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 
HLSLComponentTypeString(GLenum componentType)31 const char *HLSLComponentTypeString(GLenum componentType)
32 {
33     switch (componentType)
34     {
35         case GL_UNSIGNED_INT:
36             return "uint";
37         case GL_INT:
38             return "int";
39         case GL_UNSIGNED_NORMALIZED:
40         case GL_SIGNED_NORMALIZED:
41         case GL_FLOAT:
42             return "float";
43         default:
44             UNREACHABLE();
45             return "not-component-type";
46     }
47 }
48 
HLSLComponentTypeString(std::ostringstream & ostream,GLenum componentType,int componentCount)49 void HLSLComponentTypeString(std::ostringstream &ostream, GLenum componentType, int componentCount)
50 {
51     ostream << HLSLComponentTypeString(componentType);
52     if (componentCount > 1)
53     {
54         ostream << componentCount;
55     }
56 }
57 
HLSLMatrixTypeString(GLenum type)58 const char *HLSLMatrixTypeString(GLenum type)
59 {
60     switch (type)
61     {
62         case GL_FLOAT_MAT2:
63             return "float2x2";
64         case GL_FLOAT_MAT3:
65             return "float3x3";
66         case GL_FLOAT_MAT4:
67             return "float4x4";
68         case GL_FLOAT_MAT2x3:
69             return "float2x3";
70         case GL_FLOAT_MAT3x2:
71             return "float3x2";
72         case GL_FLOAT_MAT2x4:
73             return "float2x4";
74         case GL_FLOAT_MAT4x2:
75             return "float4x2";
76         case GL_FLOAT_MAT3x4:
77             return "float3x4";
78         case GL_FLOAT_MAT4x3:
79             return "float4x3";
80         default:
81             UNREACHABLE();
82             return "not-matrix-type";
83     }
84 }
85 
HLSLTypeString(std::ostringstream & ostream,GLenum type)86 void HLSLTypeString(std::ostringstream &ostream, GLenum type)
87 {
88     if (gl::IsMatrixType(type))
89     {
90         ostream << HLSLMatrixTypeString(type);
91         return;
92     }
93 
94     HLSLComponentTypeString(ostream, gl::VariableComponentType(type),
95                             gl::VariableComponentCount(type));
96 }
97 
FindOutputAtLocation(const std::vector<PixelShaderOutputVariable> & outputVariables,unsigned int location,size_t index=0)98 const PixelShaderOutputVariable *FindOutputAtLocation(
99     const std::vector<PixelShaderOutputVariable> &outputVariables,
100     unsigned int location,
101     size_t index = 0)
102 {
103     for (auto &outputVar : outputVariables)
104     {
105         if (outputVar.outputLocation == location && outputVar.outputIndex == index)
106         {
107             return &outputVar;
108         }
109     }
110 
111     return nullptr;
112 }
113 
WriteArrayString(std::ostringstream & strstr,unsigned int i)114 void WriteArrayString(std::ostringstream &strstr, unsigned int i)
115 {
116     static_assert(GL_INVALID_INDEX == UINT_MAX,
117                   "GL_INVALID_INDEX must be equal to the max unsigned int.");
118     if (i == UINT_MAX)
119     {
120         return;
121     }
122 
123     strstr << "[";
124     strstr << i;
125     strstr << "]";
126 }
127 
128 constexpr const char *VERTEX_ATTRIBUTE_STUB_STRING      = "@@ VERTEX ATTRIBUTES @@";
129 constexpr const char *VERTEX_OUTPUT_STUB_STRING         = "@@ VERTEX OUTPUT @@";
130 constexpr const char *PIXEL_OUTPUT_STUB_STRING          = "@@ PIXEL OUTPUT @@";
131 constexpr const char *PIXEL_MAIN_PARAMETERS_STUB_STRING = "@@ PIXEL MAIN PARAMETERS @@";
132 constexpr const char *MAIN_PROLOGUE_STUB_STRING         = "@@ MAIN PROLOGUE @@";
133 }  // anonymous namespace
134 
135 // BuiltinInfo implementation
136 
137 BuiltinInfo::BuiltinInfo()  = default;
138 BuiltinInfo::~BuiltinInfo() = default;
139 
140 // DynamicHLSL implementation
141 
DynamicHLSL(RendererD3D * const renderer)142 DynamicHLSL::DynamicHLSL(RendererD3D *const renderer) : mRenderer(renderer) {}
143 
generateVertexShaderForInputLayout(const std::string & sourceShader,const InputLayout & inputLayout,const std::vector<sh::ShaderVariable> & shaderAttributes) const144 std::string DynamicHLSL::generateVertexShaderForInputLayout(
145     const std::string &sourceShader,
146     const InputLayout &inputLayout,
147     const std::vector<sh::ShaderVariable> &shaderAttributes) const
148 {
149     std::ostringstream structStream;
150     std::ostringstream initStream;
151 
152     structStream << "struct VS_INPUT\n"
153                  << "{\n";
154 
155     int semanticIndex       = 0;
156     unsigned int inputIndex = 0;
157 
158     // If gl_PointSize is used in the shader then pointsprites rendering is expected.
159     // If the renderer does not support Geometry shaders then Instanced PointSprite emulation
160     // must be used.
161     bool usesPointSize = sourceShader.find("GL_USES_POINT_SIZE") != std::string::npos;
162     bool useInstancedPointSpriteEmulation =
163         usesPointSize && mRenderer->getFeatures().useInstancedPointSpriteEmulation.enabled;
164 
165     // Instanced PointSprite emulation requires additional entries in the
166     // VS_INPUT structure to support the vertices that make up the quad vertices.
167     // These values must be in sync with the cooresponding values added during inputlayout creation
168     // in InputLayoutCache::applyVertexBuffers().
169     //
170     // The additional entries must appear first in the VS_INPUT layout because
171     // Windows Phone 8 era devices require per vertex data to physically come
172     // before per instance data in the shader.
173     if (useInstancedPointSpriteEmulation)
174     {
175         structStream << "    float3 spriteVertexPos : SPRITEPOSITION0;\n"
176                      << "    float2 spriteTexCoord : SPRITETEXCOORD0;\n";
177     }
178 
179     for (size_t attributeIndex = 0; attributeIndex < shaderAttributes.size(); ++attributeIndex)
180     {
181         const sh::ShaderVariable &shaderAttribute = shaderAttributes[attributeIndex];
182         if (!shaderAttribute.name.empty())
183         {
184             ASSERT(inputIndex < MAX_VERTEX_ATTRIBS);
185             angle::FormatID vertexFormatID =
186                 inputIndex < inputLayout.size() ? inputLayout[inputIndex] : angle::FormatID::NONE;
187 
188             // HLSL code for input structure
189             if (IsMatrixType(shaderAttribute.type))
190             {
191                 // Matrix types are always transposed
192                 structStream << "    "
193                              << HLSLMatrixTypeString(TransposeMatrixType(shaderAttribute.type));
194             }
195             else
196             {
197                 if (shaderAttribute.name == "gl_InstanceID" ||
198                     shaderAttribute.name == "gl_VertexID")
199                 {
200                     // The input types of the instance ID and vertex ID in HLSL (uint) differs from
201                     // the ones in ESSL (int).
