ref: 516db21c2c903a7d9b0b5fc156277e9bb538ade9
parent: e42b280e1170d485af47000f411ff45d56af33bd
author: Deb Mukherjee <debargha@google.com>
date: Wed Jan 9 01:26:54 EST 2013
Further enhancements/fixes on dct/dwt hybrid txfm Fixes some scaling issues. Adds an option to only compute the dct on the low-low subband for 32x32 and 64x64 blocks using only a single 16x16 dct after 1 and 2 wavelet decomposition levels respectively. Also adds an option to use a 8x8 dct as building block. Currenlty with the 2/6 filter and with a single 16x16 dct on the low low band, the reuslts compared to full 32x32 dct is as follows: derf: -0.15% yt: -0.29% std-hd: -0.18% hd: -0.6% These are my current recommended settings, since the 2/6 filter is very simple. Results with 8x8 dct are about 0.3% worse. Change-Id: I00100cdc96e32deced591985785ef0d06f325e44
--- a/vp9/common/vp9_blockd.h
+++ b/vp9/common/vp9_blockd.h
@@ -47,6 +47,18 @@
#define MAX_MV_REFS 9
#define MAX_MV_REF_CANDIDATES 4
+#if CONFIG_DWTDCTHYBRID
+#define DWT_MAX_LENGTH 64
+#define DWT_TYPE 26 // 26/53/97
+#define DWT_PRECISION_BITS 2
+#define DWT_PRECISION_RND ((1 << DWT_PRECISION_BITS) / 2)
+
+#define DWTDCT16X16 0
+#define DWTDCT16X16_LEAN 1
+#define DWTDCT8X8 2
+#define DWTDCT_TYPE DWTDCT16X16_LEAN
+#endif
+
typedef struct {int r, c;
} POS;
--- a/vp9/common/vp9_entropy.c
+++ b/vp9/common/vp9_entropy.c
@@ -70,7 +70,8 @@
12, 13, 14, 15
};
-DECLARE_ALIGNED(64, const int, vp9_coef_bands_8x8[64]) = { 0, 1, 2, 3, 5, 4, 4, 5,+DECLARE_ALIGNED(64, const int, vp9_coef_bands_8x8[64]) = {+ 0, 1, 2, 3, 5, 4, 4, 5,
5, 3, 6, 3, 5, 4, 6, 6,
6, 5, 5, 6, 6, 6, 6, 6,
6, 6, 6, 6, 6, 6, 6, 6,
@@ -143,7 +144,215 @@
};
#if CONFIG_DWTDCTHYBRID
+
+#if DWTDCT_TYPE == DWTDCT16X16_LEAN
DECLARE_ALIGNED(16, const int, vp9_coef_bands_32x32[1024]) = {+ 0, 1, 2, 3, 5, 4, 4, 5, 5, 3, 6, 3, 5, 4, 6, 6,
+ 6, 5, 5, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6,
+ 6, 6, 6, 6, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+
+ 6, 6, 6, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+};
+
+DECLARE_ALIGNED(16, const int, vp9_default_zig_zag1d_32x32[1024]) = {+ 0, 1, 32, 64, 33, 2, 3, 34,
+ 65, 96, 128, 97, 66, 35, 4, 5,
+ 36, 67, 98, 129, 160, 192, 161, 130,
+ 99, 68, 37, 6, 7, 38, 69, 100,
+ 131, 162, 193, 224, 256, 225, 194, 163,
+ 132, 101, 70, 39, 8, 9, 40, 71,
+ 102, 133, 164, 195, 226, 257, 288, 320,
+ 289, 258, 227, 196, 165, 134, 103, 72,
+ 41, 10, 11, 42, 73, 104, 135, 166,
+ 197, 228, 259, 290, 321, 352, 384, 353,
+ 322, 291, 260, 229, 198, 167, 136, 105,
+ 74, 43, 12, 13, 44, 75, 106, 137,
+ 168, 199, 230, 261, 292, 323, 354, 385,
+ 416, 448, 417, 386, 355, 324, 293, 262,
+ 231, 200, 169, 138, 107, 76, 45, 14,
+ 15, 46, 77, 108, 139, 170, 201, 232,
+ 263, 294, 325, 356, 387, 418, 449, 480,
+ 481, 450, 419, 388, 357, 326, 295, 264,
+ 233, 202, 171, 140, 109, 78, 47, 79,
+ 110, 141, 172, 203, 234, 265, 296, 327,
+ 358, 389, 420, 451, 482, 483, 452, 421,
+ 390, 359, 328, 297, 266, 235, 204, 173,
+ 142, 111, 143, 174, 205, 236, 267, 298,
+ 329, 360, 391, 422, 453, 484, 485, 454,
+ 423, 392, 361, 330, 299, 268, 237, 206,
+ 175, 207, 238, 269, 300, 331, 362, 393,
+ 424, 455, 486, 487, 456, 425, 394, 363,
+ 332, 301, 270, 239, 271, 302, 333, 364,
+ 395, 426, 457, 488, 489, 458, 427, 396,
+ 365, 334, 303, 335, 366, 397, 428, 459,
+ 490, 491, 460, 429, 398, 367, 399, 430,
+ 461, 492, 493, 462, 431, 463, 494, 495,
+
+ 16, 512, 528, 17, 513, 529, 48, 544,
+ 560, 80, 576, 592, 49, 545, 561, 18,
+ 514, 530, 19, 515, 531, 50, 546, 562,
+ 81, 577, 593, 112, 608, 624, 144, 640,
+ 656, 113, 609, 625, 82, 578, 594, 51,
+ 547, 563, 20, 516, 532, 21, 517, 533,
+ 52, 548, 564, 83, 579, 595, 114, 610,
+ 626, 145, 641, 657, 176, 672, 688, 208,
+ 704, 720, 177, 673, 689, 146, 642, 658,
+ 115, 611, 627, 84, 580, 596, 53, 549,
