Chromium Code Reviews
chromiumcodereview-hr@appspot.gserviceaccount.com (chromiumcodereview-hr) | Please choose your nickname with Settings | Help | Chromium Project | Gerrit Changes | Sign out
(120)

Side by Side Diff: src/utils/SkTextureCompressor_ASTC.cpp

Issue 417123003: Some small ASTC optimizations (Closed) Base URL: https://skia.googlesource.com/skia.git@master
Patch Set: Created 6 years, 5 months ago
Use n/p to move between diff chunks; N/P to move between comments. Draft comments are only viewable by you.
Jump to:
View unified diff | Download patch
« no previous file with comments | « no previous file | no next file » | no next file with comments »
Toggle Intra-line Diffs ('i') | Expand Comments ('e') | Collapse Comments ('c') | Show Comments Hide Comments ('s')
OLDNEW
1 /* 1 /*
2 * Copyright 2014 Google Inc. 2 * Copyright 2014 Google Inc.
3 * 3 *
4 * Use of this source code is governed by a BSD-style license that can be 4 * Use of this source code is governed by a BSD-style license that can be
5 * found in the LICENSE file. 5 * found in the LICENSE file.
6 */ 6 */
7 7
8 #include "SkTextureCompressor_ASTC.h" 8 #include "SkTextureCompressor_ASTC.h"
9 9
10 #include "SkBlitter.h" 10 #include "SkBlitter.h"
(...skipping 19 matching lines...) Expand all
30 // if weight >= 0: 30 // if weight >= 0:
31 // total-weight += weight; 31 // total-weight += weight;
32 // total-alpha += weight * alphas[texel-x][texel-y]; 32 // total-alpha += weight * alphas[texel-x][texel-y];
33 // 33 //
34 // total-alpha /= total-weight; 34 // total-alpha /= total-weight;
35 // index = top three bits of total-alpha 35 // index = top three bits of total-alpha
36 // 36 //
37 // If the associated index does not contribute to 20 different texels (e.g. it's in 37 // If the associated index does not contribute to 20 different texels (e.g. it's in
38 // a corner), then the extra texels are stored with -1's in the table. 38 // a corner), then the extra texels are stored with -1's in the table.
39 39
40 static const int k6x5To12x12Table[30][60] = { 40 static const int8_t k6x5To12x12Table[30][60] = {
41 { 16, 0, 0, 9, 1, 0, 1, 2, 0, 10, 0, 1, 6, 1, 1, 1, 2, 1, 4, 0, 2, 2, 41 { 16, 0, 0, 9, 1, 0, 1, 2, 0, 10, 0, 1, 6, 1, 1, 1, 2, 1, 4, 0, 2, 2,
42 1, 2, -1, 0, 0, -1, 0, 0, -1, 0, 0, -1, 0, 0, -1, 0, 0, -1, 0, 0, -1, 0, 42 1, 2, -1, 0, 0, -1, 0, 0, -1, 0, 0, -1, 0, 0, -1, 0, 0, -1, 0, 0, -1, 0,
