1 /*
2 * Copyright 2013 The Android Open Source Project
3 *
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
7 *
8 * http://www.apache.org/licenses/LICENSE-2.0
9 *
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
15 */
16
17 #define ATRACE_TAG ATRACE_TAG_GRAPHICS
18
19 #include "ProgramCache.h"
20
21 #include <GLES2/gl2.h>
22 #include <GLES2/gl2ext.h>
23 #include <log/log.h>
24 #include <renderengine/private/Description.h>
25 #include <utils/String8.h>
26 #include <utils/Trace.h>
27 #include "Program.h"
28
29 ANDROID_SINGLETON_STATIC_INSTANCE(android::renderengine::gl::ProgramCache)
30
31 namespace android {
32 namespace renderengine {
33 namespace gl {
34
35 /*
36 * A simple formatter class to automatically add the endl and
37 * manage the indentation.
38 */
39
40 class Formatter;
41 static Formatter& indent(Formatter& f);
42 static Formatter& dedent(Formatter& f);
43
44 class Formatter {
45 String8 mString;
46 int mIndent;
47 typedef Formatter& (*FormaterManipFunc)(Formatter&);
48 friend Formatter& indent(Formatter& f);
49 friend Formatter& dedent(Formatter& f);
50
51 public:
Formatter()52 Formatter() : mIndent(0) {}
53
getString() const54 String8 getString() const { return mString; }
55
operator <<(Formatter & out,const char * in)56 friend Formatter& operator<<(Formatter& out, const char* in) {
57 for (int i = 0; i < out.mIndent; i++) {
58 out.mString.append(" ");
59 }
60 out.mString.append(in);
61 out.mString.append("\n");
62 return out;
63 }
operator <<(Formatter & out,const String8 & in)64 friend inline Formatter& operator<<(Formatter& out, const String8& in) {
65 return operator<<(out, in.string());
66 }
operator <<(Formatter & to,FormaterManipFunc func)67 friend inline Formatter& operator<<(Formatter& to, FormaterManipFunc func) {
68 return (*func)(to);
69 }
70 };
indent(Formatter & f)71 Formatter& indent(Formatter& f) {
72 f.mIndent++;
73 return f;
74 }
dedent(Formatter & f)75 Formatter& dedent(Formatter& f) {
76 f.mIndent--;
77 return f;
78 }
79
primeCache(EGLContext context,bool useColorManagement,bool toneMapperShaderOnly)80 void ProgramCache::primeCache(
81 EGLContext context, bool useColorManagement, bool toneMapperShaderOnly) {
82 auto& cache = mCaches[context];
83 uint32_t shaderCount = 0;
84
85 if (toneMapperShaderOnly) {
86 Key shaderKey;
87 // base settings used by HDR->SDR tonemap only
88 shaderKey.set(Key::BLEND_MASK | Key::INPUT_TRANSFORM_MATRIX_MASK |
89 Key::OUTPUT_TRANSFORM_MATRIX_MASK | Key::OUTPUT_TF_MASK |
90 Key::OPACITY_MASK | Key::ALPHA_MASK |
91 Key::ROUNDED_CORNERS_MASK | Key::TEXTURE_MASK,
92 Key::BLEND_NORMAL | Key::INPUT_TRANSFORM_MATRIX_ON |
93 Key::OUTPUT_TRANSFORM_MATRIX_ON | Key::OUTPUT_TF_SRGB |
94 Key::OPACITY_OPAQUE | Key::ALPHA_EQ_ONE |
95 Key::ROUNDED_CORNERS_OFF | Key::TEXTURE_EXT);
96 for (int i = 0; i < 4; i++) {
97 // Cache input transfer for HLG & ST2084
98 shaderKey.set(Key::INPUT_TF_MASK, (i & 1) ?
99 Key::INPUT_TF_HLG : Key::INPUT_TF_ST2084);
100
101 // Cache Y410 input on or off
102 shaderKey.set(Key::Y410_BT2020_MASK, (i & 2) ?
103 Key::Y410_BT2020_ON : Key::Y410_BT2020_OFF);
104 if (cache.count(shaderKey) == 0) {
105 cache.emplace(shaderKey, generateProgram(shaderKey));
106 shaderCount++;
107 }
108 }
109 return;
110 }
111
112 uint32_t keyMask = Key::BLEND_MASK | Key::OPACITY_MASK | Key::ALPHA_MASK | Key::TEXTURE_MASK
113 | Key::ROUNDED_CORNERS_MASK;
114 // Prime the cache for all combinations of the above masks,
115 // leaving off the experimental color matrix mask options.