+ 565, 22, 518, 534, 23, 519, 535, 54,
+ 550, 566, 85, 581, 597, 116, 612, 628,
+ 147, 643, 659, 178, 674, 690, 209, 705,
+ 721, 240, 736, 752, 272, 768, 784, 241,
+ 737, 753, 210, 706, 722, 179, 675, 691,
+ 148, 644, 660, 117, 613, 629, 86, 582,
+ 598, 55, 551, 567, 24, 520, 536, 25,
+ 521, 537, 56, 552, 568, 87, 583, 599,
+ 118, 614, 630, 149, 645, 661, 180, 676,
+ 692, 211, 707, 723, 242, 738, 754, 273,
+ 769, 785, 304, 800, 816, 336, 832, 848,
+ 305, 801, 817, 274, 770, 786, 243, 739,
+ 755, 212, 708, 724, 181, 677, 693, 150,
+ 646, 662, 119, 615, 631, 88, 584, 600,
+ 57, 553, 569, 26, 522, 538, 27, 523,
+ 539, 58, 554, 570, 89, 585, 601, 120,
+ 616, 632, 151, 647, 663, 182, 678, 694,
+ 213, 709, 725, 244, 740, 756, 275, 771,
+ 787, 306, 802, 818, 337, 833, 849, 368,
+ 864, 880, 400, 896, 912, 369, 865, 881,
+ 338, 834, 850, 307, 803, 819, 276, 772,
+ 788, 245, 741, 757, 214, 710, 726, 183,
+
+ 679, 695, 152, 648, 664, 121, 617, 633,
+ 90, 586, 602, 59, 555, 571, 28, 524,
+ 540, 29, 525, 541, 60, 556, 572, 91,
+ 587, 603, 122, 618, 634, 153, 649, 665,
+ 184, 680, 696, 215, 711, 727, 246, 742,
+ 758, 277, 773, 789, 308, 804, 820, 339,
+ 835, 851, 370, 866, 882, 401, 897, 913,
+ 432, 928, 944, 464, 960, 976, 433, 929,
+ 945, 402, 898, 914, 371, 867, 883, 340,
+ 836, 852, 309, 805, 821, 278, 774, 790,
+ 247, 743, 759, 216, 712, 728, 185, 681,
+ 697, 154, 650, 666, 123, 619, 635, 92,
+ 588, 604, 61, 557, 573, 30, 526, 542,
+ 31, 527, 543, 62, 558, 574, 93, 589,
+ 605, 124, 620, 636, 155, 651, 667, 186,
+ 682, 698, 217, 713, 729, 248, 744, 760,
+ 279, 775, 791, 310, 806, 822, 341, 837,
+ 853, 372, 868, 884, 403, 899, 915, 434,
+ 930, 946, 465, 961, 977, 496, 992, 1008,
+ 497, 993, 1009, 466, 962, 978, 435, 931,
+ 947, 404, 900, 916, 373, 869, 885, 342,
+ 838, 854, 311, 807, 823, 280, 776, 792,
+ 249, 745, 761, 218, 714, 730, 187, 683,
+ 699, 156, 652, 668, 125, 621, 637, 94,
+ 590, 606, 63, 559, 575, 95, 591, 607,
+ 126, 622, 638, 157, 653, 669, 188, 684,
+ 700, 219, 715, 731, 250, 746, 762, 281,
+ 777, 793, 312, 808, 824, 343, 839, 855,
+ 374, 870, 886, 405, 901, 917, 436, 932,
+ 948, 467, 963, 979, 498, 994, 1010, 499,
+ 995, 1011, 468, 964, 980, 437, 933, 949,
+ 406, 902, 918, 375, 871, 887, 344, 840,
+
+ 856, 313, 809, 825, 282, 778, 794, 251,
+ 747, 763, 220, 716, 732, 189, 685, 701,
+ 158, 654, 670, 127, 623, 639, 159, 655,
+ 671, 190, 686, 702, 221, 717, 733, 252,
+ 748, 764, 283, 779, 795, 314, 810, 826,
+ 345, 841, 857, 376, 872, 888, 407, 903,
+ 919, 438, 934, 950, 469, 965, 981, 500,
+ 996, 1012, 501, 997, 1013, 470, 966, 982,
+ 439, 935, 951, 408, 904, 920, 377, 873,
+ 889, 346, 842, 858, 315, 811, 827, 284,
+ 780, 796, 253, 749, 765, 222, 718, 734,
+ 191, 687, 703, 223, 719, 735, 254, 750,
+ 766, 285, 781, 797, 316, 812, 828, 347,
+ 843, 859, 378, 874, 890, 409, 905, 921,
+ 440, 936, 952, 471, 967, 983, 502, 998,
+ 1014, 503, 999, 1015, 472, 968, 984, 441,
+ 937, 953, 410, 906, 922, 379, 875, 891,
+ 348, 844, 860, 317, 813, 829, 286, 782,
+ 798, 255, 751, 767, 287, 783, 799, 318,
+ 814, 830, 349, 845, 861, 380, 876, 892,
+ 411, 907, 923, 442, 938, 954, 473, 969,
+ 985, 504, 1000, 1016, 505, 1001, 1017, 474,
+ 970, 986, 443, 939, 955, 412, 908, 924,
+ 381, 877, 893, 350, 846, 862, 319, 815,
+ 831, 351, 847, 863, 382, 878, 894, 413,
+ 909, 925, 444, 940, 956, 475, 971, 987,
+ 506, 1002, 1018, 507, 1003, 1019, 476, 972,
+ 988, 445, 941, 957, 414, 910, 926, 383,
+ 879, 895, 415, 911, 927, 446, 942, 958,
+ 477, 973, 989, 508, 1004, 1020, 509, 1005,
+ 1021, 478, 974, 990, 447, 943, 959, 479,
+ 975, 991, 510, 1006, 1022, 511, 1007, 1023,
+};
+
+#elif DWTDCT_TYPE == DWTDCT16X16
+
+DECLARE_ALIGNED(16, const int, vp9_coef_bands_32x32[1024]) = {0, 1, 2, 3, 5, 4, 4, 5, 5, 3, 6, 3, 5, 4, 6,
6, 6, 6,
6,
@@ -351,7 +560,206 @@
975, 991, 510, 1006, 1022, 511, 1007, 1023,
