Eric Hassold | 9aea642 | 2011-01-04 17:22:46 -0800 | [diff] [blame] | 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(__SSE2__) |
| 15 | #include <emmintrin.h> |
| 16 | #endif |
| 17 | |
| 18 | #if defined(__cplusplus) || defined(c_plusplus) |
| 19 | extern "C" { |
| 20 | #endif |
| 21 | |
| 22 | //----------------------------------------------------------------------------- |
| 23 | // run-time tables (~4k) |
| 24 | |
| 25 | static uint8_t abs0[255 + 255 + 1]; // abs(i) |
| 26 | static uint8_t abs1[255 + 255 + 1]; // abs(i)>>1 |
| 27 | static int8_t sclip1[1020 + 1020 + 1]; // clips [-1020, 1020] to [-128, 127] |
| 28 | static int8_t sclip2[112 + 112 + 1]; // clips [-112, 112] to [-16, 15] |
| 29 | static uint8_t clip1[255 + 510 + 1]; // clips [-255,510] to [0,255] |
| 30 | |
Vikas Arora | 03d5e34 | 2011-06-02 23:59:44 +0530 | [diff] [blame^] | 31 | // We declare this variable 'volatile' to prevent instruction reordering |
| 32 | // and make sure it's set to true _last_ (so as to be thread-safe) |
| 33 | static volatile int tables_ok = 0; |
Eric Hassold | 9aea642 | 2011-01-04 17:22:46 -0800 | [diff] [blame] | 34 | |
Vikas Arora | 03d5e34 | 2011-06-02 23:59:44 +0530 | [diff] [blame^] | 35 | void VP8DspInitTables(void) { |
Eric Hassold | 9aea642 | 2011-01-04 17:22:46 -0800 | [diff] [blame] | 36 | if (!tables_ok) { |
| 37 | int i; |
| 38 | for (i = -255; i <= 255; ++i) { |
| 39 | abs0[255 + i] = (i < 0) ? -i : i; |
| 40 | abs1[255 + i] = abs0[255 + i] >> 1; |
| 41 | } |
| 42 | for (i = -1020; i <= 1020; ++i) { |
| 43 | sclip1[1020 + i] = (i < -128) ? -128 : (i > 127) ? 127 : i; |
| 44 | } |
| 45 | for (i = -112; i <= 112; ++i) { |
| 46 | sclip2[112 + i] = (i < -16) ? -16 : (i > 15) ? 15 : i; |
| 47 | } |
| 48 | for (i = -255; i <= 255 + 255; ++i) { |
| 49 | clip1[255 + i] = (i < 0) ? 0 : (i > 255) ? 255 : i; |
| 50 | } |
| 51 | tables_ok = 1; |
| 52 | } |
| 53 | } |
| 54 | |
| 55 | static inline uint8_t clip_8b(int v) { |
| 56 | return (!(v & ~0xff)) ? v : (v < 0) ? 0 : 255; |
| 57 | } |
| 58 | |
| 59 | //----------------------------------------------------------------------------- |
| 60 | // Transforms (Paragraph 14.4) |
| 61 | |
| 62 | #define STORE(x, y, v) \ |
| 63 | dst[x + y * BPS] = clip_8b(dst[x + y * BPS] + ((v) >> 3)) |
| 64 | |
| 65 | static const int kC1 = 20091 + (1 << 16); |
| 66 | static const int kC2 = 35468; |
| 67 | #define MUL(a, b) (((a) * (b)) >> 16) |
| 68 | |
| 69 | static void Transform(const int16_t* in, uint8_t* dst) { |
| 70 | int C[4 * 4], *tmp; |
| 71 | int i; |
| 72 | tmp = C; |
| 73 | for (i = 0; i < 4; ++i) { // vertical pass |
| 74 | const int a = in[0] + in[8]; // [-4096, 4094] |
| 75 | const int b = in[0] - in[8]; // [-4095, 4095] |
| 76 | const int c = MUL(in[4], kC2) - MUL(in[12], kC1); // [-3783, 3783] |
| 77 | const int d = MUL(in[4], kC1) + MUL(in[12], kC2); // [-3785, 3781] |
| 78 | tmp[0] = a + d; // [-7881, 7875] |
| 79 | tmp[1] = b + c; // [-7878, 7878] |
| 80 | tmp[2] = b - c; // [-7878, 7878] |
| 81 | tmp[3] = a - d; // [-7877, 7879] |
| 82 | tmp += 4; |
| 83 | in++; |
| 84 | } |
| 85 | // Each pass is expanding the dynamic range by ~3.85 (upper bound). |
| 86 | // The exact value is (2. + (kC1 + kC2) / 65536). |
| 87 | // After the second pass, maximum interval is [-3794, 3794], assuming |
