1 | #include "Dither.hpp" |
2 | #include "ForceInline.hpp" |
3 | #include "ProcessDxtc.hpp" |
4 | |
5 | #include <assert.h> |
6 | #include <stdint.h> |
7 | #include <string.h> |
8 | |
9 | #ifdef __ARM_NEON |
10 | # include <arm_neon.h> |
11 | #endif |
12 | |
13 | #if defined __AVX__ && !defined __SSE4_1__ |
14 | # define __SSE4_1__ |
15 | #endif |
16 | |
17 | #if defined __SSE4_1__ || defined __AVX2__ |
18 | # ifdef _MSC_VER |
19 | # include <intrin.h> |
20 | # else |
21 | # include <x86intrin.h> |
22 | # ifndef _mm256_cvtsi256_si32 |
23 | # define _mm256_cvtsi256_si32( v ) ( _mm_cvtsi128_si32( _mm256_castsi256_si128( v ) ) ) |
24 | # endif |
25 | # endif |
26 | #endif |
27 | |
28 | |
29 | static etcpak_force_inline uint16_t to565( uint8_t r, uint8_t g, uint8_t b ) |
30 | { |
31 | return ( ( r & 0xF8 ) << 8 ) | ( ( g & 0xFC ) << 3 ) | ( b >> 3 ); |
32 | } |
33 | |
34 | static etcpak_force_inline uint16_t to565( uint32_t c ) |
35 | { |
36 | return |
37 | ( ( c & 0xF80000 ) >> 19 ) | |
38 | ( ( c & 0x00FC00 ) >> 5 ) | |
39 | ( ( c & 0x0000F8 ) << 8 ); |
40 | } |
41 | |
42 | static const uint8_t DxtcIndexTable[256] = { |
43 | 85, 87, 86, 84, 93, 95, 94, 92, 89, 91, 90, 88, 81, 83, 82, 80, |
44 | 117, 119, 118, 116, 125, 127, 126, 124, 121, 123, 122, 120, 113, 115, 114, 112, |
45 | 101, 103, 102, 100, 109, 111, 110, 108, 105, 107, 106, 104, 97, 99, 98, 96, |
46 | 69, 71, 70, 68, 77, 79, 78, 76, 73, 75, 74, 72, 65, 67, 66, 64, |
47 | 213, 215, 214, 212, 221, 223, 222, 220, 217, 219, 218, 216, 209, 211, 210, 208, |
48 | 245, 247, 246, 244, 253, 255, 254, 252, 249, 251, 250, 248, 241, 243, 242, 240, |
49 | 229, 231, 230, 228, 237, 239, 238, 236, 233, 235, 234, 232, 225, 227, 226, 224, |
50 | 197, 199, 198, 196, 205, 207, 206, 204, 201, 203, 202, 200, 193, 195, 194, 192, |
51 | 149, 151, 150, 148, 157, 159, 158, 156, 153, 155, 154, 152, 145, 147, 146, 144, |
52 | 181, 183, 182, 180, 189, 191, 190, 188, 185, 187, 186, 184, 177, 179, 178, 176, |
53 | 165, 167, 166, 164, 173, 175, 174, 172, 169, 171, 170, 168, 161, 163, 162, 160, |
54 | 133, 135, 134, 132, 141, 143, 142, 140, 137, 139, 138, 136, 129, 131, 130, 128, |
55 | 21, 23, 22, 20, 29, 31, 30, 28, 25, 27, 26, 24, 17, 19, 18, 16, |
56 | 53, 55, 54, 52, 61, 63, 62, 60, 57, 59, 58, 56, 49, 51, 50, 48, |
57 | 37, 39, 38, 36, 45, 47, 46, 44, 41, 43, 42, 40, 33, 35, 34, 32, |
58 | 5, 7, 6, 4, 13, 15, 14, 12, 9, 11, 10, 8, 1, 3, 2, 0 |
59 | }; |
60 | |
61 | static const uint8_t AlphaIndexTable_SSE[64] = { |
62 | 9, 15, 14, 13, 12, 11, 10, 8, 57, 63, 62, 61, 60, 59, 58, 56, |
63 | 49, 55, 54, 53, 52, 51, 50, 48, 41, 47, 46, 45, 44, 43, 42, 40, |
64 | 33, 39, 38, 37, 36, 35, 34, 32, 25, 31, 30, 29, 28, 27, 26, 24, |
65 | 17, 23, 22, 21, 20, 19, 18, 16, 1, 7, 6, 5, 4, 3, 2, 0, |
66 | }; |
67 | |
68 | static const uint16_t DivTable[255*3+1] = { |
69 | 0xffff, 0xffff, 0xffff, 0xffff, 0xcccc, 0xaaaa, 0x9249, 0x8000, 0x71c7, 0x6666, 0x5d17, 0x5555, 0x4ec4, 0x4924, 0x4444, 0x4000, |
70 | 0x3c3c, 0x38e3, 0x35e5, 0x3333, 0x30c3, 0x2e8b, 0x2c85, 0x2aaa, 0x28f5, 0x2762, 0x25ed, 0x2492, 0x234f, 0x2222, 0x2108, 0x2000, |
71 | 0x1f07, 0x1e1e, 0x1d41, 0x1c71, 0x1bac, 0x1af2, 0x1a41, 0x1999, 0x18f9, 0x1861, 0x17d0, 0x1745, 0x16c1, 0x1642, 0x15c9, 0x1555, |
72 | 0x14e5, 0x147a, 0x1414, 0x13b1, 0x1352, 0x12f6, 0x129e, 0x1249, 0x11f7, 0x11a7, 0x115b, 0x1111, 0x10c9, 0x1084, 0x1041, 0x1000, |
73 | 0x0fc0, 0x0f83, 0x0f48, 0x0f0f, 0x0ed7, 0x0ea0, 0x0e6c, 0x0e38, 0x0e07, 0x0dd6, 0x0da7, 0x0d79, 0x0d4c, 0x0d20, 0x0cf6, 0x0ccc, |
74 | 0x0ca4, 0x0c7c, 0x0c56, 0x0c30, 0x0c0c, 0x0be8, 0x0bc5, 0x0ba2, 0x0b81, 0x0b60, 0x0b40, 0x0b21, 0x0b02, 0x0ae4, 0x0ac7, 0x0aaa, |
75 | 0x0a8e, 0x0a72, 0x0a57, 0x0a3d, 0x0a23, 0x0a0a, 0x09f1, 0x09d8, 0x09c0, 0x09a9, 0x0991, 0x097b, 0x0964, 0x094f, 0x0939, 0x0924, |
76 | 0x090f, 0x08fb, 0x08e7, 0x08d3, 0x08c0, 0x08ad, 0x089a, 0x0888, 0x0876, 0x0864, 0x0853, 0x0842, 0x0831, 0x0820, 0x0810, 0x0800, |
77 | 0x07f0, 0x07e0, 0x07d1, 0x07c1, 0x07b3, 0x07a4, 0x0795, 0x0787, 0x0779, 0x076b, 0x075d, 0x0750, 0x0743, 0x0736, 0x0729, 0x071c, |
78 | 0x070f, 0x0703, 0x06f7, 0x06eb, 0x06df, 0x06d3, 0x06c8, 0x06bc, 0x06b1, 0x06a6, 0x069b, 0x0690, 0x0685, 0x067b, 0x0670, 0x0666, |
79 | 0x065c, 0x0652, 0x0648, 0x063e, 0x0634, 0x062b, 0x0621, 0x0618, 0x060f, 0x0606, 0x05fd, 0x05f4, 0x05eb, 0x05e2, 0x05d9, 0x05d1, |
80 | 0x05c9, 0x05c0, 0x05b8, 0x05b0, 0x05a8, 0x05a0, 0x0598, 0x0590, 0x0588, 0x0581, 0x0579, 0x0572, 0x056b, 0x0563, 0x055c, 0x0555, |
81 | 0x054e, 0x0547, 0x0540, 0x0539, 0x0532, 0x052b, 0x0525, 0x051e, 0x0518, 0x0511, 0x050b, 0x0505, 0x04fe, 0x04f8, 0x04f2, 0x04ec, |
82 | 0x04e6, 0x04e0, 0x04da, 0x04d4, 0x04ce, 0x04c8, 0x04c3, 0x04bd, 0x04b8, 0x04b2, 0x04ad, 0x04a7, 0x04a2, 0x049c, 0x0497, 0x0492, |
83 | 0x048d, 0x0487, 0x0482, 0x047d, 0x0478, 0x0473, 0x046e, 0x0469, 0x0465, 0x0460, 0x045b, 0x0456, 0x0452, 0x044d, 0x0448, 0x0444, |
84 | 0x043f, 0x043b, 0x0436, 0x0432, 0x042d, 0x0429, 0x0425, 0x0421, 0x041c, 0x0418, 0x0414, 0x0410, 0x040c, 0x0408, 0x0404, 0x0400, |
85 | 0x03fc, 0x03f8, 0x03f4, 0x03f0, 0x03ec, 0x03e8, 0x03e4, 0x03e0, 0x03dd, 0x03d9, 0x03d5, 0x03d2, 0x03ce, 0x03ca, 0x03c7, 0x03c3, |
86 | 0x03c0, 0x03bc, 0x03b9, 0x03b5, 0x03b2, 0x03ae, 0x03ab, 0x03a8, 0x03a4, 0x03a1, 0x039e, 0x039b, 0x0397, 0x0394, 0x0391, 0x038e, |
87 | 0x038b, 0x0387, 0x0384, 0x0381, 0x037e, 0x037b, 0x0378, 0x0375, 0x0372, 0x036f, 0x036c, 0x0369, 0x0366, 0x0364, 0x0361, 0x035e, |
88 | 0x035b, 0x0358, 0x0355, 0x0353, 0x0350, 0x034d, 0x034a, 0x0348, 0x0345, 0x0342, 0x0340, 0x033d, 0x033a, 0x0338, 0x0335, 0x0333, |
89 | 0x0330, 0x032e, 0x032b, 0x0329, 0x0326, 0x0324, 0x0321, 0x031f, 0x031c, 0x031a, 0x0317, 0x0315, 0x0313, 0x0310, 0x030e, 0x030c, |
90 | 0x0309, 0x0307, 0x0305, 0x0303, 0x0300, 0x02fe, 0x02fc, 0x02fa, 0x02f7, 0x02f5, 0x02f3, 0x02f1, 0x02ef, 0x02ec, 0x02ea, 0x02e8, |
91 | 0x02e6, 0x02e4, 0x02e2, 0x02e0, 0x02de, 0x02dc, 0x02da, 0x02d8, 0x02d6, 0x02d4, 0x02d2, 0x02d0, 0x02ce, 0x02cc, 0x02ca, 0x02c8, |
92 | 0x02c6, 0x02c4, 0x02c2, 0x02c0, 0x02be, 0x02bc, 0x02bb, 0x02b9, 0x02b7, 0x02b5, 0x02b3, 0x02b1, 0x02b0, 0x02ae, 0x02ac, 0x02aa, |