};
-#else // CONFIG_DWTDCTHYBRID
+#elif DWTDCT_TYPE == DWTDCT8X8
+
+DECLARE_ALIGNED(16, const int, vp9_coef_bands_32x32[1024]) = {+ 0, 1, 2, 3, 5, 4, 4, 5,
+ 5, 3, 6, 3, 5, 4, 6, 6,
+ 6, 5, 5, 6, 6, 6, 6, 6,
+ 6, 6, 6, 6, 6, 6, 6, 6,
+ 6, 6, 6, 6, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7,
+
+ 6, 6, 6, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+
+ 6, 6, 6, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+ 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+};
+
+DECLARE_ALIGNED(16, const int, vp9_default_zig_zag1d_32x32[1024]) = {+ 0, 1, 32, 64, 33, 2, 3, 34,
+ 65, 96, 128, 97, 66, 35, 4, 5,
+ 36, 67, 98, 129, 160, 192, 161, 130,
+ 99, 68, 37, 6, 7, 38, 69, 100,
+ 131, 162, 193, 224, 225, 194, 163, 132,
+ 101, 70, 39, 71, 102, 133, 164, 195,
+ 226, 227, 196, 165, 134, 103, 135, 166,
+ 197, 228, 229, 198, 167, 199, 230, 231,
+
+ 8, 256, 264, 9, 257, 265, 40, 288, 296, 72, 320, 328,
+ 41, 289, 297, 10, 258, 266, 11, 259, 267, 42, 290, 298,
+ 73, 321, 329, 104, 352, 360, 136, 384, 392, 105, 353, 361,
+ 74, 322, 330, 43, 291, 299, 12, 260, 268, 13, 261, 269,
+ 44, 292, 300, 75, 323, 331, 106, 354, 362, 137, 385, 393,
+ 168, 416, 424, 200, 448, 456, 169, 417, 425, 138, 386, 394,
+ 107, 355, 363, 76, 324, 332, 45, 293, 301, 14, 262, 270,
+ 15, 263, 271, 46, 294, 302, 77, 325, 333, 108, 356, 364,
+ 139, 387, 395, 170, 418, 426, 201, 449, 457, 232, 480, 488,
+ 233, 481, 489, 202, 450, 458, 171, 419, 427, 140, 388, 396,
+ 109, 357, 365, 78, 326, 334, 47, 295, 303, 79, 327, 335,
+ 110, 358, 366, 141, 389, 397, 172, 420, 428, 203, 451, 459,
+ 234, 482, 490, 235, 483, 491, 204, 452, 460, 173, 421, 429,
+ 142, 390, 398, 111, 359, 367, 143, 391, 399, 174, 422, 430,
+ 205, 453, 461, 236, 484, 492, 237, 485, 493, 206, 454, 462,
+ 175, 423, 431, 207, 455, 463, 238, 486, 494, 239, 487, 495,
+
+ 16, 512, 528, 17, 513, 529, 18, 514,
+ 530, 19, 515, 531, 20, 516, 532, 21,
+ 517, 533, 22, 518, 534, 23, 519, 535,
+ 24, 520, 536, 25, 521, 537, 26, 522,
+ 538, 27, 523, 539, 28, 524, 540, 29,
+ 525, 541, 30, 526, 542, 31, 527, 543,
+ 48, 544, 560, 49, 545, 561, 50, 546,
+ 562, 51, 547, 563, 52, 548, 564, 53,
+ 549, 565, 54, 550, 566, 55, 551, 567,
+ 56, 552, 568, 57, 553, 569, 58, 554,
+ 570, 59, 555, 571, 60, 556, 572, 61,
+ 557, 573, 62, 558, 574, 63, 559, 575,
+ 80, 576, 592, 81, 577, 593, 82, 578,
+ 594, 83, 579, 595, 84, 580, 596, 85,
+ 581, 597, 86, 582, 598, 87, 583, 599,
+ 88, 584, 600, 89, 585, 601, 90, 586,
+ 602, 91, 587, 603, 92, 588, 604, 93,
+ 589, 605, 94, 590, 606, 95, 591, 607,
+ 112, 608, 624, 113, 609, 625, 114, 610,
+ 626, 115, 611, 627, 116, 612, 628, 117,
+ 613, 629, 118, 614, 630, 119, 615, 631,
+ 120, 616, 632, 121, 617, 633, 122, 618,
+ 634, 123, 619, 635, 124, 620, 636, 125,
+ 621, 637, 126, 622, 638, 127, 623, 639,
+ 144, 640, 656, 145, 641, 657, 146, 642,
+ 658, 147, 643, 659, 148, 644, 660, 149,
+ 645, 661, 150, 646, 662, 151, 647, 663,
+ 152, 648, 664, 153, 649, 665, 154, 650,
+ 666, 155, 651, 667, 156, 652, 668, 157,
+ 653, 669, 158, 654, 670, 159, 655, 671,
+ 176, 672, 688, 177, 673, 689, 178, 674,
+ 690, 179, 675, 691, 180, 676, 692, 181,
+ 677, 693, 182, 678, 694, 183, 679, 695,
+ 184, 680, 696, 185, 681, 697, 186, 682,
+ 698, 187, 683, 699, 188, 684, 700, 189,
+ 685, 701, 190, 686, 702, 191, 687, 703,
+ 208, 704, 720, 209, 705, 721, 210, 706,
+ 722, 211, 707, 723, 212, 708, 724, 213,
+ 709, 725, 214, 710, 726, 215, 711, 727,
+ 216, 712, 728, 217, 713, 729, 218, 714,
+ 730, 219, 715, 731, 220, 716, 732, 221,
+ 717, 733, 222, 718, 734, 223, 719, 735,
+ 240, 736, 752, 241, 737, 753, 242, 738,
+ 754, 243, 739, 755, 244, 740, 756, 245,
+ 741, 757, 246, 742, 758, 247, 743, 759,
+ 248, 744, 760, 249, 745, 761, 250, 746,
+ 762, 251, 747, 763, 252, 748, 764, 253,
+ 749, 765, 254, 750, 766, 255, 751, 767,
