1 //
2 // Copyright (c) 2014 The ANGLE Project Authors. All rights reserved.
3 // Use of this source code is governed by a BSD-style license that can be
4 // found in the LICENSE file.
5 //
6 // DynamicHLSL.cpp: Implementation for link and run-time HLSL generation
7 //
8
9 #include "precompiled.h"
10
11 #include "libGLESv2/DynamicHLSL.h"
12 #include "libGLESv2/Shader.h"
13 #include "libGLESv2/Program.h"
14 #include "libGLESv2/renderer/Renderer.h"
15 #include "common/utilities.h"
16 #include "libGLESv2/ProgramBinary.h"
17 #include "libGLESv2/formatutils.h"
18 #include "common/blocklayout.h"
19
Str(int i)20 static std::string Str(int i)
21 {
22 char buffer[20];
23 snprintf(buffer, sizeof(buffer), "%d", i);
24 return buffer;
25 }
26
27 namespace gl_d3d
28 {
29
HLSLComponentTypeString(GLenum componentType)30 std::string HLSLComponentTypeString(GLenum componentType)
31 {
32 switch (componentType)
33 {
34 case GL_UNSIGNED_INT: return "uint";
35 case GL_INT: return "int";
36 case GL_UNSIGNED_NORMALIZED:
37 case GL_SIGNED_NORMALIZED:
38 case GL_FLOAT: return "float";
39 default: UNREACHABLE(); return "not-component-type";
40 }
41 }
42
HLSLComponentTypeString(GLenum componentType,int componentCount)43 std::string HLSLComponentTypeString(GLenum componentType, int componentCount)
44 {
45 return HLSLComponentTypeString(componentType) + (componentCount > 1 ? Str(componentCount) : "");
46 }
47
HLSLMatrixTypeString(GLenum type)48 std::string HLSLMatrixTypeString(GLenum type)
49 {
50 switch (type)
51 {
52 case GL_FLOAT_MAT2: return "float2x2";
53 case GL_FLOAT_MAT3: return "float3x3";
54 case GL_FLOAT_MAT4: return "float4x4";
55 case GL_FLOAT_MAT2x3: return "float2x3";
56 case GL_FLOAT_MAT3x2: return "float3x2";
57 case GL_FLOAT_MAT2x4: return "float2x4";
58 case GL_FLOAT_MAT4x2: return "float4x2";
59 case GL_FLOAT_MAT3x4: return "float3x4";
60 case GL_FLOAT_MAT4x3: return "float4x3";
61 default: UNREACHABLE(); return "not-matrix-type";
62 }
63 }
64
HLSLTypeString(GLenum type)65 std::string HLSLTypeString(GLenum type)
66 {
67 if (gl::IsMatrixType(type))
68 {
69 return HLSLMatrixTypeString(type);
70 }
71
72 return HLSLComponentTypeString(gl::UniformComponentType(type), gl::UniformComponentCount(type));
73 }
74
75 }
76
77 namespace gl
78 {
79
ArrayString(unsigned int i)80 std::string ArrayString(unsigned int i)
81 {
82 return (i == GL_INVALID_INDEX ? "" : "[" + Str(i) + "]");
83 }
84
85 const std::string DynamicHLSL::VERTEX_ATTRIBUTE_STUB_STRING = "@@ VERTEX ATTRIBUTES @@";
86
DynamicHLSL(rx::Renderer * const renderer)87 DynamicHLSL::DynamicHLSL(rx::Renderer *const renderer)
88 : mRenderer(renderer)
89 {
90 }
91
packVarying(PackedVarying * varying,const int maxVaryingVectors,VaryingPacking packing)92 static bool packVarying(PackedVarying *varying, const int maxVaryingVectors, VaryingPacking packing)
93 {
94 GLenum transposedType = TransposeMatrixType(varying->type);
95
96 // matrices within varying structs are not transposed
97 int registers = (varying->isStruct() ? HLSLVariableRegisterCount(*varying) : VariableRowCount(transposedType)) * varying->elementCount();
98 int elements = (varying->isStruct() ? 4 : VariableColumnCount(transposedType));
99
100 if (elements >= 2 && elements <= 4)
101 {
102 for (int r = 0; r <= maxVaryingVectors - registers; r++)
103 {
104 bool available = true;
105
106 for (int y = 0; y < registers && available; y++)
107 {
108 for (int x = 0; x < elements && available; x++)
109 {
110 if (packing[r + y][x])
111 {
112 available = false;
113 }
114 }
115 }
116
117 if (available)
118 {
119 varying->registerIndex = r;
120
121 for (int y = 0; y < registers; y++)
122 {
123 for (int x = 0; x < elements; x++)
124 {
125 packing[r + y][x] = &*varying;
126 }
127 }
128
129 return true;
130 }
131 }
132
133 if (elements == 2)
134 {
135 for (int r = maxVaryingVectors - registers; r >= 0; r--)
136 {
137 bool available = true;
138
139 for (int y = 0; y < registers && available; y++)
140 {
141 for (int x = 2; x < 4 && available; x++)
142 {
143 if (packing[r + y][x])
144 {
145 available = false;
146 }
147 }
148 }
149
150 if (available)
151 {
152 varying->registerIndex = r;
153
154 for (int y = 0; y < registers; y++)
155 {
