1 // Copyright 2016 The SwiftShader Authors. All Rights Reserved. 2 // 3 // Licensed under the Apache License, Version 2.0 (the "License"); 4 // you may not use this file except in compliance with the License. 5 // You may obtain a copy of the License at 6 // 7 // http://www.apache.org/licenses/LICENSE-2.0 8 // 9 // Unless required by applicable law or agreed to in writing, software 10 // distributed under the License is distributed on an "AS IS" BASIS, 11 // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 12 // See the License for the specific language governing permissions and 13 // limitations under the License. 14 15 #ifndef sw_Renderer_hpp 16 #define sw_Renderer_hpp 17 18 #include "VertexProcessor.hpp" 19 #include "PixelProcessor.hpp" 20 #include "SetupProcessor.hpp" 21 #include "Plane.hpp" 22 #include "Blitter.hpp" 23 #include "Common/MutexLock.hpp" 24 #include "Common/Thread.hpp" 25 #include "Main/Config.hpp" 26 27 #include <list> 28 29 namespace sw 30 { 31 class Clipper; 32 class PixelShader; 33 class VertexShader; 34 class SwiftConfig; 35 struct Task; 36 class Resource; 37 class Renderer; 38 struct Constants; 39 40 enum TranscendentalPrecision 41 { 42 APPROXIMATE, 43 PARTIAL, // 2^-10 44 ACCURATE, 45 WHQL, // 2^-21 46 IEEE // 2^-23 47 }; 48 49 extern TranscendentalPrecision logPrecision; 50 extern TranscendentalPrecision expPrecision; 51 extern TranscendentalPrecision rcpPrecision; 52 extern TranscendentalPrecision rsqPrecision; 53 extern bool perspectiveCorrection; 54 55 struct Conventions 56 { 57 bool halfIntegerCoordinates; 58 bool symmetricNormalizedDepth; 59 bool booleanFaceRegister; 60 bool fullPixelPositionRegister; 61 bool leadingVertexFirst; 62 bool secondaryColor; 63 bool colorsDefaultToZero; 64 }; 65 66 static const Conventions OpenGL = 67 { 68 true, // halfIntegerCoordinates 69 true, // symmetricNormalizedDepth 70 true, // booleanFaceRegister 71 true, // fullPixelPositionRegister 72 false, // leadingVertexFirst 73 false, // secondaryColor 74 true, // colorsDefaultToZero 75 }; 76 77 static const Conventions Direct3D = 78 { 79 false, // halfIntegerCoordinates 80 false, // symmetricNormalizedDepth 81 false, // booleanFaceRegister 82 false, // fullPixelPositionRegister 83 true, // leadingVertexFirst 84 true, // secondardyColor 85 false, // colorsDefaultToZero 86 }; 87 88 struct Query 89 { 90 enum Type { FRAGMENTS_PASSED, TRANSFORM_FEEDBACK_PRIMITIVES_WRITTEN }; 91 92 Query(Type type) : building(false), reference(0), data(0), type(type) 93 { 94 } 95 96 void begin() 97 { 98 building = true; 99 data = 0; 100 } 101 102 void end() 103 { 104 building = false; 105 } 106 107 bool building; 108 AtomicInt reference; 109 AtomicInt data; 110 111 const Type type; 112 }; 113 114 struct DrawData 115 { 116 const Constants *constants; 117 118 const void *input[MAX_VERTEX_INPUTS]; 119 unsigned int stride[MAX_VERTEX_INPUTS]; 120 Texture mipmap[TOTAL_IMAGE_UNITS]; 121 const void *indices; 122 123 struct VS 124 { 125 float4 c[VERTEX_UNIFORM_VECTORS + 1]; // One extra for indices out of range, c[VERTEX_UNIFORM_VECTORS] = {0, 0, 0, 0} 126 byte* u[MAX_UNIFORM_BUFFER_BINDINGS]; 127 byte* t[MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS]; 128 unsigned int reg[MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS]; // Offset used when reading from registers, in components 129 unsigned int row[MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS]; // Number of rows to read 130 unsigned int col[MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS]; // Number of columns to read 131 unsigned int str[MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS]; // Number of components between each varying in output buffer 132 int4 i[16]; 133 bool b[16]; 134 }; 135 