter8_C;
768 VP8VFilter8i = VFilter8i_C;
769 VP8SimpleVFilter16 = SimpleVFilter16_C;
770 VP8SimpleHFilter16 = SimpleHFilter16_C;
771 VP8SimpleVFilter16i = SimpleVFilter16i_C;
772 VP8SimpleHFilter16i = SimpleHFilter16i_C;
773 #endif
774
775 #if !WEBP_NEON_OMIT_C_CODE || WEBP_NEON_WORK_AROUND_GCC
776 VP8HFilter16i = HFilter16i_C;
777 VP8HFilter8 = HFilter8_C;
778 VP8HFilter8i = HFilter8i_C;
779 #endif
780
781 #if !WEBP_NEON_OMIT_C_CODE
782 VP8PredLuma4[0] = DC4_C;
783 VP8PredLuma4[1] = TM4_C;
784 VP8PredLuma4[2] = VE4_C;
785 VP8PredLuma4[4] = RD4_C;
786 VP8PredLuma4[6] = LD4_C;
787 #endif
788
789 VP8PredLuma4[3] = HE4_C;
790 VP8PredLuma4[5] = VR4_C;
791 VP8PredLuma4[7] = VL4_C;
792 VP8PredLuma4[8] = HD4_C;
793 VP8PredLuma4[9] = HU4_C;
794
795 #if !WEBP_NEON_OMIT_C_CODE
796 VP8PredLuma16[0] = DC16_C;
797 VP8PredLuma16[1] = TM16_C;
798 VP8PredLuma16[2] = VE16_C;
799 VP8PredLuma16[3] = HE16_C;
800 VP8PredLuma16[4] = DC16NoTop_C;
801 VP8PredLuma16[5] = DC16NoLeft_C;
802 VP8PredLuma16[6] = DC16NoTopLeft_C;
803
804 VP8PredChroma8[0] = DC8uv_C;
805 VP8PredChroma8[1] = TM8uv_C;
806 VP8PredChroma8[2] = VE8uv_C;
807 VP8PredChroma8[3] = HE8uv_C;
808 VP8PredChroma8[4] = DC8uvNoTop_C;
809 VP8PredChroma8[5] = DC8uvNoLeft_C;
810 VP8PredChroma8[6] = DC8uvNoTopLeft_C;
811 #endif
812
813 VP8DitherCombine8x8 = DitherCombine8x8_C;
814
815 // If defined, use CPUInfo() to overwrite some pointers with faster versions.
816 if (VP8GetCPUInfo != NULL) {
817 #if defined(WEBP_HAVE_SSE2)
818 if (VP8GetCPUInfo(kSSE2)) {
819 VP8DspInitSSE2();
820 #if defined(WEBP_HAVE_SSE41)
821 if (VP8GetCPUInfo(kSSE4_1)) {
822 VP8DspInitSSE41();
823 }
824 #endif
825 }
826 #endif
827 #if defined(WEBP_USE_MIPS32)
828 if (VP8GetCPUInfo(kMIPS32)) {
829 VP8DspInitMIPS32();
830 }
831 #endif
832 #if defined(WEBP_USE_MIPS_DSP_R2)
833 if (VP8GetCPUInfo(kMIPSdspR2)) {
834 VP8DspInitMIPSdspR2();
835 }
836 #endif
837 #if defined(WEBP_USE_MSA)
838 if (VP8GetCPUInfo(kMSA)) {
839 VP8DspInitMSA();
840 }
841 #endif
842 }
843
844 #if defined(WEBP_HAVE_NEON)
845 if (WEBP_NEON_OMIT_C_CODE ||
846 (VP8GetCPUInfo != NULL && VP8GetCPUInfo(kNEON))) {
847 VP8DspInitNEON();
848 }
849 #endif
850
851 assert(VP8TransformWHT != NULL);
852 assert(VP8Transform != NULL);
853 assert(VP8TransformDC != NULL);
854 assert(VP8TransformAC3 != NULL);
855 assert(VP8TransformUV != NULL);
856 assert(VP8TransformDCUV != NULL);
857 assert(VP8VFilter16 != NULL);
858 assert(VP8HFilter16 != NULL);
859 assert(VP8VFilter8 != NULL);
860 assert(VP8HFilter8 != NULL);
861 assert(VP8VFilter16i != NULL);
862 assert(VP8HFilter16i != NULL);
863 assert(VP8VFilter8i != NULL);
864 assert(VP8HFilter8i != NULL);
865 assert(VP8SimpleVFilter16 != NULL);
866 assert(VP8SimpleHFilter16 != NULL);
867 assert(VP8SimpleVFilter16i != NULL);
868 assert(VP8SimpleHFilter16i != NULL);
869 assert(VP8PredLuma4[0] != NULL);
870 assert(VP8PredLuma4[1] != NULL);
871 assert(VP8PredLuma4[2] != NULL);
872 assert(VP8PredLuma4[3] != NULL);
873 assert(VP8PredLuma4[4] != NULL);
874 assert(VP8PredLuma4[5] != NULL);
875 assert(VP8PredLuma4[6] != NULL);
876 assert(VP8PredLuma4[7] != NULL);
877 assert(VP8PredLuma4[8] != NULL);
878 assert(VP8PredLuma4[9] != NULL);
879 assert(VP8PredLuma16[0] != NULL);
880 assert(VP8PredLuma16[1] != NULL);
881 assert(VP8PredLuma16[2] != NULL);
882 assert(VP8PredLuma16[3] != NULL);
883 assert(VP8PredLuma16[4] != NULL);
884 assert(VP8PredLuma16[5] != NULL);
885 assert(VP8PredLuma16[6] != NULL);
886 assert(VP8PredChroma8[0] != NULL);
887 assert(VP8PredChroma8[1] != NULL);
888 assert(VP8PredChroma8[2] != NULL);
889 assert(VP8PredChroma8[3] != NULL);
890 assert(VP8PredChroma8[4] != NULL);
891 assert(VP8PredChroma8[5] != NULL);
892 assert(VP8PredChroma8[6] != NULL);
893 assert(VP8DitherCombine8x8 != NULL);
894 }
895