202                     structStream << " uint";
203                 }
204                 else
205                 {
206                     GLenum componentType = mRenderer->getVertexComponentType(vertexFormatID);
207 
208                     structStream << "    ";
209                     HLSLComponentTypeString(structStream, componentType,
210                                             VariableComponentCount(shaderAttribute.type));
211                 }
212             }
213 
214             structStream << " " << DecorateVariable(shaderAttribute.name) << " : ";
215 
216             if (shaderAttribute.name == "gl_InstanceID")
217             {
218                 structStream << "SV_InstanceID";
219             }
220             else if (shaderAttribute.name == "gl_VertexID")
221             {
222                 structStream << "SV_VertexID";
223             }
224             else
225             {
226                 structStream << "TEXCOORD" << semanticIndex;
227                 semanticIndex += VariableRegisterCount(shaderAttribute.type);
228             }
229 
230             structStream << ";\n";
231 
232             // HLSL code for initialization
233             initStream << "    " << DecorateVariable(shaderAttribute.name) << " = ";
234 
235             // Mismatched vertex attribute to vertex input may result in an undefined
236             // data reinterpretation (eg for pure integer->float, float->pure integer)
237             // TODO: issue warning with gl debug info extension, when supported
238             if (IsMatrixType(shaderAttribute.type) ||
239                 (mRenderer->getVertexConversionType(vertexFormatID) & VERTEX_CONVERT_GPU) != 0)
240             {
241                 GenerateAttributeConversionHLSL(vertexFormatID, shaderAttribute, initStream);
242             }
243             else
244             {
245                 initStream << "input." << DecorateVariable(shaderAttribute.name);
246             }
247 
248             if (shaderAttribute.name == "gl_VertexID")
249             {
250                 // dx_VertexID contains the firstVertex offset
251                 initStream << " + dx_VertexID";
252             }
253 
254             initStream << ";\n";
255 
256             inputIndex += VariableRowCount(TransposeMatrixType(shaderAttribute.type));
257         }
258     }
259 
260     structStream << "};\n"
261                     "\n"
262                     "void initAttributes(VS_INPUT input)\n"
263                     "{\n"
264                  << initStream.str() << "}\n";
265 
266     std::string vertexHLSL(sourceShader);
267 
268     bool success =
269         angle::ReplaceSubstring(&vertexHLSL, VERTEX_ATTRIBUTE_STUB_STRING, structStream.str());
270     ASSERT(success);
271 
272     return vertexHLSL;
273 }
274 
generatePixelShaderForOutputSignature(const std::string & sourceShader,const std::vector<PixelShaderOutputVariable> & outputVariables,bool usesFragDepth,const std::vector<GLenum> & outputLayout) const275 std::string DynamicHLSL::generatePixelShaderForOutputSignature(
276     const std::string &sourceShader,
277     const std::vector<PixelShaderOutputVariable> &outputVariables,
278     bool usesFragDepth,
279     const std::vector<GLenum> &outputLayout) const
280 {
281     const int shaderModel      = mRenderer->getMajorShaderModel();
282     std::string targetSemantic = (shaderModel >= 4) ? "SV_TARGET" : "COLOR";
283     std::string depthSemantic  = (shaderModel >= 4) ? "SV_Depth" : "DEPTH";
284 
285     std::ostringstream declarationStream;
286     std::ostringstream copyStream;
287 
288     declarationStream << "struct PS_OUTPUT\n"
289                          "{\n";
290 
291     size_t numOutputs = outputLayout.size();
292 
293     // Workaround for HLSL 3.x: We can't do a depth/stencil only render, the runtime will complain.
294     if (numOutputs == 0 && (shaderModel == 3 || !mRenderer->getShaderModelSuffix().empty()))
295     {
296         numOutputs = 1u;
297     }
298     const PixelShaderOutputVariable defaultOutput(GL_FLOAT_VEC4, "unused", "float4(0, 0, 0, 1)", 0,
299                                                   0);
300     size_t outputIndex = 0;
301 
302     for (size_t layoutIndex = 0; layoutIndex < numOutputs; ++layoutIndex)
303     {
304         GLenum binding = outputLayout.empty() ? GL_COLOR_ATTACHMENT0 : outputLayout[layoutIndex];
305 
306         if (binding != GL_NONE)
307         {
308             unsigned int location = (binding - GL_COLOR_ATTACHMENT0);
309             outputIndex =
310                 layoutIndex > 0 && binding == outputLayout[layoutIndex - 1] ? outputIndex + 1 : 0;
311 
312             const PixelShaderOutputVariable *outputVariable =
313                 outputLayout.empty() ? &defaultOutput
314                                      : FindOutputAtLocation(outputVariables, location, outputIndex);
315 
316             // OpenGL ES 3.0 spec $4.2.1
317             // If [...] not all user-defined output variables are written, the values of fragment
318             // colors corresponding to unwritten variables are similarly undefined.
319             if (outputVariable)
320             {
321                 declarationStream << "    ";
322                 HLSLTypeString(declarationStream, outputVariable->type);
323                 declarationStream << " " << outputVariable->name << " : " << targetSemantic
324                                   << static_cast<int>(layoutIndex) << ";\n";
325 
326                 copyStream << "    output." << outputVariable->name << " = "
327                            << outputVariable->source << ";\n";
328             }
329         }
330     }
331 
332     if (usesFragDepth)
333     {
334         declarationStream << "    float gl_Depth : " << depthSemantic << ";\n";
335         copyStream << "    output.gl_Depth = gl_Depth; \n";
336     }
337 
338     declarationStream << "};\n"
339                          "\n"
340                          "PS_OUTPUT generateOutput()\n"
341                          "{\n"
342                          "    PS_OUTPUT output;\n"
343                       << copyStream.str()
344                       << "    return output;\n"
345                          "}\n";
346 
347     std::string pixelHLSL(sourceShader);
348 
349     bool success =
350         angle::ReplaceSubstring(&pixelHLSL, PIXEL_OUTPUT_STUB_STRING, declarationStream.str());
351     ASSERT(success);
352 
353     return pixelHLSL;
354 }
355 
generateComputeShaderForImage2DBindSignature(const d3d::Context * context,ProgramD3D & programD3D,const gl::ProgramState & programData,std::vector<sh::ShaderVariable> & image2DUniforms,const gl::ImageUnitTextureTypeMap & image2DBindLayout) const356 std::string DynamicHLSL::generateComputeShaderForImage2DBindSignature(
357     const d3d::Context *context,
358     ProgramD3D &programD3D,
359     const gl::ProgramState &programData,
360     std::vector<sh::ShaderVariable> &image2DUniforms,
361     const gl::ImageUnitTextureTypeMap &image2DBindLayout) const
362 {
363     std::string computeHLSL(
364         programData.getAttachedShader(ShaderType::Compute)->getTranslatedSource());
365 
366     if (image2DUniforms.empty())
367     {
368         return computeHLSL;
369     }
370 
371     return GenerateComputeShaderForImage2DBindSignature(context, programD3D, programData,
372                                                         image2DUniforms, image2DBindLayout);
373 }
374 
generateVaryingLinkHLSL(const VaryingPacking & varyingPacking,const BuiltinInfo & builtins,bool programUsesPointSize,std::ostringstream & hlslStream) const375 void DynamicHLSL::generateVaryingLinkHLSL(const VaryingPacking &varyingPacking,
376                                           const BuiltinInfo &builtins,
377                                           bool programUsesPointSize,
378                                           std::ostringstream &hlslStream) const
379 {
380     ASSERT(builtins.dxPosition.enabled);
381     hlslStream << "{\n"
382                << "    float4 dx_Position : " << builtins.dxPosition.str() << ";\n";
383 
384     if (builtins.glPosition.enabled)
385     {
386         hlslStream << "    float4 gl_Position : " << builtins.glPosition.str() << ";\n";
387     }
388 
389     if (builtins.glFragCoord.enabled)
390     {
391         hlslStream << "    float4 gl_FragCoord : " << builtins.glFragCoord.str() << ";\n";
392     }
393 
394     if (builtins.glPointCoord.enabled)
395     {
396         hlslStream << "    float2 gl_PointCoord : " << builtins.glPointCoord.str() << ";\n";
397     }
398 
399     if (builtins.glPointSize.enabled)
400     {
401         hlslStream << "    float gl_PointSize : " << builtins.glPointSize.str() << ";\n";
402     }
403 
404     if (builtins.glViewIDOVR.enabled)
405     {
406         hlslStream << "    nointerpolation uint gl_ViewID_OVR : " << builtins.glViewIDOVR.str()
407                    << ";\n";
408     }
409 
410     std::string varyingSemantic =
411         GetVaryingSemantic(mRenderer->getMajorShaderModel(), programUsesPointSize);
412 
413     const auto &registerInfos = varyingPacking.getRegisterList();
414     for (GLuint registerIndex = 0u; registerIndex < registerInfos.size(); ++registerIndex)
415     {
416         const PackedVaryingRegister &registerInfo = registerInfos[registerIndex];
417         const auto &varying                       = registerInfo.packedVarying->varying();
418         ASSERT(!varying.isStruct());
419 
420         // TODO: Add checks to ensure D3D interpolation modifiers don't result in too many
421         // registers being used.
