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.
11 //
12 // Author: Skal (pascal.massimino@gmail.com)
13 
14 #include "./dsp.h"
15 #include "../dec/vp8i.h"
16 
17 //------------------------------------------------------------------------------
18 
clip_8b(int v)19 static WEBP_INLINE uint8_t clip_8b(int v) {
20   return (!(v & ~0xff)) ? v : (v < 0) ? 0 : 255;
21 }
22 
23 //------------------------------------------------------------------------------
24 // Transforms (Paragraph 14.4)
25 
26 #define STORE(x, y, v) \
27   dst[x + y * BPS] = clip_8b(dst[x + y * BPS] + ((v) >> 3))
28 
29 #define STORE2(y, dc, d, c) do {    \
30   const int DC = (dc);              \
31   STORE(0, y, DC + (d));            \
32   STORE(1, y, DC + (c));            \
33   STORE(2, y, DC - (c));            \
34   STORE(3, y, DC - (d));            \
35 } while (0)
36 
37 static const int kC1 = 20091 + (1 << 16);
38 static const int kC2 = 35468;
39 #define MUL(a, b) (((a) * (b)) >> 16)
40 
TransformOne(const int16_t * in,uint8_t * dst)41 static 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 = MUL(in[4], kC2) - MUL(in[12], kC1);   // [-3783, 3783]
49     const int d = MUL(in[4], kC1) + MUL(in[12], kC2);   // [-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. + (kC1 + kC2) / 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 = MUL(tmp[4], kC2) - MUL(tmp[12], kC1);
70     const int d = MUL(tmp[4], kC1) + MUL(tmp[12], kC2);
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
TransformAC3(const int16_t * in,uint8_t * dst)81 static void TransformAC3(const int16_t* in, uint8_t* dst) {
82   const int a = in[0] + 4;
83   const int c4 = MUL(in[4], kC2);
84   const int d4 = MUL(in[4], kC1);
85   const int c1 = MUL(in[1], kC2);
86   const int d1 = MUL(in[1], kC1);
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 MUL
93 #undef STORE2
94 
TransformTwo(const int16_t * in,uint8_t * dst,int do_two)95 static void TransformTwo(const int16_t* in, uint8_t* dst, int do_two) {
96   TransformOne(in, dst);
97   if (do_two) {
98     TransformOne(in + 16, dst + 4);
99   }
100 }
101 
TransformUV(const int16_t * in,uint8_t * dst)102 static void TransformUV(const int16_t* in, uint8_t* dst) {
103   VP8Transform(in + 0 * 16, dst, 1);
104   VP8Transform(in + 2 * 16, dst + 4 * BPS, 1);
105 }
106 
TransformDC(const int16_t * in,uint8_t * dst)107 static void TransformDC(const int16_t *in, uint8_t* dst) {
108   const int DC = in[0] + 4;
109   int i, j;
110   for (j = 0; j < 4; ++j) {
111     for (i = 0; i < 4; ++i) {
112       STORE(i, j, DC);
113     }
114   }
115 }
116 
TransformDCUV(const int16_t * in,uint8_t * dst)117 static void TransformDCUV(const int16_t* in, uint8_t* dst) {
118   if (in[0 * 16]) VP8TransformDC(in + 0 * 16, dst);
119   if (in[1 * 16]) VP8TransformDC(in + 1 * 16, dst + 4);
120   if (in[2 * 16]) VP8TransformDC(in + 2 * 16, dst + 4 * BPS);
121   if (in[3 * 16]) VP8TransformDC(in + 3 * 16, dst + 4 * BPS + 4);
122 }
123 
124 #undef STORE
125 
126 //------------------------------------------------------------------------------
127 // Paragraph 14.3
128 
TransformWHT(const int16_t * in,int16_t * out)129 static void TransformWHT(const int16_t* in, int16_t* out) {
