/*------------------------------------------------------------------------- * drawElements Quality Program OpenGL ES 3.1 Module * ------------------------------------------------- * * Copyright 2014 The Android Open Source Project * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. * You may obtain a copy of the License at * * http://www.apache.org/licenses/LICENSE-2.0 * * Unless required by applicable law or agreed to in writing, software * distributed under the License is distributed on an "AS IS" BASIS, * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. * See the License for the specific language governing permissions and * limitations under the License. * *//*! * \file * \brief Explicit uniform location tests *//*--------------------------------------------------------------------*/ #include "es31fUniformLocationTests.hpp" #include "tcuTestLog.hpp" #include "tcuTextureUtil.hpp" #include "tcuVectorUtil.hpp" #include "tcuCommandLine.hpp" #include "glsShaderLibrary.hpp" #include "glsTextureTestUtil.hpp" #include "gluShaderProgram.hpp" #include "gluTexture.hpp" #include "gluPixelTransfer.hpp" #include "gluVarType.hpp" #include "gluVarTypeUtil.hpp" #include "glwFunctions.hpp" #include "glwEnums.hpp" #include "sglrContextUtil.hpp" #include "deStringUtil.hpp" #include "deUniquePtr.hpp" #include "deString.h" #include "deRandom.hpp" #include "deInt32.h" #include #include namespace deqp { namespace gles31 { namespace Functional { namespace { using std::string; using std::vector; using std::map; using de::UniquePtr; using glu::VarType; struct UniformInfo { enum ShaderStage { SHADERSTAGE_NONE = 0, SHADERSTAGE_VERTEX = (1<<0), SHADERSTAGE_FRAGMENT= (1<<1), SHADERSTAGE_BOTH = (SHADERSTAGE_VERTEX | SHADERSTAGE_FRAGMENT), }; VarType type; ShaderStage declareLocation; // support declarations with/without layout qualifiers, needed for linkage testing ShaderStage layoutLocation; ShaderStage checkLocation; int location; // -1 for unset UniformInfo (VarType type_, ShaderStage declareLocation_, ShaderStage layoutLocation_, ShaderStage checkLocation_, int location_ = -1) : type (type_) , declareLocation (declareLocation_) , layoutLocation (layoutLocation_) , checkLocation (checkLocation_) , location (location_) { } }; class UniformLocationCase : public tcu::TestCase { public: UniformLocationCase (tcu::TestContext& context, glu::RenderContext& renderContext, const char* name, const char* desc, const vector& uniformInfo); virtual ~UniformLocationCase (void) {} virtual IterateResult iterate (void); protected: IterateResult run (const vector& uniformList); static glu::ProgramSources genShaderSources (const vector& uniformList); bool verifyLocations (const glu::ShaderProgram& program, const vector& uniformList); void render (const glu::ShaderProgram& program, const vector& uniformList); static bool verifyResult (const tcu::ConstPixelBufferAccess& access); static float getExpectedValue (glu::DataType type, int id, const char* name); de::MovePtr createTexture (glu::DataType samplerType, float redChannelValue, int binding); glu::RenderContext& m_renderCtx; const vector m_uniformInfo; enum { RENDER_SIZE = 16 }; }; string getUniformName (int ndx, const glu::VarType& type, const glu::TypeComponentVector& path) { std::ostringstream buff; buff << "uni" << ndx << glu::TypeAccessFormat(type, path); return buff.str(); } string getFirstComponentName (const glu::VarType& type) { std::ostringstream buff; if (glu::isDataTypeVector(type.getBasicType())) buff << glu::TypeAccessFormat(type, glu::SubTypeAccess(type).component(0).getPath()); else if (glu::isDataTypeMatrix(type.getBasicType())) buff << glu::TypeAccessFormat(type, glu::SubTypeAccess(type).column(0).component(0).getPath()); return buff.str(); } UniformLocationCase::UniformLocationCase (tcu::TestContext& context, glu::RenderContext& renderContext, const char* name, const char* desc, const vector& uniformInfo) : TestCase (context, name, desc) , m_renderCtx (renderContext) , m_uniformInfo (uniformInfo) { } // [from, to] std::vector shuffledRange (int from, int to, int seed) { const int count = to - from; vector retval (count); de::Random rng (seed); DE_ASSERT(count > 0); for (int ndx = 0; ndx < count; ndx++) retval[ndx] = ndx + from; rng.shuffle(retval.begin(), retval.end()); return retval; } glu::DataType getDataTypeSamplerSampleType (glu::DataType type) { using namespace glu; if (type >= TYPE_SAMPLER_1D && type <= TYPE_SAMPLER_3D) return TYPE_FLOAT_VEC4; else if (type >= TYPE_INT_SAMPLER_1D && type <= TYPE_INT_SAMPLER_3D) return TYPE_INT_VEC4; else if (type >= TYPE_UINT_SAMPLER_1D && type <= TYPE_UINT_SAMPLER_3D) return TYPE_UINT_VEC4; else if (type >= TYPE_SAMPLER_1D_SHADOW && type <= TYPE_SAMPLER_2D_ARRAY_SHADOW) return TYPE_FLOAT; else DE_FATAL("Unknown sampler type"); return TYPE_INVALID; } // A (hopefully) unique value for a uniform. For multi-component types creates only one value. Values are in the range [0,1] for floats, [-128, 127] for ints, [0,255] for uints and 0/1 for booleans. Samplers are treated according to the types they return. float UniformLocationCase::getExpectedValue (glu::DataType type, int id, const char* name) { const deUint32 hash = deStringHash(name) + deInt32Hash(id); glu::DataType adjustedType = type; if (glu::isDataTypeSampler(type)) adjustedType = getDataTypeSamplerSampleType(type); if (glu::isDataTypeIntOrIVec(adjustedType)) return float(hash%128); else if (glu::isDataTypeUintOrUVec(adjustedType)) return float(hash%255); else if (glu::isDataTypeFloatOrVec(adjustedType)) return float(hash%255)/255.0f; else if (glu::isDataTypeBoolOrBVec(adjustedType)) return float(hash%2); else DE_FATAL("Unkown primitive type"); return glu::TYPE_INVALID; } UniformLocationCase::IterateResult UniformLocationCase::iterate (void) { return run(m_uniformInfo); } UniformLocationCase::IterateResult UniformLocationCase::run (const vector& uniformList) { using gls::TextureTestUtil::RandomViewport; const glu::ProgramSources sources = genShaderSources(uniformList); const glu::ShaderProgram program (m_renderCtx, sources); const int baseSeed = m_testCtx.getCommandLine().getBaseSeed(); const glw::Functions& gl = m_renderCtx.getFunctions(); const RandomViewport viewport (m_renderCtx.getRenderTarget(), RENDER_SIZE, RENDER_SIZE, deStringHash(getName()) + baseSeed); tcu::Surface rendered (RENDER_SIZE, RENDER_SIZE); if (!verifyLocations(program, uniformList)) return STOP; gl.clearColor(0.0f, 0.0f, 0.0f, 1.0f); gl.clear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT | GL_STENCIL_BUFFER_BIT); gl.viewport(viewport.x, viewport.y, viewport.width, viewport.height); render(program, uniformList); glu::readPixels(m_renderCtx, viewport.x, viewport.y, rendered.getAccess()); if (!verifyResult(rendered.getAccess())) { m_testCtx.setTestResult(QP_TEST_RESULT_FAIL, "Shader produced incorrect result"); return STOP; } m_testCtx.setTestResult(QP_TEST_RESULT_PASS, "Pass"); return STOP; } glu::ProgramSources UniformLocationCase::genShaderSources (const vector& uniformList) { std::ostringstream vertDecl, vertMain, fragDecl, fragMain; vertDecl << "#version 310 es\n" << "precision highp float;\n" << "precision highp int;\n" << "float verify(float val, float ref) { return float(abs(val-ref) < 0.05); }\n\n" << "in highp vec4 a_position;\n" << "out highp vec4 v_color;\n"; fragDecl << "#version 310 es\n\n" << "precision highp float;\n" << "precision highp int;\n" << "float verify(float val, float ref) { return float(abs(val-ref) < 0.05); }\n\n" << "in highp vec4 v_color;\n" << "layout(location = 0) out mediump vec4 o_color;\n\n"; vertMain << "void main()\n{\n" << " gl_Position = a_position;\n" << " v_color = vec4(1.0);\n"; fragMain << "void main()\n{\n" << " o_color = v_color;\n"; std::set declaredStructs; // Declare uniforms for (int uniformNdx = 0; uniformNdx < int(uniformList.size()); uniformNdx++) { const UniformInfo& uniformInfo = uniformList[uniformNdx]; const bool declareInVert = (uniformInfo.declareLocation & UniformInfo::SHADERSTAGE_VERTEX) != 0; const bool declareInFrag = (uniformInfo.declareLocation & UniformInfo::SHADERSTAGE_FRAGMENT) != 0; const bool layoutInVert = (uniformInfo.layoutLocation & UniformInfo::SHADERSTAGE_VERTEX) != 0; const bool layoutInFrag = (uniformInfo.layoutLocation & UniformInfo::SHADERSTAGE_FRAGMENT) != 