136 struct PS 137 { 138 word4 cW[8][4]; 139 float4 c[FRAGMENT_UNIFORM_VECTORS]; 140 byte* u[MAX_UNIFORM_BUFFER_BINDINGS]; 141 int4 i[16]; 142 bool b[16]; 143 }; 144 145 union 146 { 147 VS vs; 148 VertexProcessor::FixedFunction ff; 149 }; 150 151 PS ps; 152 153 int instanceID; 154 155 VertexProcessor::PointSprite point; 156 float lineWidth; 157 158 PixelProcessor::Stencil stencil[2]; // clockwise, counterclockwise 159 PixelProcessor::Stencil stencilCCW; 160 PixelProcessor::Fog fog; 161 PixelProcessor::Factor factor; 162 unsigned int occlusion[16]; // Number of pixels passing depth test 163 164 #if PERF_PROFILE 165 int64_t cycles[PERF_TIMERS][16]; 166 #endif 167 168 TextureStage::Uniforms textureStage[8]; 169 170 float4 Wx16; 171 float4 Hx16; 172 float4 X0x16; 173 float4 Y0x16; 174 float4 XXXX; 175 float4 YYYY; 176 float4 halfPixelX; 177 float4 halfPixelY; 178 float viewportHeight; 179 float slopeDepthBias; 180 float depthRange; 181 float depthNear; 182 Plane clipPlane[6]; 183 184 unsigned int *colorBuffer[RENDERTARGETS]; 185 int colorPitchB[RENDERTARGETS]; 186 int colorSliceB[RENDERTARGETS]; 187 float *depthBuffer; 188 int depthPitchB; 189 int depthSliceB; 190 unsigned char *stencilBuffer; 191 int stencilPitchB; 192 int stencilSliceB; 193 194 int scissorX0; 195 int scissorX1; 196 int scissorY0; 197 int scissorY1; 198 199 float4 a2c0; 200 float4 a2c1; 201 float4 a2c2; 202 float4 a2c3; 203 }; 204 205 struct DrawCall 206 { 207 DrawCall(); 208 209 ~DrawCall(); 210 211 AtomicInt drawType; 212 AtomicInt batchSize; 213 214 Routine *vertexRoutine; 215 Routine *setupRoutine; 216 Routine *pixelRoutine; 217 218 VertexProcessor::RoutinePointer vertexPointer; 219 SetupProcessor::RoutinePointer setupPointer; 220 PixelProcessor::RoutinePointer pixelPointer; 221 222 int (Renderer::*setupPrimitives)(int batch, int count); 223 SetupProcessor::State setupState; 224 225 Resource *vertexStream[MAX_VERTEX_INPUTS]; 226 Resource *indexBuffer; 227 Surface *renderTarget[RENDERTARGETS]; 228 Surface *depthBuffer; 229 Surface *stencilBuffer; 230 Resource *texture[TOTAL_IMAGE_UNITS]; 231 Resource* pUniformBuffers[MAX_UNIFORM_BUFFER_BINDINGS]; 232 Resource* vUniformBuffers[MAX_UNIFORM_BUFFER_BINDINGS]; 233 Resource* transformFeedbackBuffers[MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS]; 234 235 unsigned int vsDirtyConstF; 236 unsigned int vsDirtyConstI; 237 unsigned int vsDirtyConstB; 238 239 unsigned int psDirtyConstF; 240 unsigned int psDirtyConstI; 241 unsigned int psDirtyConstB; 242 243 std::list<Query*> *queries; 244 245 AtomicInt clipFlags; 246 247 AtomicInt primitive; // Current primitive to enter pipeline 248 AtomicInt count; // Number of primitives to render 249 AtomicInt references; // Remaining references to this draw call, 0 when done drawing, -1 when resources unlocked and slot is free 250 251 DrawData *data; 252 }; 253 254 struct Viewport 255 { 256 float x0; 257 float y0; 258 float width; 259 float height; 260 float minZ; 261 float maxZ; 262 }; 263 264 class Renderer : public VertexProcessor, public PixelProcessor, public SetupProcessor 265 { 266 struct Task 267 { 268 enum Type 269 { 270 PRIMITIVES, 271 PIXELS, 272 273 RESUME, 274 SUSPEND 275 }; 276 277 AtomicInt type; 278 AtomicInt primitiveUnit; 279 AtomicInt pixelCluster; 280 }; 281 282 struct PrimitiveProgress 283 { 284 void init() 285 { 286 drawCall = 0; 287 firstPrimitive = 0; 288 primitiveCount = 0; 289 visible = 0; 290 references = 0; 291 } 292 293 AtomicInt drawCall; 294 AtomicInt firstPrimitive; 295 AtomicInt primitiveCount; 296 AtomicInt visible; 297 AtomicInt references; 298 }; 299 300 struct PixelProgress 301 { 302 void init() 303 { 304 drawCall = 0; 305 processedPrimitives = 0; 306 executing = false; 307 } 308 309 