• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // Copyright 2010 Google Inc.
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 functions.
9 //
10 // Author: Skal (pascal.massimino@gmail.com)
11 
12 #include "vp8i.h"
13 
14 #if defined(__cplusplus) || defined(c_plusplus)
15 extern "C" {
16 #endif
17 
18 //-----------------------------------------------------------------------------
19 // run-time tables (~4k)
20 
21 static uint8_t abs0[255 + 255 + 1];     // abs(i)
22 static uint8_t abs1[255 + 255 + 1];     // abs(i)>>1
23 static int8_t sclip1[1020 + 1020 + 1];  // clips [-1020, 1020] to [-128, 127]
24 static int8_t sclip2[112 + 112 + 1];    // clips [-112, 112] to [-16, 15]
25 static uint8_t clip1[255 + 510 + 1];    // clips [-255,510] to [0,255]
26 
27 // We declare this variable 'volatile' to prevent instruction reordering
28 // and make sure it's set to true _last_ (so as to be thread-safe)
29 static volatile int tables_ok = 0;
30 
VP8DspInitTables(void)31 void VP8DspInitTables(void) {
32   if (!tables_ok) {
33     int i;
34     for (i = -255; i <= 255; ++i) {
35       abs0[255 + i] = (i < 0) ? -i : i;
36       abs1[255 + i] = abs0[255 + i] >> 1;
37     }
38     for (i = -1020; i <= 1020; ++i) {
39       sclip1[1020 + i] = (i < -128) ? -128 : (i > 127) ? 127 : i;
40     }
41     for (i = -112; i <= 112; ++i) {
42       sclip2[112 + i] = (i < -16) ? -16 : (i > 15) ? 15 : i;
43     }
44     for (i = -255; i <= 255 + 255; ++i) {
45       clip1[255 + i] = (i < 0) ? 0 : (i > 255) ? 255 : i;
46     }
47     tables_ok = 1;
48   }
49 }
50 
clip_8b(int v)51 static inline uint8_t clip_8b(int v) {
52   return (!(v & ~0xff)) ? v : (v < 0) ? 0 : 255;
53 }
54 
55 //-----------------------------------------------------------------------------
56 // Transforms (Paragraph 14.4)
57 
58 #define STORE(x, y, v) \
59   dst[x + y * BPS] = clip_8b(dst[x + y * BPS] + ((v) >> 3))
60 
61 static const int kC1 = 20091 + (1 << 16);
62 static const int kC2 = 35468;
63 #define MUL(a, b) (((a) * (b)) >> 16)
64 
TransformOne(const int16_t * in,uint8_t * dst)65 static void TransformOne(const int16_t* in, uint8_t* dst) {
66   int C[4 * 4], *tmp;
67   int i;
68   tmp = C;
69   for (i = 0; i < 4; ++i) {    // vertical pass
70     const int a = in[0] + in[8];    // [-4096, 4094]
71     const int b = in[0] - in[8];    // [-4095, 4095]
72     const int c = MUL(in[4], kC2) - MUL(in[12], kC1);   // [-3783, 3783]
73     const int d = MUL(in[4], kC1) + MUL(in[12], kC2);   // [-3785, 3781]
74     tmp[0] = a + d;   // [-7881, 7875]
75     tmp[1] = b + c;   // [-7878, 7878]
76     tmp[2] = b - c;   // [-7878, 7878]
77     tmp[3] = a - d;   // [-7877, 7879]
78     tmp += 4;
79     in++;
80   }
81   // Each pass is expanding the dynamic range by ~3.85 (upper bound).
82   // The exact value is (2. + (kC1 + kC2) / 65536).
83   // After the second pass, maximum interval is [-3794, 3794], assuming
84   // an input in [-2048, 2047] interval. We then need to add a dst value
85   // in the [0, 255] range.
86   // In the worst case scenario, the input to clip_8b() can be as large as
87   // [-60713, 60968].