0; const bool checkInVert = (uniformInfo.checkLocation & UniformInfo::SHADERSTAGE_VERTEX) != 0; const bool checkInFrag = (uniformInfo.checkLocation & UniformInfo::SHADERSTAGE_FRAGMENT) != 0; const string layout = uniformInfo.location >= 0 ? "layout(location = " + de::toString(uniformInfo.location) + ") " : ""; const string uniName = "uni" + de::toString(uniformNdx); int location = uniformInfo.location; int subTypeIndex = 0; DE_ASSERT((declareInVert && layoutInVert) || !layoutInVert); // Cannot have layout without declaration DE_ASSERT((declareInFrag && layoutInFrag) || !layoutInFrag); DE_ASSERT(location<0 || (layoutInVert || layoutInFrag)); // Cannot have location without layout // struct definitions if (uniformInfo.type.isStructType()) { const glu::StructType* const structType = uniformInfo.type.getStructPtr(); if (!declaredStructs.count(structType)) { if (declareInVert) vertDecl << glu::declare(structType, 0) << ";\n"; if (declareInFrag) fragDecl << glu::declare(structType, 0) << ";\n"; declaredStructs.insert(structType); } } if (declareInVert) vertDecl << "uniform " << (layoutInVert ? layout : "") << glu::declare(uniformInfo.type, uniName) << ";\n"; if (declareInFrag) fragDecl << "uniform " << (layoutInFrag ? layout : "") << glu::declare(uniformInfo.type, uniName) << ";\n"; // Anything that needs to be done for each enclosed primitive type for (glu::BasicTypeIterator subTypeIter = glu::BasicTypeIterator::begin(&uniformInfo.type); subTypeIter != glu::BasicTypeIterator::end(&uniformInfo.type); subTypeIter++, subTypeIndex++) { const glu::VarType subType = glu::getVarType(uniformInfo.type, subTypeIter.getPath()); const glu::DataType scalarType = glu::getDataTypeScalarType(subType.getBasicType()); const char* const typeName = glu::getDataTypeName(scalarType); const string expectValue = de::floatToString(getExpectedValue(scalarType, location >= 0 ? location+subTypeIndex : -1, typeName), 3); if (glu::isDataTypeSampler(scalarType)) { if (checkInVert) vertMain << " v_color.rgb *= verify(float( texture(" << uniName << glu::TypeAccessFormat(uniformInfo.type, subTypeIter.getPath()) << ", vec2(0.5)).r), " << expectValue << ");\n"; if (checkInFrag) fragMain << " o_color.rgb *= verify(float( texture(" << uniName << glu::TypeAccessFormat(uniformInfo.type, subTypeIter.getPath()) << ", vec2(0.5)).r), " << expectValue << ");\n"; } else { if (checkInVert) vertMain << " v_color.rgb *= verify(float(" << uniName << glu::TypeAccessFormat(uniformInfo.type, subTypeIter.getPath()) << getFirstComponentName(subType) << "), " << expectValue << ");\n"; if (checkInFrag) fragMain << " o_color.rgb *= verify(float(" << uniName << glu::TypeAccessFormat(uniformInfo.type, subTypeIter.getPath()) << getFirstComponentName(subType) << "), " << expectValue << ");\n"; } } } vertMain << "}\n"; fragMain << "}\n"; return glu::makeVtxFragSources(vertDecl.str() + vertMain.str(), fragDecl.str() + fragMain.str()); } bool UniformLocationCase::verifyLocations (const glu::ShaderProgram& program, const vector& uniformList) { using tcu::TestLog; const glw::Functions& gl = m_renderCtx.getFunctions(); const bool vertexOk = program.getShaderInfo(glu::SHADERTYPE_VERTEX).compileOk; const bool fragmentOk = program.getShaderInfo(glu::SHADERTYPE_FRAGMENT).compileOk; const bool linkOk = program.getProgramInfo().linkOk; const deUint32 programID = program.getProgram(); TestLog& log = m_testCtx.getLog(); std::set usedLocations; log << program; if (!vertexOk || !fragmentOk || !linkOk) { log << TestLog::Message << "ERROR: shader failed to compile/link" << TestLog::EndMessage; m_testCtx.setTestResult(QP_TEST_RESULT_FAIL, "Shader failed to compile/link"); return false; } for (int uniformNdx = 0; uniformNdx < int(uniformList.size()); uniformNdx++) { const UniformInfo& uniformInfo = uniformList[uniformNdx]; int subTypeIndex = 0; for (glu::BasicTypeIterator subTypeIter = glu::BasicTypeIterator::begin(&uniformInfo.type); subTypeIter != glu::BasicTypeIterator::end(&uniformInfo.type); subTypeIter++, subTypeIndex++) { const string name = getUniformName(uniformNdx, uniformInfo.type, subTypeIter.getPath()); const int gotLoc = gl.getUniformLocation(programID, name.c_str()); const int expectLoc = uniformInfo.location >= 0 ? uniformInfo.location+subTypeIndex : -1; if (expectLoc >= 0) { if (uniformInfo.checkLocation == 0 && gotLoc == -1) continue; if (gotLoc != expectLoc) { log << TestLog::Message << "ERROR: found uniform " << name << " in location " << gotLoc << " when it should have been in " << expectLoc << TestLog::EndMessage; m_testCtx.setTestResult(QP_TEST_RESULT_FAIL, "Incorrect uniform location"); return false; } if (usedLocations.find(expectLoc) != usedLocations.end()) { log << TestLog::Message << "ERROR: expected uniform " << name << " in location " << gotLoc << " but it has already been used" << TestLog::EndMessage; m_testCtx.setTestResult(QP_TEST_RESULT_FAIL, "Overlapping uniform location"); return false; } usedLocations.insert(expectLoc); } else if (gotLoc >= 0) { if (usedLocations.count(gotLoc)) { log << TestLog::Message << "ERROR: found uniform " << name << " in location " << gotLoc << " which has already been used" << TestLog::EndMessage; m_testCtx.setTestResult(QP_TEST_RESULT_FAIL, "Overlapping uniform location"); return false; } usedLocations.insert(gotLoc); } } } return true; } // Check that shader output is white (or very close to it) bool UniformLocationCase::verifyResult (const tcu::ConstPixelBufferAccess& access) { using tcu::Vec4; const Vec4 threshold (0.1f, 0.1f, 0.1f, 0.1f); const Vec4 reference (1.0f, 1.0f, 1.0f, 1.0f); for (int y = 0; y < access.getHeight(); y++) { for (int x = 0; x < access.getWidth(); x++) { const Vec4 diff = abs(access.getPixel(x, y) - reference); if (!boolAll(lessThanEqual(diff, threshold))) return false; } } return true; } // get a 4 channel 8 bits each texture format that is usable by the given sampler type deUint32 getTextureFormat (glu::DataType samplerType) { using namespace glu; switch (samplerType) { case TYPE_SAMPLER_1D: case TYPE_SAMPLER_2D: case TYPE_SAMPLER_CUBE: case TYPE_SAMPLER_2D_ARRAY: case TYPE_SAMPLER_3D: return GL_RGBA8; case TYPE_INT_SAMPLER_1D: case TYPE_INT_SAMPLER_2D: case TYPE_INT_SAMPLER_CUBE: case TYPE_INT_SAMPLER_2D_ARRAY: case TYPE_INT_SAMPLER_3D: return GL_RGBA8I; case TYPE_UINT_SAMPLER_1D: case TYPE_UINT_SAMPLER_2D: case TYPE_UINT_SAMPLER_CUBE: case TYPE_UINT_SAMPLER_2D_ARRAY: case TYPE_UINT_SAMPLER_3D: return GL_RGBA8UI; default: DE_FATAL("Unsupported (sampler) type"); return 0; } } // create a texture suitable for sampling by the given sampler type and bind it de::MovePtr UniformLocationCase::createTexture (glu::DataType samplerType, float redChannelValue, int binding) { using namespace glu; const glw::Functions& gl = m_renderCtx.getFunctions(); const deUint32 format = getTextureFormat(samplerType); de::MovePtr tex; tex = de::MovePtr(new Texture2D(m_renderCtx, format, 16, 16)); tex->getRefTexture().allocLevel(0); if (format == GL_RGBA8I || format == GL_RGBA8UI) tcu::clear(tex->getRefTexture().getLevel(0), tcu::IVec4(int(redChannelValue), 0, 0, 0)); else tcu::clear(tex->getRefTexture().getLevel(0), tcu::Vec4(redChannelValue, 0.0f, 0.0f, 1.0f)); gl.activeTexture(GL_TEXTURE0 + binding); tex->upload(); gl.bindTexture(GL_TEXTURE_2D, tex->getGLTexture()); gl.texParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST); gl.texParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST); GLU_EXPECT_NO_ERROR(gl.getError(), "UniformLocationCase: texture upload"); return tex; } void UniformLocationCase::render (const glu::ShaderProgram& program, const vector& uniformList) { using glu::Texture2D; using de::MovePtr; typedef vector TextureList; const glw::Functions& gl = m_renderCtx.getFunctions(); const deUint32 programID = program.getProgram(); const deInt32 posLoc = gl.getAttribLocation(programID, "a_position"); // Vertex data. const float position[] = { -1.0f, -1.0f, 0.1f, 1.0f, -1.0f, 1.0f, 0.1f, 1.0f, 1.0f, -1.0f, 0.1f, 1.0f, 1.0f, 1.0f, 0.1f, 1.0f }; const deUint16 indices[] = { 0, 1, 2, 2, 1, 3 }; // some buffers to feed to the GPU, only the first element is relevant since the others are never verified float floatBuf[16] = {0.0f}; deInt32 intBuf[4] = {0}; deUint32 uintBuf[4] = {0}; TextureList texList; TCU_CHECK(posLoc >= 0); gl.useProgram(programID); try { // Set uniforms for (unsigned int uniformNdx = 0; uniformNdx < uniformList.size(); uniformNdx++) { const UniformInfo& uniformInfo = uniformList[uniformNdx]; int expectedLocation = uniformInfo.location; for (glu::BasicTypeIterator subTypeIter = glu::BasicTypeIterator::begin(&uniformInfo.type); subTypeIter != glu::BasicTypeIterator::end(&uniformInfo.type); subTypeIter++) { const glu::VarType type = glu::getVarType(uniformInfo.type, subTypeIter.getPath()); const string name = getUniformName(uniformNdx, uniformInfo.type, subTypeIter.getPath()); const int gotLoc = gl.getUniformLocation(programID, name.c_str()); const glu::DataType scalarType = glu::getDataTypeScalarType(type.getBasicType()); const char* const typeName = glu::getDataTypeName(scalarType); const float expectedValue = getExpectedValue(scalarType, expectedLocation, typeName); if (glu::isDataTypeSampler(scalarType)) { const int binding = (int)texList.size(); texList.push_back(createTexture(scalarType, expectedValue, binding).release()); gl.uniform1i(gotLoc, binding); } else if(gotLoc >= 0) { floatBuf[0] = expectedValue; intBuf[0] = int(expectedValue); uintBuf[0] = deUint32(expectedValue); m_testCtx.getLog() << tcu::TestLog::Message << "Set uniform " << name << " in location " << gotLoc << " to " << expectedValue << tcu::TestLog::EndMessage; switch (type.getBasicType()) { case glu::TYPE_FLOAT: gl.uniform1fv(gotLoc, 1, floatBuf); break; case glu::TYPE_FLOAT_VEC2: gl.uniform2fv(gotLoc, 1, floatBuf); break; case glu::TYPE_FLOAT_VEC3: gl.uniform3fv(gotLoc, 1, floatBuf); break; case glu::TYPE_FLOAT_VEC4: gl.uniform4fv(gotLoc, 1, floatBuf); break; case glu::TYPE_INT: gl.uniform1iv(gotLoc, 1, intBuf); break; case glu::TYPE_INT_VEC2: gl.uniform2iv(gotLoc, 1, intBuf); break; case glu::TYPE_INT_VEC3: gl.uniform3iv(gotLoc, 1, intBuf); break; case glu::TYPE_INT_VEC4: gl.uniform4iv(gotLoc, 1, intBuf); break; case glu::TYPE_UINT: gl.uniform1uiv(gotLoc, 1, uintBuf); break; case glu::TYPE_UINT_VEC2: gl.uniform2uiv(gotLoc, 1, uintBuf); break; case glu::TYPE_UINT_VEC3: gl.uniform3uiv(gotLoc, 1, uintBuf); break; case glu::TYPE_UINT_VEC4: gl.uniform4uiv(gotLoc, 1, uintBuf); break; case glu::TYPE_BOOL: gl.uniform1iv(gotLoc, 1, intBuf); break; case glu::TYPE_BOOL_VEC2: gl.uniform2iv(gotLoc, 1, intBuf); break; case glu::TYPE_BOOL_VEC3: gl.uniform3iv(gotLoc, 1, intBuf); break; case glu::TYPE_BOOL_VEC4: gl.uniform4iv(gotLoc, 1, intBuf); break; case glu::TYPE_FLOAT_MAT2: gl.uniformMatrix2fv(gotLoc, 1, false, floatBuf); break; case glu::TYPE_FLOAT_MAT2X3: gl.uniformMatrix2x3fv(gotLoc, 1, false, floatBuf); break; case glu::TYPE_FLOAT_MAT2X4: gl.uniformMatrix2x4fv(gotLoc, 1, false, floatBuf); break; case glu::TYPE_FLOAT_MAT3X2: gl.uniformMatrix3x2fv(gotLoc, 1, false, floatBuf); break; case glu::TYPE_FLOAT_MAT3: gl.uniformMatrix3fv(gotLoc, 1, false, floatBuf); break; case glu::TYPE_FLOAT_MAT3X4: gl.uniformMatrix3x4fv(gotLoc, 1, false, floatBuf); break; case glu::TYPE_FLOAT_MAT4X2: gl.uniformMatrix4x2fv(gotLoc, 1, false, floatBuf); break; case glu::TYPE_FLOAT_MAT4X3: gl.uniformMatrix4x3fv(gotLoc, 1, false, floatBuf); break; case glu::TYPE_FLOAT_MAT4: gl.uniformMatrix4fv(gotLoc, 1, false, floatBuf); break; default: DE_ASSERT(false); } } expectedLocation += expectedLocation>=0; } } gl.enableVertexAttribArray(posLoc); gl.vertexAttribPointer(posLoc, 4, GL_FLOAT, GL_FALSE, 0, &position[0]); gl.drawElements(GL_TRIANGLES, DE_LENGTH_OF_ARRAY(indices), GL_UNSIGNED_SHORT, &indices[0]); gl.disableVertexAttribArray(posLoc); } catch(...) { for (int i = 0; i < int(texList.size()); i++) delete texList[i]; throw; } for (int i = 0; i < int(texList.size()); i++) delete texList[i]; } class MaxUniformLocationCase : public UniformLocationCase { public: MaxUniformLocationCase (tcu::TestContext& context, glu::RenderContext& renderContext, const char* name, const char* desc, const vector& uniformInfo); virtual ~MaxUniformLocationCase (void) {} virtual IterateResult iterate (void); }; MaxUniformLocationCase::MaxUniformLocationCase (tcu::TestContext& context, glu::RenderContext& renderContext, const char* name, const char* desc, const