130   int tmp[16];
131   int i;
132   for (i = 0; i < 4; ++i) {
133     const int a0 = in[0 + i] + in[12 + i];
134     const int a1 = in[4 + i] + in[ 8 + i];
135     const int a2 = in[4 + i] - in[ 8 + i];
136     const int a3 = in[0 + i] - in[12 + i];
137     tmp[0  + i] = a0 + a1;
138     tmp[8  + i] = a0 - a1;
139     tmp[4  + i] = a3 + a2;
140     tmp[12 + i] = a3 - a2;
141   }
142   for (i = 0; i < 4; ++i) {
143     const int dc = tmp[0 + i * 4] + 3;    // w/ rounder
144     const int a0 = dc             + tmp[3 + i * 4];
145     const int a1 = tmp[1 + i * 4] + tmp[2 + i * 4];
146     const int a2 = tmp[1 + i * 4] - tmp[2 + i * 4];
147     const int a3 = dc             - tmp[3 + i * 4];
148     out[ 0] = (a0 + a1) >> 3;
149     out[16] = (a3 + a2) >> 3;
150     out[32] = (a0 - a1) >> 3;
151     out[48] = (a3 - a2) >> 3;
152     out += 64;
153   }
154 }
155 
156 void (*VP8TransformWHT)(const int16_t* in, int16_t* out);
157 
158 //------------------------------------------------------------------------------
159 // Intra predictions
160 
161 #define DST(x, y) dst[(x) + (y) * BPS]
162 
TrueMotion(uint8_t * dst,int size)163 static WEBP_INLINE void TrueMotion(uint8_t *dst, int size) {
164   const uint8_t* top = dst - BPS;
165   const uint8_t* const clip0 = VP8kclip1 - top[-1];
166   int y;
167   for (y = 0; y < size; ++y) {
168     const uint8_t* const clip = clip0 + dst[-1];
169     int x;
170     for (x = 0; x < size; ++x) {
171       dst[x] = clip[top[x]];
172     }
173     dst += BPS;
174   }
175 }
TM4(uint8_t * dst)176 static void TM4(uint8_t *dst)   { TrueMotion(dst, 4); }
TM8uv(uint8_t * dst)177 static void TM8uv(uint8_t *dst) { TrueMotion(dst, 8); }
TM16(uint8_t * dst)178 static void TM16(uint8_t *dst)  { TrueMotion(dst, 16); }
179 
180 //------------------------------------------------------------------------------
181 // 16x16
182 
VE16(uint8_t * dst)183 static void VE16(uint8_t *dst) {     // vertical
184   int j;
185   for (j = 0; j < 16; ++j) {
186     memcpy(dst + j * BPS, dst - BPS, 16);
187   }
188 }
189 
HE16(uint8_t * dst)190 static void HE16(uint8_t *dst) {     // horizontal
191   int j;
192   for (j = 16; j > 0; --j) {
193     memset(dst, dst[-1], 16);
194     dst += BPS;
195   }
196 }
197 
Put16(int v,uint8_t * dst)198 static WEBP_INLINE void Put16(int v, uint8_t* dst) {
199   int j;
200   for (j = 0; j < 16; ++j) {
201     memset(dst + j * BPS, v, 16);
202   }
203 }
204 
DC16(uint8_t * dst)205 static void DC16(uint8_t *dst) {    // DC
206   int DC = 16;
207   int j;
208   for (j = 0; j < 16; ++j) {
209     DC += dst[-1 + j * BPS] + dst[j - BPS];
210   }
211   Put16(DC >> 5, dst);
212 }
213 
DC16NoTop(uint8_t * dst)214 static void DC16NoTop(uint8_t *dst) {   // DC with top samples not available
215   int DC = 8;
216   int j;
217   for (j = 0; j < 16; ++j) {
218     DC += dst[-1 + j * BPS];
219   }
220   Put16(DC >> 4, dst);
221 }
222 
DC16NoLeft(uint8_t * dst)223 static void DC16NoLeft(uint8_t *dst) {  // DC with left samples not available
224   int DC = 8;
225   int i;
226   for (i = 0; i < 16; ++i) {
227     DC += dst[i - BPS];
228   }
229   Put16(DC >> 4, dst);
230 }
231 