422         // For example, if there are N registers, and we have N vec3 varyings and 1 float
423         // varying, then D3D will pack them into N registers.
424         // If the float varying has the 'nointerpolation' modifier on it then we would need
425         // N + 1 registers, and D3D compilation will fail.
426 
427         switch (registerInfo.packedVarying->interpolation)
428         {
429             case sh::INTERPOLATION_SMOOTH:
430                 hlslStream << "    ";
431                 break;
432             case sh::INTERPOLATION_FLAT:
433                 hlslStream << "    nointerpolation ";
434                 break;
435             case sh::INTERPOLATION_CENTROID:
436                 hlslStream << "    centroid ";
437                 break;
438             case sh::INTERPOLATION_SAMPLE:
439                 hlslStream << "    sample ";
440                 break;
441             default:
442                 UNREACHABLE();
443         }
444 
445         GLenum transposedType = gl::TransposeMatrixType(varying.type);
446         GLenum componentType  = gl::VariableComponentType(transposedType);
447         int columnCount       = gl::VariableColumnCount(transposedType);
448         HLSLComponentTypeString(hlslStream, componentType, columnCount);
449         hlslStream << " v" << registerIndex << " : " << varyingSemantic << registerIndex << ";\n";
450     }
451 
452     // Note that the following outputs need to be declared after the others. They are not included
453     // in pixel shader inputs even when they are in vertex/geometry shader outputs, and the pixel
454     // shader input struct must be a prefix of the vertex/geometry shader output struct.
455 
456     if (builtins.glViewportIndex.enabled)
457     {
458         hlslStream << "    nointerpolation uint gl_ViewportIndex : "
459                    << builtins.glViewportIndex.str() << ";\n";
460     }
461 
462     if (builtins.glLayer.enabled)
463     {
464         hlslStream << "    nointerpolation uint gl_Layer : " << builtins.glLayer.str() << ";\n";
465     }
466 
467     hlslStream << "};\n";
468 }
469 
generateShaderLinkHLSL(const gl::Caps & caps,const gl::ProgramState & programData,const ProgramD3DMetadata & programMetadata,const VaryingPacking & varyingPacking,const BuiltinVaryingsD3D & builtinsD3D,gl::ShaderMap<std::string> * shaderHLSL) const470 void DynamicHLSL::generateShaderLinkHLSL(const gl::Caps &caps,
471                                          const gl::ProgramState &programData,
472                                          const ProgramD3DMetadata &programMetadata,
473                                          const VaryingPacking &varyingPacking,
474                                          const BuiltinVaryingsD3D &builtinsD3D,
475                                          gl::ShaderMap<std::string> *shaderHLSL) const
476 {
477     ASSERT(shaderHLSL);
478     ASSERT((*shaderHLSL)[gl::ShaderType::Vertex].empty() &&
479            (*shaderHLSL)[gl::ShaderType::Fragment].empty());
480 
481     gl::Shader *vertexShaderGL   = programData.getAttachedShader(ShaderType::Vertex);
482     gl::Shader *fragmentShaderGL = programData.getAttachedShader(ShaderType::Fragment);
483     const int shaderModel        = mRenderer->getMajorShaderModel();
484 
485     const ShaderD3D *fragmentShader = nullptr;
486     if (fragmentShaderGL)
487     {
488         fragmentShader = GetImplAs<ShaderD3D>(fragmentShaderGL);
489     }
490 
491     // usesViewScale() isn't supported in the D3D9 renderer
492     ASSERT(shaderModel >= 4 || !programMetadata.usesViewScale());
493 
494     bool useInstancedPointSpriteEmulation =
495         programMetadata.usesPointSize() &&
496         mRenderer->getFeatures().useInstancedPointSpriteEmulation.enabled;
497 
498     // Validation done in the compiler
499     ASSERT(!fragmentShader || !fragmentShader->usesFragColor() || !fragmentShader->usesFragData());
500 
501     std::ostringstream vertexStream;
502     vertexStream << "struct VS_OUTPUT\n";
503     const auto &vertexBuiltins = builtinsD3D[gl::ShaderType::Vertex];
504     generateVaryingLinkHLSL(varyingPacking, vertexBuiltins, builtinsD3D.usesPointSize(),
505                             vertexStream);
506 
507     // Instanced PointSprite emulation requires additional entries originally generated in the
508     // GeometryShader HLSL. These include pointsize clamp values.
509     if (useInstancedPointSpriteEmulation)
510     {
511         vertexStream << "static float minPointSize = " << static_cast<int>(caps.minAliasedPointSize)
512                      << ".0f;\n"
513                      << "static float maxPointSize = " << static_cast<int>(caps.maxAliasedPointSize)
514                      << ".0f;\n";
515     }
516 
517     std::ostringstream vertexGenerateOutput;
518     vertexGenerateOutput << "VS_OUTPUT generateOutput(VS_INPUT input)\n"
519                          << "{\n"
520                          << "    VS_OUTPUT output;\n";
521 
522     if (vertexBuiltins.glPosition.enabled)
523     {
524         vertexGenerateOutput << "    output.gl_Position = gl_Position;\n";
525     }
526 
527     if (vertexBuiltins.glViewIDOVR.enabled)
528     {
529         vertexGenerateOutput << "    output.gl_ViewID_OVR = ViewID_OVR;\n";
530     }
531     if (programMetadata.hasANGLEMultiviewEnabled() && programMetadata.canSelectViewInVertexShader())
532     {
533         ASSERT(vertexBuiltins.glViewportIndex.enabled && vertexBuiltins.glLayer.enabled);
534         vertexGenerateOutput << "    if (multiviewSelectViewportIndex)\n"
535                              << "    {\n"
536                              << "         output.gl_ViewportIndex = ViewID_OVR;\n"
537                              << "    } else {\n"
538                              << "         output.gl_ViewportIndex = 0;\n"
539                              << "         output.gl_Layer = ViewID_OVR;\n"
540                              << "    }\n";
541     }
542 
543     // On D3D9 or D3D11 Feature Level 9, we need to emulate large viewports using dx_ViewAdjust.
544     if (shaderModel >= 4 && mRenderer->getShaderModelSuffix() == "")
545     {
546         vertexGenerateOutput << "    output.dx_Position.x = gl_Position.x;\n";
547 
548         if (programMetadata.usesViewScale())
549         {
550             // This code assumes that dx_ViewScale.y = -1.0f when rendering to texture, and +1.0f
551             // when rendering to the default framebuffer. No other values are valid.
552             vertexGenerateOutput << "    output.dx_Position.y = dx_ViewScale.y * gl_Position.y;\n";
553         }
554         else
555         {
556             vertexGenerateOutput << "    output.dx_Position.y = - gl_Position.y;\n";
557         }
558 
559         vertexGenerateOutput
560             << "    output.dx_Position.z = (gl_Position.z + gl_Position.w) * 0.5;\n"
561             << "    output.dx_Position.w = gl_Position.w;\n";
562     }
563     else
564     {
565         vertexGenerateOutput << "    output.dx_Position.x = gl_Position.x * dx_ViewAdjust.z + "
566                                 "dx_ViewAdjust.x * gl_Position.w;\n";
567 
568         // If usesViewScale() is true and we're using the D3D11 renderer via Feature Level 9_*,
569         // then we need to multiply the gl_Position.y by the viewScale.
570         // usesViewScale() isn't supported when using the D3D9 renderer.