88   tmp = C;
89   for (i = 0; i < 4; ++i) {    // horizontal pass
90     const int dc = tmp[0] + 4;
91     const int a =  dc +  tmp[8];
92     const int b =  dc -  tmp[8];
93     const int c = MUL(tmp[4], kC2) - MUL(tmp[12], kC1);
94     const int d = MUL(tmp[4], kC1) + MUL(tmp[12], kC2);
95     STORE(0, 0, a + d);
96     STORE(1, 0, b + c);
97     STORE(2, 0, b - c);
98     STORE(3, 0, a - d);
99     tmp++;
100     dst += BPS;
101   }
102 }
103 #undef MUL
104 
TransformTwo(const int16_t * in,uint8_t * dst,int do_two)105 static void TransformTwo(const int16_t* in, uint8_t* dst, int do_two) {
106   TransformOne(in, dst);
107   if (do_two) {
108     TransformOne(in + 16, dst + 4);
109   }
110 }
111 
TransformUV(const int16_t * in,uint8_t * dst)112 static void TransformUV(const int16_t* in, uint8_t* dst) {
113   VP8Transform(in + 0 * 16, dst, 1);
114   VP8Transform(in + 2 * 16, dst + 4 * BPS, 1);
115 }
116 
TransformDC(const int16_t * in,uint8_t * dst)117 static void TransformDC(const int16_t *in, uint8_t* 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 
TransformDCUV(const int16_t * in,uint8_t * dst)127 static void TransformDCUV(const int16_t* in, uint8_t* dst) {
128   if (in[0 * 16]) TransformDC(in + 0 * 16, dst);
129   if (in[1 * 16]) TransformDC(in + 1 * 16, dst + 4);
130   if (in[2 * 16]) TransformDC(in + 2 * 16, dst + 4 * BPS);
131   if (in[3 * 16]) TransformDC(in + 3 * 16, dst + 4 * BPS + 4);
132 }
133 
134 #undef STORE
135 
136 // default C implementations:
137 VP8Idct2 VP8Transform = TransformTwo;
138 VP8Idct VP8TransformUV = TransformUV;
139 VP8Idct VP8TransformDC = TransformDC;
140 VP8Idct VP8TransformDCUV = TransformDCUV;
141 
142 //-----------------------------------------------------------------------------
143 // Paragraph 14.3
144 
TransformWHT(const int16_t * in,int16_t * out)145 static void TransformWHT(const int16_t* in, int16_t* out) {
146   int tmp[16];
147   int i;
148   for (i = 0; i < 4; ++i) {
149     const int a0 = in[0 + i] + in[12 + i];
150     const int a1 = in[4 + i] + in[ 8 + i];
151     const int a2 = in[4 + i] - in[ 8 + i];
152     const int a3 = in[0 + i] - in[12 + i];
153     tmp[0  + i] = a0 + a1;
154     tmp[8  + i] = a0 - a1;
155     tmp[4  + i] = a3 + a2;
156     tmp[12 + i] = a3 - a2;
157   }
158   for (i = 0; i < 4; ++i) {
159     const int dc = tmp[0 + i * 4] + 3;    // w/ rounder
160     const int a0 = dc             + tmp[3 + i * 4];
161     const int a1 = tmp[1 + i * 4] + tmp[2 + i * 4];
162     const int a2 = tmp[1 + i * 4] - tmp[2 + i * 4];
163     const int a3 = dc             - tmp[3 + i * 4];
164     out[ 0] = (a0 + a1) >> 3;
165     out[16] = (a3 + a2) >> 3;
166     out[32] = (a0 - a1) >> 3;
167     out[48] = (a3 - a2) >> 3;
168     out += 64;
169   }
170 }
171 
172 void (*VP8TransformWHT)(const int16_t* in, int16_t* out) = TransformWHT;
173 
174 //-----------------------------------------------------------------------------
175 // Intra predictions
176 
177 #define OUT(x, y) dst[(x) + (y) * BPS]
178 
TrueMotion(uint8_t * dst,int size)179 static inline void TrueMotion(uint8_t *dst, int size) {
180   const uint8_t* top = dst - BPS;
181   const uint8_t* const clip0 = clip1 + 255 - 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(uint8_t * dst)192 static void TM4(uint8_t *dst)   { TrueMotion(dst, 4); }
