Reland "Add workaround for unused std140 and shared uniform blocks on MacOS"

CONTRIBUTORS

@@ -64,6 +64,10 @@ Intel Corporation
  Andy Chen
  Josh Triplett
  Sudarsana Nagineni
+ Jiajia Qin
+ Jiawei Shao
+ Jie Chen
+ Qiankun Miao
 
 Klarälvdalens Datakonsult AB
  Milian Wolff

include/GLSLANG/ShaderLang.h

@@ -49,7 +49,7 @@ typedef unsigned int GLenum;
 
 // Version number for shader translation API.
 // It is incremented every time the API changes.
-#define ANGLE_SH_VERSION 162
+#define ANGLE_SH_VERSION 163
 
 typedef enum {
     SH_GLES2_SPEC,
@@ -208,6 +208,14 @@ const ShCompileOptions SH_REWRITE_INTEGER_UNARY_MINUS_OPERATOR = UINT64_C(1) <<
 // It works by using an expression to emulate this function.
 const ShCompileOptions SH_EMULATE_ISNAN_FLOAT_FUNCTION = UINT64_C(1) << 27;
 
+// This flag will use all uniforms of unused std140 and shared uniform blocks at the
+// beginning of the vertex/fragment shader's main(). It is intended as a workaround for Mac
+// drivers with shader version 4.10. In those drivers, they will treat unused
+// std140 and shared uniform blocks' members as inactive. However, WebGL2.0 based on
+// OpenGL ES3.0.4 requires all members of a named uniform block declared with a shared or std140
+// layout qualifier to be considered active. The uniform block itself is also considered active.
+const ShCompileOptions SH_USE_UNUSED_STANDARD_SHARED_BLOCKS = UINT64_C(1) << 28;
+
 // Defines alternate strategies for implementing array index clamping.
 typedef enum {
   // Use the clamp intrinsic for array index clamping.

src/compiler.gypi

@@ -110,6 +110,8 @@
             'compiler/translator/Types.h',
             'compiler/translator/UnfoldShortCircuitAST.cpp',
             'compiler/translator/UnfoldShortCircuitAST.h',
+            'compiler/translator/UseInterfaceBlockFields.cpp',
+            'compiler/translator/UseInterfaceBlockFields.h',
             'compiler/translator/ValidateGlobalInitializer.cpp',
             'compiler/translator/ValidateGlobalInitializer.h',
             'compiler/translator/ValidateLimitations.cpp',

src/compiler/translator/Compiler.cpp

@@ -27,6 +27,7 @@
 #include "compiler/translator/RewriteDoWhile.h"
 #include "compiler/translator/ScalarizeVecAndMatConstructorArgs.h"
 #include "compiler/translator/UnfoldShortCircuitAST.h"
+#include "compiler/translator/UseInterfaceBlockFields.h"
 #include "compiler/translator/ValidateLimitations.h"
 #include "compiler/translator/ValidateMaxParameters.h"
 #include "compiler/translator/ValidateOutputs.h"
@@ -403,6 +404,10 @@ TIntermBlock *TCompiler::compileTreeImpl(const char *const shaderStrings[],
         if (success && shouldCollectVariables(compileOptions))
         {
             collectVariables(root);
+            if (compileOptions & SH_USE_UNUSED_STANDARD_SHARED_BLOCKS)
+            {
+                useAllMembersInUnusedStandardAndSharedBlocks(root);
+            }
             if (compileOptions & SH_ENFORCE_PACKING_RESTRICTIONS)
             {
                 success = enforcePackingRestrictions();
@@ -862,6 +867,22 @@ void TCompiler::initializeGLPosition(TIntermNode* root)
     InitializeVariables(root, list);
 }
 
+void TCompiler::useAllMembersInUnusedStandardAndSharedBlocks(TIntermNode *root)
+{
+    sh::InterfaceBlockList list;
+
+    for (auto block : interfaceBlocks)
+    {
+        if (!block.staticUse &&
+            (block.layout == sh::BLOCKLAYOUT_STANDARD || block.layout == sh::BLOCKLAYOUT_SHARED))
+        {
+            list.push_back(block);
+        }
+    }
+
+    sh::UseInterfaceBlockFields(root, list);
+}
+
 void TCompiler::initializeOutputVariables(TIntermNode *root)
 {
     InitVariableList list;