vector& uniformInfo) : UniformLocationCase(context, renderContext, name, desc, uniformInfo) { DE_ASSERT(!uniformInfo.empty()); } UniformLocationCase::IterateResult MaxUniformLocationCase::iterate (void) { int maxLocation = 1024; vector uniformInfo = m_uniformInfo; m_renderCtx.getFunctions().getIntegerv(GL_MAX_UNIFORM_LOCATIONS, &maxLocation); uniformInfo[0].location = maxLocation-1; return UniformLocationCase::run(uniformInfo); } } // Anonymous UniformLocationTests::UniformLocationTests (Context& context) : TestCaseGroup(context, "uniform_location", "Explicit uniform locations") { } UniformLocationTests::~UniformLocationTests (void) { for (int i = 0; i < int(structTypes.size()); i++) delete structTypes[i]; } glu::VarType createVarType (glu::DataType type) { return glu::VarType(type, glu::isDataTypeBoolOrBVec(type) ? glu::PRECISION_LAST : glu::PRECISION_HIGHP); } void UniformLocationTests::init (void) { using namespace glu; const UniformInfo::ShaderStage checkStages[] = { UniformInfo::SHADERSTAGE_VERTEX, UniformInfo::SHADERSTAGE_FRAGMENT }; const char* stageNames[] = {"vertex", "fragment"}; const int maxLocations = 1024; const int baseSeed = m_context.getTestContext().getCommandLine().getBaseSeed(); const DataType primitiveTypes[] = { TYPE_FLOAT, TYPE_FLOAT_VEC2, TYPE_FLOAT_VEC3, TYPE_FLOAT_VEC4, TYPE_INT, TYPE_INT_VEC2, TYPE_INT_VEC3, TYPE_INT_VEC4, TYPE_UINT, TYPE_UINT_VEC2, TYPE_UINT_VEC3, TYPE_UINT_VEC4, TYPE_BOOL, TYPE_BOOL_VEC2, TYPE_BOOL_VEC3, TYPE_BOOL_VEC4, TYPE_FLOAT_MAT2, TYPE_FLOAT_MAT2X3, TYPE_FLOAT_MAT2X4, TYPE_FLOAT_MAT3X2, TYPE_FLOAT_MAT3, TYPE_FLOAT_MAT3X4, TYPE_FLOAT_MAT4X2, TYPE_FLOAT_MAT4X3, TYPE_FLOAT_MAT4, TYPE_SAMPLER_2D, TYPE_INT_SAMPLER_2D, TYPE_UINT_SAMPLER_2D, }; const int maxPrimitiveTypeNdx = DE_LENGTH_OF_ARRAY(primitiveTypes) - 4; DE_ASSERT(primitiveTypes[maxPrimitiveTypeNdx] == TYPE_FLOAT_MAT4); // Primitive type cases with trivial linkage { tcu::TestCaseGroup* const group = new tcu::TestCaseGroup(m_testCtx, "basic", "Location specified with use, single shader stage"); de::Random rng (baseSeed + 0x1001); addChild(group); for (int primitiveNdx = 0; primitiveNdx < DE_LENGTH_OF_ARRAY(primitiveTypes); primitiveNdx++) { const DataType type = primitiveTypes[primitiveNdx]; for (int stageNdx = 0; stageNdx < DE_LENGTH_OF_ARRAY(checkStages); stageNdx++) { const string name = string(getDataTypeName(type)) + "_" + stageNames[stageNdx]; vector config; UniformInfo uniform (createVarType(type), checkStages[stageNdx], checkStages[stageNdx], checkStages[stageNdx], rng.getInt(0, maxLocations-1)); config.push_back(uniform); group->addChild(new UniformLocationCase (m_testCtx, m_context.getRenderContext(), name.c_str(), name.c_str(), config)); } } } // Arrays { tcu::TestCaseGroup* const group = new tcu::TestCaseGroup(m_testCtx, "array", "Array location specified with use, single shader stage"); de::Random rng (baseSeed + 0x2001); addChild(group); for (int primitiveNdx = 0; primitiveNdx < DE_LENGTH_OF_ARRAY(primitiveTypes); primitiveNdx++) { const DataType type = primitiveTypes[primitiveNdx]; for (int stageNdx = 0; stageNdx < DE_LENGTH_OF_ARRAY(checkStages); stageNdx++) { const string name = string(getDataTypeName(type)) + "_" + stageNames[stageNdx]; vector config; UniformInfo uniform (VarType(createVarType(type), 8), checkStages[stageNdx], checkStages[stageNdx], checkStages[stageNdx], rng.getInt(0, maxLocations-1-8)); config.push_back(uniform); group->addChild(new UniformLocationCase (m_testCtx, m_context.getRenderContext(), name.c_str(), name.c_str(), config)); } } } // Nested Arrays { tcu::TestCaseGroup* const group = new tcu::TestCaseGroup(m_testCtx, "nested_array", "Array location specified with use, single shader stage"); de::Random rng (baseSeed + 0x3001); addChild(group); for (int primitiveNdx = 0; primitiveNdx < DE_LENGTH_OF_ARRAY(primitiveTypes); primitiveNdx++) { const DataType type = primitiveTypes[primitiveNdx]; for (int stageNdx = 0; stageNdx < DE_LENGTH_OF_ARRAY(checkStages); stageNdx++) { const string name = string(getDataTypeName(type)) + "_" + stageNames[stageNdx]; // stay comfortably within minimum max uniform component count (896 in fragment) and sampler count with all types const int arraySize = (getDataTypeScalarSize(type) > 4 || isDataTypeSampler(type)) ? 