43 0, -1, 0, 0, -1, 0, 0, -1, 0, 0, -1, 0, 0, -1, 0, 0}, // n = 20 43 0, -1, 0, 0, -1, 0, 0, -1, 0, 0, -1, 0, 0, -1, 0, 0}, // n = 20
44 { 7, 1, 0, 15, 2, 0, 10, 3, 0, 3, 4, 0, 4, 1, 1, 9, 2, 1, 6, 3, 1, 2, 44 { 7, 1, 0, 15, 2, 0, 10, 3, 0, 3, 4, 0, 4, 1, 1, 9, 2, 1, 6, 3, 1, 2,
45 4, 1, 2, 1, 2, 4, 2, 2, 3, 3, 2, 1, 4, 2, -1, 0, 0, -1, 0, 0, -1, 0, 45 4, 1, 2, 1, 2, 4, 2, 2, 3, 3, 2, 1, 4, 2, -1, 0, 0, -1, 0, 0, -1, 0,
46 0, -1, 0, 0, -1, 0, 0, -1, 0, 0, -1, 0, 0, -1, 0, 0}, // n = 20 46 0, -1, 0, 0, -1, 0, 0, -1, 0, 0, -1, 0, 0, -1, 0, 0}, // n = 20
47 { 6, 3, 0, 13, 4, 0, 12, 5, 0, 4, 6, 0, 4, 3, 1, 8, 4, 1, 8, 5, 1, 3, 47 { 6, 3, 0, 13, 4, 0, 12, 5, 0, 4, 6, 0, 4, 3, 1, 8, 4, 1, 8, 5, 1, 3,
48 6, 1, 1, 3, 2, 3, 4, 2, 3, 5, 2, 1, 6, 2, -1, 0, 0, -1, 0, 0, -1, 0, 48 6, 1, 1, 3, 2, 3, 4, 2, 3, 5, 2, 1, 6, 2, -1, 0, 0, -1, 0, 0, -1, 0,
49 0, -1, 0, 0, -1, 0, 0, -1, 0, 0, -1, 0, 0, -1, 0, 0}, // n = 20 49 0, -1, 0, 0, -1, 0, 0, -1, 0, 0, -1, 0, 0, -1, 0, 0}, // n = 20
50 { 4, 5, 0, 12, 6, 0, 13, 7, 0, 6, 8, 0, 2, 5, 1, 7, 6, 1, 8, 7, 1, 4, 50 { 4, 5, 0, 12, 6, 0, 13, 7, 0, 6, 8, 0, 2, 5, 1, 7, 6, 1, 8, 7, 1, 4,
(...skipping 74 matching lines...) Expand 10 before | Expand all | Expand 10 after
125 { 1, 7, 9, 3, 8, 9, 4, 9, 9, 2, 10, 9, 2, 7, 10, 6, 8, 10, 9, 9, 10, 4, 125 { 1, 7, 9, 3, 8, 9, 4, 9, 9, 2, 10, 9, 2, 7, 10, 6, 8, 10, 9, 9, 10, 4,
126 10, 10, 3, 7, 11, 10, 8, 11, 15, 9, 11, 7, 10, 11, -1, 0, 0, -1, 0, 0, -1, 0, 126 10, 10, 3, 7, 11, 10, 8, 11, 15, 9, 11, 7, 10, 11, -1, 0, 0, -1, 0, 0, -1, 0,
127 0, -1, 0, 0, -1, 0, 0, -1, 0, 0, -1, 0, 0, -1, 0, 0}, // n = 20 127 0, -1, 0, 0, -1, 0, 0, -1, 0, 0, -1, 0, 0, -1, 0, 0}, // n = 20
128 { 2, 10, 9, 4, 11, 9, 1, 9, 10, 6, 10, 10, 10, 11, 10, 1, 9, 11, 9, 10, 11, 16, 128 { 2, 10, 9, 4, 11, 9, 1, 9, 10, 6, 10, 10, 10, 11, 10, 1, 9, 11, 9, 10, 11, 16,
129 11, 11, -1, 0, 0, -1, 0, 0, -1, 0, 0, -1, 0, 0, -1, 0, 0, -1, 0, 0, -1, 0, 129 11, 11, -1, 0, 0, -1, 0, 0, -1, 0, 0, -1, 0, 0, -1, 0, 0, -1, 0, 0, -1, 0,
130 0, -1, 0, 0, -1, 0, 0, -1, 0, 0, -1, 0, 0, -1, 0, 0} // n = 20 130 0, -1, 0, 0, -1, 0, 0, -1, 0, 0, -1, 0, 0, -1, 0, 0} // n = 20
131 }; 131 };
132 132
133 // Returns the alpha value of a texel at position (x, y) from src. 133 // Returns the alpha value of a texel at position (x, y) from src.
134 // (x, y) are assumed to be in the range [0, 12). 134 // (x, y) are assumed to be in the range [0, 12).