156 for (int x = 2; x < 4; x++)
157 {
158 packing[r + y][x] = &*varying;
159 }
160 }
161
162 return true;
163 }
164 }
165 }
166 }
167 else if (elements == 1)
168 {
169 int space[4] = { 0 };
170
171 for (int y = 0; y < maxVaryingVectors; y++)
172 {
173 for (int x = 0; x < 4; x++)
174 {
175 space[x] += packing[y][x] ? 0 : 1;
176 }
177 }
178
179 int column = 0;
180
181 for (int x = 0; x < 4; x++)
182 {
183 if (space[x] >= registers && space[x] < space[column])
184 {
185 column = x;
186 }
187 }
188
189 if (space[column] >= registers)
190 {
191 for (int r = 0; r < maxVaryingVectors; r++)
192 {
193 if (!packing[r][column])
194 {
195 varying->registerIndex = r;
196
197 for (int y = r; y < r + registers; y++)
198 {
199 packing[y][column] = &*varying;
200 }
201
202 break;
203 }
204 }
205
206 return true;
207 }
208 }
209 else UNREACHABLE();
210
211 return false;
212 }
213
214 // Packs varyings into generic varying registers, using the algorithm from [OpenGL ES Shading Language 1.00 rev. 17] appendix A section 7 page 111
215 // Returns the number of used varying registers, or -1 if unsuccesful
packVaryings(InfoLog & infoLog,VaryingPacking packing,FragmentShader * fragmentShader,VertexShader * vertexShader,const std::vector<std::string> & transformFeedbackVaryings)216 int DynamicHLSL::packVaryings(InfoLog &infoLog, VaryingPacking packing, FragmentShader *fragmentShader,
217 VertexShader *vertexShader, const std::vector<std::string>& transformFeedbackVaryings)
218 {
219 const int maxVaryingVectors = mRenderer->getMaxVaryingVectors();
220
221 vertexShader->resetVaryingsRegisterAssignment();
222 fragmentShader->resetVaryingsRegisterAssignment();
223
224 std::set<std::string> packedVaryings;
225
226 for (unsigned int varyingIndex = 0; varyingIndex < fragmentShader->mVaryings.size(); varyingIndex++)
227 {
228 PackedVarying *varying = &fragmentShader->mVaryings[varyingIndex];
229 if (packVarying(varying, maxVaryingVectors, packing))
230 {
231 packedVaryings.insert(varying->name);
232 }
233 else
234 {
235 infoLog.append("Could not pack varying %s", varying->name.c_str());
236 return -1;
237 }
238 }
239
240 for (unsigned int feedbackVaryingIndex = 0; feedbackVaryingIndex < transformFeedbackVaryings.size(); feedbackVaryingIndex++)
241 {
242 const std::string &transformFeedbackVarying = transformFeedbackVaryings[feedbackVaryingIndex];
243 if (packedVaryings.find(transformFeedbackVarying) == packedVaryings.end())
244 {
245 bool found = false;
246 for (unsigned int varyingIndex = 0; varyingIndex < vertexShader->mVaryings.size(); varyingIndex++)
247 {
248 PackedVarying *varying = &vertexShader->mVaryings[varyingIndex];
249 if (transformFeedbackVarying == varying->name)
250 {
251 if (!packVarying(varying, maxVaryingVectors, packing))
252 {
253 infoLog.append("Could not pack varying %s", varying->name.c_str());
254 return -1;
255 }
256
257 found = true;
258 break;
259 }
260 }
261
262 if (!found && transformFeedbackVarying != "gl_Position" && transformFeedbackVarying != "gl_PointSize")
263 {
264 infoLog.append("Transform feedback varying %s does not exist in the vertex shader.", transformFeedbackVarying.c_str());
265 return -1;
266 }
267 }
268 }
269
270 // Return the number of used registers
271 int registers = 0;
272
273 for (int r = 0; r < maxVaryingVectors; r++)
274 {
275 if (packing[r][0] || packing[r][1] || packing[r][2] || packing[r][3])
276 {
277 registers++;
278 }
279 }
280
281 return registers;
282 }
283
generateVaryingHLSL(VertexShader * shader,const std::string & varyingSemantic,std::vector<LinkedVarying> * linkedVaryings) const284 std::string DynamicHLSL::generateVaryingHLSL(VertexShader *shader, const std::string &varyingSemantic,
285 std::vector<LinkedVarying> *linkedVaryings) const
286 {
287 std::string varyingHLSL;
288
289 for (unsigned int varyingIndex = 0; varyingIndex < shader->mVaryings.size(); varyingIndex++)
290 {
291 const PackedVarying &varying = shader->mVaryings[varyingIndex];
292 if (varying.registerAssigned())
293 {
294 GLenum transposedType = TransposeMatrixType(varying.type);
295 int variableRows = (varying.isStruct() ? 1 : VariableRowCount(transposedType));
296
297 for (unsigned int elementIndex = 0; elementIndex < varying.elementCount(); elementIndex++)