+ 272, 768, 784, 273, 769, 785, 274, 770,
+ 786, 275, 771, 787, 276, 772, 788, 277,
+ 773, 789, 278, 774, 790, 279, 775, 791,
+ 280, 776, 792, 281, 777, 793, 282, 778,
+ 794, 283, 779, 795, 284, 780, 796, 285,
+ 781, 797, 286, 782, 798, 287, 783, 799,
+ 304, 800, 816, 305, 801, 817, 306, 802,
+ 818, 307, 803, 819, 308, 804, 820, 309,
+ 805, 821, 310, 806, 822, 311, 807, 823,
+ 312, 808, 824, 313, 809, 825, 314, 810,
+ 826, 315, 811, 827, 316, 812, 828, 317,
+ 813, 829, 318, 814, 830, 319, 815, 831,
+ 336, 832, 848, 337, 833, 849, 338, 834,
+ 850, 339, 835, 851, 340, 836, 852, 341,
+ 837, 853, 342, 838, 854, 343, 839, 855,
+ 344, 840, 856, 345, 841, 857, 346, 842,
+ 858, 347, 843, 859, 348, 844, 860, 349,
+ 845, 861, 350, 846, 862, 351, 847, 863,
+ 368, 864, 880, 369, 865, 881, 370, 866,
+ 882, 371, 867, 883, 372, 868, 884, 373,
+ 869, 885, 374, 870, 886, 375, 871, 887,
+ 376, 872, 888, 377, 873, 889, 378, 874,
+ 890, 379, 875, 891, 380, 876, 892, 381,
+ 877, 893, 382, 878, 894, 383, 879, 895,
+ 400, 896, 912, 401, 897, 913, 402, 898,
+ 914, 403, 899, 915, 404, 900, 916, 405,
+ 901, 917, 406, 902, 918, 407, 903, 919,
+ 408, 904, 920, 409, 905, 921, 410, 906,
+ 922, 411, 907, 923, 412, 908, 924, 413,
+ 909, 925, 414, 910, 926, 415, 911, 927,
+ 432, 928, 944, 433, 929, 945, 434, 930,
+ 946, 435, 931, 947, 436, 932, 948, 437,
+ 933, 949, 438, 934, 950, 439, 935, 951,
+ 440, 936, 952, 441, 937, 953, 442, 938,
+ 954, 443, 939, 955, 444, 940, 956, 445,
+ 941, 957, 446, 942, 958, 447, 943, 959,
+ 464, 960, 976, 465, 961, 977, 466, 962,
+ 978, 467, 963, 979, 468, 964, 980, 469,
+ 965, 981, 470, 966, 982, 471, 967, 983,
+ 472, 968, 984, 473, 969, 985, 474, 970,
+ 986, 475, 971, 987, 476, 972, 988, 477,
+ 973, 989, 478, 974, 990, 479, 975, 991,
+ 496, 992, 1008, 497, 993, 1009, 498, 994,
+ 1010, 499, 995, 1011, 500, 996, 1012, 501,
+ 997, 1013, 502, 998, 1014, 503, 999, 1015,
+ 504, 1000, 1016, 505, 1001, 1017, 506, 1002,
+ 1018, 507, 1003, 1019, 508, 1004, 1020, 509,
+ 1005, 1021, 510, 1006, 1022, 511, 1007, 1023,
+};
+#endif
+
+#else
DECLARE_ALIGNED(16, const int, vp9_coef_bands_32x32[1024]) = {0, 1, 2, 3, 5, 4, 4, 5, 5, 3, 6, 3, 5, 4, 6, 6,
--- a/vp9/common/vp9_idctllm.c
+++ b/vp9/common/vp9_idctllm.c
@@ -1536,6 +1536,7 @@
#if !CONFIG_DWTDCTHYBRID
#define DownshiftMultiplyBy2(x) x * 2
#define DownshiftMultiply(x) x
+
static void idct16(double *input, double *output, int stride) {static const double C1 = 0.995184726672197;
static const double C2 = 0.98078528040323;
@@ -1738,6 +1739,7 @@
output[stride*9] = step[6] - step[ 9];
output[stride*8] = step[7] - step[ 8];
}
+
static void butterfly_32_idct_1d(double *input, double *output, int stride) {static const double C1 = 0.998795456205; // cos(pi * 1 / 64)
static const double C3 = 0.989176509965; // cos(pi * 3 / 64)
@@ -1878,13 +1880,8 @@
vp9_clear_system_state(); // Make it simd safe : __asm emms;
}
-#else // CONFIG_DWTDCTHYBRID
+#else // !CONFIG_DWTDCTHYBRID
-#define DWT_MAX_LENGTH 32
-#define DWT_TYPE 26 // 26/53/97
-#define DWT_PRECISION_BITS 2
-#define DWT_PRECISION_RND ((1 << DWT_PRECISION_BITS) / 2)
-
#if DWT_TYPE == 53
// Note: block length must be even for this implementation
@@ -2388,6 +2385,72 @@
vp9_clear_system_state(); // Make it simd safe : __asm emms;
}
+static void idct8_1d(double *x) {+ int i, j;
+ double t[8];
+ static const double idctmat[64] = {+ 0.35355339059327, 0.49039264020162, 0.46193976625564, 0.41573480615127,
+ 0.35355339059327, 0.2777851165098, 0.19134171618254, 0.097545161008064,
+ 0.35355339059327, 0.41573480615127, 0.19134171618254, -0.097545161008064,
+ -0.35355339059327, -0.49039264020161, -0.46193976625564, -0.2777851165098,
+ 0.35355339059327, 0.2777851165098, -0.19134171618254, -0.49039264020162,
+ -0.35355339059327, 0.097545161008064, 0.46193976625564, 0.41573480615127,