TM8uv(uint8_t * dst)193 static void TM8uv(uint8_t *dst) { TrueMotion(dst, 8); }
TM16(uint8_t * dst)194 static void TM16(uint8_t *dst)  { TrueMotion(dst, 16); }
195 
196 //-----------------------------------------------------------------------------
197 // 16x16
198 
VE16(uint8_t * dst)199 static void VE16(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(uint8_t * dst)206 static void HE16(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 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(uint8_t * dst)221 static void DC16(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(uint8_t * dst)230 static void DC16NoTop(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(uint8_t * dst)239 static void DC16NoLeft(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(uint8_t * dst)248 static void DC16NoTopLeft(uint8_t *dst) {  // DC with no top and left samples
249   Put16(0x80, dst);
250 }
251 
252 //-----------------------------------------------------------------------------
253 // 4x4
254 
255 #define AVG3(a, b, c) (((a) + 2 * (b) + (c) + 2) >> 2)
256 #define AVG2(a, b) (((a) + (b) + 1) >> 1)
257 
VE4(uint8_t * dst)258 static void VE4(uint8_t *dst) {    // vertical
259   const uint8_t* top = dst - BPS;
260   const uint8_t vals[4] = {
261     AVG3(top[-1], top[0], top[1]),
262     AVG3(top[ 0], top[1], top[2]),
263     AVG3(top[ 1], top[2], top[3]),
264     AVG3(top[ 2], top[3], top[4])
265   };
266   int i;
267   for (i = 0; i < 4; ++i) {
268     memcpy(dst + i * BPS, vals, sizeof(vals));
269   }
270 }
271 
HE4(uint8_t * dst)272 static void HE4(uint8_t *dst) {    // horizontal
273   const int A = dst[-1 - BPS];
274   const int B = dst[-1];
275   const int C = dst[-1 + BPS];
276   const int D = dst[-1 + 2 * BPS];
277   const int E = dst[-1 + 3 * BPS];
278   *(uint32_t*)(dst + 0 * BPS) = 0x01010101U * AVG3(A, B, C);
279   *(uint32_t*)(dst + 1 * BPS) = 0x01010101U * AVG3(B, C, D);
280   *(uint32_t*)(dst + 2 * BPS) = 0x01010101U * AVG3(C, D, E);
281   *(uint32_t*)(dst + 3 * BPS) = 0x01010101U * AVG3(D, E, E);
282 }
283 
DC4(uint8_t * dst)284 static void DC4(uint8_t *dst) {   // DC
285   uint32_t dc = 4;
286   int i;
287   for (i = 0; i < 4; ++i) dc += dst[i - BPS] + dst[-1 + i * BPS];
288   dc >>= 3;
289   for (i = 0; i < 4; ++i) memset(dst + i * BPS, dc, 4);
290 }
291 
RD4(uint8_t * dst)292 static void RD4(uint8_t *dst) {   // Down-right
293   const int I = dst[-1 + 0 * BPS];
294   const int J = dst[-1 + 1 * BPS];
295   const int K = dst[-1 + 2 * BPS];
296   const int L = dst[-1 + 3 * BPS];
297   const int X = dst[-1 - BPS];
298   const int A = dst[0 - BPS];
299   const int B = dst[1 - BPS];
300   const int C = dst[2 - BPS];
301   const int D = dst[3 - BPS];
302   OUT(0, 3)                                     = AVG3(J, K, L);
303   OUT(0, 2) = OUT(1, 3)                         = AVG3(I, J, K);
304   OUT(0, 1) = OUT(1, 2) = OUT(2, 3)             = AVG3(X, I, J);
305   OUT(0, 0) = OUT(1, 1) = OUT(2, 2) = OUT(3, 3) = AVG3(A, X, I);
306   OUT(1, 0) = OUT(2, 1) = OUT(3, 2)             = AVG3(B, A, X);
307   OUT(2, 0) = OUT(3, 1)                         = AVG3(C, B, A);
308   OUT(3, 0)                                     = AVG3(D, C, B);
309 }
310 
LD4(uint8_t * dst)311 static void LD4(uint8_t *dst) {   // Down-Left