93 | 0x02a8, 0x02a7, 0x02a5, 0x02a3, 0x02a1, 0x02a0, 0x029e, 0x029c, 0x029b, 0x0299, 0x0297, 0x0295, 0x0294, 0x0292, 0x0291, 0x028f, |
94 | 0x028d, 0x028c, 0x028a, 0x0288, 0x0287, 0x0285, 0x0284, 0x0282, 0x0280, 0x027f, 0x027d, 0x027c, 0x027a, 0x0279, 0x0277, 0x0276, |
95 | 0x0274, 0x0273, 0x0271, 0x0270, 0x026e, 0x026d, 0x026b, 0x026a, 0x0268, 0x0267, 0x0265, 0x0264, 0x0263, 0x0261, 0x0260, 0x025e, |
96 | 0x025d, 0x025c, 0x025a, 0x0259, 0x0257, 0x0256, 0x0255, 0x0253, 0x0252, 0x0251, 0x024f, 0x024e, 0x024d, 0x024b, 0x024a, 0x0249, |
97 | 0x0247, 0x0246, 0x0245, 0x0243, 0x0242, 0x0241, 0x0240, 0x023e, 0x023d, 0x023c, 0x023b, 0x0239, 0x0238, 0x0237, 0x0236, 0x0234, |
98 | 0x0233, 0x0232, 0x0231, 0x0230, 0x022e, 0x022d, 0x022c, 0x022b, 0x022a, 0x0229, 0x0227, 0x0226, 0x0225, 0x0224, 0x0223, 0x0222, |
99 | 0x0220, 0x021f, 0x021e, 0x021d, 0x021c, 0x021b, 0x021a, 0x0219, 0x0218, 0x0216, 0x0215, 0x0214, 0x0213, 0x0212, 0x0211, 0x0210, |
100 | 0x020f, 0x020e, 0x020d, 0x020c, 0x020b, 0x020a, 0x0209, 0x0208, 0x0207, 0x0206, 0x0205, 0x0204, 0x0203, 0x0202, 0x0201, 0x0200, |
101 | 0x01ff, 0x01fe, 0x01fd, 0x01fc, 0x01fb, 0x01fa, 0x01f9, 0x01f8, 0x01f7, 0x01f6, 0x01f5, 0x01f4, 0x01f3, 0x01f2, 0x01f1, 0x01f0, |
102 | 0x01ef, 0x01ee, 0x01ed, 0x01ec, 0x01eb, 0x01ea, 0x01e9, 0x01e9, 0x01e8, 0x01e7, 0x01e6, 0x01e5, 0x01e4, 0x01e3, 0x01e2, 0x01e1, |
103 | 0x01e0, 0x01e0, 0x01df, 0x01de, 0x01dd, 0x01dc, 0x01db, 0x01da, 0x01da, 0x01d9, 0x01d8, 0x01d7, 0x01d6, 0x01d5, 0x01d4, 0x01d4, |
104 | 0x01d3, 0x01d2, 0x01d1, 0x01d0, 0x01cf, 0x01cf, 0x01ce, 0x01cd, 0x01cc, 0x01cb, 0x01cb, 0x01ca, 0x01c9, 0x01c8, 0x01c7, 0x01c7, |
105 | 0x01c6, 0x01c5, 0x01c4, 0x01c3, 0x01c3, 0x01c2, 0x01c1, 0x01c0, 0x01c0, 0x01bf, 0x01be, 0x01bd, 0x01bd, 0x01bc, 0x01bb, 0x01ba, |
106 | 0x01ba, 0x01b9, 0x01b8, 0x01b7, 0x01b7, 0x01b6, 0x01b5, 0x01b4, 0x01b4, 0x01b3, 0x01b2, 0x01b2, 0x01b1, 0x01b0, 0x01af, 0x01af, |
107 | 0x01ae, 0x01ad, 0x01ad, 0x01ac, 0x01ab, 0x01aa, 0x01aa, 0x01a9, 0x01a8, 0x01a8, 0x01a7, 0x01a6, 0x01a6, 0x01a5, 0x01a4, 0x01a4, |
108 | 0x01a3, 0x01a2, 0x01a2, 0x01a1, 0x01a0, 0x01a0, 0x019f, 0x019e, 0x019e, 0x019d, 0x019c, 0x019c, 0x019b, 0x019a, 0x019a, 0x0199, |
109 | 0x0198, 0x0198, 0x0197, 0x0197, 0x0196, 0x0195, 0x0195, 0x0194, 0x0193, 0x0193, 0x0192, 0x0192, 0x0191, 0x0190, 0x0190, 0x018f, |
110 | 0x018f, 0x018e, 0x018d, 0x018d, 0x018c, 0x018b, 0x018b, 0x018a, 0x018a, 0x0189, 0x0189, 0x0188, 0x0187, 0x0187, 0x0186, 0x0186, |
111 | 0x0185, 0x0184, 0x0184, 0x0183, 0x0183, 0x0182, 0x0182, 0x0181, 0x0180, 0x0180, 0x017f, 0x017f, 0x017e, 0x017e, 0x017d, 0x017d, |
112 | 0x017c, 0x017b, 0x017b, 0x017a, 0x017a, 0x0179, 0x0179, 0x0178, 0x0178, 0x0177, 0x0177, 0x0176, 0x0175, 0x0175, 0x0174, 0x0174, |
113 | 0x0173, 0x0173, 0x0172, 0x0172, 0x0171, 0x0171, 0x0170, 0x0170, 0x016f, 0x016f, 0x016e, 0x016e, 0x016d, 0x016d, 0x016c, 0x016c, |
114 | 0x016b, 0x016b, 0x016a, 0x016a, 0x0169, 0x0169, 0x0168, 0x0168, 0x0167, 0x0167, 0x0166, 0x0166, 0x0165, 0x0165, 0x0164, 0x0164, |
115 | 0x0163, 0x0163, 0x0162, 0x0162, 0x0161, 0x0161, 0x0160, 0x0160, 0x015f, 0x015f, 0x015e, 0x015e, 0x015d, 0x015d, 0x015d, 0x015c, |
116 | 0x015c, 0x015b, 0x015b, 0x015a, 0x015a, 0x0159, 0x0159, 0x0158, 0x0158, 0x0158, 0x0157, 0x0157, 0x0156, 0x0156 |
117 | }; |
118 | static const uint16_t DivTableNEON[255*3+1] = { |
119 | 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, |
120 | 0x0000, 0x1c71, 0x1af2, 0x1999, 0x1861, 0x1745, 0x1642, 0x1555, 0x147a, 0x13b1, 0x12f6, 0x1249, 0x11a7, 0x1111, 0x1084, 0x1000, |
121 | 0x0f83, 0x0f0f, 0x0ea0, 0x0e38, 0x0dd6, 0x0d79, 0x0d20, 0x0ccc, 0x0c7c, 0x0c30, 0x0be8, 0x0ba2, 0x0b60, 0x0b21, 0x0ae4, 0x0aaa, |
122 | 0x0a72, 0x0a3d, 0x0a0a, 0x09d8, 0x09a9, 0x097b, 0x094f, 0x0924, 0x08fb, 0x08d3, 0x08ad, 0x0888, 0x0864, 0x0842, 0x0820, 0x0800, |
123 | 0x07e0, 0x07c1, 0x07a4, 0x0787, 0x076b, 0x0750, 0x0736, 0x071c, 0x0703, 0x06eb, 0x06d3, 0x06bc, 0x06a6, 0x0690, 0x067b, 0x0666, |
124 | 0x0652, 0x063e, 0x062b, 0x0618, 0x0606, 0x05f4, 0x05e2, 0x05d1, 0x05c0, 0x05b0, 0x05a0, 0x0590, 0x0581, 0x0572, 0x0563, 0x0555, |
125 | 0x0547, 0x0539, 0x052b, 0x051e, 0x0511, 0x0505, 0x04f8, 0x04ec, 0x04e0, 0x04d4, 0x04c8, 0x04bd, 0x04b2, 0x04a7, 0x049c, 0x0492, |
126 | 0x0487, 0x047d, 0x0473, 0x0469, 0x0460, 0x0456, 0x044d, 0x0444, 0x043b, 0x0432, 0x0429, 0x0421, 0x0418, 0x0410, 0x0408, 0x0400, |
127 | 0x03f8, 0x03f0, 0x03e8, 0x03e0, 0x03d9, 0x03d2, 0x03ca, 0x03c3, 0x03bc, 0x03b5, 0x03ae, 0x03a8, 0x03a1, 0x039b, 0x0394, 0x038e, |
128 | 0x0387, 0x0381, 0x037b, 0x0375, 0x036f, 0x0369, 0x0364, 0x035e, 0x0358, 0x0353, 0x034d, 0x0348, 0x0342, 0x033d, 0x0338, 0x0333, |
129 | 0x032e, 0x0329, 0x0324, 0x031f, 0x031a, 0x0315, 0x0310, 0x030c, 0x0307, 0x0303, 0x02fe, 0x02fa, 0x02f5, 0x02f1, 0x02ec, 0x02e8, |
130 | 0x02e4, 0x02e0, 0x02dc, 0x02d8, 0x02d4, 0x02d0, 0x02cc, 0x02c8, 0x02c4, 0x02c0, 0x02bc, 0x02b9, 0x02b5, 0x02b1, 0x02ae, 0x02aa, |
131 | 0x02a7, 0x02a3, 0x02a0, 0x029c, 0x0299, 0x0295, 0x0292, 0x028f, 0x028c, 0x0288, 0x0285, 0x0282, 0x027f, 0x027c, 0x0279, 0x0276, |
132 | 0x0273, 0x0270, 0x026d, 0x026a, 0x0267, 0x0264, 0x0261, 0x025e, 0x025c, 0x0259, 0x0256, 0x0253, 0x0251, 0x024e, 0x024b, 0x0249, |
133 | 0x0246, 0x0243, 0x0241, 0x023e, 0x023c, 0x0239, 0x0237, 0x0234, 0x0232, 0x0230, 0x022d, 0x022b, 0x0229, 0x0226, 0x0224, 0x0222, |
134 | 0x021f, 0x021d, 0x021b, 0x0219, 0x0216, 0x0214, 0x0212, 0x0210, 0x020e, 0x020c, 0x020a, 0x0208, 0x0206, 0x0204, 0x0202, 0x0200, |
135 | 0x01fe, 0x01fc, 0x01fa, 0x01f8, 0x01f6, 0x01f4, 0x01f2, 0x01f0, 0x01ee, 0x01ec, 0x01ea, 0x01e9, 0x01e7, 0x01e5, 0x01e3, 0x01e1, |
136 | 0x01e0, 0x01de, 0x01dc, 0x01da, 0x01d9, 0x01d7, 0x01d5, 0x01d4, 0x01d2, 0x01d0, 0x01cf, 0x01cd, 0x01cb, 0x01ca, 0x01c8, 0x01c7, |
137 | 0x01c5, 0x01c3, 0x01c2, 0x01c0, 0x01bf, 0x01bd, 0x01bc, 0x01ba, 0x01b9, 0x01b7, 0x01b6, 0x01b4, 0x01b3, 0x01b2, 0x01b0, 0x01af, |
138 | 0x01ad, 0x01ac, 0x01aa, 0x01a9, 0x01a8, 0x01a6, 0x01a5, 0x01a4, 0x01a2, 0x01a1, 0x01a0, 0x019e, 0x019d, 0x019c, 0x019a, 0x0199, |
139 | 0x0198, 0x0197, 0x0195, 0x0194, 0x0193, 0x0192, 0x0190, 0x018f, 0x018e, 0x018d, 0x018b, 0x018a, 0x0189, 0x0188, 0x0187, 0x0186, |
140 | 0x0184, 0x0183, 0x0182, 0x0181, 0x0180, 0x017f, 0x017e, 0x017d, 0x017b, 0x017a, 0x0179, 0x0178, 0x0177, 0x0176, 0x0175, 0x0174, |