+ 0.35355339059327, 0.097545161008063, -0.46193976625564, -0.2777851165098,
+ 0.35355339059327, 0.41573480615127, -0.19134171618254, -0.49039264020162,
+ 0.35355339059327, -0.097545161008063, -0.46193976625564, 0.2777851165098,
+ 0.35355339059327, -0.41573480615127, -0.19134171618255, 0.49039264020162,
+ 0.35355339059327, -0.2777851165098, -0.19134171618254, 0.49039264020161,
+ -0.35355339059327, -0.097545161008064, 0.46193976625564, -0.41573480615127,
+ 0.35355339059327, -0.41573480615127, 0.19134171618254, 0.097545161008065,
+ -0.35355339059327, 0.49039264020162, -0.46193976625564, 0.2777851165098,
+ 0.35355339059327, -0.49039264020162, 0.46193976625564, -0.41573480615127,
+ 0.35355339059327, -0.2777851165098, 0.19134171618255, -0.097545161008064
+ };
+ for (i = 0; i < 8; ++i) {+ t[i] = 0;
+ for (j = 0; j < 8; ++j)
+ t[i] += idctmat[i * 8 + j] * x[j];
+ }
+ for (i = 0; i < 8; ++i) {+ x[i] = t[i];
+ }
+}
+
+static void vp9_short_idct8x8_c_f(int16_t *coefs, int16_t *block, int pitch,
+ int scale) {+ double X[8 * 8], Y[8];
+ int i, j;
+ int shortpitch = pitch >> 1;
+
+ vp9_clear_system_state(); // Make it simd safe : __asm emms;
+ {+ for (i = 0; i < 8; i++) {+ for (j = 0; j < 8; j++) {+ X[i * 8 + j] = (double)coefs[i * shortpitch + j];
+ }
+ }
+ for (i = 0; i < 8; i++)
+ idct8_1d(X + 8 * i);
+ for (i = 0; i < 8; i++) {+ for (j = 0; j < 8; ++j)
+ Y[j] = X[i + 8 * j];
+ idct8_1d(Y);
+ for (j = 0; j < 8; ++j)
+ X[i + 8 * j] = Y[j];
+ }
+ for (i = 0; i < 8; i++) {+ for (j = 0; j < 8; j++) {+ block[i * 8 + j] = (int16_t)round(X[i * 8 + j] / (8 >> scale));
+ }
+ }
+ }
+ vp9_clear_system_state(); // Make it simd safe : __asm emms;
+}
+
+#define multiply_bits(d, n) ((n) < 0 ? (d) >> (n) : (d) << (n))
+
+#if DWTDCT_TYPE == DWTDCT16X16_LEAN
+
void vp9_short_idct32x32_c(int16_t *input, int16_t *output, int pitch) {// assume output is a 32x32 buffer
// Temporary buffer to hold a 16x16 block for 16x16 inverse dct
@@ -2396,7 +2459,7 @@
int16_t buffer2[32 * 32];
// Note: pitch is in bytes, short_pitch is in short units
const int short_pitch = pitch >> 1;
- int i;
+ int i, j;
// TODO(debargha): Implement more efficiently by adding output pitch
// argument to the idct16x16 function
@@ -2405,6 +2468,46 @@
for (i = 0; i < 16; ++i) {vpx_memcpy(buffer2 + i * 32, buffer + i * 16, sizeof(*buffer2) * 16);
}
+ for (i = 0; i < 16; ++i) {+ for (j = 16; j < 32; ++j) {+ buffer2[i * 32 + j] =
+ multiply_bits(input[i * short_pitch + j], DWT_PRECISION_BITS - 2);
+ }
+ }
+ for (i = 16; i < 32; ++i) {+ for (j = 0; j < 32; ++j) {+ buffer2[i * 32 + j] =
+ multiply_bits(input[i * short_pitch + j], DWT_PRECISION_BITS - 2);
+ }
+ }
+#if DWT_TYPE == 26
+ dyadic_synthesize_26(1, 32, 32, buffer2, 32, output, 32);
+#elif DWT_TYPE == 97
+ dyadic_synthesize_97(1, 32, 32, buffer2, 32, output, 32);
+#elif DWT_TYPE == 53
+ dyadic_synthesize_53(1, 32, 32, buffer2, 32, output, 32);
+#endif
+}
+
+#elif DWTDCT_TYPE == DWTDCT16X16
+
+void vp9_short_idct32x32_c(int16_t *input, int16_t *output, int pitch) {+ // assume output is a 32x32 buffer
+ // Temporary buffer to hold a 16x16 block for 16x16 inverse dct
+ int16_t buffer[16 * 16];
+ // Temporary buffer to hold a 32x32 block for inverse 32x32 dwt
+ int16_t buffer2[32 * 32];
+ // Note: pitch is in bytes, short_pitch is in short units
+ const int short_pitch = pitch >> 1;
+ int i, j;
+
+ // TODO(debargha): Implement more efficiently by adding output pitch
+ // argument to the idct16x16 function
+ vp9_short_idct16x16_c_f(input, buffer, pitch,
+ 1 + DWT_PRECISION_BITS);
+ for (i = 0; i < 16; ++i) {+ vpx_memcpy(buffer2 + i * 32, buffer + i * 16, sizeof(*buffer2) * 16);
+ }
vp9_short_idct16x16_c_f(input + 16, buffer, pitch,
1 + DWT_PRECISION_BITS);
for (i = 0; i < 16; ++i) {@@ -2431,6 +2534,66 @@
#endif
}
+#elif DWTDCT_TYPE == DWTDCT8X8
+