| 88 | // an input in [-2048, 2047] interval. We then need to add a dst value |
| 89 | // in the [0, 255] range. |
| 90 | // In the worst case scenario, the input to clip_8b() can be as large as |
| 91 | // [-60713, 60968]. |
| 92 | tmp = C; |
| 93 | for (i = 0; i < 4; ++i) { // horizontal pass |
| 94 | const int dc = tmp[0] + 4; |
| 95 | const int a = dc + tmp[8]; |
| 96 | const int b = dc - tmp[8]; |
| 97 | const int c = MUL(tmp[4], kC2) - MUL(tmp[12], kC1); |
| 98 | const int d = MUL(tmp[4], kC1) + MUL(tmp[12], kC2); |
| 99 | STORE(0, 0, a + d); |
| 100 | STORE(1, 0, b + c); |
| 101 | STORE(2, 0, b - c); |
| 102 | STORE(3, 0, a - d); |
| 103 | tmp++; |
| 104 | dst += BPS; |
| 105 | } |
| 106 | } |
| 107 | #undef MUL |
| 108 | |
| 109 | static void TransformUV(const int16_t* in, uint8_t* dst) { |
| 110 | Transform(in + 0 * 16, dst); |
| 111 | Transform(in + 1 * 16, dst + 4); |
| 112 | Transform(in + 2 * 16, dst + 4 * BPS); |
| 113 | Transform(in + 3 * 16, dst + 4 * BPS + 4); |
| 114 | } |
| 115 | |
| 116 | static void TransformDC(const int16_t *in, uint8_t* dst) { |
| 117 | const int DC = in[0] + 4; |
| 118 | int i, j; |
| 119 | for (j = 0; j < 4; ++j) { |
| 120 | for (i = 0; i < 4; ++i) { |
| 121 | STORE(i, j, DC); |
| 122 | } |
| 123 | } |
| 124 | } |
| 125 | |
| 126 | static void TransformDCUV(const int16_t* in, uint8_t* dst) { |
| 127 | if (in[0 * 16]) TransformDC(in + 0 * 16, dst); |
| 128 | if (in[1 * 16]) TransformDC(in + 1 * 16, dst + 4); |
| 129 | if (in[2 * 16]) TransformDC(in + 2 * 16, dst + 4 * BPS); |
| 130 | if (in[3 * 16]) TransformDC(in + 3 * 16, dst + 4 * BPS + 4); |
| 131 | } |
| 132 | |
| 133 | #undef STORE |
| 134 | |
| 135 | // default C implementations: |
| 136 | VP8Idct VP8Transform = Transform; |
| 137 | VP8Idct VP8TransformUV = TransformUV; |
| 138 | VP8Idct VP8TransformDC = TransformDC; |
| 139 | VP8Idct VP8TransformDCUV = TransformDCUV; |
| 140 | |
| 141 | //----------------------------------------------------------------------------- |
| 142 | // Paragraph 14.3 |
| 143 | |
| 144 | static void TransformWHT(const int16_t* in, int16_t* out) { |
| 145 | int tmp[16]; |
| 146 | int i; |
| 147 | for (i = 0; i < 4; ++i) { |
| 148 | const int a0 = in[0 + i] + in[12 + i]; |
| 149 | const int a1 = in[4 + i] + in[ 8 + i]; |
| 150 | const int a2 = in[4 + i] - in[ 8 + i]; |
| 151 | const int a3 = in[0 + i] - in[12 + i]; |
| 152 | tmp[0 + i] = a0 + a1; |
| 153 | tmp[8 + i] = a0 - a1; |
| 154 | tmp[4 + i] = a3 + a2; |
| 155 | tmp[12 + i] = a3 - a2; |
| 156 | } |
| 157 | for (i = 0; i < 4; ++i) { |
| 158 | const int dc = tmp[0 + i * 4] + 3; // w/ rounder |
| 159 | const int a0 = dc + tmp[3 + i * 4]; |
| 160 | const int a1 = tmp[1 + i * 4] + tmp[2 + i * 4]; |
| 161 | const int a2 = tmp[1 + i * 4] - tmp[2 + i * 4]; |
| 162 | const int a3 = dc - tmp[3 + i * 4]; |
| 163 | out[ 0] = (a0 + a1) >> 3; |
| 164 | out[16] = (a3 + a2) >> 3; |
| 165 | out[32] = (a0 - a1) >> 3; |
| 166 | out[48] = (a3 - a2) >> 3; |
| 167 | out += 64; |
| 168 | } |
| 169 | } |
| 170 | |
| 171 | void (*VP8TransformWHT)(const int16_t* in, int16_t* out) = TransformWHT; |
| 172 | |
| 173 | //----------------------------------------------------------------------------- |
| 174 | // Intra predictions |
| 175 | |
| 176 | #define OUT(x, y) dst[(x) + (y) * BPS] |
| 177 | |