3 : 7; vector config; UniformInfo uniform (VarType(VarType(createVarType(type), arraySize), arraySize), checkStages[stageNdx], checkStages[stageNdx], checkStages[stageNdx], rng.getInt(0, maxLocations-1-arraySize*arraySize)); config.push_back(uniform); group->addChild(new UniformLocationCase (m_testCtx, m_context.getRenderContext(), name.c_str(), name.c_str(), config)); } } } // Structs { tcu::TestCaseGroup* const group = new tcu::TestCaseGroup(m_testCtx, "struct", "Struct location, random contents & declaration location"); de::Random rng (baseSeed + 0x4001); addChild(group); for (int caseNdx = 0; caseNdx < 16; caseNdx++) { typedef UniformInfo::ShaderStage Stage; const string name = "case_" + de::toString(caseNdx); const Stage layoutLoc = Stage(rng.getUint32()&0x3); const Stage declareLoc = Stage((rng.getUint32()&0x3) | layoutLoc); const Stage verifyLoc = Stage((rng.getUint32()&0x3) & declareLoc); const int location = layoutLoc ? rng.getInt(0, maxLocations-1-5) : -1; StructType* structProto = new StructType("S"); structTypes.push_back(structProto); structProto->addMember("a", createVarType(primitiveTypes[rng.getInt(0, maxPrimitiveTypeNdx)])); structProto->addMember("b", createVarType(primitiveTypes[rng.getInt(0, maxPrimitiveTypeNdx)])); structProto->addMember("c", createVarType(primitiveTypes[rng.getInt(0, maxPrimitiveTypeNdx)])); structProto->addMember("d", createVarType(primitiveTypes[rng.getInt(0, maxPrimitiveTypeNdx)])); structProto->addMember("e", createVarType(primitiveTypes[rng.getInt(0, maxPrimitiveTypeNdx)])); { vector config; config.push_back(UniformInfo(VarType(structProto), declareLoc, layoutLoc, verifyLoc, location)); group->addChild(new UniformLocationCase (m_testCtx, m_context.getRenderContext(), name.c_str(), name.c_str(), config)); } } } // Nested Structs { tcu::TestCaseGroup* const group = new tcu::TestCaseGroup(m_testCtx, "nested_struct", "Struct location specified with use, single shader stage"); de::Random rng (baseSeed + 0x5001); addChild(group); for (int caseNdx = 0; caseNdx < 16; caseNdx++) { typedef UniformInfo::ShaderStage Stage; const string name = "case_" + de::toString(caseNdx); const int baseLoc = rng.getInt(0, maxLocations-1-60); // Structs need to be added in the order of their declaration const Stage layoutLocs[]= { Stage(rng.getUint32()&0x3), Stage(rng.getUint32()&0x3), Stage(rng.getUint32()&0x3), Stage(rng.getUint32()&0x3), }; const deUint32 tempDecl[] = { (rng.getUint32()&0x3) | layoutLocs[0], (rng.getUint32()&0x3) | layoutLocs[1], (rng.getUint32()&0x3) | layoutLocs[2], (rng.getUint32()&0x3) | layoutLocs[3], }; // Component structs need to be declared if anything using them is declared const Stage declareLocs[] = { Stage(tempDecl[0] | tempDecl[1] | tempDecl[2] | tempDecl[3]), Stage(tempDecl[1] | tempDecl[2] | tempDecl[3]), Stage(tempDecl[2] | tempDecl[3]), Stage(tempDecl[3]), }; const Stage verifyLocs[] = { Stage(rng.getUint32()&0x3 & declareLocs[0]), Stage(rng.getUint32()&0x3 & declareLocs[1]), Stage(rng.getUint32()&0x3 & declareLocs[2]), Stage(rng.getUint32()&0x3 & declareLocs[3]), }; StructType* testTypes[] = { new StructType("Type0"), new StructType("Type1"), new StructType("Type2"), new StructType("Type3"), }; structTypes.push_back(testTypes[0]); structTypes.push_back(testTypes[1]); structTypes.push_back(testTypes[2]); structTypes.push_back(testTypes[3]); testTypes[0]->addMember("a", createVarType(primitiveTypes[rng.getInt(0, maxPrimitiveTypeNdx)])); testTypes[0]->addMember("b", createVarType(primitiveTypes[rng.getInt(0, maxPrimitiveTypeNdx)])); testTypes[0]->addMember("c", createVarType(primitiveTypes[rng.getInt(0, maxPrimitiveTypeNdx)])); testTypes[0]->addMember("d", createVarType(primitiveTypes[rng.getInt(0, maxPrimitiveTypeNdx)])); testTypes[0]->addMember("e", createVarType(primitiveTypes[rng.getInt(0, maxPrimitiveTypeNdx)])); testTypes[1]->addMember("a", createVarType(primitiveTypes[rng.getInt(0, maxPrimitiveTypeNdx)])); testTypes[1]->addMember("b", createVarType(primitiveTypes[rng.getInt(0, maxPrimitiveTypeNdx)])); testTypes[1]->addMember("c", createVarType(primitiveTypes[rng.getInt(0, maxPrimitiveTypeNdx)])); testTypes[1]->addMember("d", createVarType(primitiveTypes[rng.getInt(0, maxPrimitiveTypeNdx)])); testTypes[1]->addMember("e", createVarType(primitiveTypes[rng.getInt(0, maxPrimitiveTypeNdx)])); testTypes[2]->addMember("a", VarType(testTypes[0])); testTypes[2]->addMember("b", VarType(testTypes[1])); testTypes[2]->addMember("c", createVarType(primitiveTypes[rng.getInt(0, maxPrimitiveTypeNdx)])); testTypes[3]->addMember("a", VarType(testTypes[2])); { vector