298 {
299 for (int row = 0; row < variableRows; row++)
300 {
301 switch (varying.interpolation)
302 {
303 case INTERPOLATION_SMOOTH: varyingHLSL += " "; break;
304 case INTERPOLATION_FLAT: varyingHLSL += " nointerpolation "; break;
305 case INTERPOLATION_CENTROID: varyingHLSL += " centroid "; break;
306 default: UNREACHABLE();
307 }
308
309 unsigned int semanticIndex = elementIndex * variableRows + varying.registerIndex + row;
310 std::string n = Str(semanticIndex);
311
312 std::string typeString;
313
314 if (varying.isStruct())
315 {
316 // matrices within structs are not transposed, so
317 // do not use the special struct prefix "rm"
318 typeString = decorateVariable(varying.structName);
319 }
320 else
321 {
322 GLenum componentType = UniformComponentType(transposedType);
323 int columnCount = VariableColumnCount(transposedType);
324 typeString = gl_d3d::HLSLComponentTypeString(componentType, columnCount);
325 }
326 varyingHLSL += typeString + " v" + n + " : " + varyingSemantic + n + ";\n";
327 }
328 }
329
330 if (linkedVaryings)
331 {
332 linkedVaryings->push_back(LinkedVarying(varying.name, varying.type, varying.elementCount(),
333 varyingSemantic, varying.registerIndex,
334 variableRows * varying.elementCount()));
335 }
336 }
337 }
338
339 return varyingHLSL;
340 }
341
generateInputLayoutHLSL(const VertexFormat inputLayout[],const Attribute shaderAttributes[]) const342 std::string DynamicHLSL::generateInputLayoutHLSL(const VertexFormat inputLayout[], const Attribute shaderAttributes[]) const
343 {
344 std::string structHLSL, initHLSL;
345
346 int semanticIndex = 0;
347 unsigned int inputIndex = 0;
348
349 for (unsigned int attributeIndex = 0; attributeIndex < MAX_VERTEX_ATTRIBS; attributeIndex++)
350 {
351 ASSERT(inputIndex < MAX_VERTEX_ATTRIBS);
352
353 const VertexFormat &vertexFormat = inputLayout[inputIndex];
354 const Attribute &shaderAttribute = shaderAttributes[attributeIndex];
355
356 if (!shaderAttribute.name.empty())
357 {
358 // HLSL code for input structure
359 if (IsMatrixType(shaderAttribute.type))
360 {
361 // Matrix types are always transposed
362 structHLSL += " " + gl_d3d::HLSLMatrixTypeString(TransposeMatrixType(shaderAttribute.type));
363 }
364 else
365 {
366 GLenum componentType = mRenderer->getVertexComponentType(vertexFormat);
367 structHLSL += " " + gl_d3d::HLSLComponentTypeString(componentType, UniformComponentCount(shaderAttribute.type));
368 }
369
370 structHLSL += " " + decorateVariable(shaderAttribute.name) + " : TEXCOORD" + Str(semanticIndex) + ";\n";
371 semanticIndex += AttributeRegisterCount(shaderAttribute.type);
372
373 // HLSL code for initialization
374 initHLSL += " " + decorateVariable(shaderAttribute.name) + " = ";
375
376 // Mismatched vertex attribute to vertex input may result in an undefined
377 // data reinterpretation (eg for pure integer->float, float->pure integer)
378 // TODO: issue warning with gl debug info extension, when supported
379 if (IsMatrixType(shaderAttribute.type) ||
380 (mRenderer->getVertexConversionType(vertexFormat) & rx::VERTEX_CONVERT_GPU) != 0)
381 {
382 initHLSL += generateAttributeConversionHLSL(vertexFormat, shaderAttribute);
383 }
384 else
385 {
386 initHLSL += "input." + decorateVariable(shaderAttribute.name);
387 }
388
389 initHLSL += ";\n";
390
391 inputIndex += VariableRowCount(TransposeMatrixType(shaderAttribute.type));
392 }
393 }
394
395 return "struct VS_INPUT\n"
396 "{\n" +
397 structHLSL +
398 "};\n"
399 "\n"
400 "void initAttributes(VS_INPUT input)\n"
401 "{\n" +
402 initHLSL +
403 "}\n";
404 }
405
generateShaderLinkHLSL(InfoLog & infoLog,int registers,const VaryingPacking packing,std::string & pixelHLSL,std::string & vertexHLSL,FragmentShader * fragmentShader,VertexShader * vertexShader,const std::vector<std::string> & transformFeedbackVaryings,std::vector<LinkedVarying> * linkedVaryings,std::map<int,VariableLocation> * programOutputVars) const406 bool DynamicHLSL::generateShaderLinkHLSL(InfoLog &infoLog, int registers, const VaryingPacking packing,
407 std::string& pixelHLSL, std::string& vertexHLSL,
408 FragmentShader *fragmentShader, VertexShader *vertexShader,
