1 //
2 // Copyright 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 "libANGLE/renderer/d3d/DynamicHLSL.h"
10
11 #include "common/string_utils.h"
12 #include "common/utilities.h"
13 #include "compiler/translator/blocklayoutHLSL.h"
14 #include "libANGLE/Context.h"
15 #include "libANGLE/Program.h"
16 #include "libANGLE/Shader.h"
17 #include "libANGLE/VaryingPacking.h"
18 #include "libANGLE/formatutils.h"
19 #include "libANGLE/renderer/d3d/ProgramD3D.h"
20 #include "libANGLE/renderer/d3d/RendererD3D.h"
21 #include "libANGLE/renderer/d3d/ShaderD3D.h"
22
23 using namespace gl;
24
25 namespace rx
26 {
27
28 namespace
29 {
30
31 // kShaderStorageDeclarationString must be the same as outputHLSL.
32 constexpr const char kShaderStorageDeclarationString[] =
33 "// @@ SHADER STORAGE DECLARATION STRING @@";
34
HLSLComponentTypeString(GLenum componentType)35 const char *HLSLComponentTypeString(GLenum componentType)
36 {
37 switch (componentType)
38 {
39 case GL_UNSIGNED_INT:
40 return "uint";
41 case GL_INT:
42 return "int";
43 case GL_UNSIGNED_NORMALIZED:
44 case GL_SIGNED_NORMALIZED:
45 case GL_FLOAT:
46 return "float";
47 default:
48 UNREACHABLE();
49 return "not-component-type";
50 }
51 }
52
HLSLComponentTypeString(std::ostringstream & ostream,GLenum componentType,int componentCount)53 void HLSLComponentTypeString(std::ostringstream &ostream, GLenum componentType, int componentCount)
54 {
55 ostream << HLSLComponentTypeString(componentType);
56 if (componentCount > 1)
57 {
58 ostream << componentCount;
59 }
60 }
61
HLSLMatrixTypeString(GLenum type)62 const char *HLSLMatrixTypeString(GLenum type)
63 {
64 switch (type)
65 {
66 case GL_FLOAT_MAT2:
67 return "float2x2";
68 case GL_FLOAT_MAT3:
69 return "float3x3";
70 case GL_FLOAT_MAT4:
71 return "float4x4";
72 case GL_FLOAT_MAT2x3:
73 return "float2x3";
74 case GL_FLOAT_MAT3x2:
75 return "float3x2";
76 case GL_FLOAT_MAT2x4:
77 return "float2x4";
78 case GL_FLOAT_MAT4x2:
79 return "float4x2";
80 case GL_FLOAT_MAT3x4:
81 return "float3x4";
82 case GL_FLOAT_MAT4x3:
83 return "float4x3";
84 default:
85 UNREACHABLE();
86 return "not-matrix-type";
87 }
88 }
89
HLSLTypeString(std::ostringstream & ostream,GLenum type)90 void HLSLTypeString(std::ostringstream &ostream, GLenum type)
91 {
92 if (gl::IsMatrixType(type))
93 {
94 ostream << HLSLMatrixTypeString(type);
95 return;
96 }
97
98 HLSLComponentTypeString(ostream, gl::VariableComponentType(type),
99 gl::VariableComponentCount(type));
100 }
101
FindOutputAtLocation(const std::vector<PixelShaderOutputVariable> & outputVariables,unsigned int location,size_t index=0)102 const PixelShaderOutputVariable *FindOutputAtLocation(
103 const std::vector<PixelShaderOutputVariable> &outputVariables,
104 unsigned int location,
105 size_t index = 0)
106 {
107 for (auto &outputVar : outputVariables)
108 {
109 if (outputVar.outputLocation == location && outputVar.outputIndex == index)
110 {
111 return &outputVar;
112 }
113 }
114
115 return nullptr;
116 }
117
WriteArrayString(std::ostringstream & strstr,unsigned int i)118 void WriteArrayString(std::ostringstream &strstr, unsigned int i)
119 {
120 static_assert(GL_INVALID_INDEX == UINT_MAX,
121 "GL_INVALID_INDEX must be equal to the max unsigned int.");
122 if (i == UINT_MAX)
123 {
124 return;
125 }
126
127 strstr << "[";
128 strstr << i;
129 strstr << "]";
130 }
131
ReplaceShaderStorageDeclaration(const std::vector<ShaderStorageBlock> & shaderStorageBlocks,std::string * hlsl,size_t baseUAVRegister,gl::ShaderType shaderType)132 bool ReplaceShaderStorageDeclaration(const std::vector<ShaderStorageBlock> &shaderStorageBlocks,
133 std::string *hlsl,
134 size_t baseUAVRegister,
135 gl::ShaderType shaderType)
136 {
137 std::string ssboHeader;
138 std::ostringstream out(ssboHeader);
139 for (const ShaderStorageBlock &ssbo : shaderStorageBlocks)
140 {
141 size_t uavRegister = baseUAVRegister + ssbo.registerIndex;
142 std::string name = ssbo.name;
143 if (ssbo.arraySize > 0)
144 {
145 for (unsigned int arrayIndex = 0; arrayIndex < ssbo.arraySize; arrayIndex++)
146 {
147 out << "RWByteAddressBuffer "
148 << "dx_" << name << "_" << arrayIndex << ": register(u"
149 << uavRegister + arrayIndex << ");\n";
150 }
151 }
152 else
153 {
154 out << "RWByteAddressBuffer "
155 << "_" << name << ": register(u" << uavRegister << ");\n";
156 }
157 }
158 if (out.str().empty())
159 {
160 return true;
161 }
162 return angle::ReplaceSubstring(hlsl, kShaderStorageDeclarationString, out.str());
163 }
164
165 constexpr const char *VERTEX_ATTRIBUTE_STUB_STRING = "@@ VERTEX ATTRIBUTES @@";
166 constexpr const char *VERTEX_OUTPUT_STUB_STRING = "@@ VERTEX OUTPUT @@";
167 constexpr const char *PIXEL_OUTPUT_STUB_STRING = "@@ PIXEL OUTPUT @@";
168 constexpr const char *PIXEL_MAIN_PARAMETERS_STUB_STRING = "@@ PIXEL MAIN PARAMETERS @@";
169 constexpr const char *MAIN_PROLOGUE_STUB_STRING = "@@ MAIN PROLOGUE @@";
170 } // anonymous namespace
171
172 // BuiltinInfo implementation
173
174 BuiltinInfo::BuiltinInfo() = default;
175 BuiltinInfo::~BuiltinInfo() = default;
176
177 // DynamicHLSL implementation
178
DynamicHLSL(RendererD3D * const renderer)179 DynamicHLSL::DynamicHLSL(RendererD3D *const renderer) : mRenderer(renderer) {}
180
generateVertexShaderForInputLayout(const std::string & sourceShader,const InputLayout & inputLayout,const std::vector<sh::ShaderVariable> & shaderAttributes,const std::vector<rx::ShaderStorageBlock> & shaderStorageBlocks,size_t baseUAVRegister) const181 std::string DynamicHLSL::generateVertexShaderForInputLayout(
182 const std::string &sourceShader,
183 const InputLayout &inputLayout,
184 const std::vector<sh::ShaderVariable> &shaderAttributes,
185 const std::vector<rx::ShaderStorageBlock> &shaderStorageBlocks,
186 size_t baseUAVRegister) const
187 {
188 std::ostringstream structStream;
189 std::ostringstream initStream;
190
191 structStream << "struct VS_INPUT\n"
192 << "{\n";
193
194 int semanticIndex = 0;
195 unsigned int inputIndex = 0;
196
197 // If gl_PointSize is used in the shader then pointsprites rendering is expected.
198 // If the renderer does not support Geometry shaders then Instanced PointSprite emulation
199 // must be used.
200 bool usesPointSize = sourceShader.find("GL_USES_POINT_SIZE") != std::string::npos;
201 bool useInstancedPointSpriteEmulation =
202 usesPointSize && mRenderer->getFeatures().useInstancedPointSpriteEmulation.enabled;
203
204 // Instanced PointSprite emulation requires additional entries in the
205 // VS_INPUT structure to support the vertices that make up the quad vertices.
206 // These values must be in sync with the cooresponding values added during inputlayout creation
207 // in InputLayoutCache::applyVertexBuffers().