116
117 nsecs_t timeBefore = systemTime();
118 for (uint32_t keyVal = 0; keyVal <= keyMask; keyVal++) {
119 Key shaderKey;
120 shaderKey.set(keyMask, keyVal);
121 uint32_t tex = shaderKey.getTextureTarget();
122 if (tex != Key::TEXTURE_OFF && tex != Key::TEXTURE_EXT && tex != Key::TEXTURE_2D) {
123 continue;
124 }
125 if (cache.count(shaderKey) == 0) {
126 cache.emplace(shaderKey, generateProgram(shaderKey));
127 shaderCount++;
128 }
129 }
130
131 // Prime for sRGB->P3 conversion
132 if (useColorManagement) {
133 Key shaderKey;
134 shaderKey.set(Key::BLEND_MASK | Key::OUTPUT_TRANSFORM_MATRIX_MASK | Key::INPUT_TF_MASK |
135 Key::OUTPUT_TF_MASK,
136 Key::BLEND_PREMULT | Key::OUTPUT_TRANSFORM_MATRIX_ON | Key::INPUT_TF_SRGB |
137 Key::OUTPUT_TF_SRGB);
138 for (int i = 0; i < 16; i++) {
139 shaderKey.set(Key::OPACITY_MASK,
140 (i & 1) ? Key::OPACITY_OPAQUE : Key::OPACITY_TRANSLUCENT);
141 shaderKey.set(Key::ALPHA_MASK, (i & 2) ? Key::ALPHA_LT_ONE : Key::ALPHA_EQ_ONE);
142
143 // Cache rounded corners
144 shaderKey.set(Key::ROUNDED_CORNERS_MASK,
145 (i & 4) ? Key::ROUNDED_CORNERS_ON : Key::ROUNDED_CORNERS_OFF);
146
147 // Cache texture off option for window transition
148 shaderKey.set(Key::TEXTURE_MASK, (i & 8) ? Key::TEXTURE_EXT : Key::TEXTURE_OFF);
149 if (cache.count(shaderKey) == 0) {
150 cache.emplace(shaderKey, generateProgram(shaderKey));
151 shaderCount++;
152 }
153 }
154 }
155
156 nsecs_t timeAfter = systemTime();
157 float compileTimeMs = static_cast<float>(timeAfter - timeBefore) / 1.0E6;
158 ALOGD("shader cache generated - %u shaders in %f ms\n", shaderCount, compileTimeMs);
159 }
160
computeKey(const Description & description)161 ProgramCache::Key ProgramCache::computeKey(const Description& description) {
162 Key needs;
163 needs.set(Key::TEXTURE_MASK,
164 !description.textureEnabled
165 ? Key::TEXTURE_OFF
166 : description.texture.getTextureTarget() == GL_TEXTURE_EXTERNAL_OES
167 ? Key::TEXTURE_EXT
168 : description.texture.getTextureTarget() == GL_TEXTURE_2D
169 ? Key::TEXTURE_2D
170 : Key::TEXTURE_OFF)
171 .set(Key::ALPHA_MASK, (description.color.a < 1) ? Key::ALPHA_LT_ONE : Key::ALPHA_EQ_ONE)
172 .set(Key::BLEND_MASK,
173 description.isPremultipliedAlpha ? Key::BLEND_PREMULT : Key::BLEND_NORMAL)
174 .set(Key::OPACITY_MASK,
175 description.isOpaque ? Key::OPACITY_OPAQUE : Key::OPACITY_TRANSLUCENT)
176 .set(Key::Key::INPUT_TRANSFORM_MATRIX_MASK,
177 description.hasInputTransformMatrix() ? Key::INPUT_TRANSFORM_MATRIX_ON
178 : Key::INPUT_TRANSFORM_MATRIX_OFF)
179 .set(Key::Key::OUTPUT_TRANSFORM_MATRIX_MASK,
180 description.hasOutputTransformMatrix() || description.hasColorMatrix()
181 ? Key::OUTPUT_TRANSFORM_MATRIX_ON
182 : Key::OUTPUT_TRANSFORM_MATRIX_OFF)
183 .set(Key::ROUNDED_CORNERS_MASK,
184 description.cornerRadius > 0 ? Key::ROUNDED_CORNERS_ON : Key::ROUNDED_CORNERS_OFF)
185 .set(Key::SHADOW_MASK, description.drawShadows ? Key::SHADOW_ON : Key::SHADOW_OFF);
186 needs.set(Key::Y410_BT2020_MASK,
187 description.isY410BT2020 ? Key::Y410_BT2020_ON : Key::Y410_BT2020_OFF);
188
189 if (needs.hasTransformMatrix() ||
190 (description.inputTransferFunction != description.outputTransferFunction)) {
