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)
16extern "C" {
17#endif
18
19//------------------------------------------------------------------------------
20// run-time tables (~4k)
21
22static uint8_t abs0[255 + 255 + 1];     // abs(i)
23static uint8_t abs1[255 + 255 + 1];     // abs(i)>>1
24static int8_t sclip1[1020 + 1020 + 1];  // clips [-1020, 1020] to [-128, 127]
25static int8_t sclip2[112 + 112 + 1];    // clips [-112, 112] to [-16, 15]
26static 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)
30static volatile int tables_ok = 0;
31
32static 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
52static 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
62static const int kC1 = 20091 + (1 << 16);
63static const int kC2 = 35468;
64#define MUL(a, b) (((a) * (b)) >> 16)
65
66static 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
106static 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
113static 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
118static 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
128static 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
140static 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
167void (*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
174static 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}
187static void TM4(uint8_t *dst)   { TrueMotion(dst, 4); }
188static void TM8uv(uint8_t *dst) { TrueMotion(dst, 8); }
189static void TM16(uint8_t *dst)  { TrueMotion(dst, 16); }
190
191//------------------------------------------------------------------------------
192// 16x16
193
194static 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
201static 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
209static 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
216static 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
225static 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
234static 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
243static 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
253static 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
267static 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
279static 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
287static 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
306static 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
324static 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
346static 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
368static 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
383static 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
413static 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
420static 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
429static WEBP_INLINE void Put8x8uv(uint8_t value, uint8_t* dst) {
430  int j;
431#ifndef WEBP_REFERENCE_IMPLEMENTATION
432  const uint64_t v = (uint64_t)value * 0x0101010101010101ULL;
433  for (j = 0; j < 8; ++j) {
434    *(uint64_t*)(dst + j * BPS) = v;
435  }
436#else
437  for (j = 0; j < 8; ++j) memset(dst + j * BPS, value, 8);
438#endif
439}
440
441static void DC8uv(uint8_t *dst) {     // DC
442  int dc0 = 8;
443  int i;
444  for (i = 0; i < 8; ++i) {
445    dc0 += dst[i - BPS] + dst[-1 + i * BPS];
446  }
447  Put8x8uv(dc0 >> 4, dst);
448}
449
450static void DC8uvNoLeft(uint8_t *dst) {   // DC with no left samples
451  int dc0 = 4;
452  int i;
453  for (i = 0; i < 8; ++i) {
454    dc0 += dst[i - BPS];
455  }
456  Put8x8uv(dc0 >> 3, dst);
457}
458
459static void DC8uvNoTop(uint8_t *dst) {  // DC with no top samples
460  int dc0 = 4;
461  int i;
462  for (i = 0; i < 8; ++i) {
463    dc0 += dst[-1 + i * BPS];
464  }
465  Put8x8uv(dc0 >> 3, dst);
466}
467
468static void DC8uvNoTopLeft(uint8_t *dst) {    // DC with nothing
469  Put8x8uv(0x80, dst);
470}
471
472//------------------------------------------------------------------------------
473// default C implementations
474
475const VP8PredFunc VP8PredLuma4[NUM_BMODES] = {
476  DC4, TM4, VE4, HE4, RD4, VR4, LD4, VL4, HD4, HU4
477};
478
479const VP8PredFunc VP8PredLuma16[NUM_B_DC_MODES] = {
480  DC16, TM16, VE16, HE16,
481  DC16NoTop, DC16NoLeft, DC16NoTopLeft
482};
483
484const VP8PredFunc VP8PredChroma8[NUM_B_DC_MODES] = {
485  DC8uv, TM8uv, VE8uv, HE8uv,
486  DC8uvNoTop, DC8uvNoLeft, DC8uvNoTopLeft
487};
488