208 //
209 // The additional entries must appear first in the VS_INPUT layout because
210 // Windows Phone 8 era devices require per vertex data to physically come
211 // before per instance data in the shader.
212 if (useInstancedPointSpriteEmulation)
213 {
214 structStream << " float3 spriteVertexPos : SPRITEPOSITION0;\n"
215 << " float2 spriteTexCoord : SPRITETEXCOORD0;\n";
216 }
217
218 for (size_t attributeIndex = 0; attributeIndex < shaderAttributes.size(); ++attributeIndex)
219 {
220 const sh::ShaderVariable &shaderAttribute = shaderAttributes[attributeIndex];
221 if (!shaderAttribute.name.empty())
222 {
223 ASSERT(inputIndex < MAX_VERTEX_ATTRIBS);
224 angle::FormatID vertexFormatID =
225 inputIndex < inputLayout.size() ? inputLayout[inputIndex] : angle::FormatID::NONE;
226
227 // HLSL code for input structure
228 if (IsMatrixType(shaderAttribute.type))
229 {
230 // Matrix types are always transposed
231 structStream << " "
232 << HLSLMatrixTypeString(TransposeMatrixType(shaderAttribute.type));
233 }
234 else
235 {
236 if (shaderAttribute.name == "gl_InstanceID" ||
237 shaderAttribute.name == "gl_VertexID")
238 {
239 // The input types of the instance ID and vertex ID in HLSL (uint) differs from
240 // the ones in ESSL (int).
241 structStream << " uint";
242 }
243 else
244 {
245 GLenum componentType = mRenderer->getVertexComponentType(vertexFormatID);
246
247 structStream << " ";
248 HLSLComponentTypeString(structStream, componentType,
249 VariableComponentCount(shaderAttribute.type));
250 }
251 }
252
253 structStream << " " << DecorateVariable(shaderAttribute.name) << " : ";
254
255 if (shaderAttribute.name == "gl_InstanceID")
256 {
257 structStream << "SV_InstanceID";
258 }
259 else if (shaderAttribute.name == "gl_VertexID")
260 {
261 structStream << "SV_VertexID";
262 }
263 else
264 {
265 structStream << "TEXCOORD" << semanticIndex;
266 semanticIndex += VariableRegisterCount(shaderAttribute.type);
267 }
268
269 structStream << ";\n";
270
271 // HLSL code for initialization
272 initStream << " " << DecorateVariable(shaderAttribute.name) << " = ";
273
274 // Mismatched vertex attribute to vertex input may result in an undefined
275 // data reinterpretation (eg for pure integer->float, float->pure integer)
276 // TODO: issue warning with gl debug info extension, when supported
277 if (IsMatrixType(shaderAttribute.type) ||
278 (mRenderer->getVertexConversionType(vertexFormatID) & VERTEX_CONVERT_GPU) != 0)
279 {
280 GenerateAttributeConversionHLSL(vertexFormatID, shaderAttribute, initStream);
281 }
282 else
283 {
284 initStream << "input." << DecorateVariable(shaderAttribute.name);
285 }
286
287 if (shaderAttribute.name == "gl_VertexID")
288 {
289 // dx_VertexID contains the firstVertex offset
290 initStream << " + dx_VertexID";
291 }
292
293 initStream << ";\n";
294
295 inputIndex += VariableRowCount(TransposeMatrixType(shaderAttribute.type));
296 }
297 }
298
299 structStream << "};\n"
300 "\n"
301 "void initAttributes(VS_INPUT input)\n"
302 "{\n"
303 << initStream.str() << "}\n";
304
305 std::string vertexHLSL(sourceShader);
306
307 bool success =
308 angle::ReplaceSubstring(&vertexHLSL, VERTEX_ATTRIBUTE_STUB_STRING, structStream.str());
309 ASSERT(success);
310
311 success = ReplaceShaderStorageDeclaration(shaderStorageBlocks, &vertexHLSL, baseUAVRegister,
312 gl::ShaderType::Vertex);
313 ASSERT(success);
314
315 return vertexHLSL;
316 }
317
generatePixelShaderForOutputSignature(const std::string & sourceShader,const std::vector<PixelShaderOutputVariable> & outputVariables,bool usesFragDepth,const std::vector<GLenum> & outputLayout,const std::vector<ShaderStorageBlock> & shaderStorageBlocks,size_t baseUAVRegister) const318 std::string DynamicHLSL::generatePixelShaderForOutputSignature(
319 const std::string &sourceShader,
320 const std::vector<PixelShaderOutputVariable> &outputVariables,
321 bool usesFragDepth,
322 const std::vector<GLenum> &outputLayout,
323 const std::vector<ShaderStorageBlock> &shaderStorageBlocks,
324 size_t baseUAVRegister) const
325 {
326 const int shaderModel = mRenderer->getMajorShaderModel();
327 std::string targetSemantic = (shaderModel >= 4) ? "SV_TARGET" : "COLOR";
328 std::string depthSemantic = (shaderModel >= 4) ? "SV_Depth" : "DEPTH";
329
330 std::ostringstream declarationStream;
331 std::ostringstream copyStream;
332
333 declarationStream << "struct PS_OUTPUT\n"
334 "{\n";
335
336 size_t numOutputs = outputLayout.size();
337
338 // Workaround for HLSL 3.x: We can't do a depth/stencil only render, the runtime will complain.
339 if (numOutputs == 0 && (shaderModel == 3 || !mRenderer->getShaderModelSuffix().empty()))
340 {
341 numOutputs = 1u;
342 }
343 const PixelShaderOutputVariable defaultOutput(GL_FLOAT_VEC4, "unused", "float4(0, 0, 0, 1)", 0,
344 0);
345 size_t outputIndex = 0;
346
347 for (size_t layoutIndex = 0; layoutIndex < numOutputs; ++layoutIndex)
348 {
349 GLenum binding = outputLayout.empty() ? GL_COLOR_ATTACHMENT0 : outputLayout[layoutIndex];
350
351 if (binding != GL_NONE)
352 {
353 unsigned int location = (binding - GL_COLOR_ATTACHMENT0);
354 outputIndex =
355 layoutIndex > 0 && binding == outputLayout[layoutIndex - 1] ? outputIndex + 1 : 0;
356
357 const PixelShaderOutputVariable *outputVariable =
358 outputLayout.empty() ? &defaultOutput
359 : FindOutputAtLocation(outputVariables, location, outputIndex);
360
361 // OpenGL ES 3.0 spec $4.2.1
362 // If [...] not all user-defined output variables are written, the values of fragment
363 // colors corresponding to unwritten variables are similarly undefined.