191 switch (description.inputTransferFunction) {
192 case Description::TransferFunction::LINEAR:
193 default:
194 needs.set(Key::INPUT_TF_MASK, Key::INPUT_TF_LINEAR);
195 break;
196 case Description::TransferFunction::SRGB:
197 needs.set(Key::INPUT_TF_MASK, Key::INPUT_TF_SRGB);
198 break;
199 case Description::TransferFunction::ST2084:
200 needs.set(Key::INPUT_TF_MASK, Key::INPUT_TF_ST2084);
201 break;
202 case Description::TransferFunction::HLG:
203 needs.set(Key::INPUT_TF_MASK, Key::INPUT_TF_HLG);
204 break;
205 }
206
207 switch (description.outputTransferFunction) {
208 case Description::TransferFunction::LINEAR:
209 default:
210 needs.set(Key::OUTPUT_TF_MASK, Key::OUTPUT_TF_LINEAR);
211 break;
212 case Description::TransferFunction::SRGB:
213 needs.set(Key::OUTPUT_TF_MASK, Key::OUTPUT_TF_SRGB);
214 break;
215 case Description::TransferFunction::ST2084:
216 needs.set(Key::OUTPUT_TF_MASK, Key::OUTPUT_TF_ST2084);
217 break;
218 case Description::TransferFunction::HLG:
219 needs.set(Key::OUTPUT_TF_MASK, Key::OUTPUT_TF_HLG);
220 break;
221 }
222 }
223
224 return needs;
225 }
226
227 // Generate EOTF that converts signal values to relative display light,
228 // both normalized to [0, 1].
generateEOTF(Formatter & fs,const Key & needs)229 void ProgramCache::generateEOTF(Formatter& fs, const Key& needs) {
230 switch (needs.getInputTF()) {
231 case Key::INPUT_TF_SRGB:
232 fs << R"__SHADER__(
233 float EOTF_sRGB(float srgb) {
234 return srgb <= 0.04045 ? srgb / 12.92 : pow((srgb + 0.055) / 1.055, 2.4);
235 }
236
237 vec3 EOTF_sRGB(const vec3 srgb) {
238 return vec3(EOTF_sRGB(srgb.r), EOTF_sRGB(srgb.g), EOTF_sRGB(srgb.b));
239 }
240
241 vec3 EOTF(const vec3 srgb) {
242 return sign(srgb.rgb) * EOTF_sRGB(abs(srgb.rgb));
243 }
244 )__SHADER__";
245 break;
246 case Key::INPUT_TF_ST2084:
247 fs << R"__SHADER__(
248 vec3 EOTF(const highp vec3 color) {
249 const highp float m1 = (2610.0 / 4096.0) / 4.0;
250 const highp float m2 = (2523.0 / 4096.0) * 128.0;
251 const highp float c1 = (3424.0 / 4096.0);
252 const highp float c2 = (2413.0 / 4096.0) * 32.0;
253 const highp float c3 = (2392.0 / 4096.0) * 32.0;
254
255 highp vec3 tmp = pow(clamp(color, 0.0, 1.0), 1.0 / vec3(m2));
256 tmp = max(tmp - c1, 0.0) / (c2 - c3 * tmp);
257 return pow(tmp, 1.0 / vec3(m1));
258 }
259 )__SHADER__";
260 break;
261 case Key::INPUT_TF_HLG:
262 fs << R"__SHADER__(
263 highp float EOTF_channel(const highp float channel) {
264 const highp float a = 0.17883277;
265 const highp float b = 0.28466892;
266 const highp float c = 0.55991073;
267 return channel <= 0.5 ? channel * channel / 3.0 :
268 (exp((channel - c) / a) + b) / 12.0;
269 }
270
271 vec3 EOTF(const highp vec3 color) {
272 return vec3(EOTF_channel(color.r), EOTF_channel(color.g),
273 EOTF_channel(color.b));
274 }
275 )__SHADER__";
276 break;
277 default:
278 fs << R"__SHADER__(
279 vec3 EOTF(const vec3 linear) {
280 return linear;
281 }
282 )__SHADER__";
283 break;
284 }
285 }
286
generateToneMappingProcess(Formatter & fs,const Key & needs)287 void ProgramCache::generateToneMappingProcess(Formatter& fs, const Key& needs) {
288 // Convert relative light to absolute light.