312   const int A = dst[0 - BPS];
313   const int B = dst[1 - BPS];
314   const int C = dst[2 - BPS];
315   const int D = dst[3 - BPS];
316   const int E = dst[4 - BPS];
317   const int F = dst[5 - BPS];
318   const int G = dst[6 - BPS];
319   const int H = dst[7 - BPS];
320   OUT(0, 0)                                     = AVG3(A, B, C);
321   OUT(1, 0) = OUT(0, 1)                         = AVG3(B, C, D);
322   OUT(2, 0) = OUT(1, 1) = OUT(0, 2)             = AVG3(C, D, E);
323   OUT(3, 0) = OUT(2, 1) = OUT(1, 2) = OUT(0, 3) = AVG3(D, E, F);
324   OUT(3, 1) = OUT(2, 2) = OUT(1, 3)             = AVG3(E, F, G);
325   OUT(3, 2) = OUT(2, 3)                         = AVG3(F, G, H);
326   OUT(3, 3)                                     = AVG3(G, H, H);
327 }
328 
VR4(uint8_t * dst)329 static void VR4(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   OUT(0, 0) = OUT(1, 2) = AVG2(X, A);
339   OUT(1, 0) = OUT(2, 2) = AVG2(A, B);
340   OUT(2, 0) = OUT(3, 2) = AVG2(B, C);
341   OUT(3, 0)             = AVG2(C, D);
342 
343   OUT(0, 3) =             AVG3(K, J, I);
344   OUT(0, 2) =             AVG3(J, I, X);
345   OUT(0, 1) = OUT(1, 3) = AVG3(I, X, A);
346   OUT(1, 1) = OUT(2, 3) = AVG3(X, A, B);
347   OUT(2, 1) = OUT(3, 3) = AVG3(A, B, C);
348   OUT(3, 1) =             AVG3(B, C, D);
349 }
350 
VL4(uint8_t * dst)351 static void VL4(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   OUT(0, 0) =             AVG2(A, B);
361   OUT(1, 0) = OUT(0, 2) = AVG2(B, C);
362   OUT(2, 0) = OUT(1, 2) = AVG2(C, D);
363   OUT(3, 0) = OUT(2, 2) = AVG2(D, E);
364 
365   OUT(0, 1) =             AVG3(A, B, C);
366   OUT(1, 1) = OUT(0, 3) = AVG3(B, C, D);
367   OUT(2, 1) = OUT(1, 3) = AVG3(C, D, E);
368   OUT(3, 1) = OUT(2, 3) = AVG3(D, E, F);
369               OUT(3, 2) = AVG3(E, F, G);
370               OUT(3, 3) = AVG3(F, G, H);
371 }
372 
HU4(uint8_t * dst)373 static void HU4(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   OUT(0, 0) =             AVG2(I, J);
379   OUT(2, 0) = OUT(0, 1) = AVG2(J, K);
380   OUT(2, 1) = OUT(0, 2) = AVG2(K, L);
381   OUT(1, 0) =             AVG3(I, J, K);
382   OUT(3, 0) = OUT(1, 1) = AVG3(J, K, L);
383   OUT(3, 1) = OUT(1, 2) = AVG3(K, L, L);
384   OUT(3, 2) = OUT(2, 2) =
385     OUT(0, 3) = OUT(1, 3) = OUT(2, 3) = OUT(3, 3) = L;
386 }
387 
HD4(uint8_t * dst)388 static void HD4(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   OUT(0, 0) = OUT(2, 1) = AVG2(I, X);
399   OUT(0, 1) = OUT(2, 2) = AVG2(J, I);
400   OUT(0, 2) = OUT(2, 3) = AVG2(K, J);
401   OUT(0, 3)             = AVG2(L, K);
402 
403   OUT(3, 0)             = AVG3(A, B, C);
404   OUT(2, 0)             = AVG3(X, A, B);
405   OUT(1, 0) = OUT(3, 1) = AVG3(I, X, A);
406   OUT(1, 1) = OUT(3, 2) = AVG3(J, I, X);
407   OUT(1, 2) = OUT(3, 3) = AVG3(K, J, I);
408   OUT(1, 3)             = AVG3(L, K, J);
409 }
410 
411 #undef AVG3
412 #undef AVG2
413 
414 //-----------------------------------------------------------------------------
415 // Chroma
416 
VE8uv(uint8_t * dst)417 static void VE8uv(uint8_t *dst) {    // vertical
418   int j;
419   for (j = 0; j < 8; ++j) {
420     memcpy(dst + j * BPS, dst - BPS, 8);
421   }
422 }
423 
HE8uv(uint8_t * dst)424 static void HE8uv(uint8_t *dst) {    // horizontal
425   int j;