364 if (outputVariable)
365 {
366 declarationStream << " ";
367 HLSLTypeString(declarationStream, outputVariable->type);
368 declarationStream << " " << outputVariable->name << " : " << targetSemantic
369 << static_cast<int>(layoutIndex) << ";\n";
370
371 copyStream << " output." << outputVariable->name << " = "
372 << outputVariable->source << ";\n";
373 }
374 }
375 }
376
377 if (usesFragDepth)
378 {
379 declarationStream << " float gl_Depth : " << depthSemantic << ";\n";
380 copyStream << " output.gl_Depth = gl_Depth; \n";
381 }
382
383 declarationStream << "};\n"
384 "\n"
385 "PS_OUTPUT generateOutput()\n"
386 "{\n"
387 " PS_OUTPUT output;\n"
388 << copyStream.str()
389 << " return output;\n"
390 "}\n";
391
392 std::string pixelHLSL(sourceShader);
393
394 bool success =
395 angle::ReplaceSubstring(&pixelHLSL, PIXEL_OUTPUT_STUB_STRING, declarationStream.str());
396 ASSERT(success);
397
398 success = ReplaceShaderStorageDeclaration(shaderStorageBlocks, &pixelHLSL, baseUAVRegister,
399 gl::ShaderType::Fragment);
400 ASSERT(success);
401
402 return pixelHLSL;
403 }
404
generateShaderForImage2DBindSignature(const d3d::Context * context,ProgramD3D & programD3D,const gl::ProgramState & programData,gl::ShaderType shaderType,std::vector<sh::ShaderVariable> & image2DUniforms,const gl::ImageUnitTextureTypeMap & image2DBindLayout) const405 std::string DynamicHLSL::generateShaderForImage2DBindSignature(
406 const d3d::Context *context,
407 ProgramD3D &programD3D,
408 const gl::ProgramState &programData,
409 gl::ShaderType shaderType,
410 std::vector<sh::ShaderVariable> &image2DUniforms,
411 const gl::ImageUnitTextureTypeMap &image2DBindLayout) const
412 {
413 std::string shaderHLSL(programData.getAttachedShader(shaderType)->getTranslatedSource());
414
415 if (image2DUniforms.empty())
416 {
417 return shaderHLSL;
418 }
419
420 return GenerateShaderForImage2DBindSignature(context, programD3D, programData, shaderType,
421 image2DUniforms, image2DBindLayout);
422 }
423
generateVaryingLinkHLSL(const VaryingPacking & varyingPacking,const BuiltinInfo & builtins,bool programUsesPointSize,std::ostringstream & hlslStream) const424 void DynamicHLSL::generateVaryingLinkHLSL(const VaryingPacking &varyingPacking,
425 const BuiltinInfo &builtins,
426 bool programUsesPointSize,
427 std::ostringstream &hlslStream) const
428 {
429 ASSERT(builtins.dxPosition.enabled);
430 hlslStream << "{\n"
431 << " float4 dx_Position : " << builtins.dxPosition.str() << ";\n";
432
433 if (builtins.glPosition.enabled)
434 {
435 hlslStream << " float4 gl_Position : " << builtins.glPosition.str() << ";\n";
436 }
437
438 if (builtins.glFragCoord.enabled)
439 {
440 hlslStream << " float4 gl_FragCoord : " << builtins.glFragCoord.str() << ";\n";
441 }
442
443 if (builtins.glPointCoord.enabled)
444 {
445 hlslStream << " float2 gl_PointCoord : " << builtins.glPointCoord.str() << ";\n";
446 }
447
448 if (builtins.glPointSize.enabled)
449 {
450 hlslStream << " float gl_PointSize : " << builtins.glPointSize.str() << ";\n";
451 }
452
453 if (builtins.glViewIDOVR.enabled)
454 {
455 hlslStream << " nointerpolation uint gl_ViewID_OVR : " << builtins.glViewIDOVR.str()
456 << ";\n";
457 }
458
459 std::string varyingSemantic =
460 GetVaryingSemantic(mRenderer->getMajorShaderModel(), programUsesPointSize);
461
462 const auto ®isterInfos = varyingPacking.getRegisterList();
463 for (GLuint registerIndex = 0u; registerIndex < registerInfos.size(); ++registerIndex)
464 {
465 const PackedVaryingRegister ®isterInfo = registerInfos[registerIndex];
466 const auto &varying = registerInfo.packedVarying->varying();
467 ASSERT(!varying.isStruct());
468
469 // TODO: Add checks to ensure D3D interpolation modifiers don't result in too many
470 // registers being used.
471 // For example, if there are N registers, and we have N vec3 varyings and 1 float
472 // varying, then D3D will pack them into N registers.
473 // If the float varying has the 'nointerpolation' modifier on it then we would need
474 // N + 1 registers, and D3D compilation will fail.
475
476 switch (registerInfo.packedVarying->interpolation)
477 {
478 case sh::INTERPOLATION_SMOOTH:
479 hlslStream << " ";
480 break;
481 case sh::INTERPOLATION_FLAT:
482 hlslStream << " nointerpolation ";
483 break;
484 case sh::INTERPOLATION_CENTROID:
485 hlslStream << " centroid ";
486 break;
487 case sh::INTERPOLATION_SAMPLE:
488 hlslStream << " sample ";
489 break;
490 default:
491 UNREACHABLE();
492 }
493
494 GLenum transposedType = gl::TransposeMatrixType(varying.type);
495 GLenum componentType = gl::VariableComponentType(transposedType);
496 int columnCount = gl::VariableColumnCount(transposedType);
497 HLSLComponentTypeString(hlslStream, componentType, columnCount);
498 hlslStream << " v" << registerIndex << " : " << varyingSemantic << registerIndex << ";\n";
499 }
500
501 // Note that the following outputs need to be declared after the others. They are not included
502 // in pixel shader inputs even when they are in vertex/geometry shader outputs, and the pixel
503 // shader input struct must be a prefix of the vertex/geometry shader output struct.
504
505 if (builtins.glViewportIndex.enabled)
506 {
507 hlslStream << " nointerpolation uint gl_ViewportIndex : "
508 << builtins.glViewportIndex.str() << ";\n";
509 }
510
511 if (builtins.glLayer.enabled)
512 {
513 hlslStream << " nointerpolation uint gl_Layer : " << builtins.glLayer.str() << ";\n";
514 }
515
516 hlslStream << "};\n";
517 }
518
generateShaderLinkHLSL(const gl::Caps & caps,const gl::ProgramState & programData,const ProgramD3DMetadata & programMetadata,const VaryingPacking & varyingPacking,const BuiltinVaryingsD3D & builtinsD3D,gl::ShaderMap<std::string> * shaderHLSL) const519 void DynamicHLSL::generateShaderLinkHLSL(const gl::Caps &caps,
520 const gl::ProgramState &programData,
521 const ProgramD3DMetadata &programMetadata,
522 const VaryingPacking &varyingPacking,
523 const BuiltinVaryingsD3D &builtinsD3D,
524 gl::ShaderMap<std::string> *shaderHLSL) const
525 {
526 ASSERT(shaderHLSL);
527 ASSERT((*shaderHLSL)[gl::ShaderType::Vertex].empty() &&
528 (*shaderHLSL)[gl::ShaderType::Fragment].empty());
529
530 gl::Shader *vertexShaderGL = programData.getAttachedShader(ShaderType::Vertex);
531 gl::Shader *fragmentShaderGL = programData.getAttachedShader(ShaderType::Fragment);
532 const int shaderModel = mRenderer->getMajorShaderModel();
533
534 const ShaderD3D *fragmentShader = nullptr;
535 if (fragmentShaderGL)
536 {
537 fragmentShader = GetImplAs<ShaderD3D>(fragmentShaderGL);
538 }
539
540 // usesViewScale() isn't supported in the D3D9 renderer
541 ASSERT(shaderModel >= 4 || !programMetadata.usesViewScale());
542
543 bool useInstancedPointSpriteEmulation =
544 programMetadata.usesPointSize() &&
545 mRenderer->getFeatures().useInstancedPointSpriteEmulation.enabled;
546
547 // Validation done in the compiler
548 ASSERT(!fragmentShader || !fragmentShader->usesFragColor() || !fragmentShader->usesFragData());
549
550 std::ostringstream vertexStream;
551 vertexStream << "struct VS_OUTPUT\n";
552 const auto &vertexBuiltins = builtinsD3D[gl::ShaderType::Vertex];
553 generateVaryingLinkHLSL(varyingPacking, vertexBuiltins, builtinsD3D.usesPointSize(),
554 vertexStream);
555
556 // Instanced PointSprite emulation requires additional entries originally generated in the
557 // GeometryShader HLSL. These include pointsize clamp values.
558 if (useInstancedPointSpriteEmulation)
559 {
560 vertexStream << "static float minPointSize = " << static_cast<int>(caps.minAliasedPointSize)
561 << ".0f;\n"
562 << "static float maxPointSize = " << static_cast<int>(caps.maxAliasedPointSize)
563 << ".0f;\n";
564 }
565
566 std::ostringstream vertexGenerateOutput;
567 vertexGenerateOutput << "VS_OUTPUT generateOutput(VS_INPUT input)\n"
568 << "{\n"
569 << " VS_OUTPUT output;\n";
570
571 if (vertexBuiltins.glPosition.enabled)
572 {
573 vertexGenerateOutput << " output.gl_Position = gl_Position;\n";
574 }
575
576 if (vertexBuiltins.glViewIDOVR.enabled)
577 {
578 vertexGenerateOutput << " output.gl_ViewID_OVR = ViewID_OVR;\n";
579 }
580 if (programMetadata.hasANGLEMultiviewEnabled() && programMetadata.canSelectViewInVertexShader())
581 {
582 ASSERT(vertexBuiltins.glViewportIndex.enabled && vertexBuiltins.glLayer.enabled);
583 vertexGenerateOutput << " if (multiviewSelectViewportIndex)\n"
584 << " {\n"
585 << " output.gl_ViewportIndex = ViewID_OVR;\n"
586 << " } else {\n"
587 << " output.gl_ViewportIndex = 0;\n"
588 << " output.gl_Layer = ViewID_OVR;\n"
589 << " }\n";
590 }
591
592 // On D3D9 or D3D11 Feature Level 9, we need to emulate large viewports using dx_ViewAdjust.