289 switch (needs.getInputTF()) {
290 case Key::INPUT_TF_ST2084:
291 fs << R"__SHADER__(
292 highp vec3 ScaleLuminance(highp vec3 color) {
293 return color * 10000.0;
294 }
295 )__SHADER__";
296 break;
297 case Key::INPUT_TF_HLG:
298 fs << R"__SHADER__(
299 highp vec3 ScaleLuminance(highp vec3 color) {
300 // The formula is:
301 // alpha * pow(Y, gamma - 1.0) * color + beta;
302 // where alpha is 1000.0, gamma is 1.2, beta is 0.0.
303 return color * 1000.0 * pow(color.y, 0.2);
304 }
305 )__SHADER__";
306 break;
307 default:
308 fs << R"__SHADER__(
309 highp vec3 ScaleLuminance(highp vec3 color) {
310 return color * displayMaxLuminance;
311 }
312 )__SHADER__";
313 break;
314 }
315
316 // Tone map absolute light to display luminance range.
317 switch (needs.getInputTF()) {
318 case Key::INPUT_TF_ST2084:
319 case Key::INPUT_TF_HLG:
320 switch (needs.getOutputTF()) {
321 case Key::OUTPUT_TF_HLG:
322 // Right now when mixed PQ and HLG contents are presented,
323 // HLG content will always be converted to PQ. However, for
324 // completeness, we simply clamp the value to [0.0, 1000.0].
325 fs << R"__SHADER__(
326 highp vec3 ToneMap(highp vec3 color) {
327 return clamp(color, 0.0, 1000.0);
328 }
329 )__SHADER__";
330 break;
331 case Key::OUTPUT_TF_ST2084:
332 fs << R"__SHADER__(
333 highp vec3 ToneMap(highp vec3 color) {
334 return color;
335 }
336 )__SHADER__";
337 break;
338 default:
339 fs << R"__SHADER__(
340 highp vec3 ToneMap(highp vec3 color) {
341 float maxMasteringLumi = maxMasteringLuminance;
342 float maxContentLumi = maxContentLuminance;
343 float maxInLumi = min(maxMasteringLumi, maxContentLumi);
344 float maxOutLumi = displayMaxLuminance;
345
346 float nits = color.y;
347
348 // clamp to max input luminance
349 nits = clamp(nits, 0.0, maxInLumi);
350
351 // scale [0.0, maxInLumi] to [0.0, maxOutLumi]
352 if (maxInLumi <= maxOutLumi) {
353 return color * (maxOutLumi / maxInLumi);
354 } else {
355 // three control points
356 const float x0 = 10.0;
357 const float y0 = 17.0;
358 float x1 = maxOutLumi * 0.75;
359 float y1 = x1;
360 float x2 = x1 + (maxInLumi - x1) / 2.0;
361 float y2 = y1 + (maxOutLumi - y1) * 0.75;
362
363 // horizontal distances between the last three control points
364 float h12 = x2 - x1;
365 float h23 = maxInLumi - x2;
366 // tangents at the last three control points
367 float m1 = (y2 - y1) / h12;
368 float m3 = (maxOutLumi - y2) / h23;
369 float m2 = (m1 + m3) / 2.0;
370
371 if (nits < x0) {
372 // scale [0.0, x0] to [0.0, y0] linearly
373 float slope = y0 / x0;
374 return color * slope;
375 } else if (nits < x1) {
376 // scale [x0, x1] to [y0, y1] linearly
377 float slope = (y1 - y0) / (x1 - x0);
378 nits = y0 + (nits - x0) * slope;
379 } else if (nits < x2) {
380 // scale [x1, x2] to [y1, y2] using Hermite interp
381 float t = (nits - x1) / h12;
382 nits = (y1 * (1.0 + 2.0 * t) + h12 * m1 * t) * (1.0 - t) * (1.0 - t) +
383 (y2 * (3.0 - 2.0 * t) + h12 * m2 * (t - 1.0)) * t * t;
384 } else {
385 // scale [x2, maxInLumi] to [y2, maxOutLumi] using Hermite interp
386 float t = (nits - x2) / h23;
387 nits = (y2 * (1.0 + 2.0 * t) + h23 * m2 * t) * (1.0 - t) * (1.0 - t) +
388 (maxOutLumi * (3.0 - 2.0 * t) + h23 * m3 * (t - 1.0)) * t * t;
389 }
390 }
391
392 // color.y is greater than x0 and is thus non-zero
393 return color * (nits / color.y);
394 }
395 )__SHADER__";
396 break;
397 }
398 break;
399 default:
400 // inverse tone map; the output luminance can be up to maxOutLumi.