426   for (j = 0; j < 8; ++j) {
427     memset(dst, dst[-1], 8);
428     dst += BPS;
429   }
430 }
431 
432 // helper for chroma-DC predictions
Put8x8uv(uint64_t v,uint8_t * dst)433 static inline void Put8x8uv(uint64_t v, uint8_t* dst) {
434   int j;
435   for (j = 0; j < 8; ++j) {
436     *(uint64_t*)(dst + j * BPS) = v;
437   }
438 }
439 
DC8uv(uint8_t * dst)440 static void DC8uv(uint8_t *dst) {     // DC
441   int dc0 = 8;
442   int i;
443   for (i = 0; i < 8; ++i) {
444     dc0 += dst[i - BPS] + dst[-1 + i * BPS];
445   }
446   Put8x8uv((uint64_t)((dc0 >> 4) * 0x0101010101010101ULL), dst);
447 }
448 
DC8uvNoLeft(uint8_t * dst)449 static void DC8uvNoLeft(uint8_t *dst) {   // DC with no left samples
450   int dc0 = 4;
451   int i;
452   for (i = 0; i < 8; ++i) {
453     dc0 += dst[i - BPS];
454   }
455   Put8x8uv((uint64_t)((dc0 >> 3) * 0x0101010101010101ULL), dst);
456 }
457 
DC8uvNoTop(uint8_t * dst)458 static void DC8uvNoTop(uint8_t *dst) {  // DC with no top samples
459   int dc0 = 4;
460   int i;
461   for (i = 0; i < 8; ++i) {
462     dc0 += dst[-1 + i * BPS];
463   }
464   Put8x8uv((uint64_t)((dc0 >> 3) * 0x0101010101010101ULL), dst);
465 }
466 
DC8uvNoTopLeft(uint8_t * dst)467 static void DC8uvNoTopLeft(uint8_t *dst) {    // DC with nothing
468   Put8x8uv(0x8080808080808080ULL, dst);
469 }
470 
471 //-----------------------------------------------------------------------------
472 // default C implementations
473 
474 VP8PredFunc VP8PredLuma4[NUM_BMODES] = {
475   DC4, TM4, VE4, HE4, RD4, VR4, LD4, VL4, HD4, HU4
476 };
477 
478 VP8PredFunc VP8PredLuma16[NUM_B_DC_MODES] = {
479   DC16, TM16, VE16, HE16,
480   DC16NoTop, DC16NoLeft, DC16NoTopLeft
481 };
482 
483 VP8PredFunc VP8PredChroma8[NUM_B_DC_MODES] = {
484   DC8uv, TM8uv, VE8uv, HE8uv,
485   DC8uvNoTop, DC8uvNoLeft, DC8uvNoTopLeft
486 };
487 
488 //-----------------------------------------------------------------------------
489 // Edge filtering functions
490 
491 // 4 pixels in, 2 pixels out
do_filter2(uint8_t * p,int step)492 static inline void do_filter2(uint8_t* p, int step) {
493   const int p1 = p[-2*step], p0 = p[-step], q0 = p[0], q1 = p[step];
494   const int a = 3 * (q0 - p0) + sclip1[1020 + p1 - q1];
495   const int a1 = sclip2[112 + ((a + 4) >> 3)];
496   const int a2 = sclip2[112 + ((a + 3) >> 3)];
497   p[-step] = clip1[255 + p0 + a2];
498   p[    0] = clip1[255 + q0 - a1];
499 }
500 
501 // 4 pixels in, 4 pixels out
do_filter4(uint8_t * p,int step)502 static inline void do_filter4(uint8_t* p, int step) {
503   const int p1 = p[-2*step], p0 = p[-step], q0 = p[0], q1 = p[step];
504   const int a = 3 * (q0 - p0);
505   const int a1 = sclip2[112 + ((a + 4) >> 3)];
506   const int a2 = sclip2[112 + ((a + 3) >> 3)];
507   const int a3 = (a1 + 1) >> 1;
508   p[-2*step] = clip1[255 + p1 + a3];
509   p[-  step] = clip1[255 + p0 + a2];
510   p[      0] = clip1[255 + q0 - a1];
511   p[   step] = clip1[255 + q1 - a3];
512 }
513 
514 // 6 pixels in, 6 pixels out
do_filter6(uint8_t * p,int step)515 static inline void do_filter6(uint8_t* p, int step) {
516   const int p2 = p[-3*step], p1 = p[-2*step], p0 = p[-step];
517   const int q0 = p[0], q1 = p[step], q2 = p[2*step];
518   const int a = sclip1[1020 + 3 * (q0 - p0) + sclip1[1020 + p1 - q1]];