593 if (shaderModel >= 4 && mRenderer->getShaderModelSuffix() == "")
594 {
595 vertexGenerateOutput << " output.dx_Position.x = gl_Position.x;\n";
596
597 if (programMetadata.usesViewScale())
598 {
599 // This code assumes that dx_ViewScale.y = -1.0f when rendering to texture, and +1.0f
600 // when rendering to the default framebuffer. No other values are valid.
601 vertexGenerateOutput << " output.dx_Position.y = dx_ViewScale.y * gl_Position.y;\n";
602 }
603 else
604 {
605 vertexGenerateOutput
606 << " output.dx_Position.y = clipControlOrigin * gl_Position.y;\n";
607 }
608
609 vertexGenerateOutput
610 << " if (clipControlZeroToOne)\n"
611 << " {\n"
612 << " output.dx_Position.z = gl_Position.z;\n"
613 << " } else {\n"
614 << " output.dx_Position.z = (gl_Position.z + gl_Position.w) * 0.5;\n"
615 << " }\n";
616
617 vertexGenerateOutput << " output.dx_Position.w = gl_Position.w;\n";
618 }
619 else
620 {
621 vertexGenerateOutput << " output.dx_Position.x = gl_Position.x * dx_ViewAdjust.z + "
622 "dx_ViewAdjust.x * gl_Position.w;\n";
623
624 // If usesViewScale() is true and we're using the D3D11 renderer via Feature Level 9_*,
625 // then we need to multiply the gl_Position.y by the viewScale.
626 // usesViewScale() isn't supported when using the D3D9 renderer.
627 if (programMetadata.usesViewScale() &&
628 (shaderModel >= 4 && mRenderer->getShaderModelSuffix() != ""))
629 {
630 vertexGenerateOutput << " output.dx_Position.y = dx_ViewScale.y * (gl_Position.y * "
631 "dx_ViewAdjust.w + dx_ViewAdjust.y * gl_Position.w);\n";
632 }
633 else
634 {
635 vertexGenerateOutput << " output.dx_Position.y = clipControlOrigin * (gl_Position.y "
636 "* dx_ViewAdjust.w + "
637 "dx_ViewAdjust.y * gl_Position.w);\n";
638 }
639
640 vertexGenerateOutput
641 << " if (clipControlZeroToOne)\n"
642 << " {\n"
643 << " output.dx_Position.z = gl_Position.z;\n"
644 << " } else {\n"
645 << " output.dx_Position.z = (gl_Position.z + gl_Position.w) * 0.5;\n"
646 << " }\n";
647
648 vertexGenerateOutput << " output.dx_Position.w = gl_Position.w;\n";
649 }
650
651 // We don't need to output gl_PointSize if we use are emulating point sprites via instancing.
652 if (vertexBuiltins.glPointSize.enabled)
653 {
654 vertexGenerateOutput << " output.gl_PointSize = gl_PointSize;\n";
655 }
656
657 if (vertexBuiltins.glFragCoord.enabled)
658 {
659 vertexGenerateOutput << " output.gl_FragCoord = gl_Position;\n";
660 }
661
662 const auto ®isterInfos = varyingPacking.getRegisterList();
663 for (GLuint registerIndex = 0u; registerIndex < registerInfos.size(); ++registerIndex)
664 {
665 const PackedVaryingRegister ®isterInfo = registerInfos[registerIndex];
666 const auto &packedVarying = *registerInfo.packedVarying;
667 const auto &varying = *packedVarying.frontVarying.varying;
668 ASSERT(!varying.isStruct());
669
670 vertexGenerateOutput << " output.v" << registerIndex << " = ";
671
672 if (packedVarying.isStructField())
673 {
674 vertexGenerateOutput << DecorateVariable(packedVarying.frontVarying.parentStructName)
675 << ".";
676 }
677
678 vertexGenerateOutput << DecorateVariable(varying.name);
679
680 if (varying.isArray())
681 {
682 WriteArrayString(vertexGenerateOutput, registerInfo.varyingArrayIndex);
683 }
684
685 if (VariableRowCount(varying.type) > 1)
686 {
687 WriteArrayString(vertexGenerateOutput, registerInfo.varyingRowIndex);
688 }
689
690 vertexGenerateOutput << ";\n";
691 }
692
693 // Instanced PointSprite emulation requires additional entries to calculate
694 // the final output vertex positions of the quad that represents each sprite.
695 if (useInstancedPointSpriteEmulation)
696 {
697 vertexGenerateOutput
698 << "\n"
699 << " gl_PointSize = clamp(gl_PointSize, minPointSize, maxPointSize);\n";
700
701 vertexGenerateOutput
702 << " output.dx_Position.x += (input.spriteVertexPos.x * gl_PointSize / "
703 "(dx_ViewCoords.x*2)) * output.dx_Position.w;";
704
705 if (programMetadata.usesViewScale())
706 {
707 // Multiply by ViewScale to invert the rendering when appropriate
708 vertexGenerateOutput
709 << " output.dx_Position.y += (-dx_ViewScale.y * "
710 "input.spriteVertexPos.y * gl_PointSize / (dx_ViewCoords.y*2)) * "
711 "output.dx_Position.w;";
712 }
713 else
714 {
715 vertexGenerateOutput
716 << " output.dx_Position.y += (input.spriteVertexPos.y * gl_PointSize / "
717 "(dx_ViewCoords.y*2)) * output.dx_Position.w;";
718 }
719
720 vertexGenerateOutput
721 << " output.dx_Position.z += input.spriteVertexPos.z * output.dx_Position.w;\n";
722
723 if (programMetadata.usesPointCoord())
724 {
725 vertexGenerateOutput << "\n"
726 << " output.gl_PointCoord = input.spriteTexCoord;\n";
727 }
728 }
729
730 // Renderers that enable instanced pointsprite emulation require the vertex shader output member
731 // gl_PointCoord to be set to a default value if used without gl_PointSize. 0.5,0.5 is the same
732 // default value used in the generated pixel shader.
733 if (programMetadata.usesInsertedPointCoordValue())
734 {
735 ASSERT(!useInstancedPointSpriteEmulation);
736 vertexGenerateOutput << "\n"
737 << " output.gl_PointCoord = float2(0.5, 0.5);\n";
738 }
739
740 vertexGenerateOutput << "\n"
741 << " return output;\n"
742 << "}";
743
744 if (vertexShaderGL)
745 {
746 std::string vertexSource = vertexShaderGL->getTranslatedSource();
747 angle::ReplaceSubstring(&vertexSource, std::string(MAIN_PROLOGUE_STUB_STRING),
748 " initAttributes(input);\n");
749 angle::ReplaceSubstring(&vertexSource, std::string(VERTEX_OUTPUT_STUB_STRING),
750 vertexGenerateOutput.str());
751 vertexStream << vertexSource;
752 }
753
754 const auto &pixelBuiltins = builtinsD3D[gl::ShaderType::Fragment];
755
756 std::ostringstream pixelStream;
757 pixelStream << "struct PS_INPUT\n";
758 generateVaryingLinkHLSL(varyingPacking, pixelBuiltins, builtinsD3D.usesPointSize(),
759 pixelStream);
760 pixelStream << "\n";
761
762 std::ostringstream pixelPrologue;
763 if (fragmentShader && fragmentShader->usesViewID())
764 {
765 ASSERT(pixelBuiltins.glViewIDOVR.enabled);
766 pixelPrologue << " ViewID_OVR = input.gl_ViewID_OVR;\n";
767 }
768
769 if (pixelBuiltins.glFragCoord.enabled)
770 {
771 pixelPrologue << " float rhw = 1.0 / input.gl_FragCoord.w;\n";
772
773 // Certain Shader Models (4_0+ and 3_0) allow reading from dx_Position in the pixel shader.
774 // Other Shader Models (4_0_level_9_3 and 2_x) don't support this, so we emulate it using
775 // dx_ViewCoords.