409 const std::vector<std::string>& transformFeedbackVaryings,
410 std::vector<LinkedVarying> *linkedVaryings,
411 std::map<int, VariableLocation> *programOutputVars) const
412 {
413 if (pixelHLSL.empty() || vertexHLSL.empty())
414 {
415 return false;
416 }
417
418 bool usesMRT = fragmentShader->mUsesMultipleRenderTargets;
419 bool usesFragColor = fragmentShader->mUsesFragColor;
420 bool usesFragData = fragmentShader->mUsesFragData;
421 if (usesFragColor && usesFragData)
422 {
423 infoLog.append("Cannot use both gl_FragColor and gl_FragData in the same fragment shader.");
424 return false;
425 }
426
427 // Write the HLSL input/output declarations
428 const int shaderModel = mRenderer->getMajorShaderModel();
429 const int maxVaryingVectors = mRenderer->getMaxVaryingVectors();
430
431 const int registersNeeded = registers + (fragmentShader->mUsesFragCoord ? 1 : 0) + (fragmentShader->mUsesPointCoord ? 1 : 0);
432
433 // Two cases when writing to gl_FragColor and using ESSL 1.0:
434 // - with a 3.0 context, the output color is copied to channel 0
435 // - with a 2.0 context, the output color is broadcast to all channels
436 const bool broadcast = (fragmentShader->mUsesFragColor && mRenderer->getCurrentClientVersion() < 3);
437 const unsigned int numRenderTargets = (broadcast || usesMRT ? mRenderer->getMaxRenderTargets() : 1);
438
439 int shaderVersion = vertexShader->getShaderVersion();
440
441 if (registersNeeded > maxVaryingVectors)
442 {
443 infoLog.append("No varying registers left to support gl_FragCoord/gl_PointCoord");
444
445 return false;
446 }
447
448 std::string varyingSemantic = (vertexShader->mUsesPointSize && shaderModel == 3) ? "COLOR" : "TEXCOORD";
449 std::string targetSemantic = (shaderModel >= 4) ? "SV_Target" : "COLOR";
450 std::string dxPositionSemantic = (shaderModel >= 4) ? "SV_Position" : "POSITION";
451 std::string depthSemantic = (shaderModel >= 4) ? "SV_Depth" : "DEPTH";
452
453 std::string varyingHLSL = generateVaryingHLSL(vertexShader, varyingSemantic, linkedVaryings);
454
455 // special varyings that use reserved registers
456 int reservedRegisterIndex = registers;
457
458 unsigned int glPositionSemanticIndex = reservedRegisterIndex++;
459 std::string glPositionSemantic = varyingSemantic;
460
461 std::string fragCoordSemantic;
462 unsigned int fragCoordSemanticIndex = 0;
463 if (fragmentShader->mUsesFragCoord)
464 {
465 fragCoordSemanticIndex = reservedRegisterIndex++;
466 fragCoordSemantic = varyingSemantic;
467 }
468
469 std::string pointCoordSemantic;
470 unsigned int pointCoordSemanticIndex = 0;
471 if (fragmentShader->mUsesPointCoord)
472 {
473 // Shader model 3 uses a special TEXCOORD semantic for point sprite texcoords.
474 // In DX11 we compute this in the GS.
475 if (shaderModel == 3)
476 {
477 pointCoordSemanticIndex = 0;
478 pointCoordSemantic = "TEXCOORD0";
479 }
480 else if (shaderModel >= 4)
481 {
482 pointCoordSemanticIndex = reservedRegisterIndex++;
483 pointCoordSemantic = varyingSemantic;
484 }
485 }
486
487 // Add stub string to be replaced when shader is dynamically defined by its layout
488 vertexHLSL += "\n" + VERTEX_ATTRIBUTE_STUB_STRING + "\n";
489
490 vertexHLSL += "struct VS_OUTPUT\n"
491 "{\n";
492
493 if (shaderModel < 4)
494 {
495 vertexHLSL += " float4 _dx_Position : " + dxPositionSemantic + ";\n";
496 vertexHLSL += " float4 gl_Position : " + glPositionSemantic + Str(glPositionSemanticIndex) + ";\n";
497 linkedVaryings->push_back(LinkedVarying("gl_Position", GL_FLOAT_VEC4, 1, glPositionSemantic, glPositionSemanticIndex, 1));
498
499 }
500
501 vertexHLSL += varyingHLSL;
502
503 if (fragmentShader->mUsesFragCoord)
504 {
505 vertexHLSL += " float4 gl_FragCoord : " + fragCoordSemantic + Str(fragCoordSemanticIndex) + ";\n";
506 linkedVaryings->push_back(LinkedVarying("gl_FragCoord", GL_FLOAT_VEC4, 1, fragCoordSemantic, fragCoordSemanticIndex, 1));
507 }
508
509 if (vertexShader->mUsesPointSize && shaderModel >= 3)
510 {
511 vertexHLSL += " float gl_PointSize : PSIZE;\n";
512 linkedVaryings->push_back(LinkedVarying("gl_PointSize", GL_FLOAT, 1, "PSIZE", 0, 1));
513 }
514
515 if (shaderModel >= 4)
516 {
517 vertexHLSL += " float4 _dx_Position : " + dxPositionSemantic + ";\n";