AtomicInt drawCall; 310 AtomicInt processedPrimitives; 311 AtomicInt executing; 312 }; 313 314 public: 315 Renderer(Context *context, Conventions conventions, bool exactColorRounding); 316 317 virtual ~Renderer(); 318 319 void *operator new(size_t size); 320 void operator delete(void * mem); 321 322 void draw(DrawType drawType, unsigned int indexOffset, unsigned int count, bool update = true); 323 324 void clear(void *value, Format format, Surface *dest, const Rect &rect, unsigned int rgbaMask); 325 void blit(Surface *source, const SliceRectF &sRect, Surface *dest, const SliceRect &dRect, bool filter, bool isStencil = false, bool sRGBconversion = true); 326 void blit3D(Surface *source, Surface *dest); 327 328 void setIndexBuffer(Resource *indexBuffer); 329 330 void setMultiSampleMask(unsigned int mask); 331 void setTransparencyAntialiasing(TransparencyAntialiasing transparencyAntialiasing); 332 333 void setTextureResource(unsigned int sampler, Resource *resource); 334 void setTextureLevel(unsigned int sampler, unsigned int face, unsigned int level, Surface *surface, TextureType type); 335 336 void setTextureFilter(SamplerType type, int sampler, FilterType textureFilter); 337 void setMipmapFilter(SamplerType type, int sampler, MipmapType mipmapFilter); 338 void setGatherEnable(SamplerType type, int sampler, bool enable); 339 void setAddressingModeU(SamplerType type, int sampler, AddressingMode addressingMode); 340 void setAddressingModeV(SamplerType type, int sampler, AddressingMode addressingMode); 341 void setAddressingModeW(SamplerType type, int sampler, AddressingMode addressingMode); 342 void setReadSRGB(SamplerType type, int sampler, bool sRGB); 343 void setMipmapLOD(SamplerType type, int sampler, float bias); 344 void setBorderColor(SamplerType type, int sampler, const Color<float> &borderColor); 345 void setMaxAnisotropy(SamplerType type, int sampler, float maxAnisotropy); 346 void setHighPrecisionFiltering(SamplerType type, int sampler, bool highPrecisionFiltering); 347 void setSwizzleR(SamplerType type, int sampler, SwizzleType swizzleR); 348 void setSwizzleG(SamplerType type, int sampler, SwizzleType swizzleG); 349 void setSwizzleB(SamplerType type, int sampler, SwizzleType swizzleB); 350 void setSwizzleA(SamplerType type, int sampler, SwizzleType swizzleA); 351 void setCompareFunc(SamplerType type, int sampler, CompareFunc compare); 352 void setBaseLevel(SamplerType type, int sampler, int baseLevel); 353 void setMaxLevel(SamplerType type, int sampler, int maxLevel); 354 void setMinLod(SamplerType type, int sampler, float minLod); 355 void setMaxLod(SamplerType type, int sampler, float maxLod); 356 357 void setPointSpriteEnable(bool pointSpriteEnable); 358 void setPointScaleEnable(bool pointScaleEnable); 359 void setLineWidth(float width); 360 361 void setDepthBias(float bias); 362 void setSlopeDepthBias(float slopeBias); 363 364 void setRasterizerDiscard(bool rasterizerDiscard); 365 366 // Programmable pipelines 367 void setPixelShader(const PixelShader *shader); 368 void setVertexShader(const VertexShader *shader); 369 370 void setPixelShaderConstantF(unsigned int index, const float value[4], unsigned int count = 1); 371 void setPixelShaderConstantI(unsigned int index, const int value[4], unsigned int count = 1); 372 void setPixelShaderConstantB(unsigned int index, const int *boolean, unsigned int count = 1); 373 374 void setVertexShaderConstantF(unsigned int index, const float value[4], unsigned int count = 1); 375 void setVertexShaderConstantI(unsigned int index, const int value[4], unsigned int count = 1); 376 void setVertexShaderConstantB(unsigned int index, const int *boolean, unsigned int count = 1); 377 378 // Viewport & Clipper 379 void setViewport(const