776 if (shaderModel >= 4 && mRenderer->getShaderModelSuffix() == "")
777 {
778 pixelPrologue << " gl_FragCoord.x = input.dx_Position.x;\n"
779 << " gl_FragCoord.y = input.dx_Position.y;\n";
780 }
781 else if (shaderModel == 3)
782 {
783 pixelPrologue << " gl_FragCoord.x = input.dx_Position.x + 0.5;\n"
784 << " gl_FragCoord.y = input.dx_Position.y + 0.5;\n";
785 }
786 else
787 {
788 // dx_ViewCoords contains the viewport width/2, height/2, center.x and center.y. See
789 // Renderer::setViewport()
790 pixelPrologue
791 << " gl_FragCoord.x = (input.gl_FragCoord.x * rhw) * dx_ViewCoords.x + "
792 "dx_ViewCoords.z;\n"
793 << " gl_FragCoord.y = (input.gl_FragCoord.y * rhw) * dx_ViewCoords.y + "
794 "dx_ViewCoords.w;\n";
795 }
796
797 if (programMetadata.usesViewScale())
798 {
799 // For Feature Level 9_3 and below, we need to correct gl_FragCoord.y to account
800 // for dx_ViewScale. On Feature Level 10_0+, gl_FragCoord.y is calculated above using
801 // dx_ViewCoords and is always correct irrespective of dx_ViewScale's value.
802 // NOTE: usesViewScale() can only be true on D3D11 (i.e. Shader Model 4.0+).
803 if (shaderModel >= 4 && mRenderer->getShaderModelSuffix() == "")
804 {
805 // Some assumptions:
806 // - dx_ViewScale.y = -1.0f when rendering to texture
807 // - dx_ViewScale.y = +1.0f when rendering to the default framebuffer
808 // - gl_FragCoord.y has been set correctly above.
809 //
810 // When rendering to the backbuffer, the code inverts gl_FragCoord's y coordinate.
811 // This involves subtracting the y coordinate from the height of the area being
812 // rendered to.
813 //
814 // First we calculate the height of the area being rendered to:
815 // render_area_height = (2.0f / (1.0f - input.gl_FragCoord.y * rhw)) *
816 // gl_FragCoord.y
817 //
818 // Note that when we're rendering to default FB, we want our output to be
819 // equivalent to:
820 // "gl_FragCoord.y = render_area_height - gl_FragCoord.y"
821 //
822 // When we're rendering to a texture, we want our output to be equivalent to:
823 // "gl_FragCoord.y = gl_FragCoord.y;"
824 //
825 // If we set scale_factor = ((1.0f + dx_ViewScale.y) / 2.0f), then notice that
826 // - When rendering to default FB: scale_factor = 1.0f
827 // - When rendering to texture: scale_factor = 0.0f
828 //
829 // Therefore, we can get our desired output by setting:
830 // "gl_FragCoord.y = scale_factor * render_area_height - dx_ViewScale.y *
831 // gl_FragCoord.y"
832 //
833 // Simplifying, this becomes:
834 pixelPrologue
835 << " gl_FragCoord.y = (1.0f + dx_ViewScale.y) * gl_FragCoord.y /"
836 "(1.0f - input.gl_FragCoord.y * rhw) - dx_ViewScale.y * gl_FragCoord.y;\n";
837 }
838 }
839
840 pixelPrologue << " gl_FragCoord.z = (input.gl_FragCoord.z * rhw) * dx_DepthFront.x + "
841 "dx_DepthFront.y;\n"
842 << " gl_FragCoord.w = rhw;\n";
843 }
844
845 if (pixelBuiltins.glPointCoord.enabled && shaderModel >= 3)
846 {
847 pixelPrologue << " gl_PointCoord.x = input.gl_PointCoord.x;\n"
848 << " gl_PointCoord.y = 1.0 - input.gl_PointCoord.y;\n";
849 }
850
851 if (fragmentShader && fragmentShader->usesFrontFacing())
852 {
853 if (shaderModel <= 3)
854 {
855 pixelPrologue << " gl_FrontFacing = (vFace * dx_DepthFront.z >= 0.0);\n";
856 }
857 else
858 {
859 pixelPrologue << " gl_FrontFacing = isFrontFace;\n";
860 }
861 }
862
863 for (GLuint registerIndex = 0u; registerIndex < registerInfos.size(); ++registerIndex)
864 {
865 const PackedVaryingRegister ®isterInfo = registerInfos[registerIndex];
866 const auto &packedVarying = *registerInfo.packedVarying;
867
868 // Don't reference VS-only transform feedback varyings in the PS.
869 if (packedVarying.vertexOnly())
870 {
871 continue;
872 }
873
874 const auto &varying = *packedVarying.backVarying.varying;
875 ASSERT(!varying.isBuiltIn() && !varying.isStruct());
876
877 // Note that we're relying on that the active flag is set according to usage in the fragment
878 // shader.
879 if (!varying.active)
880 {
881 continue;
882 }
883
884 pixelPrologue << " ";
885
886 if (packedVarying.isStructField())
887 {
888 pixelPrologue << DecorateVariable(packedVarying.backVarying.parentStructName) << ".";
889 }
890
891 pixelPrologue << DecorateVariable(varying.name);
892
893 if (varying.isArray())
894 {
895 WriteArrayString(pixelPrologue, registerInfo.varyingArrayIndex);
896 }
897
898 GLenum transposedType = TransposeMatrixType(varying.type);
899 if (VariableRowCount(transposedType) > 1)
900 {
901 WriteArrayString(pixelPrologue, registerInfo.varyingRowIndex);
902 }
903
904 pixelPrologue << " = input.v" << registerIndex;
905
906 switch (VariableColumnCount(transposedType))
907 {
908 case 1:
909 pixelPrologue << ".x";
910 break;
911 case 2:
912 pixelPrologue << ".xy";
913 break;
914 case 3:
915 pixelPrologue << ".xyz";
916 break;
917 case 4:
918 break;
919 default:
920 UNREACHABLE();
921 }
922 pixelPrologue << ";\n";
923 }
924
925 if (fragmentShaderGL)
926 {
927 std::string pixelSource = fragmentShaderGL->getTranslatedSource();
928
929 if (fragmentShader->usesFrontFacing())
930 {
931 if (shaderModel >= 4)
932 {
933 angle::ReplaceSubstring(&pixelSource,
934 std::string(PIXEL_MAIN_PARAMETERS_STUB_STRING),
935 "PS_INPUT input, bool isFrontFace : SV_IsFrontFace");
936 }
937 else
938 {
939 angle::ReplaceSubstring(&pixelSource,
940 std::string(PIXEL_MAIN_PARAMETERS_STUB_STRING),
941 "PS_INPUT input, float vFace : VFACE");
942 }
943 }
944 else
945 {
946 angle::ReplaceSubstring(&pixelSource, std::string(PIXEL_MAIN_PARAMETERS_STUB_STRING),
947 "PS_INPUT input");
948 }
949
950 angle::ReplaceSubstring(&pixelSource, std::string(MAIN_PROLOGUE_STUB_STRING),
951 pixelPrologue.str());
952 pixelStream << pixelSource;
953 }
954
955 (*shaderHLSL)[gl::ShaderType::Vertex] = vertexStream.str();
956 (*shaderHLSL)[gl::ShaderType::Fragment] = pixelStream.str();
957 }
958
generateGeometryShaderPreamble(const VaryingPacking & varyingPacking,const BuiltinVaryingsD3D & builtinsD3D,const bool hasANGLEMultiviewEnabled,const bool selectViewInVS) const959 std::string DynamicHLSL::generateGeometryShaderPreamble(const VaryingPacking &varyingPacking,
960 const BuiltinVaryingsD3D &builtinsD3D,
961 const bool hasANGLEMultiviewEnabled,
962 const bool selectViewInVS) const
963 {
964 ASSERT(mRenderer->getMajorShaderModel() >= 4);
965
966 std::ostringstream preambleStream;
967
968 const auto &vertexBuiltins = builtinsD3D[gl::ShaderType::Vertex];
969
970 preambleStream << "struct GS_INPUT\n";
971 generateVaryingLinkHLSL(varyingPacking, vertexBuiltins, builtinsD3D.usesPointSize(),
972 preambleStream);
973 preambleStream << "\n"
974 << "struct GS_OUTPUT\n";
975 generateVaryingLinkHLSL(varyingPacking, builtinsD3D[gl::ShaderType::Geometry],
976 builtinsD3D.usesPointSize(), preambleStream);
977 preambleStream
978 << "\n"
979 << "void copyVertex(inout GS_OUTPUT output, GS_INPUT input, GS_INPUT flatinput)\n"
980 << "{\n"
981 << " output.gl_Position = input.gl_Position;\n";
982
983 if (vertexBuiltins.glPointSize.enabled)
984 {
985 preambleStream << " output.gl_PointSize = input.gl_PointSize;\n";
986 }
987
988 if (hasANGLEMultiviewEnabled)
989 {
990 preambleStream << " output.gl_ViewID_OVR = input.gl_ViewID_OVR;\n";
991 if (selectViewInVS)
992 {
993 ASSERT(builtinsD3D[gl::ShaderType::Geometry].glViewportIndex.enabled &&
994 builtinsD3D[gl::ShaderType::Geometry].glLayer.enabled);
995
996 // If the view is already selected in the VS, then we just pass the gl_ViewportIndex and
997 // gl_Layer to the output.