DC16NoTopLeft(uint8_t * dst)232 static void DC16NoTopLeft(uint8_t *dst) {  // DC with no top and left samples
233   Put16(0x80, dst);
234 }
235 
236 //------------------------------------------------------------------------------
237 // 4x4
238 
239 #define AVG3(a, b, c) (((a) + 2 * (b) + (c) + 2) >> 2)
240 #define AVG2(a, b) (((a) + (b) + 1) >> 1)
241 
VE4(uint8_t * dst)242 static void VE4(uint8_t *dst) {    // vertical
243   const uint8_t* top = dst - BPS;
244   const uint8_t vals[4] = {
245     AVG3(top[-1], top[0], top[1]),
246     AVG3(top[ 0], top[1], top[2]),
247     AVG3(top[ 1], top[2], top[3]),
248     AVG3(top[ 2], top[3], top[4])
249   };
250   int i;
251   for (i = 0; i < 4; ++i) {
252     memcpy(dst + i * BPS, vals, sizeof(vals));
253   }
254 }
255 
HE4(uint8_t * dst)256 static void HE4(uint8_t *dst) {    // horizontal
257   const int A = dst[-1 - BPS];
258   const int B = dst[-1];
259   const int C = dst[-1 + BPS];
260   const int D = dst[-1 + 2 * BPS];
261   const int E = dst[-1 + 3 * BPS];
262   *(uint32_t*)(dst + 0 * BPS) = 0x01010101U * AVG3(A, B, C);
263   *(uint32_t*)(dst + 1 * BPS) = 0x01010101U * AVG3(B, C, D);
264   *(uint32_t*)(dst + 2 * BPS) = 0x01010101U * AVG3(C, D, E);
265   *(uint32_t*)(dst + 3 * BPS) = 0x01010101U * AVG3(D, E, E);
266 }
267 
DC4(uint8_t * dst)268 static void DC4(uint8_t *dst) {   // DC
269   uint32_t dc = 4;
270   int i;
271   for (i = 0; i < 4; ++i) dc += dst[i - BPS] + dst[-1 + i * BPS];
272   dc >>= 3;
273   for (i = 0; i < 4; ++i) memset(dst + i * BPS, dc, 4);
274 }
275 
RD4(uint8_t * dst)276 static void RD4(uint8_t *dst) {   // Down-right
277   const int I = dst[-1 + 0 * BPS];
278   const int J = dst[-1 + 1 * BPS];
279   const int K = dst[-1 + 2 * BPS];
280   const int L = dst[-1 + 3 * BPS];
281   const int X = dst[-1 - BPS];
282   const int A = dst[0 - BPS];
283   const int B = dst[1 - BPS];
284   const int C = dst[2 - BPS];
285   const int D = dst[3 - BPS];
286   DST(0, 3)                                     = AVG3(J, K, L);
287   DST(0, 2) = DST(1, 3)                         = AVG3(I, J, K);
288   DST(0, 1) = DST(1, 2) = DST(2, 3)             = AVG3(X, I, J);
289   DST(0, 0) = DST(1, 1) = DST(2, 2) = DST(3, 3) = AVG3(A, X, I);
290   DST(1, 0) = DST(2, 1) = DST(3, 2)             = AVG3(B, A, X);
291   DST(2, 0) = DST(3, 1)                         = AVG3(C, B, A);
292   DST(3, 0)                                     = AVG3(D, C, B);
293 }
294 
LD4(uint8_t * dst)295 static void LD4(uint8_t *dst) {   // Down-Left
296   const int A = dst[0 - BPS];
297   const int B = dst[1 - BPS];
298   const int C = dst[2 - BPS];
299   const int D = dst[3 - BPS];
300   const int E = dst[4 - BPS];
301   const int F = dst[5 - BPS];
302   const int G = dst[6 - BPS];
303   const int H = dst[7 - BPS];
304   DST(0, 0)                                     = AVG3(A, B, C);
305   DST(1, 0) = DST(0, 1)                         = AVG3(B, C, D);
306   DST(2, 0) = DST(1, 1) = DST(0, 2)             = AVG3(C, D, E);
307   DST(3, 0) = DST(2, 1) = DST(1, 2) = DST(0, 3) = AVG3(D, E, F);
308   DST(3, 1) = DST(2, 2) = DST(1, 3)             = AVG3(E, F, G);
309   DST(3, 2) = DST(2, 3)                         = AVG3(F, G, H);