| 178 | static inline void TrueMotion(uint8_t *dst, int size) { |
| 179 | const uint8_t* top = dst - BPS; |
| 180 | const uint8_t* const clip0 = clip1 + 255 - top[-1]; |
| 181 | int y; |
| 182 | for (y = 0; y < size; ++y) { |
| 183 | const uint8_t* const clip = clip0 + dst[-1]; |
| 184 | int x; |
| 185 | for (x = 0; x < size; ++x) { |
| 186 | dst[x] = clip[top[x]]; |
| 187 | } |
| 188 | dst += BPS; |
| 189 | } |
| 190 | } |
| 191 | static void TM4(uint8_t *dst) { TrueMotion(dst, 4); } |
| 192 | static void TM8uv(uint8_t *dst) { TrueMotion(dst, 8); } |
| 193 | static void TM16(uint8_t *dst) { TrueMotion(dst, 16); } |
| 194 | |
| 195 | //----------------------------------------------------------------------------- |
| 196 | // 16x16 |
| 197 | |
| 198 | static void VE16(uint8_t *dst) { // vertical |
| 199 | int j; |
| 200 | for (j = 0; j < 16; ++j) { |
| 201 | memcpy(dst + j * BPS, dst - BPS, 16); |
| 202 | } |
| 203 | } |
| 204 | |
| 205 | static void HE16(uint8_t *dst) { // horizontal |
| 206 | int j; |
| 207 | for (j = 16; j > 0; --j) { |
| 208 | memset(dst, dst[-1], 16); |
| 209 | dst += BPS; |
| 210 | } |
| 211 | } |
| 212 | |
| 213 | static inline void Put16(int v, uint8_t* dst) { |
| 214 | int j; |
| 215 | for (j = 0; j < 16; ++j) { |
| 216 | memset(dst + j * BPS, v, 16); |
| 217 | } |
| 218 | } |
| 219 | |
| 220 | static void DC16(uint8_t *dst) { // DC |
| 221 | int DC = 16; |
| 222 | int j; |
| 223 | for (j = 0; j < 16; ++j) { |
| 224 | DC += dst[-1 + j * BPS] + dst[j - BPS]; |
| 225 | } |
| 226 | Put16(DC >> 5, dst); |
| 227 | } |
| 228 | |
| 229 | static void DC16NoTop(uint8_t *dst) { // DC with top samples not available |
| 230 | int DC = 8; |
| 231 | int j; |
| 232 | for (j = 0; j < 16; ++j) { |
| 233 | DC += dst[-1 + j * BPS]; |
| 234 | } |
| 235 | Put16(DC >> 4, dst); |
| 236 | } |
| 237 | |
| 238 | static void DC16NoLeft(uint8_t *dst) { // DC with left samples not available |
| 239 | int DC = 8; |
| 240 | int i; |
| 241 | for (i = 0; i < 16; ++i) { |
| 242 | DC += dst[i - BPS]; |
| 243 | } |
| 244 | Put16(DC >> 4, dst); |
| 245 | } |
| 246 | |
| 247 | static void DC16NoTopLeft(uint8_t *dst) { // DC with no top and left samples |
| 248 | Put16(0x80, dst); |
| 249 | } |
| 250 | |
| 251 | //----------------------------------------------------------------------------- |
| 252 | // 4x4 |
| 253 | |
| 254 | #define AVG3(a, b, c) (((a) + 2 * (b) + (c) + 2) >> 2) |
| 255 | #define AVG2(a, b) (((a) + (b) + 1) >> 1) |
| 256 | |
| 257 | static void VE4(uint8_t *dst) { // vertical |
| 258 | const uint8_t* top = dst - BPS; |
| 259 | const uint8_t vals[4] = { |
| 260 | AVG3(top[-1], top[0], top[1]), |
| 261 | AVG3(top[ 0], top[1], top[2]), |
| 262 | AVG3(top[ 1], top[2], top[3]), |
| 263 | AVG3(top[ 2], top[3], top[4]) |
| 264 | }; |
| 265 | int i; |
| 266 | for (i = 0; i < 4; ++i) { |
| 267 | memcpy(dst + i * BPS, vals, sizeof(vals)); |
| 268 | } |
| 269 | } |
| 270 | |
| 271 | static void HE4(uint8_t *dst) { // horizontal |
| 272 | const int A = dst[-1 - BPS]; |
| 273 | const int B = dst[-1]; |
| 274 | const int C = dst[-1 + BPS]; |
| 275 | const int D = dst[-1 + 2 * BPS]; |
| 276 | const int E = dst[-1 + 3 * BPS]; |
| 277 | *(uint32_t*)(dst + 0 * BPS) = 0x01010101U * AVG3(A, B, C); |
| 278 | *(uint32_t*)(dst + 1 * BPS) = 0x01010101U * AVG3(B, C, D); |