998 preambleStream << " output.gl_ViewportIndex = input.gl_ViewportIndex;\n"
999 << " output.gl_Layer = input.gl_Layer;\n";
1000 }
1001 }
1002
1003 const auto ®isterInfos = varyingPacking.getRegisterList();
1004 for (GLuint registerIndex = 0u; registerIndex < registerInfos.size(); ++registerIndex)
1005 {
1006 const PackedVaryingRegister &varyingRegister = registerInfos[registerIndex];
1007 preambleStream << " output.v" << registerIndex << " = ";
1008 if (varyingRegister.packedVarying->interpolation == sh::INTERPOLATION_FLAT)
1009 {
1010 preambleStream << "flat";
1011 }
1012 preambleStream << "input.v" << registerIndex << "; \n";
1013 }
1014
1015 if (vertexBuiltins.glFragCoord.enabled)
1016 {
1017 preambleStream << " output.gl_FragCoord = input.gl_FragCoord;\n";
1018 }
1019
1020 // Only write the dx_Position if we aren't using point sprites
1021 preambleStream << "#ifndef ANGLE_POINT_SPRITE_SHADER\n"
1022 << " output.dx_Position = input.dx_Position;\n"
1023 << "#endif // ANGLE_POINT_SPRITE_SHADER\n"
1024 << "}\n";
1025
1026 if (hasANGLEMultiviewEnabled && !selectViewInVS)
1027 {
1028 ASSERT(builtinsD3D[gl::ShaderType::Geometry].glViewportIndex.enabled &&
1029 builtinsD3D[gl::ShaderType::Geometry].glLayer.enabled);
1030
1031 // According to the HLSL reference, using SV_RenderTargetArrayIndex is only valid if the
1032 // render target is an array resource. Because of this we do not write to gl_Layer if we are
1033 // taking the side-by-side code path. We still select the viewport index in the layered code
1034 // path as that is always valid. See:
1035 // https://msdn.microsoft.com/en-us/library/windows/desktop/bb509647(v=vs.85).aspx
1036 preambleStream << "\n"
1037 << "void selectView(inout GS_OUTPUT output, GS_INPUT input)\n"
1038 << "{\n"
1039 << " if (multiviewSelectViewportIndex)\n"
1040 << " {\n"
1041 << " output.gl_ViewportIndex = input.gl_ViewID_OVR;\n"
1042 << " } else {\n"
1043 << " output.gl_ViewportIndex = 0;\n"
1044 << " output.gl_Layer = input.gl_ViewID_OVR;\n"
1045 << " }\n"
1046 << "}\n";
1047 }
1048
1049 return preambleStream.str();
1050 }
1051
generateGeometryShaderHLSL(const gl::Caps & caps,gl::PrimitiveMode primitiveType,const gl::ProgramState & programData,const bool useViewScale,const bool hasANGLEMultiviewEnabled,const bool selectViewInVS,const bool pointSpriteEmulation,const std::string & preambleString) const1052 std::string DynamicHLSL::generateGeometryShaderHLSL(const gl::Caps &caps,
1053 gl::PrimitiveMode primitiveType,
1054 const gl::ProgramState &programData,
1055 const bool useViewScale,
1056 const bool hasANGLEMultiviewEnabled,
1057 const bool selectViewInVS,
1058 const bool pointSpriteEmulation,
1059 const std::string &preambleString) const
1060 {
1061 ASSERT(mRenderer->getMajorShaderModel() >= 4);
1062
1063 std::stringstream shaderStream;
1064
1065 const bool pointSprites = (primitiveType == gl::PrimitiveMode::Points) && pointSpriteEmulation;
1066 const bool usesPointCoord = preambleString.find("gl_PointCoord") != std::string::npos;
1067
1068 const char *inputPT = nullptr;
1069 const char *outputPT = nullptr;
1070 int inputSize = 0;
1071 int maxVertexOutput = 0;
1072
1073 switch (primitiveType)
1074 {
1075 case gl::PrimitiveMode::Points:
1076 inputPT = "point";
1077 inputSize = 1;
1078
1079 if (pointSprites)
1080 {
1081 outputPT = "Triangle";
1082 maxVertexOutput = 4;
1083 }
1084 else
1085 {
1086 outputPT = "Point";
1087 maxVertexOutput = 1;
1088 }
1089
1090 break;
1091
1092 case gl::PrimitiveMode::Lines:
1093 case gl::PrimitiveMode::LineStrip:
1094 case gl::PrimitiveMode::LineLoop:
1095 inputPT = "line";
1096 outputPT = "Line";
1097 inputSize = 2;
1098 maxVertexOutput = 2;
1099 break;
1100
1101 case gl::PrimitiveMode::Triangles:
1102 case gl::PrimitiveMode::TriangleStrip:
1103 case gl::PrimitiveMode::TriangleFan:
1104 inputPT = "triangle";
1105 outputPT = "Triangle";
1106 inputSize = 3;
1107 maxVertexOutput = 3;
1108 break;
1109
1110 default:
1111 UNREACHABLE();
1112 break;
1113 }
1114
1115 if (pointSprites || hasANGLEMultiviewEnabled)
1116 {
1117 shaderStream << "cbuffer DriverConstants : register(b0)\n"
1118 "{\n";
1119
1120 if (pointSprites)
1121 {
1122 shaderStream << " float4 dx_ViewCoords : packoffset(c1);\n";
1123 if (useViewScale)
1124 {
1125 shaderStream << " float2 dx_ViewScale : packoffset(c3);\n";
1126 }
1127 }
1128
1129 if (hasANGLEMultiviewEnabled)
1130 {
1131 // We have to add a value which we can use to keep track of which multi-view code path
1132 // is to be selected in the GS.