519   const int a1 = (27 * a + 63) >> 7;  // eq. to ((3 * a + 7) * 9) >> 7
520   const int a2 = (18 * a + 63) >> 7;  // eq. to ((2 * a + 7) * 9) >> 7
521   const int a3 = (9  * a + 63) >> 7;  // eq. to ((1 * a + 7) * 9) >> 7
522   p[-3*step] = clip1[255 + p2 + a3];
523   p[-2*step] = clip1[255 + p1 + a2];
524   p[-  step] = clip1[255 + p0 + a1];
525   p[      0] = clip1[255 + q0 - a1];
526   p[   step] = clip1[255 + q1 - a2];
527   p[ 2*step] = clip1[255 + q2 - a3];
528 }
529 
hev(const uint8_t * p,int step,int thresh)530 static inline int hev(const uint8_t* p, int step, int thresh) {
531   const int p1 = p[-2*step], p0 = p[-step], q0 = p[0], q1 = p[step];
532   return (abs0[255 + p1 - p0] > thresh) || (abs0[255 + q1 - q0] > thresh);
533 }
534 
needs_filter(const uint8_t * p,int step,int thresh)535 static inline int needs_filter(const uint8_t* p, int step, int thresh) {
536   const int p1 = p[-2*step], p0 = p[-step], q0 = p[0], q1 = p[step];
537   return (2 * abs0[255 + p0 - q0] + abs1[255 + p1 - q1]) <= thresh;
538 }
539 
needs_filter2(const uint8_t * p,int step,int t,int it)540 static inline int needs_filter2(const uint8_t* p, int step, int t, int it) {
541   const int p3 = p[-4*step], p2 = p[-3*step], p1 = p[-2*step], p0 = p[-step];
542   const int q0 = p[0], q1 = p[step], q2 = p[2*step], q3 = p[3*step];
543   if ((2 * abs0[255 + p0 - q0] + abs1[255 + p1 - q1]) > t)
544     return 0;
545   return abs0[255 + p3 - p2] <= it && abs0[255 + p2 - p1] <= it &&
546          abs0[255 + p1 - p0] <= it && abs0[255 + q3 - q2] <= it &&
547          abs0[255 + q2 - q1] <= it && abs0[255 + q1 - q0] <= it;
548 }
549 
550 //-----------------------------------------------------------------------------
551 // Simple In-loop filtering (Paragraph 15.2)
552 
SimpleVFilter16(uint8_t * p,int stride,int thresh)553 static void SimpleVFilter16(uint8_t* p, int stride, int thresh) {
554   int i;
555   for (i = 0; i < 16; ++i) {
556     if (needs_filter(p + i, stride, thresh)) {
557       do_filter2(p + i, stride);
558     }
559   }
560 }
561 
SimpleHFilter16(uint8_t * p,int stride,int thresh)562 static void SimpleHFilter16(uint8_t* p, int stride, int thresh) {
563   int i;
564   for (i = 0; i < 16; ++i) {
565     if (needs_filter(p + i * stride, 1, thresh)) {
566       do_filter2(p + i * stride, 1);
567     }
568   }
569 }
570 
SimpleVFilter16i(uint8_t * p,int stride,int thresh)571 static void SimpleVFilter16i(uint8_t* p, int stride, int thresh) {
572   int k;
573   for (k = 3; k > 0; --k) {
574     p += 4 * stride;
575     SimpleVFilter16(p, stride, thresh);
576   }
577 }
578 
SimpleHFilter16i(uint8_t * p,int stride,int thresh)579 static void SimpleHFilter16i(uint8_t* p, int stride, int thresh) {
580   int k;
581   for (k = 3; k > 0; --k) {
582     p += 4;
583     SimpleHFilter16(p, stride, thresh);
584   }
585 }
586 
587 //-----------------------------------------------------------------------------
588 // Complex In-loop filtering (Paragraph 15.3)
589 
FilterLoop26(uint8_t * p,int hstride,int vstride,int size,int thresh,int ithresh,int hev_thresh)590 static inline void FilterLoop26(uint8_t* p, int hstride, int vstride, int size,
591                                 int thresh, int ithresh, int hev_thresh) {
592   while (size-- > 0) {
593     if (needs_filter2(p, hstride, thresh, ithresh)) {