| 279 | *(uint32_t*)(dst + 2 * BPS) = 0x01010101U * AVG3(C, D, E); |
| 280 | *(uint32_t*)(dst + 3 * BPS) = 0x01010101U * AVG3(D, E, E); |
| 281 | } |
| 282 | |
| 283 | static void DC4(uint8_t *dst) { // DC |
| 284 | uint32_t dc = 4; |
| 285 | int i; |
| 286 | for (i = 0; i < 4; ++i) dc += dst[i - BPS] + dst[-1 + i * BPS]; |
| 287 | dc >>= 3; |
| 288 | for (i = 0; i < 4; ++i) memset(dst + i * BPS, dc, 4); |
| 289 | } |
| 290 | |
| 291 | static void RD4(uint8_t *dst) { // Down-right |
| 292 | const int I = dst[-1 + 0 * BPS]; |
| 293 | const int J = dst[-1 + 1 * BPS]; |
| 294 | const int K = dst[-1 + 2 * BPS]; |
| 295 | const int L = dst[-1 + 3 * BPS]; |
| 296 | const int X = dst[-1 - BPS]; |
| 297 | const int A = dst[0 - BPS]; |
| 298 | const int B = dst[1 - BPS]; |
| 299 | const int C = dst[2 - BPS]; |
| 300 | const int D = dst[3 - BPS]; |
| 301 | OUT(0, 3) = AVG3(J, K, L); |
| 302 | OUT(0, 2) = OUT(1, 3) = AVG3(I, J, K); |
| 303 | OUT(0, 1) = OUT(1, 2) = OUT(2, 3) = AVG3(X, I, J); |
| 304 | OUT(0, 0) = OUT(1, 1) = OUT(2, 2) = OUT(3, 3) = AVG3(A, X, I); |
| 305 | OUT(1, 0) = OUT(2, 1) = OUT(3, 2) = AVG3(B, A, X); |
| 306 | OUT(2, 0) = OUT(3, 1) = AVG3(C, B, A); |
| 307 | OUT(3, 0) = AVG3(D, C, B); |
| 308 | } |
| 309 | |
| 310 | static void LD4(uint8_t *dst) { // Down-Left |
| 311 | const int A = dst[0 - BPS]; |
| 312 | const int B = dst[1 - BPS]; |
| 313 | const int C = dst[2 - BPS]; |
| 314 | const int D = dst[3 - BPS]; |
| 315 | const int E = dst[4 - BPS]; |
| 316 | const int F = dst[5 - BPS]; |
| 317 | const int G = dst[6 - BPS]; |
| 318 | const int H = dst[7 - BPS]; |
| 319 | OUT(0, 0) = AVG3(A, B, C); |
| 320 | OUT(1, 0) = OUT(0, 1) = AVG3(B, C, D); |
| 321 | OUT(2, 0) = OUT(1, 1) = OUT(0, 2) = AVG3(C, D, E); |
| 322 | OUT(3, 0) = OUT(2, 1) = OUT(1, 2) = OUT(0, 3) = AVG3(D, E, F); |
| 323 | OUT(3, 1) = OUT(2, 2) = OUT(1, 3) = AVG3(E, F, G); |
| 324 | OUT(3, 2) = OUT(2, 3) = AVG3(F, G, H); |
| 325 | OUT(3, 3) = AVG3(G, H, H); |
| 326 | } |
| 327 | |
| 328 | static void VR4(uint8_t *dst) { // Vertical-Right |
| 329 | const int I = dst[-1 + 0 * BPS]; |
| 330 | const int J = dst[-1 + 1 * BPS]; |
| 331 | const int K = dst[-1 + 2 * BPS]; |
| 332 | const int X = dst[-1 - BPS]; |
| 333 | const int A = dst[0 - BPS]; |
| 334 | const int B = dst[1 - BPS]; |
| 335 | const int C = dst[2 - BPS]; |
| 336 | const int D = dst[3 - BPS]; |
| 337 | OUT(0, 0) = OUT(1, 2) = AVG2(X, A); |
| 338 | OUT(1, 0) = OUT(2, 2) = AVG2(A, B); |
| 339 | OUT(2, 0) = OUT(3, 2) = AVG2(B, C); |
| 340 | OUT(3, 0) = AVG2(C, D); |
| 341 | |
| 342 | OUT(0, 3) = AVG3(K, J, I); |
| 343 | OUT(0, 2) = AVG3(J, I, X); |
| 344 | OUT(0, 1) = OUT(1, 3) = AVG3(I, X, A); |
| 345 | OUT(1, 1) = OUT(2, 3) = AVG3(X, A, B); |
| 346 | OUT(2, 1) = OUT(3, 3) = AVG3(A, B, C); |
| 347 | OUT(3, 1) = AVG3(B, C, D); |
| 348 | } |
| 349 | |
| 350 | static void VL4(uint8_t *dst) { // Vertical-Left |
| 351 | const int A = dst[0 - BPS]; |
| 352 | const int B = dst[1 - BPS]; |
| 353 | const int C = dst[2 - BPS]; |
| 354 | const int D = dst[3 - BPS]; |
| 355 | const int E = dst[4 - BPS]; |
| 356 | const int F = dst[5 - BPS]; |
| 357 | const int G = dst[6 - BPS]; |
| 358 | const int H = dst[7 - BPS]; |
| 359 | OUT(0, 0) = AVG2(A, B); |
| 360 | OUT(1, 0) = OUT(0, 2) = AVG2(B, C); |
| 361 | OUT(2, 0) = OUT(1, 2) = AVG2(C, D); |
