1// basisu_transcoder_uastc.h
2#pragma once
3#include "basisu_transcoder_internal.h"
4
5namespace basist
6{
7 struct color_quad_u8
8 {
9 uint8_t m_c[4];
10 };
11
12 const uint32_t TOTAL_UASTC_MODES = 19;
13 const uint32_t UASTC_MODE_INDEX_SOLID_COLOR = 8;
14
15 const uint32_t TOTAL_ASTC_BC7_COMMON_PARTITIONS2 = 30;
16 const uint32_t TOTAL_ASTC_BC7_COMMON_PARTITIONS3 = 11;
17 const uint32_t TOTAL_BC7_3_ASTC2_COMMON_PARTITIONS = 19;
18
19 extern const uint8_t g_uastc_mode_weight_bits[TOTAL_UASTC_MODES];
20 extern const uint8_t g_uastc_mode_weight_ranges[TOTAL_UASTC_MODES];
21 extern const uint8_t g_uastc_mode_endpoint_ranges[TOTAL_UASTC_MODES];
22 extern const uint8_t g_uastc_mode_subsets[TOTAL_UASTC_MODES];
23 extern const uint8_t g_uastc_mode_planes[TOTAL_UASTC_MODES];
24 extern const uint8_t g_uastc_mode_comps[TOTAL_UASTC_MODES];
25 extern const uint8_t g_uastc_mode_has_etc1_bias[TOTAL_UASTC_MODES];
26 extern const uint8_t g_uastc_mode_has_bc1_hint0[TOTAL_UASTC_MODES];
27 extern const uint8_t g_uastc_mode_has_bc1_hint1[TOTAL_UASTC_MODES];
28 extern const uint8_t g_uastc_mode_has_alpha[TOTAL_UASTC_MODES];
29 extern const uint8_t g_uastc_mode_is_la[TOTAL_UASTC_MODES];
30
31 struct astc_bc7_common_partition2_desc
32 {
33 uint8_t m_bc7;
34 uint16_t m_astc;
35 bool m_invert;
36 };
37
38 extern const astc_bc7_common_partition2_desc g_astc_bc7_common_partitions2[TOTAL_ASTC_BC7_COMMON_PARTITIONS2];
39
40 struct bc73_astc2_common_partition_desc
41 {
42 uint8_t m_bc73;
43 uint16_t m_astc2;
44 uint8_t k; // 0-5 - how to modify the BC7 3-subset pattern to match the ASTC pattern (LSB=invert)
45 };
46
47 extern const bc73_astc2_common_partition_desc g_bc7_3_astc2_common_partitions[TOTAL_BC7_3_ASTC2_COMMON_PARTITIONS];
48
49 struct astc_bc7_common_partition3_desc
50 {
51 uint8_t m_bc7;
52 uint16_t m_astc;
53 uint8_t m_astc_to_bc7_perm; // converts ASTC to BC7 partition using g_astc_bc7_partition_index_perm_tables[][]
54 };
55
56 extern const astc_bc7_common_partition3_desc g_astc_bc7_common_partitions3[TOTAL_ASTC_BC7_COMMON_PARTITIONS3];
57
58 extern const uint8_t g_astc_bc7_patterns2[TOTAL_ASTC_BC7_COMMON_PARTITIONS2][16];
59 extern const uint8_t g_astc_bc7_patterns3[TOTAL_ASTC_BC7_COMMON_PARTITIONS3][16];
60 extern const uint8_t g_bc7_3_astc2_patterns2[TOTAL_BC7_3_ASTC2_COMMON_PARTITIONS][16];
61
62 extern const uint8_t g_astc_bc7_pattern2_anchors[TOTAL_ASTC_BC7_COMMON_PARTITIONS2][3];
63 extern const uint8_t g_astc_bc7_pattern3_anchors[TOTAL_ASTC_BC7_COMMON_PARTITIONS3][3];
64 extern const uint8_t g_bc7_3_astc2_patterns2_anchors[TOTAL_BC7_3_ASTC2_COMMON_PARTITIONS][3];
65
66 extern const uint32_t g_uastc_mode_huff_codes[TOTAL_UASTC_MODES + 1][2];
67
68 extern const uint8_t g_astc_to_bc7_partition_index_perm_tables[6][3];
69 extern const uint8_t g_bc7_to_astc_partition_index_perm_tables[6][3]; // inverse of g_astc_to_bc7_partition_index_perm_tables
70
71 extern const uint8_t* s_uastc_to_bc1_weights[6];
72