571         if (programMetadata.usesViewScale() &&
572             (shaderModel >= 4 && mRenderer->getShaderModelSuffix() != ""))
573         {
574             vertexGenerateOutput << "    output.dx_Position.y = dx_ViewScale.y * (gl_Position.y * "
575                                     "dx_ViewAdjust.w + dx_ViewAdjust.y * gl_Position.w);\n";
576         }
577         else
578         {
579             vertexGenerateOutput
580                 << "    output.dx_Position.y = -(gl_Position.y * dx_ViewAdjust.w + "
581                    "dx_ViewAdjust.y * gl_Position.w);\n";
582         }
583 
584         vertexGenerateOutput
585             << "    output.dx_Position.z = (gl_Position.z + gl_Position.w) * 0.5;\n"
586             << "    output.dx_Position.w = gl_Position.w;\n";
587     }
588 
589     // We don't need to output gl_PointSize if we use are emulating point sprites via instancing.
590     if (vertexBuiltins.glPointSize.enabled)
591     {
592         vertexGenerateOutput << "    output.gl_PointSize = gl_PointSize;\n";
593     }
594 
595     if (vertexBuiltins.glFragCoord.enabled)
596     {
597         vertexGenerateOutput << "    output.gl_FragCoord = gl_Position;\n";
598     }
599 
600     const auto &registerInfos = varyingPacking.getRegisterList();
601     for (GLuint registerIndex = 0u; registerIndex < registerInfos.size(); ++registerIndex)
602     {
603         const PackedVaryingRegister &registerInfo = registerInfos[registerIndex];
604         const auto &packedVarying                 = *registerInfo.packedVarying;
605         const auto &varying                       = *packedVarying.frontVarying.varying;
606         ASSERT(!varying.isStruct());
607 
608         vertexGenerateOutput << "    output.v" << registerIndex << " = ";
609 
610         if (packedVarying.isStructField())
611         {
612             vertexGenerateOutput << DecorateVariable(packedVarying.frontVarying.parentStructName)
613                                  << ".";
614         }
615 
616         vertexGenerateOutput << DecorateVariable(varying.name);
617 
618         if (varying.isArray())
619         {
620             WriteArrayString(vertexGenerateOutput, registerInfo.varyingArrayIndex);
621         }
622 
623         if (VariableRowCount(varying.type) > 1)
624         {
625             WriteArrayString(vertexGenerateOutput, registerInfo.varyingRowIndex);
626         }
627 
628         vertexGenerateOutput << ";\n";
629     }
630 
631     // Instanced PointSprite emulation requires additional entries to calculate
632     // the final output vertex positions of the quad that represents each sprite.
633     if (useInstancedPointSpriteEmulation)
634     {
635         vertexGenerateOutput
636             << "\n"
637             << "    gl_PointSize = clamp(gl_PointSize, minPointSize, maxPointSize);\n";
638 
639         vertexGenerateOutput
640             << "    output.dx_Position.x += (input.spriteVertexPos.x * gl_PointSize / "
641                "(dx_ViewCoords.x*2)) * output.dx_Position.w;";
642 
643         if (programMetadata.usesViewScale())
644         {
645             // Multiply by ViewScale to invert the rendering when appropriate
646             vertexGenerateOutput
647                 << "    output.dx_Position.y += (-dx_ViewScale.y * "
648                    "input.spriteVertexPos.y * gl_PointSize / (dx_ViewCoords.y*2)) * "
649                    "output.dx_Position.w;";
650         }
651         else
652         {
653             vertexGenerateOutput
654                 << "    output.dx_Position.y += (input.spriteVertexPos.y * gl_PointSize / "
655                    "(dx_ViewCoords.y*2)) * output.dx_Position.w;";
656         }
657 
658         vertexGenerateOutput
659             << "    output.dx_Position.z += input.spriteVertexPos.z * output.dx_Position.w;\n";
660 
661         if (programMetadata.usesPointCoord())
662         {
663             vertexGenerateOutput << "\n"
664                                  << "    output.gl_PointCoord = input.spriteTexCoord;\n";
665         }
666     }
667 
668     // Renderers that enable instanced pointsprite emulation require the vertex shader output member
669     // gl_PointCoord to be set to a default value if used without gl_PointSize. 0.5,0.5 is the same
670     // default value used in the generated pixel shader.
671     if (programMetadata.usesInsertedPointCoordValue())
672     {
673         ASSERT(!useInstancedPointSpriteEmulation);
674         vertexGenerateOutput << "\n"
675                              << "    output.gl_PointCoord = float2(0.5, 0.5);\n";
676     }
677 
678     vertexGenerateOutput << "\n"
679                          << "    return output;\n"
680                          << "}";
681 
682     if (vertexShaderGL)
683     {
684         std::string vertexSource = vertexShaderGL->getTranslatedSource();
685         angle::ReplaceSubstring(&vertexSource, std::string(MAIN_PROLOGUE_STUB_STRING),
686                                 "    initAttributes(input);\n");
687         angle::ReplaceSubstring(&vertexSource, std::string(VERTEX_OUTPUT_STUB_STRING),
688                                 vertexGenerateOutput.str());
689         vertexStream << vertexSource;
690     }
691 
692     const auto &pixelBuiltins = builtinsD3D[gl::ShaderType::Fragment];
693 
694     std::ostringstream pixelStream;
695     pixelStream << "struct PS_INPUT\n";
696     generateVaryingLinkHLSL(varyingPacking, pixelBuiltins, builtinsD3D.usesPointSize(),
697                             pixelStream);
698     pixelStream << "\n";
699 
700     std::ostringstream pixelPrologue;
701     if (fragmentShader && fragmentShader->usesViewID())
702     {
703         ASSERT(pixelBuiltins.glViewIDOVR.enabled);
704         pixelPrologue << "    ViewID_OVR = input.gl_ViewID_OVR;\n";
705     }
706 
707     if (pixelBuiltins.glFragCoord.enabled)
708     {
709         pixelPrologue << "    float rhw = 1.0 / input.gl_FragCoord.w;\n";
710 
711         // Certain Shader Models (4_0+ and 3_0) allow reading from dx_Position in the pixel shader.
712         // Other Shader Models (4_0_level_9_3 and 2_x) don't support this, so we emulate it using
713         // dx_ViewCoords.
714         if (shaderModel >= 4 && mRenderer->getShaderModelSuffix() == "")
715         {
716             pixelPrologue << "    gl_FragCoord.x = input.dx_Position.x;\n"
717                           << "    gl_FragCoord.y = input.dx_Position.y;\n";
718         }
719         else if (shaderModel == 3)
720         {
721             pixelPrologue << "    gl_FragCoord.x = input.dx_Position.x + 0.5;\n"
722                           << "    gl_FragCoord.y = input.dx_Position.y + 0.5;\n";
723         }
724         else
725         {
726             // dx_ViewCoords contains the viewport width/2, height/2, center.x and center.y. See
727             // Renderer::setViewport()
728             pixelPrologue
729                 << "    gl_FragCoord.x = (input.gl_FragCoord.x * rhw) * dx_ViewCoords.x + "
730                    "dx_ViewCoords.z;\n"
731                 << "    gl_FragCoord.y = (input.gl_FragCoord.y * rhw) * dx_ViewCoords.y + "
732                    "dx_ViewCoords.w;\n";
733         }
734 
735         if (programMetadata.usesViewScale())
736         {
737             // For Feature Level 9_3 and below, we need to correct gl_FragCoord.y to account
738             // for dx_ViewScale. On Feature Level 10_0+, gl_FragCoord.y is calculated above using
739             // dx_ViewCoords and is always correct irrespective of dx_ViewScale's value.
740             // NOTE: usesViewScale() can only be true on D3D11 (i.e. Shader Model 4.0+).
741             if (shaderModel >= 4 && mRenderer->getShaderModelSuffix() == "")
742             {
743                 // Some assumptions:
744                 //  - dx_ViewScale.y = -1.0f when rendering to texture
745                 //  - dx_ViewScale.y = +1.0f when rendering to the default framebuffer
746                 //  - gl_FragCoord.y has been set correctly above.
747                 //
748                 // When rendering to the backbuffer, the code inverts gl_FragCoord's y coordinate.
749                 // This involves subtracting the y coordinate from the height of the area being
750                 // rendered to.