| 362 | OUT(3, 0) = OUT(2, 2) = AVG2(D, E); |
| 363 | |
| 364 | OUT(0, 1) = AVG3(A, B, C); |
| 365 | OUT(1, 1) = OUT(0, 3) = AVG3(B, C, D); |
| 366 | OUT(2, 1) = OUT(1, 3) = AVG3(C, D, E); |
| 367 | OUT(3, 1) = OUT(2, 3) = AVG3(D, E, F); |
| 368 | OUT(3, 2) = AVG3(E, F, G); |
| 369 | OUT(3, 3) = AVG3(F, G, H); |
| 370 | } |
| 371 | |
| 372 | static void HU4(uint8_t *dst) { // Horizontal-Up |
| 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 | OUT(0, 0) = AVG2(I, J); |
| 378 | OUT(2, 0) = OUT(0, 1) = AVG2(J, K); |
| 379 | OUT(2, 1) = OUT(0, 2) = AVG2(K, L); |
| 380 | OUT(1, 0) = AVG3(I, J, K); |
| 381 | OUT(3, 0) = OUT(1, 1) = AVG3(J, K, L); |
| 382 | OUT(3, 1) = OUT(1, 2) = AVG3(K, L, L); |
| 383 | OUT(3, 2) = OUT(2, 2) = |
| 384 | OUT(0, 3) = OUT(1, 3) = OUT(2, 3) = OUT(3, 3) = L; |
| 385 | } |
| 386 | |
| 387 | static void HD4(uint8_t *dst) { // Horizontal-Down |
| 388 | const int I = dst[-1 + 0 * BPS]; |
| 389 | const int J = dst[-1 + 1 * BPS]; |
| 390 | const int K = dst[-1 + 2 * BPS]; |
| 391 | const int L = dst[-1 + 3 * BPS]; |
| 392 | const int X = dst[-1 - BPS]; |
| 393 | const int A = dst[0 - BPS]; |
| 394 | const int B = dst[1 - BPS]; |
| 395 | const int C = dst[2 - BPS]; |
| 396 | |
| 397 | OUT(0, 0) = OUT(2, 1) = AVG2(I, X); |
| 398 | OUT(0, 1) = OUT(2, 2) = AVG2(J, I); |
| 399 | OUT(0, 2) = OUT(2, 3) = AVG2(K, J); |
| 400 | OUT(0, 3) = AVG2(L, K); |
| 401 | |
| 402 | OUT(3, 0) = AVG3(A, B, C); |
| 403 | OUT(2, 0) = AVG3(X, A, B); |
| 404 | OUT(1, 0) = OUT(3, 1) = AVG3(I, X, A); |
| 405 | OUT(1, 1) = OUT(3, 2) = AVG3(J, I, X); |
| 406 | OUT(1, 2) = OUT(3, 3) = AVG3(K, J, I); |
| 407 | OUT(1, 3) = AVG3(L, K, J); |
| 408 | } |
| 409 | |
| 410 | #undef AVG3 |
| 411 | #undef AVG2 |
| 412 | |
| 413 | //----------------------------------------------------------------------------- |
| 414 | // Chroma |
| 415 | |
| 416 | static void VE8uv(uint8_t *dst) { // vertical |
| 417 | int j; |
| 418 | for (j = 0; j < 8; ++j) { |
| 419 | memcpy(dst + j * BPS, dst - BPS, 8); |
| 420 | } |
| 421 | } |
| 422 | |
| 423 | static void HE8uv(uint8_t *dst) { // horizontal |
| 424 | int j; |
| 425 | for (j = 0; j < 8; ++j) { |
| 426 | memset(dst, dst[-1], 8); |
| 427 | dst += BPS; |
| 428 | } |
| 429 | } |
| 430 | |
| 431 | // helper for chroma-DC predictions |
| 432 | static inline void Put8x8uv(uint64_t v, uint8_t* dst) { |
| 433 | int j; |
| 434 | for (j = 0; j < 8; ++j) { |
| 435 | *(uint64_t*)(dst + j * BPS) = v; |
| 436 | } |
| 437 | } |
| 438 | |
| 439 | static void DC8uv(uint8_t *dst) { // DC |
| 440 | int dc0 = 8; |
| 441 | int i; |
| 442 | for (i = 0; i < 8; ++i) { |
| 443 | dc0 += dst[i - BPS] + dst[-1 + i * BPS]; |
| 444 | } |
| 445 | Put8x8uv((uint64_t)((dc0 >> 4) * 0x0101010101010101ULL), dst); |
| 446 | } |
| 447 | |
| 448 | static void DC8uvNoLeft(uint8_t *dst) { // DC with no left samples |
| 449 | int dc0 = 4; |
| 450 | int i; |
| 451 | for (i = 0; i < 8; ++i) { |
| 452 | dc0 += dst[i - BPS]; |
| 453 | } |
| 454 | Put8x8uv((uint64_t)((dc0 >> 3) * 0x0101010101010101ULL), dst); |
| 455 | } |
| 456 | |
| 457 | static void DC8uvNoTop(uint8_t *dst) { // DC with no top samples |
| 458 | int dc0 = 4; |
| 459 | int i; |
| 460 | for (i = 0; i < 8; ++i) { |
| 461 | dc0 += dst[-1 + i * BPS]; |
| 462 | } |
| 463 | Put8x8uv((uint64_t)((dc0 >> 3) * 0x0101010101010101ULL), dst); |