Viewport &viewport); 380 void setScissor(const Rect &scissor); 381 void setClipFlags(int flags); 382 void setClipPlane(unsigned int index, const float plane[4]); 383 384 // Partial transform 385 void setModelMatrix(const Matrix &M, int i = 0); 386 void setViewMatrix(const Matrix &V); 387 void setBaseMatrix(const Matrix &B); 388 void setProjectionMatrix(const Matrix &P); 389 390 void addQuery(Query *query); 391 void removeQuery(Query *query); 392 393 void synchronize(); 394 395 #if PERF_HUD 396 // Performance timers 397 int getThreadCount(); 398 int64_t getVertexTime(int thread); 399 int64_t getSetupTime(int thread); 400 int64_t getPixelTime(int thread); 401 void resetTimers(); 402 #endif 403 404 static int getClusterCount() { return clusterCount; } 405 406 private: 407 static void threadFunction(void *parameters); 408 void threadLoop(int threadIndex); 409 void taskLoop(int threadIndex); 410 void findAvailableTasks(); 411 void scheduleTask(int threadIndex); 412 void executeTask(int threadIndex); 413 void finishRendering(Task &pixelTask); 414 415 void processPrimitiveVertices(int unit, unsigned int start, unsigned int count, unsigned int loop, int thread); 416 417 int setupSolidTriangles(int batch, int count); 418 int setupWireframeTriangle(int batch, int count); 419 int setupVertexTriangle(int batch, int count); 420 int setupLines(int batch, int count); 421 int setupPoints(int batch, int count); 422 423 bool setupLine(Primitive &primitive, Triangle &triangle, const DrawCall &draw); 424 bool setupPoint(Primitive &primitive, Triangle &triangle, const DrawCall &draw); 425 426 bool isReadWriteTexture(int sampler); 427 void updateClipper(); 428 void updateConfiguration(bool initialUpdate = false); 429 void initializeThreads(); 430 void terminateThreads(); 431 432 void loadConstants(const VertexShader *vertexShader); 433 void loadConstants(const PixelShader *pixelShader); 434 435 Context *context; 436 Clipper *clipper; 437 Blitter *blitter; 438 Viewport viewport; 439 Rect scissor; 440 int clipFlags; 441 442 Triangle *triangleBatch[16]; 443 Primitive *primitiveBatch[16]; 444 445 // User-defined clipping planes 446 Plane userPlane[MAX_CLIP_PLANES]; 447 Plane clipPlane[MAX_CLIP_PLANES]; // Tranformed to clip space 448 bool updateClipPlanes; 449 450 AtomicInt exitThreads; 451 AtomicInt threadsAwake; 452 Thread *worker[16]; 453 Event *resume[16]; // Events for resuming threads 454 Event *suspend[16]; // Events for suspending threads 455 Event *resumeApp; // Event for resuming the application thread 456 457 PrimitiveProgress primitiveProgress[16]; 458 PixelProgress pixelProgress[16]; 459 Task task[16]; // Current tasks for threads 460 461 enum { 462 DRAW_COUNT = 16, // Number of draw calls buffered (must be power of 2) 463 DRAW_COUNT_BITS = DRAW_COUNT - 1, 464 }; 465 DrawCall *drawCall[DRAW_COUNT]; 466 DrawCall *drawList[DRAW_COUNT]; 467 468 AtomicInt currentDraw; 469 AtomicInt nextDraw; 470 471 enum { 472 TASK_COUNT = 32, // Size of the task queue (must be power of 2) 473 TASK_COUNT_BITS = TASK_COUNT - 1, 474 }; 475 Task taskQueue[TASK_COUNT]; 476 AtomicInt qHead; 477 AtomicInt qSize; 478 479 static AtomicInt unitCount; 480 static AtomicInt clusterCount; 481 482 MutexLock schedulerMutex; 483 484 #if PERF_HUD 485 int64_t vertexTime[16]; 486 int64_t setupTime[16]; 487 int64_t pixelTime[16]; 488 #endif 489 490 VertexTask *vertexTask[16]; 491 492 SwiftConfig *swiftConfig; 493 494 std::list<Query*> queries; 495 Resource *sync; 496 497 VertexProcessor::State vertexState; 498 SetupProcessor::State setupState; 499 PixelProcessor::State pixelState; 500 501 Routine *vertexRoutine; 502 Routine *setupRoutine; 503 Routine *pixelRoutine; 504 }; 505 } 506 507 #endif // sw_Renderer_hpp 508