+void vp9_short_idct32x32_c(int16_t *input, int16_t *output, int pitch) {+ // assume output is a 32x32 buffer
+ // Temporary buffer to hold a 16x16 block for 16x16 inverse dct
+ int16_t buffer[8 * 8];
+ // Temporary buffer to hold a 32x32 block for inverse 32x32 dwt
+ int16_t buffer2[32 * 32];
+ // Note: pitch is in bytes, short_pitch is in short units
+ const int short_pitch = pitch >> 1;
+ int i, j;
+
+ // TODO(debargha): Implement more efficiently by adding output pitch
+ // argument to the idct16x16 function
+ vp9_short_idct8x8_c_f(input, buffer, pitch,
+ 1 + DWT_PRECISION_BITS);
+ for (i = 0; i < 8; ++i) {+ vpx_memcpy(buffer2 + i * 32, buffer + i * 8, sizeof(*buffer2) * 8);
+ }
+ vp9_short_idct8x8_c_f(input + 8, buffer, pitch,
+ 1 + DWT_PRECISION_BITS);
+ for (i = 0; i < 8; ++i) {+ vpx_memcpy(buffer2 + i * 32 + 8, buffer + i * 8, sizeof(*buffer2) * 8);
+ }
+ vp9_short_idct8x8_c_f(input + 8 * short_pitch, buffer, pitch,
+ 1 + DWT_PRECISION_BITS);
+ for (i = 0; i < 8; ++i) {+ vpx_memcpy(buffer2 + i * 32 + 8 * 32, buffer + i * 8,
+ sizeof(*buffer2) * 8);
+ }
+ vp9_short_idct8x8_c_f(input + 8 * short_pitch + 8, buffer, pitch,
+ 1 + DWT_PRECISION_BITS);
+ for (i = 0; i < 8; ++i) {+ vpx_memcpy(buffer2 + i * 32 + 8 * 33, buffer + i * 8,
+ sizeof(*buffer2) * 8);
+ }
+ for (i = 0; i < 16; ++i) {+ for (j = 16; j < 32; ++j) {+ buffer2[i * 32 + j] =
+ multiply_bits(input[i * short_pitch + j], DWT_PRECISION_BITS - 2);
+ }
+ }
+ for (i = 16; i < 32; ++i) {+ for (j = 0; j < 32; ++j) {+ buffer2[i * 32 + j] =
+ multiply_bits(input[i * short_pitch + j], DWT_PRECISION_BITS - 2);
+ }
+ }
+#if DWT_TYPE == 26
+ dyadic_synthesize_26(2, 32, 32, buffer2, 32, output, 32);
+#elif DWT_TYPE == 97
+ dyadic_synthesize_97(2, 32, 32, buffer2, 32, output, 32);
+#elif DWT_TYPE == 53
+ dyadic_synthesize_53(2, 32, 32, buffer2, 32, output, 32);
+#endif
+}
+
+#endif
+
+#if CONFIG_TX64X64
void vp9_short_idct64x64_c(int16_t *input, int16_t *output, int pitch) {// assume output is a 64x64 buffer
// Temporary buffer to hold a 16x16 block for 16x16 inverse dct
@@ -2448,6 +2611,20 @@
for (i = 0; i < 16; ++i) {vpx_memcpy(buffer2 + i * 64, buffer + i * 16, sizeof(*buffer2) * 16);
}
+#if DWTDCT_TYPE == DWTDCT16X16_LEAN
+ for (i = 0; i < 16; ++i) {+ for (j = 16; j < 64; ++j) {+ buffer2[i * 64 + j] =
+ multiply_bits(input[i * short_pitch + j], DWT_PRECISION_BITS - 1);
+ }
+ }
+ for (i = 16; i < 64; ++i) {+ for (j = 0; j < 64; ++j) {+ buffer2[i * 64 + j] =
+ multiply_bits(input[i * short_pitch + j], DWT_PRECISION_BITS - 1);
+ }
+ }
+#elif DWTDCT_TYPE == DWTDCT16X16
vp9_short_idct16x16_c_f(input + 16, buffer, pitch,
2 + DWT_PRECISION_BITS);
for (i = 0; i < 16; ++i) {@@ -2467,33 +2644,19 @@
}
// Copying and scaling highest bands into buffer2
-#if DWT_PRECISION_BITS < 1
for (i = 0; i < 32; ++i) {- for (j = 0; j < 32; ++j) {- buffer2[i * 64 + 32 + j] =
- input[i * short_pitch + 32 + j] >> (1 - DWT_PRECISION_BITS);
- }
- }
- for (i = 0; i < 32; ++i) {- for (j = 0; j < 64; ++j) {+ for (j = 32; j < 64; ++j) {buffer2[i * 64 + j] =
- input[(i + 32) * short_pitch + j] >> (1 - DWT_PRECISION_BITS);
+ multiply_bits(input[i * short_pitch + j], DWT_PRECISION_BITS - 1);
}
}
-#else
- for (i = 0; i < 32; ++i) {- for (j = 0; j < 32; ++j) {- buffer2[i * 64 + 32 + j] =
- input[i * short_pitch + 32 + j] << (DWT_PRECISION_BITS - 1);
- }
- }
- for (i = 0; i < 32; ++i) {+ for (i = 32; i < 64; ++i) { for (j = 0; j < 64; ++j) {buffer2[i * 64 + j] =
- input[(i + 32) * short_pitch + j] << (DWT_PRECISION_BITS - 1);
+ multiply_bits(input[i * short_pitch + j], DWT_PRECISION_BITS - 1);
}
}
-#endif
+#endif // DWTDCT_TYPE
#if DWT_TYPE == 26
dyadic_synthesize_26(2, 64, 64, buffer2, 64, output, 64);
@@ -2503,4 +2666,5 @@
dyadic_synthesize_53(2, 64, 64, buffer2, 64, output, 64);
#endif
}
-#endif // CONFIG_DWTDCTHYBRID
+#endif // CONFIG_TX64X64
+#endif // !CONFIG_DWTDCTHYBRID