594       if (hev(p, hstride, hev_thresh)) {
595         do_filter2(p, hstride);
596       } else {
597         do_filter6(p, hstride);
598       }
599     }
600     p += vstride;
601   }
602 }
603 
FilterLoop24(uint8_t * p,int hstride,int vstride,int size,int thresh,int ithresh,int hev_thresh)604 static inline void FilterLoop24(uint8_t* p, int hstride, int vstride, int size,
605                                 int thresh, int ithresh, int hev_thresh) {
606   while (size-- > 0) {
607     if (needs_filter2(p, hstride, thresh, ithresh)) {
608       if (hev(p, hstride, hev_thresh)) {
609         do_filter2(p, hstride);
610       } else {
611         do_filter4(p, hstride);
612       }
613     }
614     p += vstride;
615   }
616 }
617 
618 // on macroblock edges
VFilter16(uint8_t * p,int stride,int thresh,int ithresh,int hev_thresh)619 static void VFilter16(uint8_t* p, int stride,
620                       int thresh, int ithresh, int hev_thresh) {
621   FilterLoop26(p, stride, 1, 16, thresh, ithresh, hev_thresh);
622 }
623 
HFilter16(uint8_t * p,int stride,int thresh,int ithresh,int hev_thresh)624 static void HFilter16(uint8_t* p, int stride,
625                       int thresh, int ithresh, int hev_thresh) {
626   FilterLoop26(p, 1, stride, 16, thresh, ithresh, hev_thresh);
627 }
628 
629 // on three inner edges
VFilter16i(uint8_t * p,int stride,int thresh,int ithresh,int hev_thresh)630 static void VFilter16i(uint8_t* p, int stride,
631                        int thresh, int ithresh, int hev_thresh) {
632   int k;
633   for (k = 3; k > 0; --k) {
634     p += 4 * stride;
635     FilterLoop24(p, stride, 1, 16, thresh, ithresh, hev_thresh);
636   }
637 }
638 
HFilter16i(uint8_t * p,int stride,int thresh,int ithresh,int hev_thresh)639 static void HFilter16i(uint8_t* p, int stride,
640                        int thresh, int ithresh, int hev_thresh) {
641   int k;
642   for (k = 3; k > 0; --k) {
643     p += 4;
644     FilterLoop24(p, 1, stride, 16, thresh, ithresh, hev_thresh);
645   }
646 }
647 
648 // 8-pixels wide variant, for chroma filtering
VFilter8(uint8_t * u,uint8_t * v,int stride,int thresh,int ithresh,int hev_thresh)649 static void VFilter8(uint8_t* u, uint8_t* v, int stride,
650                      int thresh, int ithresh, int hev_thresh) {
651   FilterLoop26(u, stride, 1, 8, thresh, ithresh, hev_thresh);
652   FilterLoop26(v, stride, 1, 8, thresh, ithresh, hev_thresh);
653 }
654 
HFilter8(uint8_t * u,uint8_t * v,int stride,int thresh,int ithresh,int hev_thresh)655 static void HFilter8(uint8_t* u, uint8_t* v, int stride,
656                      int thresh, int ithresh, int hev_thresh) {
657   FilterLoop26(u, 1, stride, 8, thresh, ithresh, hev_thresh);
658   FilterLoop26(v, 1, stride, 8, thresh, ithresh, hev_thresh);
659 }
660 
VFilter8i(uint8_t * u,uint8_t * v,int stride,int thresh,int ithresh,int hev_thresh)661 static void VFilter8i(uint8_t* u, uint8_t* v, int stride,
662                       int thresh, int ithresh, int hev_thresh) {
663   FilterLoop24(u + 4 * stride, stride, 1, 8, thresh, ithresh, hev_thresh);
664   FilterLoop24(v + 4 * stride, stride, 1, 8, thresh, ithresh, hev_thresh);
665 }
666 
HFilter8i(uint8_t * u,uint8_t * v,int stride,int thresh,int ithresh,int hev_thresh)667 static void HFilter8i(uint8_t* u, uint8_t* v, int stride,