751                 //
752                 // First we calculate the height of the area being rendered to:
753                 //    render_area_height = (2.0f / (1.0f - input.gl_FragCoord.y * rhw)) *
754                 //    gl_FragCoord.y
755                 //
756                 // Note that when we're rendering to default FB, we want our output to be
757                 // equivalent to:
758                 //    "gl_FragCoord.y = render_area_height - gl_FragCoord.y"
759                 //
760                 // When we're rendering to a texture, we want our output to be equivalent to:
761                 //    "gl_FragCoord.y = gl_FragCoord.y;"
762                 //
763                 // If we set scale_factor = ((1.0f + dx_ViewScale.y) / 2.0f), then notice that
764                 //  - When rendering to default FB: scale_factor = 1.0f
765                 //  - When rendering to texture:    scale_factor = 0.0f
766                 //
767                 // Therefore, we can get our desired output by setting:
768                 //    "gl_FragCoord.y = scale_factor * render_area_height - dx_ViewScale.y *
769                 //    gl_FragCoord.y"
770                 //
771                 // Simplifying, this becomes:
772                 pixelPrologue
773                     << "    gl_FragCoord.y = (1.0f + dx_ViewScale.y) * gl_FragCoord.y /"
774                        "(1.0f - input.gl_FragCoord.y * rhw)  - dx_ViewScale.y * gl_FragCoord.y;\n";
775             }
776         }
777 
778         pixelPrologue << "    gl_FragCoord.z = (input.gl_FragCoord.z * rhw) * dx_DepthFront.x + "
779                          "dx_DepthFront.y;\n"
780                       << "    gl_FragCoord.w = rhw;\n";
781     }
782 
783     if (pixelBuiltins.glPointCoord.enabled && shaderModel >= 3)
784     {
785         pixelPrologue << "    gl_PointCoord.x = input.gl_PointCoord.x;\n"
786                       << "    gl_PointCoord.y = 1.0 - input.gl_PointCoord.y;\n";
787     }
788 
789     if (fragmentShader && fragmentShader->usesFrontFacing())
790     {
791         if (shaderModel <= 3)
792         {
793             pixelPrologue << "    gl_FrontFacing = (vFace * dx_DepthFront.z >= 0.0);\n";
794         }
795         else
796         {
797             pixelPrologue << "    gl_FrontFacing = isFrontFace;\n";
798         }
799     }
800 
801     for (GLuint registerIndex = 0u; registerIndex < registerInfos.size(); ++registerIndex)
802     {
803         const PackedVaryingRegister &registerInfo = registerInfos[registerIndex];
804         const auto &packedVarying                 = *registerInfo.packedVarying;
805 
806         // Don't reference VS-only transform feedback varyings in the PS.
807         if (packedVarying.vertexOnly())
808         {
809             continue;
810         }
811 
812         const auto &varying = *packedVarying.backVarying.varying;
813         ASSERT(!varying.isBuiltIn() && !varying.isStruct());
814 
815         // Note that we're relying on that the active flag is set according to usage in the fragment
816         // shader.
817         if (!varying.active)
818         {
819             continue;
820         }
821 
822         pixelPrologue << "    ";
823 
824         if (packedVarying.isStructField())
825         {
826             pixelPrologue << DecorateVariable(packedVarying.backVarying.parentStructName) << ".";
827         }
828 
829         pixelPrologue << DecorateVariable(varying.name);
830 
831         if (varying.isArray())
832         {
833             WriteArrayString(pixelPrologue, registerInfo.varyingArrayIndex);
834         }
835 
836         GLenum transposedType = TransposeMatrixType(varying.type);
837         if (VariableRowCount(transposedType) > 1)
838         {
839             WriteArrayString(pixelPrologue, registerInfo.varyingRowIndex);
840         }
841 
842         pixelPrologue << " = input.v" << registerIndex;
843 
844         switch (VariableColumnCount(transposedType))
845         {
846             case 1:
847                 pixelPrologue << ".x";
848                 break;
849             case 2:
850                 pixelPrologue << ".xy";
851                 break;
852             case 3:
853                 pixelPrologue << ".xyz";
854                 break;
855             case 4:
856                 break;
857             default:
858                 UNREACHABLE();
859         }
860         pixelPrologue << ";\n";
861     }
862 
863     if (fragmentShaderGL)
864     {
865         std::string pixelSource = fragmentShaderGL->getTranslatedSource();
866 
867         if (fragmentShader->usesFrontFacing())
868         {
869             if (shaderModel >= 4)
870             {
871                 angle::ReplaceSubstring(&pixelSource,
872                                         std::string(PIXEL_MAIN_PARAMETERS_STUB_STRING),
873                                         "PS_INPUT input, bool isFrontFace : SV_IsFrontFace");
874             }
875             else
876             {
877                 angle::ReplaceSubstring(&pixelSource,
878                                         std::string(PIXEL_MAIN_PARAMETERS_STUB_STRING),
879                                         "PS_INPUT input, float vFace : VFACE");
880             }
881         }
882         else
883         {
884             angle::ReplaceSubstring(&pixelSource, std::string(PIXEL_MAIN_PARAMETERS_STUB_STRING),
885                                     "PS_INPUT input");
886         }
887 
888         angle::ReplaceSubstring(&pixelSource, std::string(MAIN_PROLOGUE_STUB_STRING),
889                                 pixelPrologue.str());
890         pixelStream << pixelSource;
891     }
892 
893     (*shaderHLSL)[gl::ShaderType::Vertex]   = vertexStream.str();
894     (*shaderHLSL)[gl::ShaderType::Fragment] = pixelStream.str();
895 }
896 
generateGeometryShaderPreamble(const VaryingPacking & varyingPacking,const BuiltinVaryingsD3D & builtinsD3D,const bool hasANGLEMultiviewEnabled,const bool selectViewInVS) const897 std::string DynamicHLSL::generateGeometryShaderPreamble(const VaryingPacking &varyingPacking,
898                                                         const BuiltinVaryingsD3D &builtinsD3D,
899                                                         const bool hasANGLEMultiviewEnabled,
900                                                         const bool selectViewInVS) const
901 {
902     ASSERT(mRenderer->getMajorShaderModel() >= 4);
903 
904     std::ostringstream preambleStream;
905 
906     const auto &vertexBuiltins = builtinsD3D[gl::ShaderType::Vertex];
907 
908     preambleStream << "struct GS_INPUT\n";
909     generateVaryingLinkHLSL(varyingPacking, vertexBuiltins, builtinsD3D.usesPointSize(),
910                             preambleStream);
911     preambleStream << "\n"
912                    << "struct GS_OUTPUT\n";
913     generateVaryingLinkHLSL(varyingPacking, builtinsD3D[gl::ShaderType::Geometry],
914                             builtinsD3D.usesPointSize(), preambleStream);
915     preambleStream
916         << "\n"
917         << "void copyVertex(inout GS_OUTPUT output, GS_INPUT input, GS_INPUT flatinput)\n"
918         << "{\n"
919         << "    output.gl_Position = input.gl_Position;\n";
920 
921     if (vertexBuiltins.glPointSize.enabled)
922     {
923         preambleStream << "    output.gl_PointSize = input.gl_PointSize;\n";
924     }
925 
926     if (hasANGLEMultiviewEnabled)
927     {
928         preambleStream << "    output.gl_ViewID_OVR = input.gl_ViewID_OVR;\n";
929         if (selectViewInVS)
930         {
931             ASSERT(builtinsD3D[gl::ShaderType::Geometry].glViewportIndex.enabled &&
932                    builtinsD3D[gl::ShaderType::Geometry].glLayer.enabled);
933 
934             // If the view is already selected in the VS, then we just pass the gl_ViewportIndex and
935             // gl_Layer to the output.