135 uint8_t get_alpha(const uint8_t *src, int rowBytes, int x, int y) { 135 static inline uint8_t get_alpha(const uint8_t *src, int rowBytes, int x, int y) {
136 SkASSERT(x >= 0 && x < 12); 136 SkASSERT(x >= 0 && x < 12);
137 SkASSERT(y >= 0 && y < 12); 137 SkASSERT(y >= 0 && y < 12);
138 SkASSERT(rowBytes >= 12); 138 SkASSERT(rowBytes >= 12);
139 return *(src + y*rowBytes + x); 139 return *(src + y*rowBytes + x);
140 } 140 }
141 141
142 // Output the 16 bytes stored in top and bottom and advance the pointer. The byt es 142 // Output the 16 bytes stored in top and bottom and advance the pointer. The byt es
143 // are stored as the integers are represented in memory, so they should be swapp ed 143 // are stored as the integers are represented in memory, so they should be swapp ed
144 // if necessary. 144 // if necessary.
145 static inline void send_packing(uint8_t** dst, const uint64_t top, const uint64_ t bottom) { 145 static inline void send_packing(uint8_t** dst, const uint64_t top, const uint64_ t bottom) {
(...skipping 27 matching lines...) Expand all
173 // v0 = 255, v1 = 0, so everything will be opaque. 173 // v0 = 255, v1 = 0, so everything will be opaque.
174 send_packing(dst, SkTEndian_SwapLE64(0x000000000001FE0173ULL), 0); 174 send_packing(dst, SkTEndian_SwapLE64(0x000000000001FE0173ULL), 0);
175 return; 175 return;
176 } 176 }
177 } 177 }
178 178
179 uint8_t indices[30]; // 6x5 index grid 179 uint8_t indices[30]; // 6x5 index grid
180 for (int idx = 0; idx < 30; ++idx) { 180 for (int idx = 0; idx < 30; ++idx) {
181 int weightTot = 0; 181 int weightTot = 0;
182 int alphaTot = 0; 182 int alphaTot = 0;
183
184 for (int w = 0; w < 20; ++w) { 183 for (int w = 0; w < 20; ++w) {
185 const int weight = k6x5To12x12Table[idx][w*3]; 184 const int8_t weight = k6x5To12x12Table[idx][w*3];
186 if (weight > 0) { 185 if (weight > 0) {
187 const int x = k6x5To12x12Table[idx][w*3 + 1]; 186 const int x = k6x5To12x12Table[idx][w*3 + 1];
188 const int y = k6x5To12x12Table[idx][w*3 + 2]; 187 const int y = k6x5To12x12Table[idx][w*3 + 2];
189 weightTot += weight; 188 weightTot += weight;
190 alphaTot += weight * get_alpha(src, rowBytes, x, y); 189 alphaTot += weight * get_alpha(src, rowBytes, x, y);
190 } else {
robertphillips 2014/07/25 16:11:41 // why is this valid?
krajcevski 2014/07/25 17:12:05 Done.
191 break;
191 } 192 }
192 } 193 }
193 194
194 indices[idx] = (alphaTot + (weightTot/2)) / weightTot; 195 indices[idx] = (alphaTot / weightTot) >> 5;
195 } 196 }
196 197
197 // Pack indices... The ASTC block layout is fairly complicated. An extensive 198 // Pack indices... The ASTC block layout is fairly complicated. An extensive
198 // description can be found here: 199 // description can be found here:
199 // https://www.opengl.org/registry/specs/KHR/texture_compression_astc_hdr.tx t 200 // https://www.opengl.org/registry/specs/KHR/texture_compression_astc_hdr.tx t
200 // 201 //
201 // Here is a summary of the options that we've chosen: 202 // Here is a summary of the options that we've chosen:
202 // 1. Block mode: 0b00101110011 203 // 1. Block mode: 0b00101110011
203 // - 6x5 texel grid 204 // - 6x5 texel grid
204 // - Single plane 205 // - Single plane
205 // - Low-precision index values 206 // - Low-precision index values
206 // - Index range 0-7 (three bits per index) 207 // - Index range 0-7 (three bits per index)
207 // 2. Partitions: 0b00 208 // 2. Partitions: 0b00
208 // - One partition 209 // - One partition
209 // 3. Color Endpoint Mode: 0b0000 210 // 3. Color Endpoint Mode: 0b0000
210 // - Direct luminance -- e0=(v0,v0,v0,0xFF); e1=(v1,v1,v1,0xFF); 211 // - Direct luminance -- e0=(v0,v0,v0,0xFF); e1=(v1,v1,v1,0xFF);
211 // 4. 8-bit endpoints: 212 // 4. 8-bit endpoints:
212 // v0 = 0, v1 = 255 213 // v0 = 0, v1 = 255
213 // 214 //
214 // The rest of the block contains the 30 index values from before, which 215 // The rest of the block contains the 30 index values from before, which
215 // are currently stored in the indices variable. 216 // are currently stored in the indices variable.