668                       int thresh, int ithresh, int hev_thresh) {
669   FilterLoop24(u + 4, 1, stride, 8, thresh, ithresh, hev_thresh);
670   FilterLoop24(v + 4, 1, stride, 8, thresh, ithresh, hev_thresh);
671 }
672 
673 //-----------------------------------------------------------------------------
674 
675 void (*VP8VFilter16)(uint8_t*, int, int, int, int) = VFilter16;
676 void (*VP8HFilter16)(uint8_t*, int, int, int, int) = HFilter16;
677 void (*VP8VFilter8)(uint8_t*, uint8_t*, int, int, int, int) = VFilter8;
678 void (*VP8HFilter8)(uint8_t*, uint8_t*, int, int, int, int) = HFilter8;
679 void (*VP8VFilter16i)(uint8_t*, int, int, int, int) = VFilter16i;
680 void (*VP8HFilter16i)(uint8_t*, int, int, int, int) = HFilter16i;
681 void (*VP8VFilter8i)(uint8_t*, uint8_t*, int, int, int, int) = VFilter8i;
682 void (*VP8HFilter8i)(uint8_t*, uint8_t*, int, int, int, int) = HFilter8i;
683 
684 void (*VP8SimpleVFilter16)(uint8_t*, int, int) = SimpleVFilter16;
685 void (*VP8SimpleHFilter16)(uint8_t*, int, int) = SimpleHFilter16;
686 void (*VP8SimpleVFilter16i)(uint8_t*, int, int) = SimpleVFilter16i;
687 void (*VP8SimpleHFilter16i)(uint8_t*, int, int) = SimpleHFilter16i;
688 
689 //-----------------------------------------------------------------------------
690 // SSE2 detection.
691 //
692 
693 #if defined(__pic__) && defined(__i386__)
GetCPUInfo(int cpu_info[4],int info_type)694 static inline void GetCPUInfo(int cpu_info[4], int info_type) {
695   __asm__ volatile (
696     "mov %%ebx, %%edi\n"
697     "cpuid\n"
698     "xchg %%edi, %%ebx\n"
699     : "=a"(cpu_info[0]), "=D"(cpu_info[1]), "=c"(cpu_info[2]), "=d"(cpu_info[3])
700     : "a"(info_type));
701 }
702 #elif defined(__i386__) || defined(__x86_64__)
GetCPUInfo(int cpu_info[4],int info_type)703 static inline void GetCPUInfo(int cpu_info[4], int info_type) {
704   __asm__ volatile (
705     "cpuid\n"
706     : "=a"(cpu_info[0]), "=b"(cpu_info[1]), "=c"(cpu_info[2]), "=d"(cpu_info[3])
707     : "a"(info_type));
708 }
709 #elif defined(_MSC_VER)  // Visual C++
710 #define GetCPUInfo __cpuid
711 #endif
712 
713 #if defined(__i386__) || defined(__x86_64__) || defined(_MSC_VER)
x86CPUInfo(CPUFeature feature)714 static int x86CPUInfo(CPUFeature feature) {
715   int cpu_info[4];
716   GetCPUInfo(cpu_info, 1);
717   if (feature == kSSE2) {
718     return 0 != (cpu_info[3] & 0x04000000);
719   }
720   if (feature == kSSE3) {
721     return 0 != (cpu_info[2] & 0x00000001);
722   }
723   return 0;
724 }
725 VP8CPUInfo VP8DecGetCPUInfo = x86CPUInfo;
726 #else
727 VP8CPUInfo VP8DecGetCPUInfo = NULL;
728 #endif
729 
730 //-----------------------------------------------------------------------------
731 
732 extern void VP8DspInitSSE2(void);
733 
VP8DspInit(void)734 void VP8DspInit(void) {
735   // If defined, use CPUInfo() to overwrite some pointers with faster versions.
736   if (VP8DecGetCPUInfo) {
737     if (VP8DecGetCPUInfo(kSSE2)) {
738 #if defined(__SSE2__) || defined(_MSC_VER)
739       VP8DspInitSSE2();
740 #endif
741     }
742     if (VP8DecGetCPUInfo(kSSE3)) {
743       // later we'll plug some SSE3 variant here
744     }
745   }
746 }
747 
748 #if defined(__cplusplus) || defined(c_plusplus)
749 }    // extern "C"
750 #endif
751