936             preambleStream << "    output.gl_ViewportIndex = input.gl_ViewportIndex;\n"
937                            << "    output.gl_Layer = input.gl_Layer;\n";
938         }
939     }
940 
941     const auto &registerInfos = varyingPacking.getRegisterList();
942     for (GLuint registerIndex = 0u; registerIndex < registerInfos.size(); ++registerIndex)
943     {
944         const PackedVaryingRegister &varyingRegister = registerInfos[registerIndex];
945         preambleStream << "    output.v" << registerIndex << " = ";
946         if (varyingRegister.packedVarying->interpolation == sh::INTERPOLATION_FLAT)
947         {
948             preambleStream << "flat";
949         }
950         preambleStream << "input.v" << registerIndex << "; \n";
951     }
952 
953     if (vertexBuiltins.glFragCoord.enabled)
954     {
955         preambleStream << "    output.gl_FragCoord = input.gl_FragCoord;\n";
956     }
957 
958     // Only write the dx_Position if we aren't using point sprites
959     preambleStream << "#ifndef ANGLE_POINT_SPRITE_SHADER\n"
960                    << "    output.dx_Position = input.dx_Position;\n"
961                    << "#endif  // ANGLE_POINT_SPRITE_SHADER\n"
962                    << "}\n";
963 
964     if (hasANGLEMultiviewEnabled && !selectViewInVS)
965     {
966         ASSERT(builtinsD3D[gl::ShaderType::Geometry].glViewportIndex.enabled &&
967                builtinsD3D[gl::ShaderType::Geometry].glLayer.enabled);
968 
969         // According to the HLSL reference, using SV_RenderTargetArrayIndex is only valid if the
970         // render target is an array resource. Because of this we do not write to gl_Layer if we are
971         // taking the side-by-side code path. We still select the viewport index in the layered code
972         // path as that is always valid. See:
973         // https://msdn.microsoft.com/en-us/library/windows/desktop/bb509647(v=vs.85).aspx
974         preambleStream << "\n"
975                        << "void selectView(inout GS_OUTPUT output, GS_INPUT input)\n"
976                        << "{\n"
977                        << "    if (multiviewSelectViewportIndex)\n"
978                        << "    {\n"
979                        << "        output.gl_ViewportIndex = input.gl_ViewID_OVR;\n"
980                        << "    } else {\n"
981                        << "        output.gl_ViewportIndex = 0;\n"
982                        << "        output.gl_Layer = input.gl_ViewID_OVR;\n"
983                        << "    }\n"
984                        << "}\n";
985     }
986 
987     return preambleStream.str();
988 }
989 
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) const990 std::string DynamicHLSL::generateGeometryShaderHLSL(const gl::Caps &caps,
991                                                     gl::PrimitiveMode primitiveType,
992                                                     const gl::ProgramState &programData,
993                                                     const bool useViewScale,
994                                                     const bool hasANGLEMultiviewEnabled,
995                                                     const bool selectViewInVS,
996                                                     const bool pointSpriteEmulation,
997                                                     const std::string &preambleString) const
998 {
999     ASSERT(mRenderer->getMajorShaderModel() >= 4);
1000 
1001     std::stringstream shaderStream;
1002 
1003     const bool pointSprites = (primitiveType == gl::PrimitiveMode::Points) && pointSpriteEmulation;
1004     const bool usesPointCoord = preambleString.find("gl_PointCoord") != std::string::npos;
1005 
1006     const char *inputPT  = nullptr;
1007     const char *outputPT = nullptr;
1008     int inputSize        = 0;
1009     int maxVertexOutput  = 0;
1010 
1011     switch (primitiveType)
1012     {
1013         case gl::PrimitiveMode::Points:
1014             inputPT   = "point";
1015             inputSize = 1;
1016 
1017             if (pointSprites)
1018             {
1019                 outputPT        = "Triangle";
1020                 maxVertexOutput = 4;
1021             }
1022             else
1023             {
1024                 outputPT        = "Point";
1025                 maxVertexOutput = 1;
1026             }
1027 
1028             break;
1029 
1030         case gl::PrimitiveMode::Lines:
1031         case gl::PrimitiveMode::LineStrip:
1032         case gl::PrimitiveMode::LineLoop:
1033             inputPT         = "line";
1034             outputPT        = "Line";
1035             inputSize       = 2;
1036             maxVertexOutput = 2;
1037             break;
1038 
1039         case gl::PrimitiveMode::Triangles:
1040         case gl::PrimitiveMode::TriangleStrip:
1041         case gl::PrimitiveMode::TriangleFan:
1042             inputPT         = "triangle";
1043             outputPT        = "Triangle";
1044             inputSize       = 3;
1045             maxVertexOutput = 3;
1046             break;
1047 
1048         default:
1049             UNREACHABLE();
1050             break;
1051     }
1052 
1053     if (pointSprites || hasANGLEMultiviewEnabled)
1054     {
1055         shaderStream << "cbuffer DriverConstants : register(b0)\n"
1056                         "{\n";
1057 
1058         if (pointSprites)
1059         {
1060             shaderStream << "    float4 dx_ViewCoords : packoffset(c1);\n";
1061             if (useViewScale)
1062             {
1063                 shaderStream << "    float2 dx_ViewScale : packoffset(c3);\n";
1064             }
1065         }
1066 
1067         if (hasANGLEMultiviewEnabled)
1068         {
1069             // We have to add a value which we can use to keep track of which multi-view code path
1070             // is to be selected in the GS.
1071             shaderStream << "    float multiviewSelectViewportIndex : packoffset(c3.z);\n";
1072         }
1073 
1074         shaderStream << "};\n\n";
1075     }
1076 
1077     if (pointSprites)
1078     {
1079         shaderStream << "#define ANGLE_POINT_SPRITE_SHADER\n"
1080                         "\n"
1081                         "static float2 pointSpriteCorners[] = \n"
1082                         "{\n"
1083                         "    float2( 0.5f, -0.5f),\n"
1084                         "    float2( 0.5f,  0.5f),\n"
1085                         "    float2(-0.5f, -0.5f),\n"
1086                         "    float2(-0.5f,  0.5f)\n"
1087                         "};\n"
1088                         "\n"
1089                         "static float2 pointSpriteTexcoords[] = \n"
1090                         "{\n"
1091                         "    float2(1.0f, 1.0f),\n"
1092                         "    float2(1.0f, 0.0f),\n"
1093                         "    float2(0.0f, 1.0f),\n"
1094                         "    float2(0.0f, 0.0f)\n"
1095                         "};\n"
1096                         "\n"
1097                         "static float minPointSize = "
1098                      << static_cast<int>(caps.minAliasedPointSize)
1099                      << ".0f;\n"
1100                         "static float maxPointSize = "
1101                      << static_cast<int>(caps.maxAliasedPointSize) << ".0f;\n"
1102                      << "\n";
1103     }
1104 
1105     shaderStream << preambleString << "\n"
1106                  << "[maxvertexcount(" << maxVertexOutput << ")]\n"
1107                  << "void main(" << inputPT << " GS_INPUT input[" << inputSize << "], ";
1108 
1109     if (primitiveType == gl::PrimitiveMode::TriangleStrip)
1110     {
1111         shaderStream << "uint primitiveID : SV_PrimitiveID, ";
1112     }
1113 
1114     shaderStream << " inout " << outputPT << "Stream<GS_OUTPUT> outStream)\n"
1115                  << "{\n"
1116                  << "    GS_OUTPUT output = (GS_OUTPUT)0;\n";
1117 
1118     if (primitiveType == gl::PrimitiveMode::TriangleStrip)
1119     {
1120         shaderStream << "    uint lastVertexIndex = (primitiveID % 2 == 0 ? 2 : 1);\n";
1121     }
1122     else
1123     {
1124         shaderStream << "    uint lastVertexIndex = " << (inputSize - 1) << ";\n";
1125     }
1126 
1127     for (int vertexIndex = 0; vertexIndex < inputSize; ++vertexIndex)
1128     {
1129         shaderStream << "    copyVertex(output, input[" << vertexIndex
1130                      << "], input[lastVertexIndex]);\n";
1131         if (hasANGLEMultiviewEnabled && !selectViewInVS)
1132         {
1133             shaderStream << "   selectView(output, input[" << vertexIndex << "]);\n";