518 vertexHLSL += " float4 gl_Position : " + glPositionSemantic + Str(glPositionSemanticIndex) + ";\n";
519 linkedVaryings->push_back(LinkedVarying("gl_Position", GL_FLOAT_VEC4, 1, glPositionSemantic, glPositionSemanticIndex, 1));
520 }
521
522 vertexHLSL += "};\n"
523 "\n"
524 "VS_OUTPUT main(VS_INPUT input)\n"
525 "{\n"
526 " initAttributes(input);\n";
527
528 if (shaderModel >= 4)
529 {
530 vertexHLSL += "\n"
531 " gl_main();\n"
532 "\n"
533 " VS_OUTPUT output;\n"
534 " output.gl_Position = gl_Position;\n"
535 " output._dx_Position.x = gl_Position.x;\n"
536 " output._dx_Position.y = -gl_Position.y;\n"
537 " output._dx_Position.z = (gl_Position.z + gl_Position.w) * 0.5;\n"
538 " output._dx_Position.w = gl_Position.w;\n";
539 }
540 else
541 {
542 vertexHLSL += "\n"
543 " gl_main();\n"
544 "\n"
545 " VS_OUTPUT output;\n"
546 " output.gl_Position = gl_Position;\n"
547 " output._dx_Position.x = gl_Position.x * dx_ViewAdjust.z + dx_ViewAdjust.x * gl_Position.w;\n"
548 " output._dx_Position.y = -(gl_Position.y * dx_ViewAdjust.w + dx_ViewAdjust.y * gl_Position.w);\n"
549 " output._dx_Position.z = (gl_Position.z + gl_Position.w) * 0.5;\n"
550 " output._dx_Position.w = gl_Position.w;\n";
551 }
552
553 if (vertexShader->mUsesPointSize && shaderModel >= 3)
554 {
555 vertexHLSL += " output.gl_PointSize = gl_PointSize;\n";
556 }
557
558 if (fragmentShader->mUsesFragCoord)
559 {
560 vertexHLSL += " output.gl_FragCoord = gl_Position;\n";
561 }
562
563 for (unsigned int vertVaryingIndex = 0; vertVaryingIndex < vertexShader->mVaryings.size(); vertVaryingIndex++)
564 {
565 const PackedVarying &varying = vertexShader->mVaryings[vertVaryingIndex];
566 if (varying.registerAssigned())
567 {
568 for (unsigned int elementIndex = 0; elementIndex < varying.elementCount(); elementIndex++)
569 {
570 int variableRows = (varying.isStruct() ? 1 : VariableRowCount(TransposeMatrixType(varying.type)));
571
572 for (int row = 0; row < variableRows; row++)
573 {
574 int r = varying.registerIndex + elementIndex * variableRows + row;
575 vertexHLSL += " output.v" + Str(r);
576
577 bool sharedRegister = false; // Register used by multiple varyings
578
579 for (int x = 0; x < 4; x++)
580 {
581 if (packing[r][x] && packing[r][x] != packing[r][0])
582 {
583 sharedRegister = true;
584 break;
585 }
586 }
587
588 if(sharedRegister)
589 {
590 vertexHLSL += ".";
591
592 for (int x = 0; x < 4; x++)
593 {
594 if (packing[r][x] == &varying)
595 {
596 switch(x)
597 {
598 case 0: vertexHLSL += "x"; break;
599 case 1: vertexHLSL += "y"; break;
600 case 2: vertexHLSL += "z"; break;
601 case 3: vertexHLSL += "w"; break;
602 }
603 }
604 }
605 }
606
607 vertexHLSL += " = _" + varying.name;
608
609 if (varying.isArray())
610 {
611 vertexHLSL += ArrayString(elementIndex);
612 }
613
614 if (variableRows > 1)
615 {
616 vertexHLSL += ArrayString(row);
617 }
618
619 vertexHLSL += ";\n";
620 }
621 }
622 }
623 }
624
625 vertexHLSL += "\n"
626 " return output;\n"
627 "}\n";
628
629 pixelHLSL += "struct PS_INPUT\n"
630 "{\n";
631
632 pixelHLSL += varyingHLSL;
633
634 if (fragmentShader->mUsesFragCoord)
635 {
636 pixelHLSL += " float4 gl_FragCoord : " + fragCoordSemantic + Str(fragCoordSemanticIndex) + ";\n";
637 }
638
639 if (fragmentShader->mUsesPointCoord && shaderModel >= 3)
640 {
641 pixelHLSL += " float2 gl_PointCoord : " + pointCoordSemantic + Str(pointCoordSemanticIndex) + ";\n";
642 }
643
644 // Must consume the PSIZE element if the geometry shader is not active
645 // We won't know if we use a GS until we draw
646 if (vertexShader->mUsesPointSize && shaderModel >= 4)
647 {
648 pixelHLSL += " float gl_PointSize : PSIZE;\n";
649 }
650
651 if (fragmentShader->mUsesFragCoord)
652 {
653 if (shaderModel >= 4)
654 {
655 pixelHLSL += " float4 dx_VPos : SV_Position;\n";
656 }
657 else if (shaderModel >= 3)
658 {
659 pixelHLSL += " float2 dx_VPos : VPOS;\n";
660 }
661 }
662
663 pixelHLSL += "};\n"
664 "\n"
665 "struct PS_OUTPUT\n"
666 "{\n";
667
668 if (shaderVersion < 300)
669 {
670 for (unsigned int renderTargetIndex = 0; renderTargetIndex < numRenderTargets; renderTargetIndex++)
671 {
672 pixelHLSL += " float4 gl_Color" + Str(renderTargetIndex) + " : " + targetSemantic + Str(renderTargetIndex) + ";\n";