401 fs << R"__SHADER__(
402 highp vec3 ToneMap(highp vec3 color) {
403 const float maxOutLumi = 3000.0;
404
405 const float x0 = 5.0;
406 const float y0 = 2.5;
407 float x1 = displayMaxLuminance * 0.7;
408 float y1 = maxOutLumi * 0.15;
409 float x2 = displayMaxLuminance * 0.9;
410 float y2 = maxOutLumi * 0.45;
411 float x3 = displayMaxLuminance;
412 float y3 = maxOutLumi;
413
414 float c1 = y1 / 3.0;
415 float c2 = y2 / 2.0;
416 float c3 = y3 / 1.5;
417
418 float nits = color.y;
419
420 float scale;
421 if (nits <= x0) {
422 // scale [0.0, x0] to [0.0, y0] linearly
423 const float slope = y0 / x0;
424 return color * slope;
425 } else if (nits <= x1) {
426 // scale [x0, x1] to [y0, y1] using a curve
427 float t = (nits - x0) / (x1 - x0);
428 nits = (1.0 - t) * (1.0 - t) * y0 + 2.0 * (1.0 - t) * t * c1 + t * t * y1;
429 } else if (nits <= x2) {
430 // scale [x1, x2] to [y1, y2] using a curve
431 float t = (nits - x1) / (x2 - x1);
432 nits = (1.0 - t) * (1.0 - t) * y1 + 2.0 * (1.0 - t) * t * c2 + t * t * y2;
433 } else {
434 // scale [x2, x3] to [y2, y3] using a curve
435 float t = (nits - x2) / (x3 - x2);
436 nits = (1.0 - t) * (1.0 - t) * y2 + 2.0 * (1.0 - t) * t * c3 + t * t * y3;
437 }
438
439 // color.y is greater than x0 and is thus non-zero
440 return color * (nits / color.y);
441 }
442 )__SHADER__";
443 break;
444 }
445
446 // convert absolute light to relative light.
447 switch (needs.getOutputTF()) {
448 case Key::OUTPUT_TF_ST2084:
449 fs << R"__SHADER__(
450 highp vec3 NormalizeLuminance(highp vec3 color) {
451 return color / 10000.0;
452 }
453 )__SHADER__";
454 break;
455 case Key::OUTPUT_TF_HLG:
456 fs << R"__SHADER__(
457 highp vec3 NormalizeLuminance(highp vec3 color) {
458 return color / 1000.0 * pow(color.y / 1000.0, -0.2 / 1.2);
459 }
460 )__SHADER__";
461 break;
462 default:
463 fs << R"__SHADER__(
464 highp vec3 NormalizeLuminance(highp vec3 color) {
465 return color / displayMaxLuminance;
466 }
467 )__SHADER__";
468 break;
469 }
470 }
471
472 // Generate OOTF that modifies the relative scence light to relative display light.