1134         }
1135         if (!pointSprites)
1136         {
1137             ASSERT(inputSize == maxVertexOutput);
1138             shaderStream << "    outStream.Append(output);\n";
1139         }
1140     }
1141 
1142     if (pointSprites)
1143     {
1144         shaderStream << "\n"
1145                         "    float4 dx_Position = input[0].dx_Position;\n"
1146                         "    float gl_PointSize = clamp(input[0].gl_PointSize, minPointSize, "
1147                         "maxPointSize);\n"
1148                         "    float2 viewportScale = float2(1.0f / dx_ViewCoords.x, 1.0f / "
1149                         "dx_ViewCoords.y) * dx_Position.w;\n";
1150 
1151         for (int corner = 0; corner < 4; corner++)
1152         {
1153             if (useViewScale)
1154             {
1155                 shaderStream << "    \n"
1156                                 "    output.dx_Position = dx_Position + float4(1.0f, "
1157                                 "-dx_ViewScale.y, 1.0f, 1.0f)"
1158                                 "        * float4(pointSpriteCorners["
1159                              << corner << "] * viewportScale * gl_PointSize, 0.0f, 0.0f);\n";
1160             }
1161             else
1162             {
1163                 shaderStream << "\n"
1164                                 "    output.dx_Position = dx_Position + float4(pointSpriteCorners["
1165                              << corner << "] * viewportScale * gl_PointSize, 0.0f, 0.0f);\n";
1166             }
1167 
1168             if (usesPointCoord)
1169             {
1170                 shaderStream << "    output.gl_PointCoord = pointSpriteTexcoords[" << corner
1171                              << "];\n";
1172             }
1173 
1174             shaderStream << "    outStream.Append(output);\n";
1175         }
1176     }
1177 
1178     shaderStream << "    \n"
1179                     "    outStream.RestartStrip();\n"
1180                     "}\n";
1181 
1182     return shaderStream.str();
1183 }
1184 
1185 // static
GenerateAttributeConversionHLSL(angle::FormatID vertexFormatID,const sh::ShaderVariable & shaderAttrib,std::ostringstream & outStream)1186 void DynamicHLSL::GenerateAttributeConversionHLSL(angle::FormatID vertexFormatID,
1187                                                   const sh::ShaderVariable &shaderAttrib,
1188                                                   std::ostringstream &outStream)
1189 {
1190     // Matrix
1191     if (IsMatrixType(shaderAttrib.type))
1192     {
1193         outStream << "transpose(input." << DecorateVariable(shaderAttrib.name) << ")";
1194         return;
1195     }
1196 
1197     GLenum shaderComponentType           = VariableComponentType(shaderAttrib.type);
1198     int shaderComponentCount             = VariableComponentCount(shaderAttrib.type);
1199     const gl::VertexFormat &vertexFormat = gl::GetVertexFormatFromID(vertexFormatID);
1200 
1201     // Perform integer to float conversion (if necessary)
1202     if (shaderComponentType == GL_FLOAT && vertexFormat.type != GL_FLOAT)
1203     {
1204         // TODO: normalization for 32-bit integer formats
1205         ASSERT(!vertexFormat.normalized && !vertexFormat.pureInteger);
1206         outStream << "float" << shaderComponentCount << "(input."
1207                   << DecorateVariable(shaderAttrib.name) << ")";
1208         return;
1209     }
1210 
1211     // No conversion necessary
1212     outStream << "input." << DecorateVariable(shaderAttrib.name);
1213 }
1214 
getPixelShaderOutputKey(const gl::State & data,const gl::ProgramState & programData,const ProgramD3DMetadata & metadata,std::vector<PixelShaderOutputVariable> * outPixelShaderKey)1215 void DynamicHLSL::getPixelShaderOutputKey(const gl::State &data,
1216                                           const gl::ProgramState &programData,
1217                                           const ProgramD3DMetadata &metadata,
1218                                           std::vector<PixelShaderOutputVariable> *outPixelShaderKey)
1219 {
1220     // Two cases when writing to gl_FragColor and using ESSL 1.0:
1221     // - with a 3.0 context, the output color is copied to channel 0
1222     // - with a 2.0 context, the output color is broadcast to all channels
1223     bool broadcast = metadata.usesBroadcast(data);
1224     const unsigned int numRenderTargets =
1225         (broadcast || metadata.usesMultipleFragmentOuts()
1226              ? static_cast<unsigned int>(data.getCaps().maxDrawBuffers)
1227              : 1);
1228 
1229     if (!metadata.usesCustomOutVars())
1230     {
1231         for (unsigned int renderTargetIndex = 0; renderTargetIndex < numRenderTargets;
1232              renderTargetIndex++)
1233         {
1234             PixelShaderOutputVariable outputKeyVariable;
1235             outputKeyVariable.type = GL_FLOAT_VEC4;
1236             outputKeyVariable.name = "gl_Color" + Str(renderTargetIndex);
1237             outputKeyVariable.source =
1238                 broadcast ? "gl_Color[0]" : "gl_Color[" + Str(renderTargetIndex) + "]";
1239             outputKeyVariable.outputLocation = renderTargetIndex;
1240 
1241             outPixelShaderKey->push_back(outputKeyVariable);
1242         }
1243 
1244         if (metadata.usesSecondaryColor())
1245         {
1246             for (unsigned int secondaryIndex = 0;
1247                  secondaryIndex < data.getExtensions().maxDualSourceDrawBuffers; secondaryIndex++)
1248             {
1249                 PixelShaderOutputVariable outputKeyVariable;
1250                 outputKeyVariable.type           = GL_FLOAT_VEC4;
1251                 outputKeyVariable.name           = "gl_SecondaryColor" + Str(secondaryIndex);
1252                 outputKeyVariable.source         = "gl_SecondaryColor[" + Str(secondaryIndex) + "]";
1253                 outputKeyVariable.outputLocation = secondaryIndex;
1254                 outputKeyVariable.outputIndex    = 1;
1255 
1256                 outPixelShaderKey->push_back(outputKeyVariable);
1257             }
1258         }
1259     }
1260     else
1261     {
1262         const ShaderD3D *fragmentShader = metadata.getFragmentShader();
1263 
1264         if (!fragmentShader)
1265         {
1266             return;
1267         }
1268 
1269         const auto &shaderOutputVars = fragmentShader->getState().getActiveOutputVariables();
1270 
1271         for (size_t outputLocationIndex = 0u;
1272              outputLocationIndex < programData.getOutputLocations().size(); ++outputLocationIndex)
1273         {
1274             const VariableLocation &outputLocation =
1275                 programData.getOutputLocations().at(outputLocationIndex);
1276             if (!outputLocation.used())
1277             {
1278                 continue;
1279             }
1280             const sh::ShaderVariable &outputVariable = shaderOutputVars[outputLocation.index];
1281             const std::string &variableName          = "out_" + outputVariable.name;
1282 
1283             // Fragment outputs can't be arrays of arrays. ESSL 3.10 section 4.3.6.
1284             const std::string &elementString =
1285                 (outputVariable.isArray() ? Str(outputLocation.arrayIndex) : "");
1286 
1287             ASSERT(outputVariable.active);
1288 
1289             PixelShaderOutputVariable outputKeyVariable;
1290             outputKeyVariable.type = outputVariable.type;
1291             outputKeyVariable.name = variableName + elementString;
1292             outputKeyVariable.source =
1293                 variableName +
1294                 (outputVariable.isArray() ? ArrayString(outputLocation.arrayIndex) : "");
1295             outputKeyVariable.outputLocation = outputLocationIndex;
1296 
1297             outPixelShaderKey->push_back(outputKeyVariable);
1298         }
1299 
1300         // Now generate any secondary outputs...
1301         for (size_t outputLocationIndex = 0u;
1302              outputLocationIndex < programData.getSecondaryOutputLocations().size();
1303              ++outputLocationIndex)
1304         {
1305             const VariableLocation &outputLocation =
1306                 programData.getSecondaryOutputLocations().at(outputLocationIndex);
1307             if (!outputLocation.used())
1308             {
1309                 continue;
1310             }
1311             const sh::ShaderVariable &outputVariable = shaderOutputVars[outputLocation.index];
1312             const std::string &variableName          = "out_" + outputVariable.name;
1313 
1314             // Fragment outputs can't be arrays of arrays. ESSL 3.10 section 4.3.6.