73 uint32_t bc7_convert_partition_index_3_to_2(uint32_t p, uint32_t k);
74
75 inline uint32_t astc_interpolate(uint32_t l, uint32_t h, uint32_t w, bool srgb)
76 {
77 if (srgb)
78 {
79 l = (l << 8) | 0x80;
80 h = (h << 8) | 0x80;
81 }
82 else
83 {
84 l = (l << 8) | l;
85 h = (h << 8) | h;
86 }
87
88 uint32_t k = (l * (64 - w) + h * w + 32) >> 6;
89
90 return k >> 8;
91 }
92
93 struct astc_block_desc
94 {
95 int m_weight_range; // weight BISE range
96
97 int m_subsets; // number of ASTC partitions
98 int m_partition_seed; // partition pattern seed
99 int m_cem; // color endpoint mode used by all subsets
100
101 int m_ccs; // color component selector (dual plane only)
102 bool m_dual_plane; // true if dual plane
103
104 // Weight and endpoint BISE values.
105 // Note these values are NOT linear, they must be BISE encoded. See Table 97 and Table 107.
106 uint8_t m_endpoints[18]; // endpoint values, in RR GG BB etc. order
107 uint8_t m_weights[64]; // weight index values, raster order, in P0 P1, P0 P1, etc. or P0, P0, P0, P0, etc. order
108 };
109
110 const uint32_t BC7ENC_TOTAL_ASTC_RANGES = 21;
111
112 // See tables 81, 93, 18.13.Endpoint Unquantization
113 const uint32_t TOTAL_ASTC_RANGES = 21;
114 extern const int g_astc_bise_range_table[TOTAL_ASTC_RANGES][3];
115
116 struct astc_quant_bin
117 {
118 uint8_t m_unquant; // unquantized value
119 uint8_t m_index; // sorted index
120 };
121
122 extern astc_quant_bin g_astc_unquant[BC7ENC_TOTAL_ASTC_RANGES][256]; // [ASTC encoded endpoint index]
123
124 int astc_get_levels(int range);
125 bool astc_is_valid_endpoint_range(uint32_t range);
126 uint32_t unquant_astc_endpoint(uint32_t packed_bits, uint32_t packed_trits, uint32_t packed_quints, uint32_t range);
127 uint32_t unquant_astc_endpoint_val(uint32_t packed_val, uint32_t range);
128
129 const uint8_t* get_anchor_indices(uint32_t subsets, uint32_t mode, uint32_t common_pattern, const uint8_t*& pPartition_pattern);
130
131 // BC7
132 const uint32_t BC7ENC_BLOCK_SIZE = 16;
133
134 struct bc7_block
135 {
136 uint64_t m_qwords[2];
137 };
138
139 struct bc7_optimization_results
140 {
141 uint32_t m_mode;
142 uint32_t m_partition;
143 uint8_t m_selectors[16];
144 uint8_t m_alpha_selectors[16];
145 color_quad_u8 m_low[3];
146 color_quad_u8 m_high[3];
147 uint32_t m_pbits[3][2];
148 uint32_t m_index_selector;
149 uint32_t m_rotation;
150 };
151
152 extern const uint32_t g_bc7_weights1[2];
153 extern const uint32_t g_bc7_weights2[4];
154 extern const uint32_t g_bc7_weights3[8];
155 extern const uint32_t g_bc7_weights4[16];
156 extern const uint32_t g_astc_weights4[16];
157 extern const uint32_t g_astc_weights5[32];
158 extern const uint32_t g_astc_weights_3levels[3];
159 extern const uint8_t g_bc7_partition1[16];
160 extern const uint8_t g_bc7_partition2[64 * 16];
161 extern const uint8_t g_bc7_partition3[64 * 16];
162 extern const uint8_t g_bc7_table_anchor_index_second_subset[64];
163 extern const uint8_t g_bc7_table_anchor_index_third_subset_1[64];