generateOOTF(Formatter & fs,const ProgramCache::Key & needs)473 void ProgramCache::generateOOTF(Formatter& fs, const ProgramCache::Key& needs) {
474 if (!needs.needsToneMapping()) {
475 fs << R"__SHADER__(
476 highp vec3 OOTF(const highp vec3 color) {
477 return color;
478 }
479 )__SHADER__";
480 } else {
481 generateToneMappingProcess(fs, needs);
482 fs << R"__SHADER__(
483 highp vec3 OOTF(const highp vec3 color) {
484 return NormalizeLuminance(ToneMap(ScaleLuminance(color)));
485 }
486 )__SHADER__";
487 }
488 }
489
490 // Generate OETF that converts relative display light to signal values,
491 // both normalized to [0, 1]
generateOETF(Formatter & fs,const Key & needs)492 void ProgramCache::generateOETF(Formatter& fs, const Key& needs) {
493 switch (needs.getOutputTF()) {
494 case Key::OUTPUT_TF_SRGB:
495 fs << R"__SHADER__(
496 float OETF_sRGB(const float linear) {
497 return linear <= 0.0031308 ?
498 linear * 12.92 : (pow(linear, 1.0 / 2.4) * 1.055) - 0.055;
499 }
500
501 vec3 OETF_sRGB(const vec3 linear) {
502 return vec3(OETF_sRGB(linear.r), OETF_sRGB(linear.g), OETF_sRGB(linear.b));
503 }
504
505 vec3 OETF(const vec3 linear) {
506 return sign(linear.rgb) * OETF_sRGB(abs(linear.rgb));
507 }
508 )__SHADER__";
509 break;
510 case Key::OUTPUT_TF_ST2084:
511 fs << R"__SHADER__(
512 vec3 OETF(const vec3 linear) {
513 const highp float m1 = (2610.0 / 4096.0) / 4.0;
514 const highp float m2 = (2523.0 / 4096.0) * 128.0;
515 const highp float c1 = (3424.0 / 4096.0);
516 const highp float c2 = (2413.0 / 4096.0) * 32.0;
517 const highp float c3 = (2392.0 / 4096.0) * 32.0;
518
519 highp vec3 tmp = pow(linear, vec3(m1));
520 tmp = (c1 + c2 * tmp) / (1.0 + c3 * tmp);
521 return pow(tmp, vec3(m2));
522 }
523 )__SHADER__";
524 break;
525 case Key::OUTPUT_TF_HLG:
526 fs << R"__SHADER__(
527 highp float OETF_channel(const highp float channel) {
528 const highp float a = 0.17883277;
529 const highp float b = 0.28466892;
530 const highp float c = 0.55991073;
531 return channel <= 1.0 / 12.0 ? sqrt(3.0 * channel) :
532 a * log(12.0 * channel - b) + c;
533 }
534
535 vec3 OETF(const highp vec3 color) {
536 return vec3(OETF_channel(color.r), OETF_channel(color.g),
537 OETF_channel(color.b));
538 }
539 )__SHADER__";
540 break;
541 default:
542 fs << R"__SHADER__(
543 vec3 OETF(const vec3 linear) {
544 return linear;
545 }
546 )__SHADER__";
547 break;
548 }
549 }
550
generateVertexShader(const Key & needs)551 String8 ProgramCache::generateVertexShader(const Key& needs) {
552 Formatter vs;
553 if (needs.hasTextureCoords()) {
554 vs << "attribute vec4 texCoords;"
555 << "varying vec2 outTexCoords;";
556 }
557 if (needs.hasRoundedCorners()) {
558 vs << "attribute lowp vec4 cropCoords;";
559 vs << "varying lowp vec2 outCropCoords;";
560 }
561 if (needs.drawShadows()) {
562 vs << "attribute lowp vec4 shadowColor;";
563 vs << "varying lowp vec4 outShadowColor;";
564 vs << "attribute lowp vec4 shadowParams;";
565 vs << "varying lowp vec3 outShadowParams;";
566 }
567 vs << "attribute vec4 position;"
568 << "uniform mat4 projection;"
569 << "uniform mat4 texture;"
570 << "void main(void) {" << indent << "gl_Position = projection * position;";
571 if (needs.hasTextureCoords()) {
572 vs << "outTexCoords = (texture * texCoords).st;";
573 }
574 if (needs.hasRoundedCorners()) {
575 vs << "outCropCoords = cropCoords.st;";
576 }
577 if (needs.drawShadows()) {
578 vs << "outShadowColor = shadowColor;";
579 vs << "outShadowParams = shadowParams.xyz;";
580 }
581 vs << dedent << "}";
582 return vs.getString();
583 }
584
generateFragmentShader(const Key & needs)585 String8 ProgramCache::generateFragmentShader(const Key& needs) {
586 Formatter fs;
587 if (needs.getTextureTarget() == Key::TEXTURE_EXT) {
588 fs << "#extension GL_OES_EGL_image_external : require";
589 }
590
591 // default precision is required-ish in fragment shaders
592 fs << "precision mediump float;";
593
594 if (needs.getTextureTarget() == Key::TEXTURE_EXT) {
595 fs << "uniform samplerExternalOES sampler;";
596 } else if (needs.getTextureTarget() == Key::TEXTURE_2D) {
597 fs << "uniform sampler2D sampler;";
598 }
599
600 if (needs.hasTextureCoords()) {
601 fs << "varying vec2 outTexCoords;";
602 }
603
604 if (needs.hasRoundedCorners()) {
605 // Rounded corners implementation using a signed distance function.