673 }
674
675 if (fragmentShader->mUsesFragDepth)
676 {
677 pixelHLSL += " float gl_Depth : " + depthSemantic + ";\n";
678 }
679 }
680 else
681 {
682 defineOutputVariables(fragmentShader, programOutputVars);
683
684 const std::vector<Attribute> &shaderOutputVars = fragmentShader->getOutputVariables();
685 for (auto locationIt = programOutputVars->begin(); locationIt != programOutputVars->end(); locationIt++)
686 {
687 const VariableLocation &outputLocation = locationIt->second;
688 const ShaderVariable &outputVariable = shaderOutputVars[outputLocation.index];
689 const std::string &elementString = (outputLocation.element == GL_INVALID_INDEX ? "" : Str(outputLocation.element));
690
691 pixelHLSL += " " + gl_d3d::HLSLTypeString(outputVariable.type) +
692 " out_" + outputLocation.name + elementString +
693 " : " + targetSemantic + Str(locationIt->first) + ";\n";
694 }
695 }
696
697 pixelHLSL += "};\n"
698 "\n";
699
700 if (fragmentShader->mUsesFrontFacing)
701 {
702 if (shaderModel >= 4)
703 {
704 pixelHLSL += "PS_OUTPUT main(PS_INPUT input, bool isFrontFace : SV_IsFrontFace)\n"
705 "{\n";
706 }
707 else
708 {
709 pixelHLSL += "PS_OUTPUT main(PS_INPUT input, float vFace : VFACE)\n"
710 "{\n";
711 }
712 }
713 else
714 {
715 pixelHLSL += "PS_OUTPUT main(PS_INPUT input)\n"
716 "{\n";
717 }
718
719 if (fragmentShader->mUsesFragCoord)
720 {
721 pixelHLSL += " float rhw = 1.0 / input.gl_FragCoord.w;\n";
722
723 if (shaderModel >= 4)
724 {
725 pixelHLSL += " gl_FragCoord.x = input.dx_VPos.x;\n"
726 " gl_FragCoord.y = input.dx_VPos.y;\n";
727 }
728 else if (shaderModel >= 3)
729 {
730 pixelHLSL += " gl_FragCoord.x = input.dx_VPos.x + 0.5;\n"
731 " gl_FragCoord.y = input.dx_VPos.y + 0.5;\n";
732 }
733 else
734 {
735 // dx_ViewCoords contains the viewport width/2, height/2, center.x and center.y. See Renderer::setViewport()
736 pixelHLSL += " gl_FragCoord.x = (input.gl_FragCoord.x * rhw) * dx_ViewCoords.x + dx_ViewCoords.z;\n"
737 " gl_FragCoord.y = (input.gl_FragCoord.y * rhw) * dx_ViewCoords.y + dx_ViewCoords.w;\n";
738 }
739
740 pixelHLSL += " gl_FragCoord.z = (input.gl_FragCoord.z * rhw) * dx_DepthFront.x + dx_DepthFront.y;\n"
741 " gl_FragCoord.w = rhw;\n";
742 }
743
744 if (fragmentShader->mUsesPointCoord && shaderModel >= 3)
745 {
746 pixelHLSL += " gl_PointCoord.x = input.gl_PointCoord.x;\n";
747 pixelHLSL += " gl_PointCoord.y = 1.0 - input.gl_PointCoord.y;\n";
748 }
749
750 if (fragmentShader->mUsesFrontFacing)
751 {
752 if (shaderModel <= 3)
753 {
754 pixelHLSL += " gl_FrontFacing = (vFace * dx_DepthFront.z >= 0.0);\n";
755 }
756 else
757 {
758 pixelHLSL += " gl_FrontFacing = isFrontFace;\n";
759 }
760 }
761
762 for (unsigned int varyingIndex = 0; varyingIndex < fragmentShader->mVaryings.size(); varyingIndex++)
763 {
764 const PackedVarying &varying = fragmentShader->mVaryings[varyingIndex];
765 if (varying.registerAssigned())
766 {
767 for (unsigned int elementIndex = 0; elementIndex < varying.elementCount(); elementIndex++)
768 {
769 GLenum transposedType = TransposeMatrixType(varying.type);
770 int variableRows = (varying.isStruct() ? 1 : VariableRowCount(transposedType));
771 for (int row = 0; row < variableRows; row++)
772 {
773 std::string n = Str(varying.registerIndex + elementIndex * variableRows + row);
774 pixelHLSL += " _" + varying.name;
775
776 if (varying.isArray())
777 {
778 pixelHLSL += ArrayString(elementIndex);
779 }
780
781 if (variableRows > 1)
782 {
783 pixelHLSL += ArrayString(row);
784 }
785
786 if (varying.isStruct())
787 {
788 pixelHLSL += " = input.v" + n + ";\n"; break;
789 }
790 else
791 {
792 switch (VariableColumnCount(transposedType))
793 {
794 case 1: pixelHLSL += " = input.v" + n + ".x;\n"; break;
795 case 2: pixelHLSL += " = input.v" + n + ".xy;\n"; break;
796 case 3: pixelHLSL += " = input.v" + n + ".xyz;\n"; break;
797 case 4: pixelHLSL += " = input.v" + n + ";\n"; break;
798 default: UNREACHABLE();
799 }
800 }
801 }
802 }
803 }
804 else UNREACHABLE();
805 }
806
807 pixelHLSL += "\n"
808 " gl_main();\n"
809 "\n"
810 " PS_OUTPUT output;\n";
811
812 if (shaderVersion < 300)
813 {
814 for (unsigned int renderTargetIndex = 0; renderTargetIndex < numRenderTargets; renderTargetIndex++)