| 464 | } |
| 465 | |
| 466 | static void DC8uvNoTopLeft(uint8_t *dst) { // DC with nothing |
| 467 | Put8x8uv(0x8080808080808080ULL, dst); |
| 468 | } |
| 469 | |
| 470 | //----------------------------------------------------------------------------- |
| 471 | // default C implementations |
| 472 | |
Vikas Arora | 03d5e34 | 2011-06-02 23:59:44 +0530 | [diff] [blame^] | 473 | VP8PredFunc VP8PredLuma4[NUM_BMODES] = { |
Eric Hassold | 9aea642 | 2011-01-04 17:22:46 -0800 | [diff] [blame] | 474 | DC4, TM4, VE4, HE4, RD4, VR4, LD4, VL4, HD4, HU4 |
| 475 | }; |
| 476 | |
Vikas Arora | 03d5e34 | 2011-06-02 23:59:44 +0530 | [diff] [blame^] | 477 | VP8PredFunc VP8PredLuma16[NUM_B_DC_MODES] = { |
Eric Hassold | 9aea642 | 2011-01-04 17:22:46 -0800 | [diff] [blame] | 478 | DC16, TM16, VE16, HE16, |
| 479 | DC16NoTop, DC16NoLeft, DC16NoTopLeft |
| 480 | }; |
| 481 | |
Vikas Arora | 03d5e34 | 2011-06-02 23:59:44 +0530 | [diff] [blame^] | 482 | VP8PredFunc VP8PredChroma8[NUM_B_DC_MODES] = { |
Eric Hassold | 9aea642 | 2011-01-04 17:22:46 -0800 | [diff] [blame] | 483 | DC8uv, TM8uv, VE8uv, HE8uv, |
| 484 | DC8uvNoTop, DC8uvNoLeft, DC8uvNoTopLeft |
| 485 | }; |
| 486 | |
| 487 | //----------------------------------------------------------------------------- |
| 488 | // Edge filtering functions |
| 489 | |
| 490 | // 4 pixels in, 2 pixels out |
| 491 | static inline void do_filter2(uint8_t* p, int step) { |
| 492 | const int p1 = p[-2*step], p0 = p[-step], q0 = p[0], q1 = p[step]; |
| 493 | const int a = 3 * (q0 - p0) + sclip1[1020 + p1 - q1]; |
| 494 | const int a1 = sclip2[112 + ((a + 4) >> 3)]; |
| 495 | const int a2 = sclip2[112 + ((a + 3) >> 3)]; |
| 496 | p[-step] = clip1[255 + p0 + a2]; |
| 497 | p[ 0] = clip1[255 + q0 - a1]; |
| 498 | } |
| 499 | |
| 500 | // 4 pixels in, 4 pixels out |
| 501 | static inline void do_filter4(uint8_t* p, int step) { |
| 502 | const int p1 = p[-2*step], p0 = p[-step], q0 = p[0], q1 = p[step]; |
| 503 | const int a = 3 * (q0 - p0); |
| 504 | const int a1 = sclip2[112 + ((a + 4) >> 3)]; |
| 505 | const int a2 = sclip2[112 + ((a + 3) >> 3)]; |
| 506 | const int a3 = (a1 + 1) >> 1; |
| 507 | p[-2*step] = clip1[255 + p1 + a3]; |
| 508 | p[- step] = clip1[255 + p0 + a2]; |
| 509 | p[ 0] = clip1[255 + q0 - a1]; |
| 510 | p[ step] = clip1[255 + q1 - a3]; |
| 511 | } |
| 512 | |
| 513 | // 6 pixels in, 6 pixels out |
| 514 | static inline void do_filter6(uint8_t* p, int step) { |
| 515 | const int p2 = p[-3*step], p1 = p[-2*step], p0 = p[-step]; |
| 516 | const int q0 = p[0], q1 = p[step], q2 = p[2*step]; |
| 517 | const int a = sclip1[1020 + 3 * (q0 - p0) + sclip1[1020 + p1 - q1]]; |
| 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] = clip1[255 + p2 + a3]; |
| 522 | p[-2*step] = clip1[255 + p1 + a2]; |
| 523 | p[- step] = clip1[255 + p0 + a1]; |
| 524 | p[ 0] = clip1[255 + q0 - a1]; |
| 525 | p[ step] = clip1[255 + q1 - a2]; |
| 526 | p[ 2*step] = clip1[255 + q2 - a3]; |
| 527 | } |
| 528 | |
| 529 | static 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 (abs0[255 + p1 - p0] > thresh) || (abs0[255 + q1 - q0] > thresh); |
| 532 | } |
| 533 | |
| 534 | static inline int needs_filter(const uint8_t* p, int step, int thresh) { |
| 535 | const int p1 = p[-2*step], p0 = p[-step], q0 = p[0], q1 = p[step]; |