141 | 0x0173, 0x0172, 0x0171, 0x0170, 0x016f, 0x016e, 0x016d, 0x016c, 0x016b, 0x016a, 0x0169, 0x0168, 0x0167, 0x0166, 0x0165, 0x0164, |
142 | 0x0163, 0x0162, 0x0161, 0x0160, 0x015f, 0x015e, 0x015d, 0x015c, 0x015b, 0x015a, 0x0159, 0x0158, 0x0158, 0x0157, 0x0156, 0x0155, |
143 | 0x0154, 0x0153, 0x0152, 0x0151, 0x0150, 0x0150, 0x014f, 0x014e, 0x014d, 0x014c, 0x014b, 0x014a, 0x014a, 0x0149, 0x0148, 0x0147, |
144 | 0x0146, 0x0146, 0x0145, 0x0144, 0x0143, 0x0142, 0x0142, 0x0141, 0x0140, 0x013f, 0x013e, 0x013e, 0x013d, 0x013c, 0x013b, 0x013b, |
145 | 0x013a, 0x0139, 0x0138, 0x0138, 0x0137, 0x0136, 0x0135, 0x0135, 0x0134, 0x0133, 0x0132, 0x0132, 0x0131, 0x0130, 0x0130, 0x012f, |
146 | 0x012e, 0x012e, 0x012d, 0x012c, 0x012b, 0x012b, 0x012a, 0x0129, 0x0129, 0x0128, 0x0127, 0x0127, 0x0126, 0x0125, 0x0125, 0x0124, |
147 | 0x0123, 0x0123, 0x0122, 0x0121, 0x0121, 0x0120, 0x0120, 0x011f, 0x011e, 0x011e, 0x011d, 0x011c, 0x011c, 0x011b, 0x011b, 0x011a, |
148 | 0x0119, 0x0119, 0x0118, 0x0118, 0x0117, 0x0116, 0x0116, 0x0115, 0x0115, 0x0114, 0x0113, 0x0113, 0x0112, 0x0112, 0x0111, 0x0111, |
149 | 0x0110, 0x010f, 0x010f, 0x010e, 0x010e, 0x010d, 0x010d, 0x010c, 0x010c, 0x010b, 0x010a, 0x010a, 0x0109, 0x0109, 0x0108, 0x0108, |
150 | 0x0107, 0x0107, 0x0106, 0x0106, 0x0105, 0x0105, 0x0104, 0x0104, 0x0103, 0x0103, 0x0102, 0x0102, 0x0101, 0x0101, 0x0100, 0x0100, |
151 | 0x00ff, 0x00ff, 0x00fe, 0x00fe, 0x00fd, 0x00fd, 0x00fc, 0x00fc, 0x00fb, 0x00fb, 0x00fa, 0x00fa, 0x00f9, 0x00f9, 0x00f8, 0x00f8, |
152 | 0x00f7, 0x00f7, 0x00f6, 0x00f6, 0x00f5, 0x00f5, 0x00f4, 0x00f4, 0x00f4, 0x00f3, 0x00f3, 0x00f2, 0x00f2, 0x00f1, 0x00f1, 0x00f0, |
153 | 0x00f0, 0x00f0, 0x00ef, 0x00ef, 0x00ee, 0x00ee, 0x00ed, 0x00ed, 0x00ed, 0x00ec, 0x00ec, 0x00eb, 0x00eb, 0x00ea, 0x00ea, 0x00ea, |
154 | 0x00e9, 0x00e9, 0x00e8, 0x00e8, 0x00e7, 0x00e7, 0x00e7, 0x00e6, 0x00e6, 0x00e5, 0x00e5, 0x00e5, 0x00e4, 0x00e4, 0x00e3, 0x00e3, |
155 | 0x00e3, 0x00e2, 0x00e2, 0x00e1, 0x00e1, 0x00e1, 0x00e0, 0x00e0, 0x00e0, 0x00df, 0x00df, 0x00de, 0x00de, 0x00de, 0x00dd, 0x00dd, |
156 | 0x00dd, 0x00dc, 0x00dc, 0x00db, 0x00db, 0x00db, 0x00da, 0x00da, 0x00da, 0x00d9, 0x00d9, 0x00d9, 0x00d8, 0x00d8, 0x00d7, 0x00d7, |
157 | 0x00d7, 0x00d6, 0x00d6, 0x00d6, 0x00d5, 0x00d5, 0x00d5, 0x00d4, 0x00d4, 0x00d4, 0x00d3, 0x00d3, 0x00d3, 0x00d2, 0x00d2, 0x00d2, |
158 | 0x00d1, 0x00d1, 0x00d1, 0x00d0, 0x00d0, 0x00d0, 0x00cf, 0x00cf, 0x00cf, 0x00ce, 0x00ce, 0x00ce, 0x00cd, 0x00cd, 0x00cd, 0x00cc, |
159 | 0x00cc, 0x00cc, 0x00cb, 0x00cb, 0x00cb, 0x00ca, 0x00ca, 0x00ca, 0x00c9, 0x00c9, 0x00c9, 0x00c9, 0x00c8, 0x00c8, 0x00c8, 0x00c7, |
160 | 0x00c7, 0x00c7, 0x00c6, 0x00c6, 0x00c6, 0x00c5, 0x00c5, 0x00c5, 0x00c5, 0x00c4, 0x00c4, 0x00c4, 0x00c3, 0x00c3, 0x00c3, 0x00c3, |
161 | 0x00c2, 0x00c2, 0x00c2, 0x00c1, 0x00c1, 0x00c1, 0x00c1, 0x00c0, 0x00c0, 0x00c0, 0x00bf, 0x00bf, 0x00bf, 0x00bf, 0x00be, 0x00be, |
162 | 0x00be, 0x00bd, 0x00bd, 0x00bd, 0x00bd, 0x00bc, 0x00bc, 0x00bc, 0x00bc, 0x00bb, 0x00bb, 0x00bb, 0x00ba, 0x00ba, 0x00ba, 0x00ba, |
163 | 0x00b9, 0x00b9, 0x00b9, 0x00b9, 0x00b8, 0x00b8, 0x00b8, 0x00b8, 0x00b7, 0x00b7, 0x00b7, 0x00b7, 0x00b6, 0x00b6, 0x00b6, 0x00b6, |
164 | 0x00b5, 0x00b5, 0x00b5, 0x00b5, 0x00b4, 0x00b4, 0x00b4, 0x00b4, 0x00b3, 0x00b3, 0x00b3, 0x00b3, 0x00b2, 0x00b2, 0x00b2, 0x00b2, |
165 | 0x00b1, 0x00b1, 0x00b1, 0x00b1, 0x00b0, 0x00b0, 0x00b0, 0x00b0, 0x00af, 0x00af, 0x00af, 0x00af, 0x00ae, 0x00ae, 0x00ae, 0x00ae, |
166 | 0x00ae, 0x00ad, 0x00ad, 0x00ad, 0x00ad, 0x00ac, 0x00ac, 0x00ac, 0x00ac, 0x00ac, 0x00ab, 0x00ab, 0x00ab, 0x00ab, |
167 | }; |
168 | |
169 | static const uint16_t DivTableAlpha[256] = { |
170 | 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xe38e, 0xcccc, 0xba2e, 0xaaaa, 0x9d89, 0x9249, 0x8888, 0x8000, |
171 | 0x7878, 0x71c7, 0x6bca, 0x6666, 0x6186, 0x5d17, 0x590b, 0x5555, 0x51eb, 0x4ec4, 0x4bda, 0x4924, 0x469e, 0x4444, 0x4210, 0x4000, |
172 | 0x3e0f, 0x3c3c, 0x3a83, 0x38e3, 0x3759, 0x35e5, 0x3483, 0x3333, 0x31f3, 0x30c3, 0x2fa0, 0x2e8b, 0x2d82, 0x2c85, 0x2b93, 0x2aaa, |
173 | 0x29cb, 0x28f5, 0x2828, 0x2762, 0x26a4, 0x25ed, 0x253c, 0x2492, 0x23ee, 0x234f, 0x22b6, 0x2222, 0x2192, 0x2108, 0x2082, 0x2000, |
174 | 0x1f81, 0x1f07, 0x1e91, 0x1e1e, 0x1dae, 0x1d41, 0x1cd8, 0x1c71, 0x1c0e, 0x1bac, 0x1b4e, 0x1af2, 0x1a98, 0x1a41, 0x19ec, 0x1999, |
175 | 0x1948, 0x18f9, 0x18ac, 0x1861, 0x1818, 0x17d0, 0x178a, 0x1745, 0x1702, 0x16c1, 0x1681, 0x1642, 0x1605, 0x15c9, 0x158e, 0x1555, |
176 | 0x151d, 0x14e5, 0x14af, 0x147a, 0x1446, 0x1414, 0x13e2, 0x13b1, 0x1381, 0x1352, 0x1323, 0x12f6, 0x12c9, 0x129e, 0x1273, 0x1249, |
177 | 0x121f, 0x11f7, 0x11cf, 0x11a7, 0x1181, 0x115b, 0x1135, 0x1111, 0x10ec, 0x10c9, 0x10a6, 0x1084, 0x1062, 0x1041, 0x1020, 0x1000, |
178 | 0x0fe0, 0x0fc0, 0x0fa2, 0x0f83, 0x0f66, 0x0f48, 0x0f2b, 0x0f0f, 0x0ef2, 0x0ed7, 0x0ebb, 0x0ea0, 0x0e86, 0x0e6c, 0x0e52, 0x0e38, |
179 | 0x0e1f, 0x0e07, 0x0dee, 0x0dd6, 0x0dbe, 0x0da7, 0x0d90, 0x0d79, 0x0d62, 0x0d4c, 0x0d36, 0x0d20, 0x0d0b, 0x0cf6, 0x0ce1, 0x0ccc, |
180 | 0x0cb8, 0x0ca4, 0x0c90, 0x0c7c, 0x0c69, 0x0c56, 0x0c43, 0x0c30, 0x0c1e, 0x0c0c, 0x0bfa, 0x0be8, 0x0bd6, 0x0bc5, 0x0bb3, 0x0ba2, |
181 | 0x0b92, 0x0b81, 0x0b70, 0x0b60, 0x0b50, 0x0b40, 0x0b30, 0x0b21, 0x0b11, 0x0b02, 0x0af3, 0x0ae4, 0x0ad6, 0x0ac7, 0x0ab8, 0x0aaa, |
182 | 0x0a9c, 0x0a8e, 0x0a80, 0x0a72, 0x0a65, 0x0a57, 0x0a4a, 0x0a3d, 0x0a30, 0x0a23, 0x0a16, 0x0a0a, 0x09fd, 0x09f1, 0x09e4, 0x09d8, |
183 | 0x09cc, 0x09c0, 0x09b4, 0x09a9, 0x099d, 0x0991, 0x0986, 0x097b, 0x0970, 0x0964, 0x095a, 0x094f, 0x0944, 0x0939, 0x092f, 0x0924, |
184 | 0x091a, 0x090f, 0x0905, 0x08fb, 0x08f1, 0x08e7, 0x08dd, 0x08d3, 0x08ca, 0x08c0, 0x08b7, 0x08ad, 0x08a4, 0x089a, 0x0891, 0x0888, |
185 | 0x087f, 0x0876, 0x086d, 0x0864, 0x085b, 0x0853, 0x084a, 0x0842, 0x0839, 0x0831, 0x0828, 0x0820, 0x0818, 0x0810, 0x0808, 0x0800, |
186 | }; |
187 | |
188 | static etcpak_force_inline uint64_t ProcessRGB( const uint8_t* src ) |
189 | { |
190 | #ifdef __SSE4_1__ |
191 | __m128i px0 = _mm_loadu_si128(((__m128i*)src) + 0); |
192 | __m128i px1 = _mm_loadu_si128(((__m128i*)src) + 1); |
193 | __m128i px2 = _mm_loadu_si128(((__m128i*)src) + 2); |