164 extern const uint8_t g_bc7_table_anchor_index_third_subset_2[64];
165 extern const uint8_t g_bc7_num_subsets[8];
166 extern const uint8_t g_bc7_partition_bits[8];
167 extern const uint8_t g_bc7_color_index_bitcount[8];
168 extern const uint8_t g_bc7_mode_has_p_bits[8];
169 extern const uint8_t g_bc7_mode_has_shared_p_bits[8];
170 extern const uint8_t g_bc7_color_precision_table[8];
171 extern const int8_t g_bc7_alpha_precision_table[8];
172 extern const uint8_t g_bc7_alpha_index_bitcount[8];
173
174 inline bool get_bc7_mode_has_seperate_alpha_selectors(int mode) { return (mode == 4) || (mode == 5); }
175 inline int get_bc7_color_index_size(int mode, int index_selection_bit) { return g_bc7_color_index_bitcount[mode] + index_selection_bit; }
176 inline int get_bc7_alpha_index_size(int mode, int index_selection_bit) { return g_bc7_alpha_index_bitcount[mode] - index_selection_bit; }
177
178 struct endpoint_err
179 {
180 uint16_t m_error; uint8_t m_lo; uint8_t m_hi;
181 };
182
183 extern endpoint_err g_bc7_mode_6_optimal_endpoints[256][2]; // [c][pbit]
184 const uint32_t BC7ENC_MODE_6_OPTIMAL_INDEX = 5;
185
186 extern endpoint_err g_bc7_mode_5_optimal_endpoints[256]; // [c]
187 const uint32_t BC7ENC_MODE_5_OPTIMAL_INDEX = 1;
188
189 // Packs a BC7 block from a high-level description. Handles all BC7 modes.
190 void encode_bc7_block(void* pBlock, const bc7_optimization_results* pResults);
191
192 // Packs an ASTC block
193 // Constraints: Always 4x4, all subset CEM's must be equal, only tested with LDR CEM's.
194 bool pack_astc_block(uint32_t* pDst, const astc_block_desc* pBlock, uint32_t mode);
195
196 void pack_astc_solid_block(void* pDst_block, const color32& color);
197
198#ifdef _DEBUG
199 int astc_compute_texel_partition(int seed, int x, int y, int z, int partitioncount, bool small_block);
200#endif
201
202 struct uastc_block
203 {
204 union
205 {
206 uint8_t m_bytes[16];
207 uint32_t m_dwords[4];
208 };
209 };
210
211 struct unpacked_uastc_block
212 {
213 astc_block_desc m_astc;
214
215 uint32_t m_mode;
216 uint32_t m_common_pattern;
217
218 color32 m_solid_color;
219
220 bool m_bc1_hint0;
221 bool m_bc1_hint1;
222
223 bool m_etc1_flip;
224 bool m_etc1_diff;
225 uint32_t m_etc1_inten0;
226 uint32_t m_etc1_inten1;
227
228 uint32_t m_etc1_bias;
229
230 uint32_t m_etc2_hints;
231
232 uint32_t m_etc1_selector;
233 uint32_t m_etc1_r, m_etc1_g, m_etc1_b;
234 };
235
236 color32 apply_etc1_bias(const color32 &block_color, uint32_t bias, uint32_t limit, uint32_t subblock);
237
238 struct decoder_etc_block;
239 struct eac_block;
240
241 bool unpack_uastc(uint32_t mode, uint32_t common_pattern, const color32& solid_color, const astc_block_desc& astc, color32* pPixels, bool srgb);
242 bool unpack_uastc(const unpacked_uastc_block& unpacked_blk, color32* pPixels, bool srgb);
243
244 bool unpack_uastc(const uastc_block& blk, color32* pPixels, bool srgb);