| 536 | return (2 * abs0[255 + p0 - q0] + abs1[255 + p1 - q1]) <= thresh; |
| 537 | } |
| 538 | |
| 539 | static inline int needs_filter2(const uint8_t* p, int step, int t, int it) { |
| 540 | const int p3 = p[-4*step], p2 = p[-3*step], p1 = p[-2*step], p0 = p[-step]; |
| 541 | const int q0 = p[0], q1 = p[step], q2 = p[2*step], q3 = p[3*step]; |
| 542 | if ((2 * abs0[255 + p0 - q0] + abs1[255 + p1 - q1]) > t) |
| 543 | return 0; |
| 544 | return abs0[255 + p3 - p2] <= it && abs0[255 + p2 - p1] <= it && |
| 545 | abs0[255 + p1 - p0] <= it && abs0[255 + q3 - q2] <= it && |
| 546 | abs0[255 + q2 - q1] <= it && abs0[255 + q1 - q0] <= it; |
| 547 | } |
| 548 | |
| 549 | //----------------------------------------------------------------------------- |
| 550 | // Simple In-loop filtering (Paragraph 15.2) |
| 551 | |
| 552 | static void SimpleVFilter16(uint8_t* p, int stride, int thresh) { |
| 553 | int i; |
| 554 | for (i = 0; i < 16; ++i) { |
| 555 | if (needs_filter(p + i, stride, thresh)) { |
| 556 | do_filter2(p + i, stride); |
| 557 | } |
| 558 | } |
| 559 | } |
| 560 | |
| 561 | static void SimpleHFilter16(uint8_t* p, int stride, int thresh) { |
| 562 | int i; |
| 563 | for (i = 0; i < 16; ++i) { |
| 564 | if (needs_filter(p + i * stride, 1, thresh)) { |
| 565 | do_filter2(p + i * stride, 1); |
| 566 | } |
| 567 | } |
| 568 | } |
| 569 | |
| 570 | static void SimpleVFilter16i(uint8_t* p, int stride, int thresh) { |
| 571 | int k; |
| 572 | for (k = 3; k > 0; --k) { |
| 573 | p += 4 * stride; |
| 574 | SimpleVFilter16(p, stride, thresh); |
| 575 | } |
| 576 | } |
| 577 | |
| 578 | static void SimpleHFilter16i(uint8_t* p, int stride, int thresh) { |
| 579 | int k; |
| 580 | for (k = 3; k > 0; --k) { |
| 581 | p += 4; |
| 582 | SimpleHFilter16(p, stride, thresh); |
| 583 | } |
| 584 | } |
| 585 | |
| 586 | //----------------------------------------------------------------------------- |
| 587 | // Complex In-loop filtering (Paragraph 15.3) |
| 588 | |
| 589 | static inline void FilterLoop26(uint8_t* p, int hstride, int vstride, int size, |
| 590 | int thresh, int ithresh, int hev_thresh) { |
| 591 | while (size-- > 0) { |
| 592 | if (needs_filter2(p, hstride, thresh, ithresh)) { |
| 593 | if (hev(p, hstride, hev_thresh)) { |
| 594 | do_filter2(p, hstride); |
| 595 | } else { |
| 596 | do_filter6(p, hstride); |
| 597 | } |
| 598 | } |
| 599 | p += vstride; |
| 600 | } |
| 601 | } |
| 602 | |
| 603 | static inline void FilterLoop24(uint8_t* p, int hstride, int vstride, int size, |
| 604 | int thresh, int ithresh, int hev_thresh) { |
| 605 | while (size-- > 0) { |
| 606 | if (needs_filter2(p, hstride, thresh, ithresh)) { |
| 607 | if (hev(p, hstride, hev_thresh)) { |
| 608 | do_filter2(p, hstride); |
| 609 | } else { |
| 610 | do_filter4(p, hstride); |
| 611 | } |
| 612 | } |
| 613 | p += vstride; |
| 614 | } |
| 615 | } |
| 616 | |
| 617 | // on macroblock edges |
| 618 | static void VFilter16(uint8_t* p, int stride, |
| 619 | int thresh, int ithresh, int hev_thresh) { |
| 620 | FilterLoop26(p, stride, 1, 16, thresh, ithresh, hev_thresh); |
| 621 | } |
| 622 | |
| 623 | static void HFilter16(uint8_t* p, int stride, |
| 624 | int thresh, int ithresh, int hev_thresh) { |
| 625 | FilterLoop26(p, 1, stride, 16, thresh, ithresh, hev_thresh); |
| 626 | } |
| 627 | |
| 628 | // on three inner edges |