194 | __m128i px3 = _mm_loadu_si128(((__m128i*)src) + 3); |
195 | |
196 | __m128i smask = _mm_set1_epi32( 0xF8FCF8 ); |
197 | __m128i sd0 = _mm_and_si128( px0, smask ); |
198 | __m128i sd1 = _mm_and_si128( px1, smask ); |
199 | __m128i sd2 = _mm_and_si128( px2, smask ); |
200 | __m128i sd3 = _mm_and_si128( px3, smask ); |
201 | |
202 | __m128i sc = _mm_shuffle_epi32(sd0, _MM_SHUFFLE(0, 0, 0, 0)); |
203 | |
204 | __m128i sc0 = _mm_cmpeq_epi8(sd0, sc); |
205 | __m128i sc1 = _mm_cmpeq_epi8(sd1, sc); |
206 | __m128i sc2 = _mm_cmpeq_epi8(sd2, sc); |
207 | __m128i sc3 = _mm_cmpeq_epi8(sd3, sc); |
208 | |
209 | __m128i sm0 = _mm_and_si128(sc0, sc1); |
210 | __m128i sm1 = _mm_and_si128(sc2, sc3); |
211 | __m128i sm = _mm_and_si128(sm0, sm1); |
212 | |
213 | if( _mm_testc_si128(sm, _mm_set1_epi32(-1)) ) |
214 | { |
215 | uint32_t c; |
216 | memcpy( &c, src, 4 ); |
217 | return uint64_t( to565( c ) ) << 16; |
218 | } |
219 | |
220 | __m128i min0 = _mm_min_epu8( px0, px1 ); |
221 | __m128i min1 = _mm_min_epu8( px2, px3 ); |
222 | __m128i min2 = _mm_min_epu8( min0, min1 ); |
223 | |
224 | __m128i max0 = _mm_max_epu8( px0, px1 ); |
225 | __m128i max1 = _mm_max_epu8( px2, px3 ); |
226 | __m128i max2 = _mm_max_epu8( max0, max1 ); |
227 | |
228 | __m128i min3 = _mm_shuffle_epi32( min2, _MM_SHUFFLE( 2, 3, 0, 1 ) ); |
229 | __m128i max3 = _mm_shuffle_epi32( max2, _MM_SHUFFLE( 2, 3, 0, 1 ) ); |
230 | __m128i min4 = _mm_min_epu8( min2, min3 ); |
231 | __m128i max4 = _mm_max_epu8( max2, max3 ); |
232 | |
233 | __m128i min5 = _mm_shuffle_epi32( min4, _MM_SHUFFLE( 0, 0, 2, 2 ) ); |
234 | __m128i max5 = _mm_shuffle_epi32( max4, _MM_SHUFFLE( 0, 0, 2, 2 ) ); |
235 | __m128i rmin = _mm_min_epu8( min4, min5 ); |
236 | __m128i rmax = _mm_max_epu8( max4, max5 ); |
237 | |
238 | __m128i range1 = _mm_subs_epu8( rmax, rmin ); |
239 | __m128i range2 = _mm_sad_epu8( rmax, rmin ); |
240 | |
241 | uint32_t vrange = _mm_cvtsi128_si32( range2 ) >> 1; |
242 | __m128i range = _mm_set1_epi16( DivTable[vrange] ); |
243 | |
244 | __m128i inset1 = _mm_srli_epi16( range1, 4 ); |
245 | __m128i inset = _mm_and_si128( inset1, _mm_set1_epi8( 0xF ) ); |
246 | __m128i min = _mm_adds_epu8( rmin, inset ); |
247 | __m128i max = _mm_subs_epu8( rmax, inset ); |
248 | |
249 | __m128i c0 = _mm_subs_epu8( px0, rmin ); |
250 | __m128i c1 = _mm_subs_epu8( px1, rmin ); |
251 | __m128i c2 = _mm_subs_epu8( px2, rmin ); |
252 | __m128i c3 = _mm_subs_epu8( px3, rmin ); |
253 | |
254 | __m128i is0 = _mm_maddubs_epi16( c0, _mm_set1_epi8( 1 ) ); |
255 | __m128i is1 = _mm_maddubs_epi16( c1, _mm_set1_epi8( 1 ) ); |
256 | __m128i is2 = _mm_maddubs_epi16( c2, _mm_set1_epi8( 1 ) ); |
257 | __m128i is3 = _mm_maddubs_epi16( c3, _mm_set1_epi8( 1 ) ); |
258 | |
259 | __m128i s0 = _mm_hadd_epi16( is0, is1 ); |
260 | __m128i s1 = _mm_hadd_epi16( is2, is3 ); |
261 | |
262 | __m128i m0 = _mm_mulhi_epu16( s0, range ); |
263 | __m128i m1 = _mm_mulhi_epu16( s1, range ); |
264 | |
265 | __m128i p0 = _mm_packus_epi16( m0, m1 ); |
266 | |
267 | __m128i p1 = _mm_or_si128( _mm_srai_epi32( p0, 6 ), _mm_srai_epi32( p0, 12 ) ); |
268 | __m128i p2 = _mm_or_si128( _mm_srai_epi32( p0, 18 ), p0 ); |
269 | __m128i p3 = _mm_or_si128( p1, p2 ); |
270 | __m128i p =_mm_shuffle_epi8( p3, _mm_set1_epi32( 0x0C080400 ) ); |
271 | |
272 | uint32_t vmin = _mm_cvtsi128_si32( min ); |
273 | uint32_t vmax = _mm_cvtsi128_si32( max ); |
274 | uint32_t vp = _mm_cvtsi128_si32( p ); |
275 | |
276 | return uint64_t( ( uint64_t( to565( vmin ) ) << 16 ) | to565( vmax ) | ( uint64_t( vp ) << 32 ) ); |
277 | #elif defined __ARM_NEON |
278 | # ifdef __aarch64__ |
279 | uint8x16x4_t px = vld4q_u8( src ); |
280 | |
281 | uint8x16_t lr = px.val[0]; |
282 | uint8x16_t lg = px.val[1]; |
283 | uint8x16_t lb = px.val[2]; |
284 | |
285 | uint8_t rmaxr = vmaxvq_u8( lr ); |
286 | uint8_t rmaxg = vmaxvq_u8( lg ); |
287 | uint8_t rmaxb = vmaxvq_u8( lb ); |
288 | |
289 | uint8_t rminr = vminvq_u8( lr ); |
290 | uint8_t rming = vminvq_u8( lg ); |
291 | uint8_t rminb = vminvq_u8( lb ); |
292 | |
293 | int rr = rmaxr - rminr; |
294 | int rg = rmaxg - rming; |
295 | int rb = rmaxb - rminb; |
296 | |
297 | int vrange1 = rr + rg + rb; |
298 | uint16_t vrange2 = DivTableNEON[vrange1]; |
299 | |
300 | uint8_t insetr = rr >> 4; |
301 | uint8_t insetg = rg >> 4; |
302 | uint8_t insetb = rb >> 4; |
303 | |
304 | uint8_t minr = rminr + insetr; |
305 | uint8_t ming = rming + insetg; |
306 | uint8_t minb = rminb + insetb; |
307 | |
308 | uint8_t maxr = rmaxr - insetr; |
309 | uint8_t maxg = rmaxg - insetg; |
310 | uint8_t maxb = rmaxb - insetb; |
311 | |
312 | uint8x16_t cr = vsubq_u8( lr, vdupq_n_u8( rminr ) ); |
313 | uint8x16_t cg = vsubq_u8( lg, vdupq_n_u8( rming ) ); |
314 | uint8x16_t cb = vsubq_u8( lb, vdupq_n_u8( rminb ) ); |
315 | |
316 | uint16x8_t is0l = vaddl_u8( vget_low_u8( cr ), vget_low_u8( cg ) ); |
317 | uint16x8_t is0h = vaddl_u8( vget_high_u8( cr ), vget_high_u8( cg ) ); |
318 | uint16x8_t is1l = vaddw_u8( is0l, vget_low_u8( cb ) ); |
319 | uint16x8_t is1h = vaddw_u8( is0h, vget_high_u8( cb ) ); |
320 | |
321 | int16x8_t range = vdupq_n_s16( vrange2 ); |
322 | uint16x8_t m0 = vreinterpretq_u16_s16( vqdmulhq_s16( vreinterpretq_s16_u16( is1l ), range ) ); |
323 | uint16x8_t m1 = vreinterpretq_u16_s16( vqdmulhq_s16( vreinterpretq_s16_u16( is1h ), range ) ); |
324 | |
325 | uint8x8_t p00 = vmovn_u16( m0 ); |
326 | uint8x8_t p01 = vmovn_u16( m1 ); |
327 | uint8x16_t p0 = vcombine_u8( p00, p01 ); |
328 | |
329 | uint32x4_t p1 = vaddq_u32( vshrq_n_u32( vreinterpretq_u32_u8( p0 ), 6 ), vshrq_n_u32( vreinterpretq_u32_u8( p0 ), 12 ) ); |
330 | uint32x4_t p2 = vaddq_u32( vshrq_n_u32( vreinterpretq_u32_u8( p0 ), 18 ), vreinterpretq_u32_u8( p0 ) ); |
331 | uint32x4_t p3 = vaddq_u32( p1, p2 ); |
332 | |
333 | uint16x4x2_t p4 = vuzp_u16( vget_low_u16( vreinterpretq_u16_u32( p3 ) ), vget_high_u16( vreinterpretq_u16_u32( p3 ) ) ); |
334 | uint8x8x2_t p = vuzp_u8( vreinterpret_u8_u16( p4.val[0] ), vreinterpret_u8_u16( p4.val[0] ) ); |
335 | |
336 | uint32_t vp; |
337 | vst1_lane_u32( &vp, vreinterpret_u32_u8( p.val[0] ), 0 ); |
338 | |