310   DST(3, 3)                                     = AVG3(G, H, H);
311 }
312 
VR4(uint8_t * dst)313 static void VR4(uint8_t *dst) {   // Vertical-Right
314   const int I = dst[-1 + 0 * BPS];
315   const int J = dst[-1 + 1 * BPS];
316   const int K = dst[-1 + 2 * BPS];
317   const int X = dst[-1 - BPS];
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   DST(0, 0) = DST(1, 2) = AVG2(X, A);
323   DST(1, 0) = DST(2, 2) = AVG2(A, B);
324   DST(2, 0) = DST(3, 2) = AVG2(B, C);
325   DST(3, 0)             = AVG2(C, D);
326 
327   DST(0, 3) =             AVG3(K, J, I);
328   DST(0, 2) =             AVG3(J, I, X);
329   DST(0, 1) = DST(1, 3) = AVG3(I, X, A);
330   DST(1, 1) = DST(2, 3) = AVG3(X, A, B);
331   DST(2, 1) = DST(3, 3) = AVG3(A, B, C);
332   DST(3, 1) =             AVG3(B, C, D);
333 }
334 
VL4(uint8_t * dst)335 static void VL4(uint8_t *dst) {   // Vertical-Left
336   const int A = dst[0 - BPS];
337   const int B = dst[1 - BPS];
338   const int C = dst[2 - BPS];
339   const int D = dst[3 - BPS];
340   const int E = dst[4 - BPS];
341   const int F = dst[5 - BPS];
342   const int G = dst[6 - BPS];
343   const int H = dst[7 - BPS];
344   DST(0, 0) =             AVG2(A, B);
345   DST(1, 0) = DST(0, 2) = AVG2(B, C);
346   DST(2, 0) = DST(1, 2) = AVG2(C, D);
347   DST(3, 0) = DST(2, 2) = AVG2(D, E);
348 
349   DST(0, 1) =             AVG3(A, B, C);
350   DST(1, 1) = DST(0, 3) = AVG3(B, C, D);
351   DST(2, 1) = DST(1, 3) = AVG3(C, D, E);
352   DST(3, 1) = DST(2, 3) = AVG3(D, E, F);
353               DST(3, 2) = AVG3(E, F, G);
354               DST(3, 3) = AVG3(F, G, H);
355 }
356 
HU4(uint8_t * dst)357 static void HU4(uint8_t *dst) {   // Horizontal-Up
358   const int I = dst[-1 + 0 * BPS];
359   const int J = dst[-1 + 1 * BPS];
360   const int K = dst[-1 + 2 * BPS];
361   const int L = dst[-1 + 3 * BPS];
362   DST(0, 0) =             AVG2(I, J);
363   DST(2, 0) = DST(0, 1) = AVG2(J, K);
364   DST(2, 1) = DST(0, 2) = AVG2(K, L);
365   DST(1, 0) =             AVG3(I, J, K);
366   DST(3, 0) = DST(1, 1) = AVG3(J, K, L);
367   DST(3, 1) = DST(1, 2) = AVG3(K, L, L);
368   DST(3, 2) = DST(2, 2) =
369     DST(0, 3) = DST(1, 3) = DST(2, 3) = DST(3, 3) = L;
370 }
371 
HD4(uint8_t * dst)372 static void HD4(uint8_t *dst) {  // Horizontal-Down
373   const int I = dst[-1 + 0 * BPS];
374   const int J = dst[-1 + 1 * BPS];
375   const int K = dst[-1 + 2 * BPS];
376   const int L = dst[-1 + 3 * BPS];
377   const int X = dst[-1 - BPS];
378   const int A = dst[0 - BPS];
379   const int B = dst[1 - BPS];
380   const int C = dst[2 - BPS];
381 
382   DST(0, 0) = DST(2, 1) = AVG2(I, X);
383   DST(0, 1) = DST(2, 2) = AVG2(J, I);
384   DST(0, 2) = DST(2, 3) = AVG2(K, J);
385   DST(0, 3)             = AVG2(L, K);
386 
387   DST(3, 0)             = AVG3(A, B, C);
388   DST(2, 0)             = AVG3(X, A, B);
389   DST(1, 0) = DST(3, 1) = AVG3(I, X, A);
390   DST(1, 1) = DST(3, 2) = AVG3(J, I, X);
391   DST(1, 2) = DST(3, 3) = AVG3(K, J, I);
392   DST(1, 3)             = AVG3(L, K, J);
393 }
394 
395 #undef DST
396 #undef AVG3
397 #undef AVG2
398 
399 //------------------------------------------------------------------------------
400 // Chroma
401 