1315             const std::string &elementString =
1316                 (outputVariable.isArray() ? Str(outputLocation.arrayIndex) : "");
1317 
1318             ASSERT(outputVariable.active);
1319 
1320             PixelShaderOutputVariable outputKeyVariable;
1321             outputKeyVariable.type = outputVariable.type;
1322             outputKeyVariable.name = variableName + elementString;
1323             outputKeyVariable.source =
1324                 variableName +
1325                 (outputVariable.isArray() ? ArrayString(outputLocation.arrayIndex) : "");
1326             outputKeyVariable.outputLocation = outputLocationIndex;
1327             outputKeyVariable.outputIndex    = 1;
1328 
1329             outPixelShaderKey->push_back(outputKeyVariable);
1330         }
1331     }
1332 }
1333 
1334 // BuiltinVarying Implementation.
BuiltinVarying()1335 BuiltinVarying::BuiltinVarying() : enabled(false), index(0), systemValue(false) {}
1336 
str() const1337 std::string BuiltinVarying::str() const
1338 {
1339     return (systemValue ? semantic : (semantic + Str(index)));
1340 }
1341 
enableSystem(const std::string & systemValueSemantic)1342 void BuiltinVarying::enableSystem(const std::string &systemValueSemantic)
1343 {
1344     enabled     = true;
1345     semantic    = systemValueSemantic;
1346     systemValue = true;
1347 }
1348 
enable(const std::string & semanticVal,unsigned int indexVal)1349 void BuiltinVarying::enable(const std::string &semanticVal, unsigned int indexVal)
1350 {
1351     enabled  = true;
1352     semantic = semanticVal;
1353     index    = indexVal;
1354 }
1355 
1356 // BuiltinVaryingsD3D Implementation.
BuiltinVaryingsD3D(const ProgramD3DMetadata & metadata,const VaryingPacking & packing)1357 BuiltinVaryingsD3D::BuiltinVaryingsD3D(const ProgramD3DMetadata &metadata,
1358                                        const VaryingPacking &packing)
1359 {
1360     updateBuiltins(gl::ShaderType::Vertex, metadata, packing);
1361     updateBuiltins(gl::ShaderType::Fragment, metadata, packing);
1362     int shaderModel = metadata.getRendererMajorShaderModel();
1363     if (shaderModel >= 4)
1364     {
1365         updateBuiltins(gl::ShaderType::Geometry, metadata, packing);
1366     }
1367     // In shader model >= 4, some builtins need to be the same in vertex and pixel shaders - input
1368     // struct needs to be a prefix of output struct.
1369     ASSERT(shaderModel < 4 || mBuiltinInfo[gl::ShaderType::Vertex].glPosition.enabled ==
1370                                   mBuiltinInfo[gl::ShaderType::Fragment].glPosition.enabled);
1371     ASSERT(shaderModel < 4 || mBuiltinInfo[gl::ShaderType::Vertex].glFragCoord.enabled ==
1372                                   mBuiltinInfo[gl::ShaderType::Fragment].glFragCoord.enabled);
1373     ASSERT(shaderModel < 4 || mBuiltinInfo[gl::ShaderType::Vertex].glPointCoord.enabled ==
1374                                   mBuiltinInfo[gl::ShaderType::Fragment].glPointCoord.enabled);
1375     ASSERT(shaderModel < 4 || mBuiltinInfo[gl::ShaderType::Vertex].glPointSize.enabled ==
1376                                   mBuiltinInfo[gl::ShaderType::Fragment].glPointSize.enabled);
1377     ASSERT(shaderModel < 4 || mBuiltinInfo[gl::ShaderType::Vertex].glViewIDOVR.enabled ==
1378                                   mBuiltinInfo[gl::ShaderType::Fragment].glViewIDOVR.enabled);
1379 }
1380 
1381 BuiltinVaryingsD3D::~BuiltinVaryingsD3D() = default;
1382 
updateBuiltins(gl::ShaderType shaderType,const ProgramD3DMetadata & metadata,const VaryingPacking & packing)1383 void BuiltinVaryingsD3D::updateBuiltins(gl::ShaderType shaderType,
1384                                         const ProgramD3DMetadata &metadata,
1385                                         const VaryingPacking &packing)
1386 {
1387     const std::string &userSemantic = GetVaryingSemantic(metadata.getRendererMajorShaderModel(),
1388                                                          metadata.usesSystemValuePointSize());
1389 
1390     // Note that when enabling builtins only for specific shader stages in shader model >= 4, the
1391     // code needs to ensure that the input struct of the shader stage is a prefix of the output
1392     // struct of the previous stage.
1393 
1394     unsigned int reservedSemanticIndex = packing.getMaxSemanticIndex();
1395 
1396     BuiltinInfo *builtins = &mBuiltinInfo[shaderType];
1397 
1398     if (metadata.getRendererMajorShaderModel() >= 4)
1399     {
1400         builtins->dxPosition.enableSystem("SV_Position");
1401     }
1402     else if (shaderType == gl::ShaderType::Fragment)
1403     {
1404         builtins->dxPosition.enableSystem("VPOS");
1405     }
1406     else
1407     {
1408         builtins->dxPosition.enableSystem("POSITION");
1409     }
1410 
1411     if (metadata.usesTransformFeedbackGLPosition())
1412     {
1413         builtins->glPosition.enable(userSemantic, reservedSemanticIndex++);
1414     }
1415 
1416     if (metadata.usesFragCoord())
1417     {
1418         builtins->glFragCoord.enable(userSemantic, reservedSemanticIndex++);
1419     }
1420 
1421     if (shaderType == gl::ShaderType::Vertex ? metadata.addsPointCoordToVertexShader()
1422                                              : metadata.usesPointCoord())
1423     {
1424         // SM3 reserves the TEXCOORD semantic for point sprite texcoords (gl_PointCoord)
1425         // In D3D11 we manually compute gl_PointCoord in the GS.
1426         if (metadata.getRendererMajorShaderModel() >= 4)
1427         {
1428             builtins->glPointCoord.enable(userSemantic, reservedSemanticIndex++);
1429         }
1430         else
1431         {
1432             builtins->glPointCoord.enable("TEXCOORD", 0);
1433         }
1434     }
1435 
1436     if (metadata.hasANGLEMultiviewEnabled())
1437     {
1438         // Although it is possible to compute gl_ViewID_OVR from the value of
1439         // SV_ViewportArrayIndex or SV_RenderTargetArrayIndex and the multi-view state in the
1440         // driver constant buffer, it is easier and cleaner to always pass it as a varying.
1441         builtins->glViewIDOVR.enable(userSemantic, reservedSemanticIndex++);
1442 
1443         if (shaderType == gl::ShaderType::Vertex)
1444         {
1445             if (metadata.canSelectViewInVertexShader())
1446             {
1447                 builtins->glViewportIndex.enableSystem("SV_ViewportArrayIndex");
1448                 builtins->glLayer.enableSystem("SV_RenderTargetArrayIndex");
1449             }
1450         }
1451 
1452         if (shaderType == gl::ShaderType::Geometry)
1453         {
1454             // gl_Layer and gl_ViewportIndex are necessary so that we can write to either based on
1455             // the multiview state in the driver constant buffer.
1456             builtins->glViewportIndex.enableSystem("SV_ViewportArrayIndex");
1457             builtins->glLayer.enableSystem("SV_RenderTargetArrayIndex");
1458         }
1459     }
1460 
1461     // Special case: do not include PSIZE semantic in HLSL 3 pixel shaders
1462     if (metadata.usesSystemValuePointSize() &&
1463         (shaderType != gl::ShaderType::Fragment || metadata.getRendererMajorShaderModel() >= 4))
1464     {
1465         builtins->glPointSize.enableSystem("PSIZE");
1466     }
1467 }
1468 
1469 }  // namespace rx
1470