339 | return uint64_t( ( uint64_t( to565( minr, ming, minb ) ) << 16 ) | to565( maxr, maxg, maxb ) | ( uint64_t( vp ) << 32 ) ); |
340 | # else |
341 | uint32x4_t px0 = vld1q_u32( (uint32_t*)src ); |
342 | uint32x4_t px1 = vld1q_u32( (uint32_t*)src + 4 ); |
343 | uint32x4_t px2 = vld1q_u32( (uint32_t*)src + 8 ); |
344 | uint32x4_t px3 = vld1q_u32( (uint32_t*)src + 12 ); |
345 | |
346 | uint32x4_t smask = vdupq_n_u32( 0xF8FCF8 ); |
347 | uint32x4_t sd0 = vandq_u32( smask, px0 ); |
348 | uint32x4_t sd1 = vandq_u32( smask, px1 ); |
349 | uint32x4_t sd2 = vandq_u32( smask, px2 ); |
350 | uint32x4_t sd3 = vandq_u32( smask, px3 ); |
351 | |
352 | uint32x4_t sc = vdupq_n_u32( sd0[0] ); |
353 | |
354 | uint32x4_t sc0 = vceqq_u32( sd0, sc ); |
355 | uint32x4_t sc1 = vceqq_u32( sd1, sc ); |
356 | uint32x4_t sc2 = vceqq_u32( sd2, sc ); |
357 | uint32x4_t sc3 = vceqq_u32( sd3, sc ); |
358 | |
359 | uint32x4_t sm0 = vandq_u32( sc0, sc1 ); |
360 | uint32x4_t sm1 = vandq_u32( sc2, sc3 ); |
361 | int64x2_t sm = vreinterpretq_s64_u32( vandq_u32( sm0, sm1 ) ); |
362 | |
363 | if( sm[0] == -1 && sm[1] == -1 ) |
364 | { |
365 | return uint64_t( to565( src[0], src[1], src[2] ) ) << 16; |
366 | } |
367 | |
368 | uint32x4_t mask = vdupq_n_u32( 0xFFFFFF ); |
369 | uint8x16_t l0 = vreinterpretq_u8_u32( vandq_u32( mask, px0 ) ); |
370 | uint8x16_t l1 = vreinterpretq_u8_u32( vandq_u32( mask, px1 ) ); |
371 | uint8x16_t l2 = vreinterpretq_u8_u32( vandq_u32( mask, px2 ) ); |
372 | uint8x16_t l3 = vreinterpretq_u8_u32( vandq_u32( mask, px3 ) ); |
373 | |
374 | uint8x16_t min0 = vminq_u8( l0, l1 ); |
375 | uint8x16_t min1 = vminq_u8( l2, l3 ); |
376 | uint8x16_t min2 = vminq_u8( min0, min1 ); |
377 | |
378 | uint8x16_t max0 = vmaxq_u8( l0, l1 ); |
379 | uint8x16_t max1 = vmaxq_u8( l2, l3 ); |
380 | uint8x16_t max2 = vmaxq_u8( max0, max1 ); |
381 | |
382 | uint8x16_t min3 = vreinterpretq_u8_u32( vrev64q_u32( vreinterpretq_u32_u8( min2 ) ) ); |
383 | uint8x16_t max3 = vreinterpretq_u8_u32( vrev64q_u32( vreinterpretq_u32_u8( max2 ) ) ); |
384 | |
385 | uint8x16_t min4 = vminq_u8( min2, min3 ); |
386 | uint8x16_t max4 = vmaxq_u8( max2, max3 ); |
387 | |
388 | uint8x16_t min5 = vcombine_u8( vget_high_u8( min4 ), vget_low_u8( min4 ) ); |
389 | uint8x16_t max5 = vcombine_u8( vget_high_u8( max4 ), vget_low_u8( max4 ) ); |
390 | |
391 | uint8x16_t rmin = vminq_u8( min4, min5 ); |
392 | uint8x16_t rmax = vmaxq_u8( max4, max5 ); |
393 | |
394 | uint8x16_t range1 = vsubq_u8( rmax, rmin ); |
395 | uint8x8_t range2 = vget_low_u8( range1 ); |
396 | uint8x8x2_t range3 = vzip_u8( range2, vdup_n_u8( 0 ) ); |
397 | uint16x4_t range4 = vreinterpret_u16_u8( range3.val[0] ); |
398 | |
399 | uint16_t vrange1; |
400 | uint16x4_t range5 = vpadd_u16( range4, range4 ); |
401 | uint16x4_t range6 = vpadd_u16( range5, range5 ); |
402 | vst1_lane_u16( &vrange1, range6, 0 ); |
403 | |
404 | uint32_t vrange2 = ( 2 << 16 ) / uint32_t( vrange1 + 1 ); |
405 | uint16x8_t range = vdupq_n_u16( vrange2 ); |
406 | |
407 | uint8x16_t inset = vshrq_n_u8( range1, 4 ); |
408 | uint8x16_t min = vaddq_u8( rmin, inset ); |
409 | uint8x16_t max = vsubq_u8( rmax, inset ); |
410 | |
411 | uint8x16_t c0 = vsubq_u8( l0, rmin ); |
412 | uint8x16_t c1 = vsubq_u8( l1, rmin ); |
413 | uint8x16_t c2 = vsubq_u8( l2, rmin ); |
414 | uint8x16_t c3 = vsubq_u8( l3, rmin ); |
415 | |
416 | uint16x8_t is0 = vpaddlq_u8( c0 ); |
417 | uint16x8_t is1 = vpaddlq_u8( c1 ); |
418 | uint16x8_t is2 = vpaddlq_u8( c2 ); |
419 | uint16x8_t is3 = vpaddlq_u8( c3 ); |
420 | |
421 | uint16x4_t is4 = vpadd_u16( vget_low_u16( is0 ), vget_high_u16( is0 ) ); |
422 | uint16x4_t is5 = vpadd_u16( vget_low_u16( is1 ), vget_high_u16( is1 ) ); |
423 | uint16x4_t is6 = vpadd_u16( vget_low_u16( is2 ), vget_high_u16( is2 ) ); |
424 | uint16x4_t is7 = vpadd_u16( vget_low_u16( is3 ), vget_high_u16( is3 ) ); |
425 | |
426 | uint16x8_t s0 = vcombine_u16( is4, is5 ); |
427 | uint16x8_t s1 = vcombine_u16( is6, is7 ); |
428 | |
429 | uint16x8_t m0 = vreinterpretq_u16_s16( vqdmulhq_s16( vreinterpretq_s16_u16( s0 ), vreinterpretq_s16_u16( range ) ) ); |
430 | uint16x8_t m1 = vreinterpretq_u16_s16( vqdmulhq_s16( vreinterpretq_s16_u16( s1 ), vreinterpretq_s16_u16( range ) ) ); |
431 | |
432 | uint8x8_t p00 = vmovn_u16( m0 ); |
433 | uint8x8_t p01 = vmovn_u16( m1 ); |
434 | uint8x16_t p0 = vcombine_u8( p00, p01 ); |
435 | |
436 | uint32x4_t p1 = vaddq_u32( vshrq_n_u32( vreinterpretq_u32_u8( p0 ), 6 ), vshrq_n_u32( vreinterpretq_u32_u8( p0 ), 12 ) ); |
437 | uint32x4_t p2 = vaddq_u32( vshrq_n_u32( vreinterpretq_u32_u8( p0 ), 18 ), vreinterpretq_u32_u8( p0 ) ); |
438 | uint32x4_t p3 = vaddq_u32( p1, p2 ); |
439 | |
440 | uint16x4x2_t p4 = vuzp_u16( vget_low_u16( vreinterpretq_u16_u32( p3 ) ), vget_high_u16( vreinterpretq_u16_u32( p3 ) ) ); |
441 | uint8x8x2_t p = vuzp_u8( vreinterpret_u8_u16( p4.val[0] ), vreinterpret_u8_u16( p4.val[0] ) ); |
442 | |
443 | uint32_t vmin, vmax, vp; |
444 | vst1q_lane_u32( &vmin, vreinterpretq_u32_u8( min ), 0 ); |
445 | vst1q_lane_u32( &vmax, vreinterpretq_u32_u8( max ), 0 ); |
446 | vst1_lane_u32( &vp, vreinterpret_u32_u8( p.val[0] ), 0 ); |
447 | |
448 | return uint64_t( ( uint64_t( to565( vmin ) ) << 16 ) | to565( vmax ) | ( uint64_t( vp ) << 32 ) ); |
449 | # endif |
450 | #else |
451 | uint32_t ref; |
452 | memcpy( &ref, src, 4 ); |
453 | uint32_t refMask = ref & 0xF8FCF8; |
454 | auto stmp = src + 4; |
455 | for( int i=1; i<16; i++ ) |
456 | { |
457 | uint32_t px; |
458 | memcpy( &px, stmp, 4 ); |
459 | if( ( px & 0xF8FCF8 ) != refMask ) break; |
460 | stmp += 4; |
461 | } |
462 | if( stmp == src + 64 ) |
463 | { |
464 | return uint64_t( to565( ref ) ) << 16; |
465 | } |
466 | |
467 | uint8_t min[3] = { src[0], src[1], src[2] }; |
468 | uint8_t max[3] = { src[0], src[1], src[2] }; |
469 | auto tmp = src + 4; |
470 | for( int i=1; i<16; i++ ) |
471 | { |
472 | for( int j=0; j<3; j++ ) |
473 | { |
474 | if( tmp[j] < min[j] ) min[j] = tmp[j]; |
475 | else if( tmp[j] > max[j] ) max[j] = tmp[j]; |
476 | } |
477 | tmp += 4; |
478 | } |
479 | |
480 | const uint32_t range = DivTable[max[0] - min[0] + max[1] - min[1] + max[2] - min[2]]; |