1133 shaderStream << " float multiviewSelectViewportIndex : packoffset(c3.z);\n";
1134 }
1135
1136 shaderStream << "};\n\n";
1137 }
1138
1139 if (pointSprites)
1140 {
1141 shaderStream << "#define ANGLE_POINT_SPRITE_SHADER\n"
1142 "\n"
1143 "static float2 pointSpriteCorners[] = \n"
1144 "{\n"
1145 " float2( 0.5f, -0.5f),\n"
1146 " float2( 0.5f, 0.5f),\n"
1147 " float2(-0.5f, -0.5f),\n"
1148 " float2(-0.5f, 0.5f)\n"
1149 "};\n"
1150 "\n"
1151 "static float2 pointSpriteTexcoords[] = \n"
1152 "{\n"
1153 " float2(1.0f, 1.0f),\n"
1154 " float2(1.0f, 0.0f),\n"
1155 " float2(0.0f, 1.0f),\n"
1156 " float2(0.0f, 0.0f)\n"
1157 "};\n"
1158 "\n"
1159 "static float minPointSize = "
1160 << static_cast<int>(caps.minAliasedPointSize)
1161 << ".0f;\n"
1162 "static float maxPointSize = "
1163 << static_cast<int>(caps.maxAliasedPointSize) << ".0f;\n"
1164 << "\n";
1165 }
1166
1167 shaderStream << preambleString << "\n"
1168 << "[maxvertexcount(" << maxVertexOutput << ")]\n"
1169 << "void main(" << inputPT << " GS_INPUT input[" << inputSize << "], ";
1170
1171 if (primitiveType == gl::PrimitiveMode::TriangleStrip)
1172 {
1173 shaderStream << "uint primitiveID : SV_PrimitiveID, ";
1174 }
1175
1176 shaderStream << " inout " << outputPT << "Stream<GS_OUTPUT> outStream)\n"
1177 << "{\n"
1178 << " GS_OUTPUT output = (GS_OUTPUT)0;\n";
1179
1180 if (primitiveType == gl::PrimitiveMode::TriangleStrip)
1181 {
1182 shaderStream << " uint lastVertexIndex = (primitiveID % 2 == 0 ? 2 : 1);\n";
1183 }
1184 else
1185 {
1186 shaderStream << " uint lastVertexIndex = " << (inputSize - 1) << ";\n";
1187 }
1188
1189 for (int vertexIndex = 0; vertexIndex < inputSize; ++vertexIndex)
1190 {
1191 shaderStream << " copyVertex(output, input[" << vertexIndex
1192 << "], input[lastVertexIndex]);\n";
1193 if (hasANGLEMultiviewEnabled && !selectViewInVS)
1194 {
1195 shaderStream << " selectView(output, input[" << vertexIndex << "]);\n";
1196 }
1197 if (!pointSprites)
1198 {
1199 ASSERT(inputSize == maxVertexOutput);
1200 shaderStream << " outStream.Append(output);\n";
1201 }
1202 }
1203
1204 if (pointSprites)
1205 {
1206 shaderStream << "\n"
1207 " float4 dx_Position = input[0].dx_Position;\n"
1208 " float gl_PointSize = clamp(input[0].gl_PointSize, minPointSize, "
1209 "maxPointSize);\n"
1210 " float2 viewportScale = float2(1.0f / dx_ViewCoords.x, 1.0f / "
1211 "dx_ViewCoords.y) * dx_Position.w;\n";
1212
1213 for (int corner = 0; corner < 4; corner++)
1214 {
1215 if (useViewScale)
1216 {
1217 shaderStream << " \n"
1218 " output.dx_Position = dx_Position + float4(1.0f, "
1219 "-dx_ViewScale.y, 1.0f, 1.0f)"
1220 " * float4(pointSpriteCorners["
1221 << corner << "] * viewportScale * gl_PointSize, 0.0f, 0.0f);\n";
1222 }
1223 else
1224 {
1225 shaderStream << "\n"
1226 " output.dx_Position = dx_Position + float4(pointSpriteCorners["
1227 << corner << "] * viewportScale * gl_PointSize, 0.0f, 0.0f);\n";
1228 }
1229
1230 if (usesPointCoord)
1231 {
1232 shaderStream << " output.gl_PointCoord = pointSpriteTexcoords[" << corner
1233 << "];\n";
1234 }
1235
1236 shaderStream << " outStream.Append(output);\n";
1237 }
1238 }
1239
1240 shaderStream << " \n"
1241 " outStream.RestartStrip();\n"
1242 "}\n";
1243
1244 return shaderStream.str();
1245 }
1246
1247 // static
GenerateAttributeConversionHLSL(angle::FormatID vertexFormatID,const sh::ShaderVariable & shaderAttrib,std::ostringstream & outStream)1248 void DynamicHLSL::GenerateAttributeConversionHLSL(angle::FormatID vertexFormatID,
1249 const sh::ShaderVariable &shaderAttrib,
1250 std::ostringstream &outStream)
1251 {
1252 // Matrix
1253 if (IsMatrixType(shaderAttrib.type))
1254 {
1255 outStream << "transpose(input." << DecorateVariable(shaderAttrib.name) << ")";
1256 return;
1257 }
1258
1259 GLenum shaderComponentType = VariableComponentType(shaderAttrib.type);
1260 int shaderComponentCount = VariableComponentCount(shaderAttrib.type);
1261 const gl::VertexFormat &vertexFormat = gl::GetVertexFormatFromID(vertexFormatID);
1262
1263 // Perform integer to float conversion (if necessary)
1264 if (shaderComponentType == GL_FLOAT && vertexFormat.type != GL_FLOAT)
1265 {
1266 // TODO: normalization for 32-bit integer formats
1267 ASSERT(!vertexFormat.normalized && !vertexFormat.pureInteger);
1268 outStream << "float" << shaderComponentCount << "(input."
1269 << DecorateVariable(shaderAttrib.name) << ")";
1270 return;
1271 }
1272
1273 // No conversion necessary
1274 outStream << "input." << DecorateVariable(shaderAttrib.name);
1275 }
1276
getPixelShaderOutputKey(const gl::State & data,const gl::ProgramState & programData,const ProgramD3DMetadata & metadata,std::vector<PixelShaderOutputVariable> * outPixelShaderKey)1277 void DynamicHLSL::getPixelShaderOutputKey(const gl::State &data,
1278 const gl::ProgramState &programData,
1279 const ProgramD3DMetadata &metadata,
1280 std::vector<PixelShaderOutputVariable> *outPixelShaderKey)
1281 {
1282 // Two cases when writing to gl_FragColor and using ESSL 1.0:
1283 // - with a 3.0 context, the output color is copied to channel 0
1284 // - with a 2.0 context, the output color is broadcast to all channels
1285 bool broadcast = metadata.usesBroadcast(data);
1286 const unsigned int numRenderTargets =
1287 (broadcast || metadata.usesMultipleFragmentOuts()
1288 ? static_cast<unsigned int>(data.getCaps().maxDrawBuffers)
1289 : 1);
1290
1291 if (!metadata.usesCustomOutVars())
1292 {
1293 for (unsigned int renderTargetIndex = 0; renderTargetIndex < numRenderTargets;
1294 renderTargetIndex++)
1295 {
1296 PixelShaderOutputVariable outputKeyVariable;
1297 outputKeyVariable.type = GL_FLOAT_VEC4;
1298 outputKeyVariable.name = "gl_Color" + Str(renderTargetIndex);
1299 outputKeyVariable.source =
1300 broadcast ? "gl_Color[0]" : "gl_Color[" + Str(renderTargetIndex) + "]";
1301 outputKeyVariable.outputLocation = renderTargetIndex;
1302
1303 outPixelShaderKey->push_back(outputKeyVariable);
1304 }
1305
1306 if (metadata.usesSecondaryColor())
1307 {
1308 for (unsigned int secondaryIndex = 0;
1309 secondaryIndex < data.getCaps().maxDualSourceDrawBuffers; secondaryIndex++)
1310 {
1311 PixelShaderOutputVariable outputKeyVariable;
1312 outputKeyVariable.type = GL_FLOAT_VEC4;
1313 outputKeyVariable.name = "gl_SecondaryColor" + Str(secondaryIndex);
1314 outputKeyVariable.source = "gl_SecondaryColor[" + Str(secondaryIndex) + "]";
1315 outputKeyVariable.outputLocation = secondaryIndex;
1316 outputKeyVariable.outputIndex = 1;
1317
1318 outPixelShaderKey->push_back(outputKeyVariable);
1319 }
1320 }
1321 }
1322 else
1323 {
1324 const ShaderD3D *fragmentShader = metadata.getFragmentShader();
1325
1326 if (!fragmentShader)
1327 {
1328 return;
1329 }
1330
1331 const auto &shaderOutputVars = fragmentShader->getState().getActiveOutputVariables();
1332
1333 for (size_t outputLocationIndex = 0u;
1334 outputLocationIndex < programData.getOutputLocations().size(); ++outputLocationIndex)
1335 {
1336 const VariableLocation &outputLocation =
1337 programData.getOutputLocations().at(outputLocationIndex);
1338 if (!outputLocation.used())
1339 {
1340 continue;
1341 }
1342 const sh::ShaderVariable &outputVariable = shaderOutputVars[outputLocation.index];
1343 const std::string &variableName = "out_" + outputVariable.name;
1344
1345 // Fragment outputs can't be arrays of arrays. ESSL 3.10 section 4.3.6.
1346 const std::string &elementString =
1347 (outputVariable.isArray() ? Str(outputLocation.arrayIndex) : "");
1348
1349 ASSERT(outputVariable.active);
1350
1351 PixelShaderOutputVariable outputKeyVariable;
1352 outputKeyVariable.type = outputVariable.type;
1353 outputKeyVariable.name = variableName + elementString;
1354 outputKeyVariable.source =
1355 variableName +
1356 (outputVariable.isArray() ? ArrayString(outputLocation.arrayIndex) : "");
1357 outputKeyVariable.outputLocation = outputLocationIndex;
1358
1359 outPixelShaderKey->push_back(outputKeyVariable);
1360 }
1361
1362 // Now generate any secondary outputs...