VE8uv(uint8_t * dst)402 static void VE8uv(uint8_t *dst) {    // vertical
403   int j;
404   for (j = 0; j < 8; ++j) {
405     memcpy(dst + j * BPS, dst - BPS, 8);
406   }
407 }
408 
HE8uv(uint8_t * dst)409 static void HE8uv(uint8_t *dst) {    // horizontal
410   int j;
411   for (j = 0; j < 8; ++j) {
412     memset(dst, dst[-1], 8);
413     dst += BPS;
414   }
415 }
416 
417 // helper for chroma-DC predictions
Put8x8uv(uint8_t value,uint8_t * dst)418 static WEBP_INLINE void Put8x8uv(uint8_t value, uint8_t* dst) {
419   int j;
420   for (j = 0; j < 8; ++j) {
421     memset(dst + j * BPS, value, 8);
422   }
423 }
424 
DC8uv(uint8_t * dst)425 static void DC8uv(uint8_t *dst) {     // DC
426   int dc0 = 8;
427   int i;
428   for (i = 0; i < 8; ++i) {
429     dc0 += dst[i - BPS] + dst[-1 + i * BPS];
430   }
431   Put8x8uv(dc0 >> 4, dst);
432 }
433 
DC8uvNoLeft(uint8_t * dst)434 static void DC8uvNoLeft(uint8_t *dst) {   // DC with no left samples
435   int dc0 = 4;
436   int i;
437   for (i = 0; i < 8; ++i) {
438     dc0 += dst[i - BPS];
439   }
440   Put8x8uv(dc0 >> 3, dst);
441 }
442 
DC8uvNoTop(uint8_t * dst)443 static void DC8uvNoTop(uint8_t *dst) {  // DC with no top samples
444   int dc0 = 4;
445   int i;
446   for (i = 0; i < 8; ++i) {
447     dc0 += dst[-1 + i * BPS];
448   }
449   Put8x8uv(dc0 >> 3, dst);
450 }
451 
DC8uvNoTopLeft(uint8_t * dst)452 static void DC8uvNoTopLeft(uint8_t *dst) {    // DC with nothing
453   Put8x8uv(0x80, dst);
454 }
455 
456 //------------------------------------------------------------------------------
457 // default C implementations
458 
459 const VP8PredFunc VP8PredLuma4[NUM_BMODES] = {
460   DC4, TM4, VE4, HE4, RD4, VR4, LD4, VL4, HD4, HU4
461 };
462 
463 const VP8PredFunc VP8PredLuma16[NUM_B_DC_MODES] = {
464   DC16, TM16, VE16, HE16,
465   DC16NoTop, DC16NoLeft, DC16NoTopLeft
466 };
467 
468 const VP8PredFunc VP8PredChroma8[NUM_B_DC_MODES] = {
469   DC8uv, TM8uv, VE8uv, HE8uv,
470   DC8uvNoTop, DC8uvNoLeft, DC8uvNoTopLeft
471 };
472 
473 //------------------------------------------------------------------------------
474 // Edge filtering functions
475 
476 // 4 pixels in, 2 pixels out
do_filter2(uint8_t * p,int step)477 static WEBP_INLINE void do_filter2(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) + VP8ksclip1[p1 - q1];  // in [-893,892]
480   const int a1 = VP8ksclip2[(a + 4) >> 3];            // in [-16,15]
481   const int a2 = VP8ksclip2[(a + 3) >> 3];
482   p[-step] = VP8kclip1[p0 + a2];
483   p[    0] = VP8kclip1[q0 - a1];
484 }
485 
486 // 4 pixels in, 4 pixels out
do_filter4(uint8_t * p,int step)487 static WEBP_INLINE void do_filter4(uint8_t* p, int step) {
488   const int p1 = p[-2*step], p0 = p[-step], q0 = p[0], q1 = p[step];
489   const int a = 3 * (q0 - p0);
490   const int a1 = VP8ksclip2[(a + 4) >> 3];
491   const int a2 = VP8ksclip2[(a + 3) >> 3];
492   const int a3 = (a1 + 1) >> 1;
493   p[-2*step] = VP8kclip1[p1 + a3];
494   p[-  step] = VP8kclip1[p0 + a2];
495   p[      0] = VP8kclip1[q0 - a1];
496   p[   step] = VP8kclip1[q1 - a3];
497 }
498 
499 // 6 pixels in, 6 pixels out
do_filter6(uint8_t * p,int step)500 static WEBP_INLINE void do_filter6(uint8_t* p, int step) {