481 | const uint32_t rmin = min[0] + min[1] + min[2]; |
482 | for( int i=0; i<3; i++ ) |
483 | { |
484 | const uint8_t inset = ( max[i] - min[i] ) >> 4; |
485 | min[i] += inset; |
486 | max[i] -= inset; |
487 | } |
488 | |
489 | uint32_t data = 0; |
490 | for( int i=0; i<16; i++ ) |
491 | { |
492 | const uint32_t c = src[0] + src[1] + src[2] - rmin; |
493 | const uint8_t idx = ( c * range ) >> 16; |
494 | data |= idx << (i*2); |
495 | src += 4; |
496 | } |
497 | |
498 | return uint64_t( ( uint64_t( to565( min[0], min[1], min[2] ) ) << 16 ) | to565( max[0], max[1], max[2] ) | ( uint64_t( data ) << 32 ) ); |
499 | #endif |
500 | } |
501 | |
502 | #ifdef __AVX2__ |
503 | static etcpak_force_inline void ProcessRGB_AVX( const uint8_t* src, char*& dst ) |
504 | { |
505 | __m256i px0 = _mm256_loadu_si256(((__m256i*)src) + 0); |
506 | __m256i px1 = _mm256_loadu_si256(((__m256i*)src) + 1); |
507 | __m256i px2 = _mm256_loadu_si256(((__m256i*)src) + 2); |
508 | __m256i px3 = _mm256_loadu_si256(((__m256i*)src) + 3); |
509 | |
510 | __m256i smask = _mm256_set1_epi32( 0xF8FCF8 ); |
511 | __m256i sd0 = _mm256_and_si256( px0, smask ); |
512 | __m256i sd1 = _mm256_and_si256( px1, smask ); |
513 | __m256i sd2 = _mm256_and_si256( px2, smask ); |
514 | __m256i sd3 = _mm256_and_si256( px3, smask ); |
515 | |
516 | __m256i sc = _mm256_shuffle_epi32(sd0, _MM_SHUFFLE(0, 0, 0, 0)); |
517 | |
518 | __m256i sc0 = _mm256_cmpeq_epi8(sd0, sc); |
519 | __m256i sc1 = _mm256_cmpeq_epi8(sd1, sc); |
520 | __m256i sc2 = _mm256_cmpeq_epi8(sd2, sc); |
521 | __m256i sc3 = _mm256_cmpeq_epi8(sd3, sc); |
522 | |
523 | __m256i sm0 = _mm256_and_si256(sc0, sc1); |
524 | __m256i sm1 = _mm256_and_si256(sc2, sc3); |
525 | __m256i sm = _mm256_and_si256(sm0, sm1); |
526 | |
527 | const int64_t solid0 = 1 - _mm_testc_si128( _mm256_castsi256_si128( sm ), _mm_set1_epi32( -1 ) ); |
528 | const int64_t solid1 = 1 - _mm_testc_si128( _mm256_extracti128_si256( sm, 1 ), _mm_set1_epi32( -1 ) ); |
529 | |
530 | if( solid0 + solid1 == 0 ) |
531 | { |
532 | const auto c0 = uint64_t( to565( src[0], src[1], src[2] ) ); |
533 | const auto c1 = uint64_t( to565( src[16], src[17], src[18] ) ); |
534 | memcpy( dst, &c0, 8 ); |
535 | memcpy( dst+8, &c1, 8 ); |
536 | dst += 16; |
537 | return; |
538 | } |
539 | |
540 | __m256i min0 = _mm256_min_epu8( px0, px1 ); |
541 | __m256i min1 = _mm256_min_epu8( px2, px3 ); |
542 | __m256i min2 = _mm256_min_epu8( min0, min1 ); |
543 | |
544 | __m256i max0 = _mm256_max_epu8( px0, px1 ); |
545 | __m256i max1 = _mm256_max_epu8( px2, px3 ); |
546 | __m256i max2 = _mm256_max_epu8( max0, max1 ); |
547 | |
548 | __m256i min3 = _mm256_shuffle_epi32( min2, _MM_SHUFFLE( 2, 3, 0, 1 ) ); |
549 | __m256i max3 = _mm256_shuffle_epi32( max2, _MM_SHUFFLE( 2, 3, 0, 1 ) ); |
550 | __m256i min4 = _mm256_min_epu8( min2, min3 ); |
551 | __m256i max4 = _mm256_max_epu8( max2, max3 ); |
552 | |
553 | __m256i min5 = _mm256_shuffle_epi32( min4, _MM_SHUFFLE( 0, 0, 2, 2 ) ); |
554 | __m256i max5 = _mm256_shuffle_epi32( max4, _MM_SHUFFLE( 0, 0, 2, 2 ) ); |
555 | __m256i rmin = _mm256_min_epu8( min4, min5 ); |
556 | __m256i rmax = _mm256_max_epu8( max4, max5 ); |
557 | |
558 | __m256i range1 = _mm256_subs_epu8( rmax, rmin ); |
559 | __m256i range2 = _mm256_sad_epu8( rmax, rmin ); |
560 | |
561 | uint16_t vrange0 = DivTable[_mm256_cvtsi256_si32( range2 ) >> 1]; |
562 | uint16_t vrange1 = DivTable[_mm256_extract_epi16( range2, 8 ) >> 1]; |
563 | __m256i range00 = _mm256_set1_epi16( vrange0 ); |
564 | __m256i range = _mm256_inserti128_si256( range00, _mm_set1_epi16( vrange1 ), 1 ); |
565 | |
566 | __m256i inset1 = _mm256_srli_epi16( range1, 4 ); |
567 | __m256i inset = _mm256_and_si256( inset1, _mm256_set1_epi8( 0xF ) ); |
568 | __m256i min = _mm256_adds_epu8( rmin, inset ); |
569 | __m256i max = _mm256_subs_epu8( rmax, inset ); |
570 | |
571 | __m256i c0 = _mm256_subs_epu8( px0, rmin ); |
572 | __m256i c1 = _mm256_subs_epu8( px1, rmin ); |
573 | __m256i c2 = _mm256_subs_epu8( px2, rmin ); |
574 | __m256i c3 = _mm256_subs_epu8( px3, rmin ); |
575 | |
576 | __m256i is0 = _mm256_maddubs_epi16( c0, _mm256_set1_epi8( 1 ) ); |
577 | __m256i is1 = _mm256_maddubs_epi16( c1, _mm256_set1_epi8( 1 ) ); |
578 | __m256i is2 = _mm256_maddubs_epi16( c2, _mm256_set1_epi8( 1 ) ); |
579 | __m256i is3 = _mm256_maddubs_epi16( c3, _mm256_set1_epi8( 1 ) ); |
580 | |
581 | __m256i s0 = _mm256_hadd_epi16( is0, is1 ); |
582 | __m256i s1 = _mm256_hadd_epi16( is2, is3 ); |
583 | |
584 | __m256i m0 = _mm256_mulhi_epu16( s0, range ); |
585 | __m256i m1 = _mm256_mulhi_epu16( s1, range ); |
586 | |
587 | __m256i p0 = _mm256_packus_epi16( m0, m1 ); |
588 | |
589 | __m256i p1 = _mm256_or_si256( _mm256_srai_epi32( p0, 6 ), _mm256_srai_epi32( p0, 12 ) ); |
590 | __m256i p2 = _mm256_or_si256( _mm256_srai_epi32( p0, 18 ), p0 ); |
591 | __m256i p3 = _mm256_or_si256( p1, p2 ); |
592 | __m256i p =_mm256_shuffle_epi8( p3, _mm256_set1_epi32( 0x0C080400 ) ); |
593 | |
594 | __m256i mm0 = _mm256_unpacklo_epi8( _mm256_setzero_si256(), min ); |
595 | __m256i mm1 = _mm256_unpacklo_epi8( _mm256_setzero_si256(), max ); |
596 | __m256i mm2 = _mm256_unpacklo_epi64( mm1, mm0 ); |
597 | __m256i mmr = _mm256_slli_epi64( _mm256_srli_epi64( mm2, 11 ), 11 ); |
598 | __m256i mmg = _mm256_slli_epi64( _mm256_srli_epi64( mm2, 26 ), 5 ); |
599 | __m256i mmb = _mm256_srli_epi64( _mm256_slli_epi64( mm2, 16 ), 59 ); |
600 | __m256i mm3 = _mm256_or_si256( mmr, mmg ); |
601 | __m256i mm4 = _mm256_or_si256( mm3, mmb ); |
602 | __m256i mm5 = _mm256_shuffle_epi8( mm4, _mm256_set1_epi32( 0x09080100 ) ); |
603 | |
604 | __m256i d0 = _mm256_unpacklo_epi32( mm5, p ); |
605 | __m256i d1 = _mm256_permute4x64_epi64( d0, _MM_SHUFFLE( 3, 2, 2, 0 ) ); |
606 | __m128i d2 = _mm256_castsi256_si128( d1 ); |
607 | |
608 | __m128i mask = _mm_set_epi64x( 0xFFFF0000 | -solid1, 0xFFFF0000 | -solid0 ); |
609 | __m128i d3 = _mm_and_si128( d2, mask ); |
610 | _mm_storeu_si128( (__m128i*)dst, d3 ); |
611 | |
612 | for( int j=4; j<8; j++ ) dst[j] = (char)DxtcIndexTable[(uint8_t)dst[j]]; |
613 | for( int j=12; j<16; j++ ) dst[j] = (char)DxtcIndexTable[(uint8_t)dst[j]]; |