--- a/vp9/encoder/vp9_dct.c
+++ b/vp9/encoder/vp9_dct.c
@@ -1686,11 +1686,6 @@
#else // CONFIG_DWTDCTHYBRID
-#define DWT_MAX_LENGTH 64
-#define DWT_TYPE 26 // 26/53/97
-#define DWT_PRECISION_BITS 2
-#define DWT_PRECISION_RND ((1 << DWT_PRECISION_BITS) / 2)
-
#if DWT_TYPE == 53
// Note: block length must be even for this implementation
@@ -2139,10 +2134,97 @@
vp9_clear_system_state(); // Make it simd safe : __asm emms;
}
+void vp9_short_fdct8x8_c_f(short *block, short *coefs, int pitch, int scale) {+ int j1, i, j, k;
+ static int count = 0;
+ short x[8 * 8];
+ float b[8];
+ float b1[8];
+ float d[8][8];
+ float f0 = (float) .7071068;
+ float f1 = (float) .4903926;
+ float f2 = (float) .4619398;
+ float f3 = (float) .4157348;
+ float f4 = (float) .3535534;
+ float f5 = (float) .2777851;
+ float f6 = (float) .1913417;
+ float f7 = (float) .0975452;
+ pitch = pitch / 2;
+ for (i = 0, k = 0; i < 8; i++, k += pitch) {+ for (j = 0; j < 8; j++) {+ b[j] = (float)(block[k + j] << (3 - scale));
+ }
+ /* Horizontal transform */
+ for (j = 0; j < 4; j++) {+ j1 = 7 - j;
+ b1[j] = b[j] + b[j1];
+ b1[j1] = b[j] - b[j1];
+ }
+ b[0] = b1[0] + b1[3];
+ b[1] = b1[1] + b1[2];
+ b[2] = b1[1] - b1[2];
+ b[3] = b1[0] - b1[3];
+ b[4] = b1[4];
+ b[5] = (b1[6] - b1[5]) * f0;
+ b[6] = (b1[6] + b1[5]) * f0;
+ b[7] = b1[7];
+ d[i][0] = (b[0] + b[1]) * f4;
+ d[i][4] = (b[0] - b[1]) * f4;
+ d[i][2] = b[2] * f6 + b[3] * f2;
+ d[i][6] = b[3] * f6 - b[2] * f2;
+ b1[4] = b[4] + b[5];
+ b1[7] = b[7] + b[6];
+ b1[5] = b[4] - b[5];
+ b1[6] = b[7] - b[6];
+ d[i][1] = b1[4] * f7 + b1[7] * f1;
+ d[i][5] = b1[5] * f3 + b1[6] * f5;
+ d[i][7] = b1[7] * f7 - b1[4] * f1;
+ d[i][3] = b1[6] * f3 - b1[5] * f5;
+ }
+ /* Vertical transform */
+ for (i = 0; i < 8; i++) {+ for (j = 0; j < 4; j++) {+ j1 = 7 - j;
+ b1[j] = d[j][i] + d[j1][i];
+ b1[j1] = d[j][i] - d[j1][i];
+ }
+ b[0] = b1[0] + b1[3];
+ b[1] = b1[1] + b1[2];
+ b[2] = b1[1] - b1[2];
+ b[3] = b1[0] - b1[3];
+ b[4] = b1[4];
+ b[5] = (b1[6] - b1[5]) * f0;
+ b[6] = (b1[6] + b1[5]) * f0;
+ b[7] = b1[7];
+ d[0][i] = (b[0] + b[1]) * f4;
+ d[4][i] = (b[0] - b[1]) * f4;
+ d[2][i] = b[2] * f6 + b[3] * f2;
+ d[6][i] = b[3] * f6 - b[2] * f2;
+ b1[4] = b[4] + b[5];
+ b1[7] = b[7] + b[6];
+ b1[5] = b[4] - b[5];
+ b1[6] = b[7] - b[6];
+ d[1][i] = b1[4] * f7 + b1[7] * f1;
+ d[5][i] = b1[5] * f3 + b1[6] * f5;
+ d[7][i] = b1[7] * f7 - b1[4] * f1;
+ d[3][i] = b1[6] * f3 - b1[5] * f5;
+ }
+ for (i = 0; i < 8; i++) {+ for (j = 0; j < 8; j++) {+ *(coefs + j + i * 8) = (short) floor(d[i][j] + 0.5);
+ }
+ }
+ return;
+}
+
+#define divide_bits(d, n) ((n) < 0 ? (d) << (n) : (d) >> (n))
+
+#if DWTDCT_TYPE == DWTDCT16X16_LEAN
+
void vp9_short_fdct32x32_c(short *input, short *out, int pitch) {// assume out is a 32x32 buffer
short buffer[16 * 16];
- int i;
+ int i, j;
const int short_pitch = pitch >> 1;
#if DWT_TYPE == 26
dyadic_analyze_26(1, 32, 32, input, short_pitch, out, 32);
@@ -2156,7 +2238,37 @@
vp9_short_fdct16x16_c_f(out, buffer, 64, 1 + DWT_PRECISION_BITS);
for (i = 0; i < 16; ++i)
vpx_memcpy(out + i * 32, buffer + i * 16, sizeof(short) * 16);
+ for (i = 0; i < 16; ++i) {+ for (j = 16; j < 32; ++j) {+ out[i * 32 + j] = divide_bits(out[i * 32 + j], DWT_PRECISION_BITS - 2);
+ }
+ }
+ for (i = 16; i < 32; ++i) {+ for (j = 0; j < 32; ++j) {+ out[i * 32 + j] = divide_bits(out[i * 32 + j], DWT_PRECISION_BITS - 2);
+ }
+ }
+}
+#elif DWTDCT_TYPE == DWTDCT16X16
+
+void vp9_short_fdct32x32_c(short *input, short *out, int pitch) {+ // assume out is a 32x32 buffer
+ short buffer[16 * 16];
+ int i, j;
+ const int short_pitch = pitch >> 1;
+#if DWT_TYPE == 26
+ dyadic_analyze_26(1, 32, 32, input, short_pitch, out, 32);
+#elif DWT_TYPE == 97
+ dyadic_analyze_97(1, 32, 32, input, short_pitch, out, 32);
+#elif DWT_TYPE == 53
+ dyadic_analyze_53(1, 32, 32, input, short_pitch, out, 32);