606 fs << R"__SHADER__(
607 uniform float cornerRadius;
608 uniform vec2 cropCenter;
609 varying vec2 outCropCoords;
610
611 /**
612 * This function takes the current crop coordinates and calculates an alpha value based
613 * on the corner radius and distance from the crop center.
614 */
615 float applyCornerRadius(vec2 cropCoords)
616 {
617 vec2 position = cropCoords - cropCenter;
618 // Scale down the dist vector here, as otherwise large corner
619 // radii can cause floating point issues when computing the norm
620 vec2 dist = (abs(position) - cropCenter + vec2(cornerRadius)) / 16.0;
621 // Once we've found the norm, then scale back up.
622 float plane = length(max(dist, vec2(0.0))) * 16.0;
623 return 1.0 - clamp(plane - cornerRadius, 0.0, 1.0);
624 }
625 )__SHADER__";
626 }
627
628 if (needs.drawShadows()) {
629 fs << R"__SHADER__(
630 varying lowp vec4 outShadowColor;
631 varying lowp vec3 outShadowParams;
632
633 /**
634 * Returns the shadow color.
635 */
636 vec4 getShadowColor()
637 {
638 lowp float d = length(outShadowParams.xy);
639 vec2 uv = vec2(outShadowParams.z * (1.0 - d), 0.5);
640 lowp float factor = texture2D(sampler, uv).a;
641 return outShadowColor * factor;
642 }
643 )__SHADER__";
644 }
645
646 if (needs.getTextureTarget() == Key::TEXTURE_OFF || needs.hasAlpha()) {
647 fs << "uniform vec4 color;";
648 }
649
650 if (needs.isY410BT2020()) {
651 fs << R"__SHADER__(
652 vec3 convertY410BT2020(const vec3 color) {
653 const vec3 offset = vec3(0.0625, 0.5, 0.5);
654 const mat3 transform = mat3(
655 vec3(1.1678, 1.1678, 1.1678),
656 vec3( 0.0, -0.1878, 2.1481),
657 vec3(1.6836, -0.6523, 0.0));
658 // Y is in G, U is in R, and V is in B
659 return clamp(transform * (color.grb - offset), 0.0, 1.0);
660 }
661 )__SHADER__";
662 }
663
664 if (needs.hasTransformMatrix() || (needs.getInputTF() != needs.getOutputTF())) {
665 if (needs.needsToneMapping()) {
666 fs << "uniform float displayMaxLuminance;";
667 fs << "uniform float maxMasteringLuminance;";
668 fs << "uniform float maxContentLuminance;";
669 }
670
671 if (needs.hasInputTransformMatrix()) {
672 fs << "uniform mat4 inputTransformMatrix;";
673 fs << R"__SHADER__(
674 highp vec3 InputTransform(const highp vec3 color) {
675 return clamp(vec3(inputTransformMatrix * vec4(color, 1.0)), 0.0, 1.0);
676 }
677 )__SHADER__";
678 } else {
679 fs << R"__SHADER__(
680 highp vec3 InputTransform(const highp vec3 color) {
681 return color;
682 }
683 )__SHADER__";
684 }
685
686 // the transformation from a wider colorspace to a narrower one can
687 // result in >1.0 or <0.0 pixel values
688 if (needs.hasOutputTransformMatrix()) {
689 fs << "uniform mat4 outputTransformMatrix;";
690 fs << R"__SHADER__(
691 highp vec3 OutputTransform(const highp vec3 color) {
692 return clamp(vec3(outputTransformMatrix * vec4(color, 1.0)), 0.0, 1.0);
693 }
694 )__SHADER__";
695 } else {
696 fs << R"__SHADER__(
697 highp vec3 OutputTransform(const highp vec3 color) {
698 return clamp(color, 0.0, 1.0);
699 }
700 )__SHADER__";
701 }
702
703 generateEOTF(fs, needs);
704 generateOOTF(fs, needs);
705 generateOETF(fs, needs);
706 }
707