config; config.push_back(UniformInfo(VarType(testTypes[0]), declareLocs[0], layoutLocs[0], verifyLocs[0], layoutLocs[0] ? baseLoc : -1)); config.push_back(UniformInfo(VarType(testTypes[1]), declareLocs[1], layoutLocs[1], verifyLocs[1], layoutLocs[1] ? baseLoc+5 : -1)); config.push_back(UniformInfo(VarType(testTypes[2]), declareLocs[2], layoutLocs[2], verifyLocs[2], layoutLocs[2] ? baseLoc+16 : -1)); config.push_back(UniformInfo(VarType(testTypes[3]), declareLocs[3], layoutLocs[3], verifyLocs[3], layoutLocs[3] ? baseLoc+27 : -1)); group->addChild(new UniformLocationCase (m_testCtx, m_context.getRenderContext(), name.c_str(), name.c_str(), config)); } } } // Min/Max location { tcu::TestCaseGroup* const group = new tcu::TestCaseGroup(m_testCtx, "min_max", "Maximum & minimum location"); addChild(group); for (int primitiveNdx = 0; primitiveNdx < DE_LENGTH_OF_ARRAY(primitiveTypes); primitiveNdx++) { const DataType type = primitiveTypes[primitiveNdx]; for (int stageNdx = 0; stageNdx < DE_LENGTH_OF_ARRAY(checkStages); stageNdx++) { const string name = string(getDataTypeName(type)) + "_" + stageNames[stageNdx]; vector config; config.push_back(UniformInfo(createVarType(type), checkStages[stageNdx], checkStages[stageNdx], checkStages[stageNdx], 0)); group->addChild(new UniformLocationCase (m_testCtx, m_context.getRenderContext(), (name+"_min").c_str(), (name+"_min").c_str(), config)); group->addChild(new MaxUniformLocationCase (m_testCtx, m_context.getRenderContext(), (name+"_max").c_str(), (name+"_max").c_str(), config)); } } } // Link { tcu::TestCaseGroup* const group = new tcu::TestCaseGroup(m_testCtx, "link", "Location specified independently from use"); de::Random rng (baseSeed + 0x82e1); addChild(group); for (int caseNdx = 0; caseNdx < 10; caseNdx++) { const string name = "case_" + de::toString(caseNdx); vector config; vector locations = shuffledRange(0, maxLocations, 0x1234 + caseNdx*100); for (int count = 0; count < 32; count++) { typedef UniformInfo::ShaderStage Stage; const Stage layoutLoc = Stage(rng.getUint32()&0x3); const Stage declareLoc = Stage((rng.getUint32()&0x3) | layoutLoc); const Stage verifyLoc = Stage((rng.getUint32()&0x3) & declareLoc); const UniformInfo uniform (createVarType(primitiveTypes[rng.getInt(0, maxPrimitiveTypeNdx)]), declareLoc, layoutLoc, verifyLoc, (layoutLoc!=0) ? locations.back() : -1); config.push_back(uniform); locations.pop_back(); } group->addChild(new UniformLocationCase (m_testCtx, m_context.getRenderContext(), name.c_str(), name.c_str(), config)); } } // Negative { de::MovePtr negativeGroup (new tcu::TestCaseGroup(m_testCtx, "negative", "Negative tests")); { de::MovePtr es31Group (new tcu::TestCaseGroup(m_testCtx, "es31", "GLSL ES 3.1 Negative tests")); gls::ShaderLibrary shaderLibrary (m_testCtx, m_context.getRenderContext(), m_context.getContextInfo()); const vector negativeCases = shaderLibrary.loadShaderFile("shaders/es31/uniform_location.test"); for (int ndx = 0; ndx < int(negativeCases.size()); ndx++) es31Group->addChild(negativeCases[ndx]); negativeGroup->addChild(es31Group.release()); } { de::MovePtr es32Group (new tcu::TestCaseGroup(m_testCtx, "es32", "GLSL ES 3.2 Negative tests")); gls::ShaderLibrary shaderLibrary (m_testCtx, m_context.getRenderContext(), m_context.getContextInfo()); const vector negativeCases = shaderLibrary.loadShaderFile("shaders/es32/uniform_location.test"); for (int ndx = 0; ndx < int(negativeCases.size()); ndx++) es32Group->addChild(negativeCases[ndx]); negativeGroup->addChild(es32Group.release()); } addChild(negativeGroup.release()); } } } // Functional } // gles31 } // deqp