815 {
816 unsigned int sourceColorIndex = broadcast ? 0 : renderTargetIndex;
817
818 pixelHLSL += " output.gl_Color" + Str(renderTargetIndex) + " = gl_Color[" + Str(sourceColorIndex) + "];\n";
819 }
820
821 if (fragmentShader->mUsesFragDepth)
822 {
823 pixelHLSL += " output.gl_Depth = gl_Depth;\n";
824 }
825 }
826 else
827 {
828 for (auto locationIt = programOutputVars->begin(); locationIt != programOutputVars->end(); locationIt++)
829 {
830 const VariableLocation &outputLocation = locationIt->second;
831 const std::string &variableName = "out_" + outputLocation.name;
832 const std::string &outVariableName = variableName + (outputLocation.element == GL_INVALID_INDEX ? "" : Str(outputLocation.element));
833 const std::string &staticVariableName = variableName + ArrayString(outputLocation.element);
834
835 pixelHLSL += " output." + outVariableName + " = " + staticVariableName + ";\n";
836 }
837 }
838
839 pixelHLSL += "\n"
840 " return output;\n"
841 "}\n";
842
843 return true;
844 }
845
defineOutputVariables(FragmentShader * fragmentShader,std::map<int,VariableLocation> * programOutputVars) const846 void DynamicHLSL::defineOutputVariables(FragmentShader *fragmentShader, std::map<int, VariableLocation> *programOutputVars) const
847 {
848 const std::vector<Attribute> &shaderOutputVars = fragmentShader->getOutputVariables();
849
850 for (unsigned int outputVariableIndex = 0; outputVariableIndex < shaderOutputVars.size(); outputVariableIndex++)
851 {
852 const Attribute &outputVariable = shaderOutputVars[outputVariableIndex];
853 const int baseLocation = outputVariable.location == -1 ? 0 : outputVariable.location;
854
855 if (outputVariable.arraySize > 0)
856 {
857 for (unsigned int elementIndex = 0; elementIndex < outputVariable.arraySize; elementIndex++)
858 {
859 const int location = baseLocation + elementIndex;
860 ASSERT(programOutputVars->count(location) == 0);
861 (*programOutputVars)[location] = VariableLocation(outputVariable.name, elementIndex, outputVariableIndex);
862 }
863 }
864 else
865 {
866 ASSERT(programOutputVars->count(baseLocation) == 0);
867 (*programOutputVars)[baseLocation] = VariableLocation(outputVariable.name, GL_INVALID_INDEX, outputVariableIndex);
868 }
869 }
870 }
871
generateGeometryShaderHLSL(int registers,FragmentShader * fragmentShader,VertexShader * vertexShader) const872 std::string DynamicHLSL::generateGeometryShaderHLSL(int registers, FragmentShader *fragmentShader, VertexShader *vertexShader) const
873 {
874 // for now we only handle point sprite emulation
875 ASSERT(vertexShader->mUsesPointSize && mRenderer->getMajorShaderModel() >= 4);
876 return generatePointSpriteHLSL(registers, fragmentShader, vertexShader);
877 }
878
generatePointSpriteHLSL(int registers,FragmentShader * fragmentShader,VertexShader * vertexShader) const879 std::string DynamicHLSL::generatePointSpriteHLSL(int registers, FragmentShader *fragmentShader, VertexShader *vertexShader) const
880 {
881 ASSERT(registers >= 0);
882 ASSERT(vertexShader->mUsesPointSize);
883 ASSERT(mRenderer->getMajorShaderModel() >= 4);
884
885 std::string geomHLSL;
886
887 std::string varyingSemantic = "TEXCOORD";
888
889 std::string fragCoordSemantic;
890 std::string pointCoordSemantic;
891
892 int reservedRegisterIndex = registers;
893
894 if (fragmentShader->mUsesFragCoord)
895 {
896 fragCoordSemantic = varyingSemantic + Str(reservedRegisterIndex++);
897 }
898
899 if (fragmentShader->mUsesPointCoord)
900 {
901 pointCoordSemantic = varyingSemantic + Str(reservedRegisterIndex++);
902 }
903
904 geomHLSL += "uniform float4 dx_ViewCoords : register(c1);\n"
905 "\n"
906 "struct GS_INPUT\n"
907 "{\n";
908
909 std::string varyingHLSL = generateVaryingHLSL(vertexShader, varyingSemantic, NULL);
910
911 geomHLSL += varyingHLSL;
912
913 if (fragmentShader->mUsesFragCoord)
914 {
915 geomHLSL += " float4 gl_FragCoord : " + fragCoordSemantic + ";\n";
916 }
917
918 geomHLSL += " float gl_PointSize : PSIZE;\n"
919 " float4 gl_Position : SV_Position;\n"
920 "};\n"
921 "\n"
922 "struct GS_OUTPUT\n"
923 "{\n";
924
925 geomHLSL += varyingHLSL;
926
927 if (fragmentShader->mUsesFragCoord)
928 {
929 geomHLSL += " float4 gl_FragCoord : " + fragCoordSemantic + ";\n";