708 fs << "void main(void) {" << indent;
709 if (needs.drawShadows()) {
710 fs << "gl_FragColor = getShadowColor();";
711 } else {
712 if (needs.isTexturing()) {
713 fs << "gl_FragColor = texture2D(sampler, outTexCoords);";
714 if (needs.isY410BT2020()) {
715 fs << "gl_FragColor.rgb = convertY410BT2020(gl_FragColor.rgb);";
716 }
717 } else {
718 fs << "gl_FragColor.rgb = color.rgb;";
719 fs << "gl_FragColor.a = 1.0;";
720 }
721 if (needs.isOpaque()) {
722 fs << "gl_FragColor.a = 1.0;";
723 }
724 if (needs.hasAlpha()) {
725 // modulate the current alpha value with alpha set
726 if (needs.isPremultiplied()) {
727 // ... and the color too if we're premultiplied
728 fs << "gl_FragColor *= color.a;";
729 } else {
730 fs << "gl_FragColor.a *= color.a;";
731 }
732 }
733 }
734
735 if (needs.hasTransformMatrix() || (needs.getInputTF() != needs.getOutputTF())) {
736 if (!needs.isOpaque() && needs.isPremultiplied()) {
737 // un-premultiply if needed before linearization
738 // avoid divide by 0 by adding 0.5/256 to the alpha channel
739 fs << "gl_FragColor.rgb = gl_FragColor.rgb / (gl_FragColor.a + 0.0019);";
740 }
741 fs << "gl_FragColor.rgb = "
742 "OETF(OutputTransform(OOTF(InputTransform(EOTF(gl_FragColor.rgb)))));";
743 if (!needs.isOpaque() && needs.isPremultiplied()) {
744 // and re-premultiply if needed after gamma correction
745 fs << "gl_FragColor.rgb = gl_FragColor.rgb * (gl_FragColor.a + 0.0019);";
746 }
747 }
748
749 if (needs.hasRoundedCorners()) {
750 if (needs.isPremultiplied()) {
751 fs << "gl_FragColor *= vec4(applyCornerRadius(outCropCoords));";
752 } else {
753 fs << "gl_FragColor.a *= applyCornerRadius(outCropCoords);";
754 }
755 }
756
757 fs << dedent << "}";
758 return fs.getString();
759 }
760
generateProgram(const Key & needs)761 std::unique_ptr<Program> ProgramCache::generateProgram(const Key& needs) {
762 ATRACE_CALL();
763
764 // vertex shader
765 String8 vs = generateVertexShader(needs);
766
767 // fragment shader
768 String8 fs = generateFragmentShader(needs);
769
770 return std::make_unique<Program>(needs, vs.string(), fs.string());
771 }
772
useProgram(EGLContext context,const Description & description)773 void ProgramCache::useProgram(EGLContext context, const Description& description) {
774 // generate the key for the shader based on the description
775 Key needs(computeKey(description));
776
777 // look-up the program in the cache
778 auto& cache = mCaches[context];
779 auto it = cache.find(needs);
780 if (it == cache.end()) {
781 // we didn't find our program, so generate one...
782 nsecs_t time = systemTime();
783 it = cache.emplace(needs, generateProgram(needs)).first;
784 time = systemTime() - time;
785
786 ALOGV(">>> generated new program for context %p: needs=%08X, time=%u ms (%zu programs)",
787 context, needs.mKey, uint32_t(ns2ms(time)), cache.size());
788 }
789
790 // here we have a suitable program for this description
791 std::unique_ptr<Program>& program = it->second;
792 if (program->isValid()) {
793 program->use();
794 program->setUniforms(description);
795 }
796 }
797
798 } // namespace gl
799 } // namespace renderengine
800 } // namespace android
801