1363 for (size_t outputLocationIndex = 0u;
1364 outputLocationIndex < programData.getSecondaryOutputLocations().size();
1365 ++outputLocationIndex)
1366 {
1367 const VariableLocation &outputLocation =
1368 programData.getSecondaryOutputLocations().at(outputLocationIndex);
1369 if (!outputLocation.used())
1370 {
1371 continue;
1372 }
1373 const sh::ShaderVariable &outputVariable = shaderOutputVars[outputLocation.index];
1374 const std::string &variableName = "out_" + outputVariable.name;
1375
1376 // Fragment outputs can't be arrays of arrays. ESSL 3.10 section 4.3.6.
1377 const std::string &elementString =
1378 (outputVariable.isArray() ? Str(outputLocation.arrayIndex) : "");
1379
1380 ASSERT(outputVariable.active);
1381
1382 PixelShaderOutputVariable outputKeyVariable;
1383 outputKeyVariable.type = outputVariable.type;
1384 outputKeyVariable.name = variableName + elementString;
1385 outputKeyVariable.source =
1386 variableName +
1387 (outputVariable.isArray() ? ArrayString(outputLocation.arrayIndex) : "");
1388 outputKeyVariable.outputLocation = outputLocationIndex;
1389 outputKeyVariable.outputIndex = 1;
1390
1391 outPixelShaderKey->push_back(outputKeyVariable);
1392 }
1393 }
1394 }
1395
1396 // BuiltinVarying Implementation.
BuiltinVarying()1397 BuiltinVarying::BuiltinVarying() : enabled(false), index(0), systemValue(false) {}
1398
str() const1399 std::string BuiltinVarying::str() const
1400 {
1401 return (systemValue ? semantic : (semantic + Str(index)));
1402 }
1403
enableSystem(const std::string & systemValueSemantic)1404 void BuiltinVarying::enableSystem(const std::string &systemValueSemantic)
1405 {
1406 enabled = true;
1407 semantic = systemValueSemantic;
1408 systemValue = true;
1409 }
1410
enable(const std::string & semanticVal,unsigned int indexVal)1411 void BuiltinVarying::enable(const std::string &semanticVal, unsigned int indexVal)
1412 {
1413 enabled = true;
1414 semantic = semanticVal;
1415 index = indexVal;
1416 }
1417
1418 // BuiltinVaryingsD3D Implementation.
BuiltinVaryingsD3D(const ProgramD3DMetadata & metadata,const VaryingPacking & packing)1419 BuiltinVaryingsD3D::BuiltinVaryingsD3D(const ProgramD3DMetadata &metadata,
1420 const VaryingPacking &packing)
1421 {
1422 updateBuiltins(gl::ShaderType::Vertex, metadata, packing);
1423 updateBuiltins(gl::ShaderType::Fragment, metadata, packing);
1424 int shaderModel = metadata.getRendererMajorShaderModel();
1425 if (shaderModel >= 4)
1426 {
1427 updateBuiltins(gl::ShaderType::Geometry, metadata, packing);
1428 }
1429 // In shader model >= 4, some builtins need to be the same in vertex and pixel shaders - input
1430 // struct needs to be a prefix of output struct.
1431 ASSERT(shaderModel < 4 || mBuiltinInfo[gl::ShaderType::Vertex].glPosition.enabled ==
1432 mBuiltinInfo[gl::ShaderType::Fragment].glPosition.enabled);
1433 ASSERT(shaderModel < 4 || mBuiltinInfo[gl::ShaderType::Vertex].glFragCoord.enabled ==
1434 mBuiltinInfo[gl::ShaderType::Fragment].glFragCoord.enabled);
1435 ASSERT(shaderModel < 4 || mBuiltinInfo[gl::ShaderType::Vertex].glPointCoord.enabled ==
1436 mBuiltinInfo[gl::ShaderType::Fragment].glPointCoord.enabled);
1437 ASSERT(shaderModel < 4 || mBuiltinInfo[gl::ShaderType::Vertex].glPointSize.enabled ==
1438 mBuiltinInfo[gl::ShaderType::Fragment].glPointSize.enabled);
1439 ASSERT(shaderModel < 4 || mBuiltinInfo[gl::ShaderType::Vertex].glViewIDOVR.enabled ==
1440 mBuiltinInfo[gl::ShaderType::Fragment].glViewIDOVR.enabled);
1441 }
1442
1443 BuiltinVaryingsD3D::~BuiltinVaryingsD3D() = default;
1444
updateBuiltins(gl::ShaderType shaderType,const ProgramD3DMetadata & metadata,const VaryingPacking & packing)1445 void BuiltinVaryingsD3D::updateBuiltins(gl::ShaderType shaderType,
1446 const ProgramD3DMetadata &metadata,
1447 const VaryingPacking &packing)
1448 {
1449 const std::string &userSemantic = GetVaryingSemantic(metadata.getRendererMajorShaderModel(),
1450 metadata.usesSystemValuePointSize());
1451
1452 // Note that when enabling builtins only for specific shader stages in shader model >= 4, the
1453 // code needs to ensure that the input struct of the shader stage is a prefix of the output
1454 // struct of the previous stage.
1455
1456 unsigned int reservedSemanticIndex = packing.getMaxSemanticIndex();
1457
1458 BuiltinInfo *builtins = &mBuiltinInfo[shaderType];
1459
1460 if (metadata.getRendererMajorShaderModel() >= 4)
1461 {
1462 builtins->dxPosition.enableSystem("SV_Position");
1463 }
1464 else if (shaderType == gl::ShaderType::Fragment)
1465 {
1466 builtins->dxPosition.enableSystem("VPOS");
1467 }
1468 else
1469 {
1470 builtins->dxPosition.enableSystem("POSITION");
1471 }
1472
1473 if (metadata.usesTransformFeedbackGLPosition())
1474 {
1475 builtins->glPosition.enable(userSemantic, reservedSemanticIndex++);
1476 }
1477
1478 if (metadata.usesFragCoord())
1479 {
1480 builtins->glFragCoord.enable(userSemantic, reservedSemanticIndex++);
1481 }
1482
1483 if (shaderType == gl::ShaderType::Vertex ? metadata.addsPointCoordToVertexShader()
1484 : metadata.usesPointCoord())
1485 {
1486 // SM3 reserves the TEXCOORD semantic for point sprite texcoords (gl_PointCoord)
1487 // In D3D11 we manually compute gl_PointCoord in the GS.
1488 if (metadata.getRendererMajorShaderModel() >= 4)
1489 {
1490 builtins->glPointCoord.enable(userSemantic, reservedSemanticIndex++);
1491 }
1492 else
1493 {
1494 builtins->glPointCoord.enable("TEXCOORD", 0);
1495 }
1496 }
1497
1498 if (metadata.hasANGLEMultiviewEnabled())
1499 {
1500 // Although it is possible to compute gl_ViewID_OVR from the value of
1501 // SV_ViewportArrayIndex or SV_RenderTargetArrayIndex and the multi-view state in the
1502 // driver constant buffer, it is easier and cleaner to always pass it as a varying.
1503 builtins->glViewIDOVR.enable(userSemantic, reservedSemanticIndex++);
1504
1505 if (shaderType == gl::ShaderType::Vertex)
1506 {
1507 if (metadata.canSelectViewInVertexShader())
1508 {
1509 builtins->glViewportIndex.enableSystem("SV_ViewportArrayIndex");
1510 builtins->glLayer.enableSystem("SV_RenderTargetArrayIndex");
1511 }
1512 }
1513
1514 if (shaderType == gl::ShaderType::Geometry)
1515 {
1516 // gl_Layer and gl_ViewportIndex are necessary so that we can write to either based on
1517 // the multiview state in the driver constant buffer.
1518 builtins->glViewportIndex.enableSystem("SV_ViewportArrayIndex");
1519 builtins->glLayer.enableSystem("SV_RenderTargetArrayIndex");
1520 }
1521 }
1522
1523 // Special case: do not include PSIZE semantic in HLSL 3 pixel shaders
1524 if (metadata.usesSystemValuePointSize() &&
1525 (shaderType != gl::ShaderType::Fragment || metadata.getRendererMajorShaderModel() >= 4))
1526 {
1527 builtins->glPointSize.enableSystem("PSIZE");
1528 }
1529 }
1530
1531 } // namespace rx
1532