+#endif
+ // TODO(debargha): Implement more efficiently by adding output pitch
+ // argument to the dct16x16 function
+ vp9_short_fdct16x16_c_f(out, buffer, 64, 1 + DWT_PRECISION_BITS);
+ for (i = 0; i < 16; ++i)
+ vpx_memcpy(out + i * 32, buffer + i * 16, sizeof(short) * 16);
vp9_short_fdct16x16_c_f(out + 16, buffer, 64, 1 + DWT_PRECISION_BITS);
for (i = 0; i < 16; ++i)
vpx_memcpy(out + i * 32 + 16, buffer + i * 16, sizeof(short) * 16);
@@ -2170,6 +2282,52 @@
vpx_memcpy(out + i * 32 + 33 * 16, buffer + i * 16, sizeof(short) * 16);
}
+#elif DWTDCT_TYPE == DWTDCT8X8
+
+void vp9_short_fdct32x32_c(short *input, short *out, int pitch) {+ // assume out is a 32x32 buffer
+ short buffer[8 * 8];
+ int i, j;
+ const int short_pitch = pitch >> 1;
+#if DWT_TYPE == 26
+ dyadic_analyze_26(2, 32, 32, input, short_pitch, out, 32);
+#elif DWT_TYPE == 97
+ dyadic_analyze_97(2, 32, 32, input, short_pitch, out, 32);
+#elif DWT_TYPE == 53
+ dyadic_analyze_53(2, 32, 32, input, short_pitch, out, 32);
+#endif
+ // TODO(debargha): Implement more efficiently by adding output pitch
+ // argument to the dct16x16 function
+ vp9_short_fdct8x8_c_f(out, buffer, 64, 1 + DWT_PRECISION_BITS);
+ for (i = 0; i < 8; ++i)
+ vpx_memcpy(out + i * 32, buffer + i * 8, sizeof(short) * 8);
+
+ vp9_short_fdct8x8_c_f(out + 8, buffer, 64, 1 + DWT_PRECISION_BITS);
+ for (i = 0; i < 8; ++i)
+ vpx_memcpy(out + i * 32 + 8, buffer + i * 8, sizeof(short) * 8);
+
+ vp9_short_fdct8x8_c_f(out + 32 * 8, buffer, 64, 1 + DWT_PRECISION_BITS);
+ for (i = 0; i < 8; ++i)
+ vpx_memcpy(out + i * 32 + 32 * 8, buffer + i * 8, sizeof(short) * 8);
+
+ vp9_short_fdct8x8_c_f(out + 33 * 8, buffer, 64, 1 + DWT_PRECISION_BITS);
+ for (i = 0; i < 8; ++i)
+ vpx_memcpy(out + i * 32 + 33 * 8, buffer + i * 8, sizeof(short) * 8);
+
+ for (i = 0; i < 16; ++i) {+ for (j = 16; j < 32; ++j) {+ out[i * 32 + j] = divide_bits(out[i * 32 + j], DWT_PRECISION_BITS - 2);
+ }
+ }
+ for (i = 16; i < 32; ++i) {+ for (j = 0; j < 32; ++j) {+ out[i * 32 + j] = divide_bits(out[i * 32 + j], DWT_PRECISION_BITS - 2);
+ }
+ }
+}
+
+#endif
+
#if CONFIG_TX64X64
void vp9_short_fdct64x64_c(short *input, short *out, int pitch) {// assume out is a 64x64 buffer
@@ -2189,6 +2347,18 @@
for (i = 0; i < 16; ++i)
vpx_memcpy(out + i * 64, buffer + i * 16, sizeof(short) * 16);
+#if DWTDCT_TYPE == DWTDCT16X16_LEAN
+ for (i = 0; i < 16; ++i) {+ for (j = 16; j < 48; ++j) {+ out[i * 64 + j] = divide_bits(out[i * 64 + j], DWT_PRECISION_BITS - 1);
+ }
+ }
+ for (i = 16; i < 64; ++i) {+ for (j = 0; j < 64; ++j) {+ out[i * 64 + j] = divide_bits(out[i * 64 + j], DWT_PRECISION_BITS - 1);
+ }
+ }
+#elif DWTDCT_TYPE == DWTDCT16X16
vp9_short_fdct16x16_c_f(out + 16, buffer, 128, 2 + DWT_PRECISION_BITS);
for (i = 0; i < 16; ++i)
vpx_memcpy(out + i * 64 + 16, buffer + i * 16, sizeof(short) * 16);
@@ -2204,29 +2374,17 @@
// There is no dct used on the highest bands for now.
// Need to scale these coeffs by a factor of 2/2^DWT_PRECISION_BITS
// TODO(debargha): experiment with turning these coeffs to 0
-#if DWT_PRECISION_BITS < 1
for (i = 0; i < 32; ++i) {- for (j = 0; j < 32; ++j) {- out[i * 64 + 32 + j] <<= (1 - DWT_PRECISION_BITS);
+ for (j = 32; j < 64; ++j) {+ out[i * 64 + j] = divide_bits(out[i * 64 + j], DWT_PRECISION_BITS - 1);
}
}
- for (i = 0; i < 32; ++i) {+ for (i = 32; i < 64; ++i) { for (j = 0; j < 64; ++j) {- out[i * 64 + j] <<= (1 - DWT_PRECISION_BITS);
+ out[i * 64 + j] = divide_bits(out[i * 64 + j], DWT_PRECISION_BITS - 1);
}
}
-#else
- for (i = 0; i < 32; ++i) {- for (j = 0; j < 32; ++j) {- out[i * 64 + 32 + j] >>= (DWT_PRECISION_BITS - 1);
- }
- }
- for (i = 0; i < 32; ++i) {- for (j = 0; j < 64; ++j) {- out[i * 64 + j] >>= (DWT_PRECISION_BITS - 1);
- }
- }
-#endif
+#endif // DWTDCT_TYPE
}
#endif // CONFIG_TX64X64
#endif // CONFIG_DWTDCTHYBRID
--
⑨