501   const int p2 = p[-3*step], p1 = p[-2*step], p0 = p[-step];
502   const int q0 = p[0], q1 = p[step], q2 = p[2*step];
503   const int a = VP8ksclip1[3 * (q0 - p0) + VP8ksclip1[p1 - q1]];
504   // a is in [-128,127], a1 in [-27,27], a2 in [-18,18] and a3 in [-9,9]
505   const int a1 = (27 * a + 63) >> 7;  // eq. to ((3 * a + 7) * 9) >> 7
506   const int a2 = (18 * a + 63) >> 7;  // eq. to ((2 * a + 7) * 9) >> 7
507   const int a3 = (9  * a + 63) >> 7;  // eq. to ((1 * a + 7) * 9) >> 7
508   p[-3*step] = VP8kclip1[p2 + a3];
509   p[-2*step] = VP8kclip1[p1 + a2];
510   p[-  step] = VP8kclip1[p0 + a1];
511   p[      0] = VP8kclip1[q0 - a1];
512   p[   step] = VP8kclip1[q1 - a2];
513   p[ 2*step] = VP8kclip1[q2 - a3];
514 }
515 
hev(const uint8_t * p,int step,int thresh)516 static WEBP_INLINE int hev(const uint8_t* p, int step, int thresh) {
517   const int p1 = p[-2*step], p0 = p[-step], q0 = p[0], q1 = p[step];
518   return (VP8kabs0[p1 - p0] > thresh) || (VP8kabs0[q1 - q0] > thresh);
519 }
520 
needs_filter(const uint8_t * p,int step,int t)521 static WEBP_INLINE int needs_filter(const uint8_t* p, int step, int t) {
522   const int p1 = p[-2 * step], p0 = p[-step], q0 = p[0], q1 = p[step];
523   return ((4 * VP8kabs0[p0 - q0] + VP8kabs0[p1 - q1]) <= t);
524 }
525 
needs_filter2(const uint8_t * p,int step,int t,int it)526 static WEBP_INLINE int needs_filter2(const uint8_t* p,
527                                      int step, int t, int it) {
528   const int p3 = p[-4 * step], p2 = p[-3 * step], p1 = p[-2 * step];
529   const int p0 = p[-step], q0 = p[0];
530   const int q1 = p[step], q2 = p[2 * step], q3 = p[3 * step];
531   if ((4 * VP8kabs0[p0 - q0] + VP8kabs0[p1 - q1]) > t) return 0;
532   return VP8kabs0[p3 - p2] <= it && VP8kabs0[p2 - p1] <= it &&
533          VP8kabs0[p1 - p0] <= it && VP8kabs0[q3 - q2] <= it &&
534          VP8kabs0[q2 - q1] <= it && VP8kabs0[q1 - q0] <= it;
535 }
536 
537 //------------------------------------------------------------------------------
538 // Simple In-loop filtering (Paragraph 15.2)
539 
SimpleVFilter16(uint8_t * p,int stride,int thresh)540 static void SimpleVFilter16(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, thresh2)) {
545       do_filter2(p + i, stride);
546     }
547   }
548 }
549 
SimpleHFilter16(uint8_t * p,int stride,int thresh)550 static void SimpleHFilter16(uint8_t* p, int stride, int thresh) {
551   int i;
552   const int thresh2 = 2 * thresh + 1;
553   for (i = 0; i < 16; ++i) {
554     if (needs_filter(p + i * stride, 1, thresh2)) {
555       do_filter2(p + i * stride, 1);
556     }
557   }
558 }
559 
SimpleVFilter16i(uint8_t * p,int stride,int thresh)560 static void SimpleVFilter16i(uint8_t* p, int stride, int thresh) {
561   int k;
562   for (k = 3; k > 0; --k) {
563     p += 4 * stride;
564     SimpleVFilter16(p, stride, thresh);
565   }
566 }
567 
SimpleHFilter16i(uint8_t * p,int stride,int thresh)568 static void SimpleHFilter16i(uint8_t* p, int stride, int thresh) {
569   int k;
570   for (k = 3; k > 0; --k) {
571     p += 4;
572     SimpleHFilter16(p, stride, thresh);
573   }
574 }
575 
576 //------------------------------------------------------------------------------