614 | |
615 | dst += 16; |
616 | } |
617 | #endif |
618 | |
619 | static const uint8_t AlphaIndexTable[8] = { 1, 7, 6, 5, 4, 3, 2, 0 }; |
620 | |
621 | static etcpak_force_inline uint64_t ProcessAlpha( const uint8_t* src ) |
622 | { |
623 | uint8_t solid8 = *src; |
624 | uint16_t solid16 = uint16_t( solid8 ) | ( uint16_t( solid8 ) << 8 ); |
625 | uint32_t solid32 = uint32_t( solid16 ) | ( uint32_t( solid16 ) << 16 ); |
626 | uint64_t solid64 = uint64_t( solid32 ) | ( uint64_t( solid32 ) << 32 ); |
627 | if( memcmp( src, &solid64, 8 ) == 0 && memcmp( src+8, &solid64, 8 ) == 0 ) |
628 | { |
629 | return solid8; |
630 | } |
631 | |
632 | uint8_t min = src[0]; |
633 | uint8_t max = min; |
634 | for( int i=1; i<16; i++ ) |
635 | { |
636 | const auto v = src[i]; |
637 | if( v > max ) max = v; |
638 | else if( v < min ) min = v; |
639 | } |
640 | |
641 | uint32_t range = ( 8 << 13 ) / ( 1 + max - min ); |
642 | uint64_t data = 0; |
643 | for( int i=0; i<16; i++ ) |
644 | { |
645 | uint8_t a = src[i] - min; |
646 | uint64_t idx = AlphaIndexTable[( a * range ) >> 13]; |
647 | data |= idx << (i*3); |
648 | } |
649 | |
650 | return max | ( min << 8 ) | ( data << 16 ); |
651 | } |
652 | |
653 | #ifdef __SSE4_1__ |
654 | static etcpak_force_inline uint64_t ProcessRGB_SSE( __m128i px0, __m128i px1, __m128i px2, __m128i px3 ) |
655 | { |
656 | __m128i smask = _mm_set1_epi32( 0xF8FCF8 ); |
657 | __m128i sd0 = _mm_and_si128( px0, smask ); |
658 | __m128i sd1 = _mm_and_si128( px1, smask ); |
659 | __m128i sd2 = _mm_and_si128( px2, smask ); |
660 | __m128i sd3 = _mm_and_si128( px3, smask ); |
661 | |
662 | __m128i sc = _mm_shuffle_epi32(sd0, _MM_SHUFFLE(0, 0, 0, 0)); |
663 | |
664 | __m128i sc0 = _mm_cmpeq_epi8(sd0, sc); |
665 | __m128i sc1 = _mm_cmpeq_epi8(sd1, sc); |
666 | __m128i sc2 = _mm_cmpeq_epi8(sd2, sc); |
667 | __m128i sc3 = _mm_cmpeq_epi8(sd3, sc); |
668 | |
669 | __m128i sm0 = _mm_and_si128(sc0, sc1); |
670 | __m128i sm1 = _mm_and_si128(sc2, sc3); |
671 | __m128i sm = _mm_and_si128(sm0, sm1); |
672 | |
673 | if( _mm_testc_si128(sm, _mm_set1_epi32(-1)) ) |
674 | { |
675 | return uint64_t( to565( _mm_cvtsi128_si32( px0 ) ) ) << 16; |
676 | } |
677 | |
678 | px0 = _mm_and_si128( px0, _mm_set1_epi32( 0xFFFFFF ) ); |
679 | px1 = _mm_and_si128( px1, _mm_set1_epi32( 0xFFFFFF ) ); |
680 | px2 = _mm_and_si128( px2, _mm_set1_epi32( 0xFFFFFF ) ); |
681 | px3 = _mm_and_si128( px3, _mm_set1_epi32( 0xFFFFFF ) ); |
682 | |
683 | __m128i min0 = _mm_min_epu8( px0, px1 ); |
684 | __m128i min1 = _mm_min_epu8( px2, px3 ); |
685 | __m128i min2 = _mm_min_epu8( min0, min1 ); |
686 | |
687 | __m128i max0 = _mm_max_epu8( px0, px1 ); |
688 | __m128i max1 = _mm_max_epu8( px2, px3 ); |
689 | __m128i max2 = _mm_max_epu8( max0, max1 ); |
690 | |
691 | __m128i min3 = _mm_shuffle_epi32( min2, _MM_SHUFFLE( 2, 3, 0, 1 ) ); |
692 | __m128i max3 = _mm_shuffle_epi32( max2, _MM_SHUFFLE( 2, 3, 0, 1 ) ); |
693 | __m128i min4 = _mm_min_epu8( min2, min3 ); |
694 | __m128i max4 = _mm_max_epu8( max2, max3 ); |
695 | |
696 | __m128i min5 = _mm_shuffle_epi32( min4, _MM_SHUFFLE( 0, 0, 2, 2 ) ); |
697 | __m128i max5 = _mm_shuffle_epi32( max4, _MM_SHUFFLE( 0, 0, 2, 2 ) ); |
698 | __m128i rmin = _mm_min_epu8( min4, min5 ); |
699 | __m128i rmax = _mm_max_epu8( max4, max5 ); |
700 | |
701 | __m128i range1 = _mm_subs_epu8( rmax, rmin ); |
702 | __m128i range2 = _mm_sad_epu8( rmax, rmin ); |
703 | |
704 | uint32_t vrange = _mm_cvtsi128_si32( range2 ) >> 1; |
705 | __m128i range = _mm_set1_epi16( DivTable[vrange] ); |
706 | |
707 | __m128i inset1 = _mm_srli_epi16( range1, 4 ); |
708 | __m128i inset = _mm_and_si128( inset1, _mm_set1_epi8( 0xF ) ); |
709 | __m128i min = _mm_adds_epu8( rmin, inset ); |
710 | __m128i max = _mm_subs_epu8( rmax, inset ); |
711 | |
712 | __m128i c0 = _mm_subs_epu8( px0, rmin ); |
713 | __m128i c1 = _mm_subs_epu8( px1, rmin ); |
714 | __m128i c2 = _mm_subs_epu8( px2, rmin ); |
715 | __m128i c3 = _mm_subs_epu8( px3, rmin ); |
716 | |
717 | __m128i is0 = _mm_maddubs_epi16( c0, _mm_set1_epi8( 1 ) ); |
718 | __m128i is1 = _mm_maddubs_epi16( c1, _mm_set1_epi8( 1 ) ); |
719 | __m128i is2 = _mm_maddubs_epi16( c2, _mm_set1_epi8( 1 ) ); |
720 | __m128i is3 = _mm_maddubs_epi16( c3, _mm_set1_epi8( 1 ) ); |
721 | |
722 | __m128i s0 = _mm_hadd_epi16( is0, is1 ); |
723 | __m128i s1 = _mm_hadd_epi16( is2, is3 ); |
724 | |
725 | __m128i m0 = _mm_mulhi_epu16( s0, range ); |
726 | __m128i m1 = _mm_mulhi_epu16( s1, range ); |
727 | |
728 | __m128i p0 = _mm_packus_epi16( m0, m1 ); |
729 | |
730 | __m128i p1 = _mm_or_si128( _mm_srai_epi32( p0, 6 ), _mm_srai_epi32( p0, 12 ) ); |
731 | __m128i p2 = _mm_or_si128( _mm_srai_epi32( p0, 18 ), p0 ); |
732 | __m128i p3 = _mm_or_si128( p1, p2 ); |
733 | __m128i p =_mm_shuffle_epi8( p3, _mm_set1_epi32( 0x0C080400 ) ); |
734 | |
735 | uint32_t vmin = _mm_cvtsi128_si32( min ); |
736 | uint32_t vmax = _mm_cvtsi128_si32( max ); |
737 | uint32_t vp = _mm_cvtsi128_si32( p ); |
738 | |
739 | return uint64_t( ( uint64_t( to565( vmin ) ) << 16 ) | to565( vmax ) | ( uint64_t( vp ) << 32 ) ); |
740 | } |
741 | |
742 | static etcpak_force_inline uint64_t ProcessAlpha_SSE( __m128i px0, __m128i px1, __m128i px2, __m128i px3 ) |
743 | { |
744 | __m128i mask = _mm_setr_epi32( 0x0f0b0703, -1, -1, -1 ); |
745 | |
746 | __m128i m0 = _mm_shuffle_epi8( px0, mask ); |
747 | __m128i m1 = _mm_shuffle_epi8( px1, _mm_shuffle_epi32( mask, _MM_SHUFFLE( 3, 3, 0, 3 ) ) ); |
748 | __m128i m2 = _mm_shuffle_epi8( px2, _mm_shuffle_epi32( mask, _MM_SHUFFLE( 3, 0, 3, 3 ) ) ); |
749 | __m128i m3 = _mm_shuffle_epi8( px3, _mm_shuffle_epi32( mask, _MM_SHUFFLE( 0, 3, 3, 3 ) ) ); |
750 | __m128i m4 = _mm_or_si128( m0, m1 ); |
751 | __m128i m5 = _mm_or_si128( m2, m3 ); |
752 | __m128i a = _mm_or_si128( m4, m5 ); |
753 | |
754 | __m128i solidCmp = _mm_shuffle_epi8( a, _mm_setzero_si128() ); |
755 | __m128i cmpRes = _mm_cmpeq_epi8( a, solidCmp ); |
756 | if( _mm_testc_si128( cmpRes, _mm_set1_epi32( -1 ) ) ) |
757 | { |
758 | return _mm_cvtsi128_si32( a ) & 0xFF; |
759 | } |
760 | |
761 | __m128i a1 = _mm_shuffle_epi32( a, _MM_SHUFFLE( 2, 3, 0, 1 ) ); |