489//------------------------------------------------------------------------------
490// Edge filtering functions
491
492// 4 pixels in, 2 pixels out
493static WEBP_INLINE void do_filter2(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) + sclip1[1020 + p1 - q1];
496  const int a1 = sclip2[112 + ((a + 4) >> 3)];
497  const int a2 = sclip2[112 + ((a + 3) >> 3)];
498  p[-step] = clip1[255 + p0 + a2];
499  p[    0] = clip1[255 + q0 - a1];
500}
501
502// 4 pixels in, 4 pixels out
503static WEBP_INLINE void do_filter4(uint8_t* p, int step) {
504  const int p1 = p[-2*step], p0 = p[-step], q0 = p[0], q1 = p[step];
505  const int a = 3 * (q0 - p0);
506  const int a1 = sclip2[112 + ((a + 4) >> 3)];
507  const int a2 = sclip2[112 + ((a + 3) >> 3)];
508  const int a3 = (a1 + 1) >> 1;
509  p[-2*step] = clip1[255 + p1 + a3];
510  p[-  step] = clip1[255 + p0 + a2];
511  p[      0] = clip1[255 + q0 - a1];
512  p[   step] = clip1[255 + q1 - a3];
513}
514
515// 6 pixels in, 6 pixels out
516static WEBP_INLINE void do_filter6(uint8_t* p, int step) {
517  const int p2 = p[-3*step], p1 = p[-2*step], p0 = p[-step];
518  const int q0 = p[0], q1 = p[step], q2 = p[2*step];
519  const int a = sclip1[1020 + 3 * (q0 - p0) + sclip1[1020 + p1 - q1]];
520  const int a1 = (27 * a + 63) >> 7;  // eq. to ((3 * a + 7) * 9) >> 7
521  const int a2 = (18 * a + 63) >> 7;  // eq. to ((2 * a + 7) * 9) >> 7
522  const int a3 = (9  * a + 63) >> 7;  // eq. to ((1 * a + 7) * 9) >> 7
523  p[-3*step] = clip1[255 + p2 + a3];
524  p[-2*step] = clip1[255 + p1 + a2];
525  p[-  step] = clip1[255 + p0 + a1];
526  p[      0] = clip1[255 + q0 - a1];
527  p[   step] = clip1[255 + q1 - a2];
528  p[ 2*step] = clip1[255 + q2 - a3];
529}
530
531static WEBP_INLINE int hev(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 (abs0[255 + p1 - p0] > thresh) || (abs0[255 + q1 - q0] > thresh);
534}
535
536static WEBP_INLINE int needs_filter(const uint8_t* p, int step, int thresh) {
537  const int p1 = p[-2*step], p0 = p[-step], q0 = p[0], q1 = p[step];
538  return (2 * abs0[255 + p0 - q0] + abs1[255 + p1 - q1]) <= thresh;
539}
540
541static WEBP_INLINE int needs_filter2(const uint8_t* p,
542                                     int step, int t, int it) {
543  const int p3 = p[-4*step], p2 = p[-3*step], p1 = p[-2*step], p0 = p[-step];
544  const int q0 = p[0], q1 = p[step], q2 = p[2*step], q3 = p[3*step];
545  if ((2 * abs0[255 + p0 - q0] + abs1[255 + p1 - q1]) > t)
546    return 0;
547  return abs0[255 + p3 - p2] <= it && abs0[255 + p2 - p1] <= it &&
548         abs0[255 + p1 - p0] <= it && abs0[255 + q3 - q2] <= it &&
549         abs0[255 + q2 - q1] <= it && abs0[255 + q1 - q0] <= it;
550}
551
552//------------------------------------------------------------------------------
553// Simple In-loop filtering (Paragraph 15.2)
554
555static void SimpleVFilter16(uint8_t* p, int stride, int thresh) {
556  int i;
557  for (i = 0; i < 16; ++i) {
558    if (needs_filter(p + i, stride, thresh)) {
559      do_filter2(p + i, stride);
560    }
561  }
562}
563
564static void SimpleHFilter16(uint8_t* p, int stride, int thresh) {
565  int i;
566  for (i = 0; i < 16; ++i) {
567    if (needs_filter(p + i * stride, 1, thresh)) {
568      do_filter2(p + i * stride, 1);
569    }
570  }
571}
572
573static void SimpleVFilter16i(uint8_t* p, int stride, int thresh) {
574  int k;
575  for (k = 3; k > 0; --k) {
576    p += 4 * stride;
577    SimpleVFilter16(p, stride, thresh);
578  }
579}
580
581static void SimpleHFilter16i(uint8_t* p, int stride, int thresh) {
582  int k;
583  for (k = 3; k > 0; --k) {
584    p += 4;
585    SimpleHFilter16(p, stride, thresh);
586  }
587}
588
589//------------------------------------------------------------------------------
590// Complex In-loop filtering (Paragraph 15.3)
591
592static WEBP_INLINE void FilterLoop26(uint8_t* p,
593                                     int hstride, int vstride, int size,
594                                     int thresh, int ithresh, int hev_thresh) {
595  while (size-- > 0) {
596    if (needs_filter2(p, hstride, thresh, ithresh)) {
597      if (hev(p, hstride, hev_thresh)) {
598        do_filter2(p, hstride);
599      } else {
600        do_filter6(p, hstride);
601      }
602    }
603    p += vstride;
604  }
605}
606
607static WEBP_INLINE void FilterLoop24(uint8_t* p,