577 // Complex In-loop filtering (Paragraph 15.3)
578 
FilterLoop26(uint8_t * p,int hstride,int vstride,int size,int thresh,int ithresh,int hev_thresh)579 static WEBP_INLINE void FilterLoop26(uint8_t* p,
580                                      int hstride, int vstride, int size,
581                                      int thresh, int ithresh, int hev_thresh) {
582   const int thresh2 = 2 * thresh + 1;
583   while (size-- > 0) {
584     if (needs_filter2(p, hstride, thresh2, ithresh)) {
585       if (hev(p, hstride, hev_thresh)) {
586         do_filter2(p, hstride);
587       } else {
588         do_filter6(p, hstride);
589       }
590     }
591     p += vstride;
592   }
593 }
594 
FilterLoop24(uint8_t * p,int hstride,int vstride,int size,int thresh,int ithresh,int hev_thresh)595 static WEBP_INLINE void FilterLoop24(uint8_t* p,
596                                      int hstride, int vstride, int size,
597                                      int thresh, int ithresh, int hev_thresh) {
598   const int thresh2 = 2 * thresh + 1;
599   while (size-- > 0) {
600     if (needs_filter2(p, hstride, thresh2, ithresh)) {
601       if (hev(p, hstride, hev_thresh)) {
602         do_filter2(p, hstride);
603       } else {
604         do_filter4(p, hstride);
605       }
606     }
607     p += vstride;
608   }
609 }
610 
611 // on macroblock edges
VFilter16(uint8_t * p,int stride,int thresh,int ithresh,int hev_thresh)612 static void VFilter16(uint8_t* p, int stride,
613                       int thresh, int ithresh, int hev_thresh) {
614   FilterLoop26(p, stride, 1, 16, thresh, ithresh, hev_thresh);
615 }
616 
HFilter16(uint8_t * p,int stride,int thresh,int ithresh,int hev_thresh)617 static void HFilter16(uint8_t* p, int stride,
618                       int thresh, int ithresh, int hev_thresh) {
619   FilterLoop26(p, 1, stride, 16, thresh, ithresh, hev_thresh);
620 }
621 
622 // on three inner edges
VFilter16i(uint8_t * p,int stride,int thresh,int ithresh,int hev_thresh)623 static void VFilter16i(uint8_t* p, int stride,
624                        int thresh, int ithresh, int hev_thresh) {
625   int k;
626   for (k = 3; k > 0; --k) {
627     p += 4 * stride;
628     FilterLoop24(p, stride, 1, 16, thresh, ithresh, hev_thresh);
629   }
630 }
631 
HFilter16i(uint8_t * p,int stride,int thresh,int ithresh,int hev_thresh)632 static void HFilter16i(uint8_t* p, int stride,
633                        int thresh, int ithresh, int hev_thresh) {
634   int k;
635   for (k = 3; k > 0; --k) {
636     p += 4;
637     FilterLoop24(p, 1, stride, 16, thresh, ithresh, hev_thresh);
638   }
639 }
640 
641 // 8-pixels wide variant, for chroma filtering
VFilter8(uint8_t * u,uint8_t * v,int stride,int thresh,int ithresh,int hev_thresh)642 static void VFilter8(uint8_t* u, uint8_t* v, int stride,
643                      int thresh, int ithresh, int hev_thresh) {
644   FilterLoop26(u, stride, 1, 8, thresh, ithresh, hev_thresh);
645   FilterLoop26(v, stride, 1, 8, thresh, ithresh, hev_thresh);
646 }
647 
HFilter8(uint8_t * u,uint8_t * v,int stride,int thresh,int ithresh,int hev_thresh)648 static void HFilter8(uint8_t* u, uint8_t* v, int stride,
649                      int thresh, int ithresh, int hev_thresh) {