245 bool unpack_uastc(const uastc_block& blk, unpacked_uastc_block& unpacked, bool undo_blue_contract, bool read_hints = true);
246
247 bool transcode_uastc_to_astc(const uastc_block& src_blk, void* pDst);
248
249 bool transcode_uastc_to_bc7(const unpacked_uastc_block& unpacked_src_blk, bc7_optimization_results& dst_blk);
250 bool transcode_uastc_to_bc7(const uastc_block& src_blk, bc7_optimization_results& dst_blk);
251 bool transcode_uastc_to_bc7(const uastc_block& src_blk, void* pDst);
252
253 void transcode_uastc_to_etc1(unpacked_uastc_block& unpacked_src_blk, color32 block_pixels[4][4], void* pDst);
254 bool transcode_uastc_to_etc1(const uastc_block& src_blk, void* pDst);
255 bool transcode_uastc_to_etc1(const uastc_block& src_blk, void* pDst, uint32_t channel);
256
257 void transcode_uastc_to_etc2_eac_a8(unpacked_uastc_block& unpacked_src_blk, color32 block_pixels[4][4], void* pDst);
258 bool transcode_uastc_to_etc2_rgba(const uastc_block& src_blk, void* pDst);
259
260 // Packs 16 scalar values to BC4. Same PSNR as stb_dxt's BC4 encoder, around 13% faster.
261 void encode_bc4(void* pDst, const uint8_t* pPixels, uint32_t stride);
262
263 void encode_bc1_solid_block(void* pDst, uint32_t fr, uint32_t fg, uint32_t fb);
264
265 enum
266 {
267 cEncodeBC1HighQuality = 1,
268 cEncodeBC1HigherQuality = 2,
269 cEncodeBC1UseSelectors = 4,
270 };
271 void encode_bc1(void* pDst, const uint8_t* pPixels, uint32_t flags);
272
273 // Alternate PCA-free encoder, around 15% faster, same (or slightly higher) avg. PSNR
274 void encode_bc1_alt(void* pDst, const uint8_t* pPixels, uint32_t flags);
275
276 void transcode_uastc_to_bc1_hint0(const unpacked_uastc_block& unpacked_src_blk, void* pDst);
277 void transcode_uastc_to_bc1_hint1(const unpacked_uastc_block& unpacked_src_blk, const color32 block_pixels[4][4], void* pDst, bool high_quality);
278
279 bool transcode_uastc_to_bc1(const uastc_block& src_blk, void* pDst, bool high_quality);
280 bool transcode_uastc_to_bc3(const uastc_block& src_blk, void* pDst, bool high_quality);
281 bool transcode_uastc_to_bc4(const uastc_block& src_blk, void* pDst, bool high_quality, uint32_t chan0);
282 bool transcode_uastc_to_bc5(const uastc_block& src_blk, void* pDst, bool high_quality, uint32_t chan0, uint32_t chan1);
283
284 bool transcode_uastc_to_etc2_eac_r11(const uastc_block& src_blk, void* pDst, bool high_quality, uint32_t chan0);
285 bool transcode_uastc_to_etc2_eac_rg11(const uastc_block& src_blk, void* pDst, bool high_quality, uint32_t chan0, uint32_t chan1);
286
287 bool transcode_uastc_to_pvrtc1_4_rgb(const uastc_block* pSrc_blocks, void* pDst_blocks, uint32_t num_blocks_x, uint32_t num_blocks_y, bool high_quality, bool from_alpha);
288 bool transcode_uastc_to_pvrtc1_4_rgba(const uastc_block* pSrc_blocks, void* pDst_blocks, uint32_t num_blocks_x, uint32_t num_blocks_y, bool high_quality);
289
290 // uastc_init() MUST be called before using this module.
291 void uastc_init();
292
293} // namespace basist
294