930 }
931
932 if (fragmentShader->mUsesPointCoord)
933 {
934 geomHLSL += " float2 gl_PointCoord : " + pointCoordSemantic + ";\n";
935 }
936
937 geomHLSL += " float gl_PointSize : PSIZE;\n"
938 " float4 gl_Position : SV_Position;\n"
939 "};\n"
940 "\n"
941 "static float2 pointSpriteCorners[] = \n"
942 "{\n"
943 " float2( 0.5f, -0.5f),\n"
944 " float2( 0.5f, 0.5f),\n"
945 " float2(-0.5f, -0.5f),\n"
946 " float2(-0.5f, 0.5f)\n"
947 "};\n"
948 "\n"
949 "static float2 pointSpriteTexcoords[] = \n"
950 "{\n"
951 " float2(1.0f, 1.0f),\n"
952 " float2(1.0f, 0.0f),\n"
953 " float2(0.0f, 1.0f),\n"
954 " float2(0.0f, 0.0f)\n"
955 "};\n"
956 "\n"
957 "static float minPointSize = " + Str(ALIASED_POINT_SIZE_RANGE_MIN) + ".0f;\n"
958 "static float maxPointSize = " + Str(mRenderer->getMaxPointSize()) + ".0f;\n"
959 "\n"
960 "[maxvertexcount(4)]\n"
961 "void main(point GS_INPUT input[1], inout TriangleStream<GS_OUTPUT> outStream)\n"
962 "{\n"
963 " GS_OUTPUT output = (GS_OUTPUT)0;\n"
964 " output.gl_PointSize = input[0].gl_PointSize;\n";
965
966 for (int r = 0; r < registers; r++)
967 {
968 geomHLSL += " output.v" + Str(r) + " = input[0].v" + Str(r) + ";\n";
969 }
970
971 if (fragmentShader->mUsesFragCoord)
972 {
973 geomHLSL += " output.gl_FragCoord = input[0].gl_FragCoord;\n";
974 }
975
976 geomHLSL += " \n"
977 " float gl_PointSize = clamp(input[0].gl_PointSize, minPointSize, maxPointSize);\n"
978 " float4 gl_Position = input[0].gl_Position;\n"
979 " float2 viewportScale = float2(1.0f / dx_ViewCoords.x, 1.0f / dx_ViewCoords.y) * gl_Position.w;\n";
980
981 for (int corner = 0; corner < 4; corner++)
982 {
983 geomHLSL += " \n"
984 " output.gl_Position = gl_Position + float4(pointSpriteCorners[" + Str(corner) + "] * viewportScale * gl_PointSize, 0.0f, 0.0f);\n";
985
986 if (fragmentShader->mUsesPointCoord)
987 {
988 geomHLSL += " output.gl_PointCoord = pointSpriteTexcoords[" + Str(corner) + "];\n";
989 }
990
991 geomHLSL += " outStream.Append(output);\n";
992 }
993
994 geomHLSL += " \n"
995 " outStream.RestartStrip();\n"
996 "}\n";
997
998 return geomHLSL;
999 }
1000
1001 // This method needs to match OutputHLSL::decorate
decorateVariable(const std::string & name)1002 std::string DynamicHLSL::decorateVariable(const std::string &name)
1003 {
1004 if (name.compare(0, 3, "gl_") != 0 && name.compare(0, 3, "dx_") != 0)
1005 {
1006 return "_" + name;
1007 }
1008
1009 return name;
1010 }
1011
generateAttributeConversionHLSL(const VertexFormat & vertexFormat,const ShaderVariable & shaderAttrib) const1012 std::string DynamicHLSL::generateAttributeConversionHLSL(const VertexFormat &vertexFormat, const ShaderVariable &shaderAttrib) const
1013 {
1014 std::string attribString = "input." + decorateVariable(shaderAttrib.name);
1015
1016 // Matrix
1017 if (IsMatrixType(shaderAttrib.type))
1018 {
1019 return "transpose(" + attribString + ")";
1020 }
1021
1022 GLenum shaderComponentType = UniformComponentType(shaderAttrib.type);
1023 int shaderComponentCount = UniformComponentCount(shaderAttrib.type);
1024
1025 // Perform integer to float conversion (if necessary)
1026 bool requiresTypeConversion = (shaderComponentType == GL_FLOAT && vertexFormat.mType != GL_FLOAT);
1027
1028 if (requiresTypeConversion)
1029 {
1030 // TODO: normalization for 32-bit integer formats
1031 ASSERT(!vertexFormat.mNormalized && !vertexFormat.mPureInteger);
1032 return "float" + Str(shaderComponentCount) + "(" + attribString + ")";
1033 }
1034
1035 // No conversion necessary
1036 return attribString;
1037 }
1038
getInputLayoutSignature(const VertexFormat inputLayout[],GLenum signature[]) const1039 void DynamicHLSL::getInputLayoutSignature(const VertexFormat inputLayout[], GLenum signature[]) const
1040 {
1041 for (size_t inputIndex = 0; inputIndex < MAX_VERTEX_ATTRIBS; inputIndex++)
1042 {
1043 const VertexFormat &vertexFormat = inputLayout[inputIndex];
1044
1045 if (vertexFormat.mType == GL_NONE)
1046 {
1047 signature[inputIndex] = GL_NONE;
1048 }
1049 else
1050 {
1051 bool gpuConverted = ((mRenderer->getVertexConversionType(vertexFormat) & rx::VERTEX_CONVERT_GPU) != 0);
1052 signature[inputIndex] = (gpuConverted ? GL_TRUE : GL_FALSE);
1053 }
1054 }
1055 }
1056
1057 }
1058