762 | __m128i max1 = _mm_max_epu8( a, a1 ); |
763 | __m128i min1 = _mm_min_epu8( a, a1 ); |
764 | __m128i amax2 = _mm_shuffle_epi32( max1, _MM_SHUFFLE( 0, 0, 2, 2 ) ); |
765 | __m128i amin2 = _mm_shuffle_epi32( min1, _MM_SHUFFLE( 0, 0, 2, 2 ) ); |
766 | __m128i max2 = _mm_max_epu8( max1, amax2 ); |
767 | __m128i min2 = _mm_min_epu8( min1, amin2 ); |
768 | __m128i amax3 = _mm_alignr_epi8( max2, max2, 2 ); |
769 | __m128i amin3 = _mm_alignr_epi8( min2, min2, 2 ); |
770 | __m128i max3 = _mm_max_epu8( max2, amax3 ); |
771 | __m128i min3 = _mm_min_epu8( min2, amin3 ); |
772 | __m128i amax4 = _mm_alignr_epi8( max3, max3, 1 ); |
773 | __m128i amin4 = _mm_alignr_epi8( min3, min3, 1 ); |
774 | __m128i max = _mm_max_epu8( max3, amax4 ); |
775 | __m128i min = _mm_min_epu8( min3, amin4 ); |
776 | __m128i minmax = _mm_unpacklo_epi8( max, min ); |
777 | |
778 | __m128i r = _mm_sub_epi8( max, min ); |
779 | int range = _mm_cvtsi128_si32( r ) & 0xFF; |
780 | __m128i rv = _mm_set1_epi16( DivTableAlpha[range] ); |
781 | |
782 | __m128i v = _mm_sub_epi8( a, min ); |
783 | |
784 | __m128i lo16 = _mm_unpacklo_epi8( v, _mm_setzero_si128() ); |
785 | __m128i hi16 = _mm_unpackhi_epi8( v, _mm_setzero_si128() ); |
786 | |
787 | __m128i lomul = _mm_mulhi_epu16( lo16, rv ); |
788 | __m128i himul = _mm_mulhi_epu16( hi16, rv ); |
789 | |
790 | __m128i p0 = _mm_packus_epi16( lomul, himul ); |
791 | __m128i p1 = _mm_or_si128( _mm_and_si128( p0, _mm_set1_epi16( 0x3F ) ), _mm_srai_epi16( _mm_and_si128( p0, _mm_set1_epi16( 0x3F00 ) ), 5 ) ); |
792 | __m128i p2 = _mm_packus_epi16( p1, p1 ); |
793 | |
794 | uint64_t pi = _mm_cvtsi128_si64( p2 ); |
795 | uint64_t data = 0; |
796 | for( int i=0; i<8; i++ ) |
797 | { |
798 | uint64_t idx = AlphaIndexTable_SSE[(pi>>(i*8)) & 0x3F]; |
799 | data |= idx << (i*6); |
800 | } |
801 | return (uint64_t)(uint16_t)_mm_cvtsi128_si32( minmax ) | ( data << 16 ); |
802 | } |
803 | #endif |
804 | |
805 | void CompressDxt1( const uint32_t* src, uint64_t* dst, uint32_t blocks, size_t width ) |
806 | { |
807 | #ifdef __AVX2__ |
808 | if( width%8 == 0 ) |
809 | { |
810 | blocks /= 2; |
811 | uint32_t buf[8*4]; |
812 | int i = 0; |
813 | char* dst8 = (char*)dst; |
814 | |
815 | do |
816 | { |
817 | auto tmp = (char*)buf; |
818 | memcpy( tmp, src + width * 0, 8*4 ); |
819 | memcpy( tmp + 8*4, src + width * 1, 8*4 ); |
820 | memcpy( tmp + 16*4, src + width * 2, 8*4 ); |
821 | memcpy( tmp + 24*4, src + width * 3, 8*4 ); |
822 | src += 8; |
823 | if( ++i == width/8 ) |
824 | { |
825 | src += width * 3; |
826 | i = 0; |
827 | } |
828 | |
829 | ProcessRGB_AVX( (uint8_t*)buf, dst8 ); |
830 | } |
831 | while( --blocks ); |
832 | } |
833 | else |
834 | #endif |
835 | { |
836 | uint32_t buf[4*4]; |
837 | int i = 0; |
838 | |
839 | auto ptr = dst; |
840 | do |
841 | { |
842 | auto tmp = (char*)buf; |
843 | memcpy( tmp, src + width * 0, 4*4 ); |
844 | memcpy( tmp + 4*4, src + width * 1, 4*4 ); |
845 | memcpy( tmp + 8*4, src + width * 2, 4*4 ); |
846 | memcpy( tmp + 12*4, src + width * 3, 4*4 ); |
847 | src += 4; |
848 | if( ++i == width/4 ) |
849 | { |
850 | src += width * 3; |
851 | i = 0; |
852 | } |
853 | |
854 | const auto c = ProcessRGB( (uint8_t*)buf ); |
855 | uint8_t fix[8]; |
856 | memcpy( fix, &c, 8 ); |
857 | for( int j=4; j<8; j++ ) fix[j] = DxtcIndexTable[fix[j]]; |
858 | memcpy( ptr, fix, sizeof( uint64_t ) ); |
859 | ptr++; |
860 | } |
861 | while( --blocks ); |
862 | } |
863 | } |
864 | |
865 | void CompressDxt1Dither( const uint32_t* src, uint64_t* dst, uint32_t blocks, size_t width ) |
866 | { |
867 | uint32_t buf[4*4]; |
868 | int i = 0; |
869 | |
870 | auto ptr = dst; |
871 | do |
872 | { |
873 | auto tmp = (char*)buf; |
874 | memcpy( tmp, src + width * 0, 4*4 ); |
875 | memcpy( tmp + 4*4, src + width * 1, 4*4 ); |
876 | memcpy( tmp + 8*4, src + width * 2, 4*4 ); |
877 | memcpy( tmp + 12*4, src + width * 3, 4*4 ); |
878 | src += 4; |
879 | if( ++i == width/4 ) |
880 | { |
881 | src += width * 3; |
882 | i = 0; |
883 | } |
884 | |
885 | Dither( (uint8_t*)buf ); |
886 | |
887 | const auto c = ProcessRGB( (uint8_t*)buf ); |
888 | uint8_t fix[8]; |
889 | memcpy( fix, &c, 8 ); |
890 | for( int j=4; j<8; j++ ) fix[j] = DxtcIndexTable[fix[j]]; |
891 | memcpy( ptr, fix, sizeof( uint64_t ) ); |
892 | ptr++; |
893 | } |
894 | while( --blocks ); |
895 | } |
896 | |
897 | void CompressDxt5( const uint32_t* src, uint64_t* dst, uint32_t blocks, size_t width ) |
898 | { |
899 | int i = 0; |
900 | auto ptr = dst; |
901 | do |
902 | { |
903 | #ifdef __SSE4_1__ |
904 | __m128i px0 = _mm_loadu_si128( (__m128i*)( src + width * 0 ) ); |
905 | __m128i px1 = _mm_loadu_si128( (__m128i*)( src + width * 1 ) ); |
906 | __m128i px2 = _mm_loadu_si128( (__m128i*)( src + width * 2 ) ); |
907 | __m128i px3 = _mm_loadu_si128( (__m128i*)( src + width * 3 ) ); |
908 | |
909 | src += 4; |
910 | if( ++i == width/4 ) |
911 | { |
912 | src += width * 3; |
913 | i = 0; |
914 | } |
915 | |
916 | *ptr++ = ProcessAlpha_SSE( px0, px1, px2, px3 ); |
917 | |
918 | const auto c = ProcessRGB_SSE( px0, px1, px2, px3 ); |
919 | uint8_t fix[8]; |
920 | memcpy( fix, &c, 8 ); |
921 | for( int j=4; j<8; j++ ) fix[j] = DxtcIndexTable[fix[j]]; |
922 | memcpy( ptr, fix, sizeof( uint64_t ) ); |
923 | ptr++; |
924 | #else |
925 | uint32_t rgba[4*4]; |
926 | uint8_t alpha[4*4]; |
927 | |
928 | auto tmp = (char*)rgba; |
929 | memcpy( tmp, src + width * 0, 4*4 ); |
930 | memcpy( tmp + 4*4, src + width * 1, 4*4 ); |
931 | memcpy( tmp + 8*4, src + width * 2, 4*4 ); |
932 | memcpy( tmp + 12*4, src + width * 3, 4*4 ); |
933 | src += 4; |
934 | if( ++i == width/4 ) |
935 | { |
936 | src += width * 3; |
937 | i = 0; |
938 | } |
939 | |
940 | for( int i=0; i<16; i++ ) |
941 | { |
942 | alpha[i] = rgba[i] >> 24; |
943 | rgba[i] &= 0xFFFFFF; |
944 | } |
945 | *ptr++ = ProcessAlpha( alpha ); |
946 | |
947 | const auto c = ProcessRGB( (uint8_t*)rgba ); |
948 | uint8_t fix[8]; |
949 | memcpy( fix, &c, 8 ); |
950 | for( int j=4; j<8; j++ ) fix[j] = DxtcIndexTable[fix[j]]; |
951 | memcpy( ptr, fix, sizeof( uint64_t ) ); |
952 | ptr++; |
953 | #endif |
954 | } |
955 | while( --blocks ); |
956 | } |
957 | |