650   FilterLoop26(u, 1, stride, 8, thresh, ithresh, hev_thresh);
651   FilterLoop26(v, 1, stride, 8, thresh, ithresh, hev_thresh);
652 }
653 
VFilter8i(uint8_t * u,uint8_t * v,int stride,int thresh,int ithresh,int hev_thresh)654 static void VFilter8i(uint8_t* u, uint8_t* v, int stride,
655                       int thresh, int ithresh, int hev_thresh) {
656   FilterLoop24(u + 4 * stride, stride, 1, 8, thresh, ithresh, hev_thresh);
657   FilterLoop24(v + 4 * stride, stride, 1, 8, thresh, ithresh, hev_thresh);
658 }
659 
HFilter8i(uint8_t * u,uint8_t * v,int stride,int thresh,int ithresh,int hev_thresh)660 static void HFilter8i(uint8_t* u, uint8_t* v, int stride,
661                       int thresh, int ithresh, int hev_thresh) {
662   FilterLoop24(u + 4, 1, stride, 8, thresh, ithresh, hev_thresh);
663   FilterLoop24(v + 4, 1, stride, 8, thresh, ithresh, hev_thresh);
664 }
665 
666 //------------------------------------------------------------------------------
667 
668 VP8DecIdct2 VP8Transform;
669 VP8DecIdct VP8TransformAC3;
670 VP8DecIdct VP8TransformUV;
671 VP8DecIdct VP8TransformDC;
672 VP8DecIdct VP8TransformDCUV;
673 
674 VP8LumaFilterFunc VP8VFilter16;
675 VP8LumaFilterFunc VP8HFilter16;
676 VP8ChromaFilterFunc VP8VFilter8;
677 VP8ChromaFilterFunc VP8HFilter8;
678 VP8LumaFilterFunc VP8VFilter16i;
679 VP8LumaFilterFunc VP8HFilter16i;
680 VP8ChromaFilterFunc VP8VFilter8i;
681 VP8ChromaFilterFunc VP8HFilter8i;
682 VP8SimpleFilterFunc VP8SimpleVFilter16;
683 VP8SimpleFilterFunc VP8SimpleHFilter16;
684 VP8SimpleFilterFunc VP8SimpleVFilter16i;
685 VP8SimpleFilterFunc VP8SimpleHFilter16i;
686 
687 extern void VP8DspInitSSE2(void);
688 extern void VP8DspInitNEON(void);
689 extern void VP8DspInitMIPS32(void);
690 
691 static volatile VP8CPUInfo dec_last_cpuinfo_used =
692     (VP8CPUInfo)&dec_last_cpuinfo_used;
693 
VP8DspInit(void)694 void VP8DspInit(void) {
695   if (dec_last_cpuinfo_used == VP8GetCPUInfo) return;
696 
697   VP8InitClipTables();
698 
699   VP8TransformWHT = TransformWHT;
700   VP8Transform = TransformTwo;
701   VP8TransformUV = TransformUV;
702   VP8TransformDC = TransformDC;
703   VP8TransformDCUV = TransformDCUV;
704   VP8TransformAC3 = TransformAC3;
705 
706   VP8VFilter16 = VFilter16;
707   VP8HFilter16 = HFilter16;
708   VP8VFilter8 = VFilter8;
709   VP8HFilter8 = HFilter8;
710   VP8VFilter16i = VFilter16i;
711   VP8HFilter16i = HFilter16i;
712   VP8VFilter8i = VFilter8i;
713   VP8HFilter8i = HFilter8i;
714   VP8SimpleVFilter16 = SimpleVFilter16;
715   VP8SimpleHFilter16 = SimpleHFilter16;
716   VP8SimpleVFilter16i = SimpleVFilter16i;
717   VP8SimpleHFilter16i = SimpleHFilter16i;
718 
719   // If defined, use CPUInfo() to overwrite some pointers with faster versions.
720   if (VP8GetCPUInfo != NULL) {
721 #if defined(WEBP_USE_SSE2)
722     if (VP8GetCPUInfo(kSSE2)) {
723       VP8DspInitSSE2();
724     }
725 #elif defined(WEBP_USE_NEON)
726     if (VP8GetCPUInfo(kNEON)) {
727       VP8DspInitNEON();
728     }
729 #elif defined(WEBP_USE_MIPS32)
730     if (VP8GetCPUInfo(kMIPS32)) {
731       VP8DspInitMIPS32();
732     }
733 #endif
734   }
735   dec_last_cpuinfo_used = VP8GetCPUInfo;
736 }
737