608                                     int hstride, int vstride, int size,
609                                     int thresh, int ithresh, int hev_thresh) {
610  while (size-- > 0) {
611    if (needs_filter2(p, hstride, thresh, ithresh)) {
612      if (hev(p, hstride, hev_thresh)) {
613        do_filter2(p, hstride);
614      } else {
615        do_filter4(p, hstride);
616      }
617    }
618    p += vstride;
619  }
620}
621
622// on macroblock edges
623static void VFilter16(uint8_t* p, int stride,
624                      int thresh, int ithresh, int hev_thresh) {
625  FilterLoop26(p, stride, 1, 16, thresh, ithresh, hev_thresh);
626}
627
628static void HFilter16(uint8_t* p, int stride,
629                      int thresh, int ithresh, int hev_thresh) {
630  FilterLoop26(p, 1, stride, 16, thresh, ithresh, hev_thresh);
631}
632
633// on three inner edges
634static void VFilter16i(uint8_t* p, int stride,
635                       int thresh, int ithresh, int hev_thresh) {
636  int k;
637  for (k = 3; k > 0; --k) {
638    p += 4 * stride;
639    FilterLoop24(p, stride, 1, 16, thresh, ithresh, hev_thresh);
640  }
641}
642
643static void HFilter16i(uint8_t* p, int stride,
644                       int thresh, int ithresh, int hev_thresh) {
645  int k;
646  for (k = 3; k > 0; --k) {
647    p += 4;
648    FilterLoop24(p, 1, stride, 16, thresh, ithresh, hev_thresh);
649  }
650}
651
652// 8-pixels wide variant, for chroma filtering
653static void VFilter8(uint8_t* u, uint8_t* v, int stride,
654                     int thresh, int ithresh, int hev_thresh) {
655  FilterLoop26(u, stride, 1, 8, thresh, ithresh, hev_thresh);
656  FilterLoop26(v, stride, 1, 8, thresh, ithresh, hev_thresh);
657}
658
659static void HFilter8(uint8_t* u, uint8_t* v, int stride,
660                     int thresh, int ithresh, int hev_thresh) {
661  FilterLoop26(u, 1, stride, 8, thresh, ithresh, hev_thresh);
662  FilterLoop26(v, 1, stride, 8, thresh, ithresh, hev_thresh);
663}
664
665static void VFilter8i(uint8_t* u, uint8_t* v, int stride,
666                      int thresh, int ithresh, int hev_thresh) {
667  FilterLoop24(u + 4 * stride, stride, 1, 8, thresh, ithresh, hev_thresh);
668  FilterLoop24(v + 4 * stride, stride, 1, 8, thresh, ithresh, hev_thresh);
669}
670
671static void HFilter8i(uint8_t* u, uint8_t* v, int stride,
672                      int thresh, int ithresh, int hev_thresh) {
673  FilterLoop24(u + 4, 1, stride, 8, thresh, ithresh, hev_thresh);
674  FilterLoop24(v + 4, 1, stride, 8, thresh, ithresh, hev_thresh);
675}
676
677//------------------------------------------------------------------------------
678
679VP8DecIdct2 VP8Transform;
680VP8DecIdct VP8TransformUV;
681VP8DecIdct VP8TransformDC;
682VP8DecIdct VP8TransformDCUV;
683
684VP8LumaFilterFunc VP8VFilter16;
685VP8LumaFilterFunc VP8HFilter16;
686VP8ChromaFilterFunc VP8VFilter8;
687VP8ChromaFilterFunc VP8HFilter8;
688VP8LumaFilterFunc VP8VFilter16i;
689VP8LumaFilterFunc VP8HFilter16i;
690VP8ChromaFilterFunc VP8VFilter8i;
691VP8ChromaFilterFunc VP8HFilter8i;
692VP8SimpleFilterFunc VP8SimpleVFilter16;
693VP8SimpleFilterFunc VP8SimpleHFilter16;
694VP8SimpleFilterFunc VP8SimpleVFilter16i;
695VP8SimpleFilterFunc VP8SimpleHFilter16i;
696
697extern void VP8DspInitSSE2(void);
698extern void VP8DspInitNEON(void);
699
700void VP8DspInit(void) {
701  DspInitTables();
702
703  VP8Transform = TransformTwo;
704  VP8TransformUV = TransformUV;
705  VP8TransformDC = TransformDC;
706  VP8TransformDCUV = TransformDCUV;
707
708  VP8VFilter16 = VFilter16;
709  VP8HFilter16 = HFilter16;
710  VP8VFilter8 = VFilter8;
711  VP8HFilter8 = HFilter8;
712  VP8VFilter16i = VFilter16i;
713  VP8HFilter16i = HFilter16i;
714  VP8VFilter8i = VFilter8i;
715  VP8HFilter8i = HFilter8i;
716  VP8SimpleVFilter16 = SimpleVFilter16;
717  VP8SimpleHFilter16 = SimpleHFilter16;
718  VP8SimpleVFilter16i = SimpleVFilter16i;
719  VP8SimpleHFilter16i = SimpleHFilter16i;
720
721  // If defined, use CPUInfo() to overwrite some pointers with faster versions.
722  if (VP8GetCPUInfo) {
723#if defined(WEBP_USE_SSE2)
724    if (VP8GetCPUInfo(kSSE2)) {
725      VP8DspInitSSE2();
726    }
727#elif defined(WEBP_USE_NEON)
728    if (VP8GetCPUInfo(kNEON)) {
729      VP8DspInitNEON();
730    }
731#endif
732  }
733}
734
735#if defined(__cplusplus) || defined(c_plusplus)
736}    // extern "C"
737#endif
738