| 629 | static void VFilter16i(uint8_t* p, int stride, |
| 630 | int thresh, int ithresh, int hev_thresh) { |
| 631 | int k; |
| 632 | for (k = 3; k > 0; --k) { |
| 633 | p += 4 * stride; |
| 634 | FilterLoop24(p, stride, 1, 16, thresh, ithresh, hev_thresh); |
| 635 | } |
| 636 | } |
| 637 | |
| 638 | static void HFilter16i(uint8_t* p, int stride, |
| 639 | int thresh, int ithresh, int hev_thresh) { |
| 640 | int k; |
| 641 | for (k = 3; k > 0; --k) { |
| 642 | p += 4; |
| 643 | FilterLoop24(p, 1, stride, 16, thresh, ithresh, hev_thresh); |
| 644 | } |
| 645 | } |
| 646 | |
| 647 | // 8-pixels wide variant, for chroma filtering |
| 648 | static void VFilter8(uint8_t* u, uint8_t* v, int stride, |
| 649 | int thresh, int ithresh, int hev_thresh) { |
| 650 | FilterLoop26(u, stride, 1, 8, thresh, ithresh, hev_thresh); |
| 651 | FilterLoop26(v, stride, 1, 8, thresh, ithresh, hev_thresh); |
| 652 | } |
| 653 | |
| 654 | static void HFilter8(uint8_t* u, uint8_t* v, int stride, |
| 655 | int thresh, int ithresh, int hev_thresh) { |
| 656 | FilterLoop26(u, 1, stride, 8, thresh, ithresh, hev_thresh); |
| 657 | FilterLoop26(v, 1, stride, 8, thresh, ithresh, hev_thresh); |
| 658 | } |
| 659 | |
| 660 | static void VFilter8i(uint8_t* u, uint8_t* v, int stride, |
| 661 | int thresh, int ithresh, int hev_thresh) { |
| 662 | FilterLoop24(u + 4 * stride, stride, 1, 8, thresh, ithresh, hev_thresh); |
| 663 | FilterLoop24(v + 4 * stride, stride, 1, 8, thresh, ithresh, hev_thresh); |
| 664 | } |
| 665 | |
| 666 | static void HFilter8i(uint8_t* u, uint8_t* v, int stride, |
| 667 | int thresh, int ithresh, int hev_thresh) { |
| 668 | FilterLoop24(u + 4, 1, stride, 8, thresh, ithresh, hev_thresh); |
| 669 | FilterLoop24(v + 4, 1, stride, 8, thresh, ithresh, hev_thresh); |
| 670 | } |
| 671 | |
| 672 | //----------------------------------------------------------------------------- |
| 673 | |
| 674 | void (*VP8VFilter16)(uint8_t*, int, int, int, int) = VFilter16; |
| 675 | void (*VP8HFilter16)(uint8_t*, int, int, int, int) = HFilter16; |
| 676 | void (*VP8VFilter8)(uint8_t*, uint8_t*, int, int, int, int) = VFilter8; |
| 677 | void (*VP8HFilter8)(uint8_t*, uint8_t*, int, int, int, int) = HFilter8; |
| 678 | void (*VP8VFilter16i)(uint8_t*, int, int, int, int) = VFilter16i; |
| 679 | void (*VP8HFilter16i)(uint8_t*, int, int, int, int) = HFilter16i; |
| 680 | void (*VP8VFilter8i)(uint8_t*, uint8_t*, int, int, int, int) = VFilter8i; |
| 681 | void (*VP8HFilter8i)(uint8_t*, uint8_t*, int, int, int, int) = HFilter8i; |
| 682 | |
| 683 | void (*VP8SimpleVFilter16)(uint8_t*, int, int) = SimpleVFilter16; |
| 684 | void (*VP8SimpleHFilter16)(uint8_t*, int, int) = SimpleHFilter16; |
| 685 | void (*VP8SimpleVFilter16i)(uint8_t*, int, int) = SimpleVFilter16i; |
| 686 | void (*VP8SimpleHFilter16i)(uint8_t*, int, int) = SimpleHFilter16i; |
| 687 | |
| 688 | //----------------------------------------------------------------------------- |
| 689 | |
Vikas Arora | 03d5e34 | 2011-06-02 23:59:44 +0530 | [diff] [blame^] | 690 | void VP8DspInit(void) { |
Eric Hassold | 9aea642 | 2011-01-04 17:22:46 -0800 | [diff] [blame] | 691 | // later we'll plug some SSE2 variant here |
| 692 | } |
| 693 | |
| 694 | #if defined(__cplusplus) || defined(c_plusplus) |
| 695 | } // extern "C" |
| 696 | #endif |