216 217
217 uint64_t top = 0x0000000001FE000173ULL; 218 uint64_t top = 0x0000000001FE000173ULL;
218 uint64_t bottom = 0; 219 uint64_t bottom = 0;
219 220
220 for (int idx = 0; idx <= 20; ++idx) { 221 for (int idx = 0; idx <= 20; ++idx) {
221 const uint8_t index = (indices[idx] & 0xE0) >> 5; 222 const uint8_t index = indices[idx];
222 bottom |= static_cast<uint64_t>(index) << (61-(idx*3)); 223 bottom |= static_cast<uint64_t>(index) << (61-(idx*3));
223 } 224 }
224 225
225 // index 21 straddles top and bottom 226 // index 21 straddles top and bottom
226 { 227 {
227 const uint8_t index = (indices[21] & 0xE0) >> 5; 228 const uint8_t index = indices[21];
228 bottom |= index & 1; 229 bottom |= index & 1;
229 top |= static_cast<uint64_t>((index >> 2) | (index & 2)) << 62; 230 top |= static_cast<uint64_t>((index >> 2) | (index & 2)) << 62;
230 } 231 }
231 232
232 for (int idx = 22; idx < 30; ++idx) { 233 for (int idx = 22; idx < 30; ++idx) {
233 const uint8_t index = (indices[idx] & 0xE0) >> 5; 234 const uint8_t index = indices[idx];
234 top |= static_cast<uint64_t>(index) << (59-(idx-22)*3); 235 top |= static_cast<uint64_t>(index) << (59-(idx-22)*3);
235 } 236 }
236 237
237 // Reverse each 3-bit index since indices are read in reverse order... 238 // Reverse each 3-bit index since indices are read in reverse order...
238 uint64_t t = (bottom ^ (bottom >> 2)) & 0x2492492492492492ULL; 239 uint64_t t = (bottom ^ (bottom >> 2)) & 0x2492492492492492ULL;
239 bottom = bottom ^ t ^ (t << 2); 240 bottom = bottom ^ t ^ (t << 2);
240 241
241 t = (top ^ (top >> 2)) & 0x0924924000000000ULL; 242 t = (top ^ (top >> 2)) & 0x0924924000000000ULL;
242 top = top ^ t ^ (t << 2); 243 top = top ^ t ^ (t << 2);
243 244
(...skipping 18 matching lines...) Expand all
262 263
263 return true; 264 return true;
264 } 265 }
265 266
266 SkBlitter* CreateASTCBlitter(int width, int height, void* outputBuffer) { 267 SkBlitter* CreateASTCBlitter(int width, int height, void* outputBuffer) {
267 // TODO (krajcevski) 268 // TODO (krajcevski)
268 return NULL; 269 return NULL;
269 } 270 }
270 271
271 } // SkTextureCompressor 272 } // SkTextureCompressor
OLDNEW
« no previous file with comments | « no previous file | no next file » | no next file with comments »

Powered by Google App Engine
This is Rietveld 408576698