src/compiler/translator/Compiler.h

@@ -132,6 +132,10 @@ class TCompiler : public TShHandleBase
     // Returns true if, after applying the packing rules in the GLSL 1.017 spec
     // Appendix A, section 7, the shader does not use too many uniforms.
     bool enforcePackingRestrictions();
+    // Insert statements to reference all members in unused uniform blocks with standard and shared
+    // layout. This is to work around a Mac driver that treats unused standard/shared
+    // uniform blocks as inactive.
+    void useAllMembersInUnusedStandardAndSharedBlocks(TIntermNode *root);
     // Insert statements to initialize output variables in the beginning of main().
     // This is to avoid undefined behaviors.
     void initializeOutputVariables(TIntermNode *root);

src/compiler/translator/UseInterfaceBlockFields.cpp

+//
+// Copyright 2016 The ANGLE Project Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+//
+
+// UseInterfaceBlockFields.cpp: insert statements to reference all members in InterfaceBlock list at
+// the beginning of main. This is to work around a Mac driver that treats unused standard/shared
+// uniform blocks as inactive.
+
+#include "compiler/translator/UseInterfaceBlockFields.h"
+
+#include "compiler/translator/IntermNode.h"
+#include "compiler/translator/util.h"
+
+namespace sh
+{
+
+namespace
+{
+
+class UseUniformBlockMembers : public TIntermTraverser
+{
+  public:
+    UseUniformBlockMembers(const InterfaceBlockList &blocks)
+        : TIntermTraverser(true, false, false), mBlocks(blocks), mCodeInserted(false)
+    {
+    }
+
+  protected:
+    bool visitAggregate(Visit visit, TIntermAggregate *node) override;
+
+  private:
+    void insertUseCode(TIntermSequence *sequence);
+    void AddFieldUseStatements(const ShaderVariable &var, TIntermSequence *sequence);
+
+    const InterfaceBlockList &mBlocks;
+    bool mCodeInserted;
+};
+
+bool UseUniformBlockMembers::visitAggregate(Visit visit, TIntermAggregate *node)
+{
+    bool visitChildren = !mCodeInserted;
+    switch (node->getOp())
+    {
+        case EOpFunction:
+        {
+            ASSERT(visit == PreVisit);
+            if (node->getFunctionSymbolInfo()->isMain())
+            {
+                TIntermSequence *sequence = node->getSequence();
+                ASSERT(sequence->size() == 2);
+                TIntermBlock *body = (*sequence)[1]->getAsBlock();
+                ASSERT(body);
+                insertUseCode(body->getSequence());
+                mCodeInserted = true;
+                visitChildren = false;
+            }
+            break;
+        }
+        default:
+            visitChildren = false;
+            break;
+    }
+    return visitChildren;
+}
+
+void UseUniformBlockMembers::AddFieldUseStatements(const ShaderVariable &var,
+                                                   TIntermSequence *sequence)
+{
+    TString name = TString(var.name.c_str());
+    TType type   = GetShaderVariableType(var);
+
+    if (var.isArray())
+    {
+        size_t pos = name.find_last_of('[');
+        if (pos != TString::npos)
+        {
+            name = name.substr(0, pos);
+        }
+        TType elementType = type;
+        elementType.clearArrayness();
+
+        TIntermSymbol *arraySymbol = new TIntermSymbol(0, name, type);
+        for (unsigned int i = 0; i < var.arraySize; ++i)
+        {
+            TIntermBinary *element =
+                new TIntermBinary(EOpIndexDirect, arraySymbol, TIntermTyped::CreateIndexNode(i));
+
+            sequence->insert(sequence->begin(), element);
+        }
+    }
+    else if (var.isStruct())
+    {
+        TIntermSymbol *structSymbol = new TIntermSymbol(0, name, type);
+        for (unsigned int i = 0; i < var.fields.size(); ++i)
+        {
+            TIntermBinary *element = new TIntermBinary(EOpIndexDirectStruct, structSymbol,
+                                                       TIntermTyped::CreateIndexNode(i));
+
+            sequence->insert(sequence->begin(), element);
+        }
+    }
+    else
+    {
+        TIntermSymbol *symbol = new TIntermSymbol(0, name, type);
+
+        sequence->insert(sequence->begin(), symbol);
+    }
+}
+
+void UseUniformBlockMembers::insertUseCode(TIntermSequence *sequence)
+{
+    for (const auto &block : mBlocks)
+    {
+        if (block.instanceName.empty())
+        {
+            for (const auto &var : block.fields)
+            {
+                AddFieldUseStatements(var, sequence);
+            }
+        }
+        else if (block.arraySize > 0)
+        {
+            TType type                 = GetInterfaceBlockType(block);
+            TString name               = TString(block.instanceName.c_str());
+            TIntermSymbol *arraySymbol = new TIntermSymbol(0, name, type);
+            for (unsigned int i = 0; i < block.arraySize; ++i)
+            {
+                TIntermBinary *instanceSymbol = new TIntermBinary(EOpIndexDirect, arraySymbol,
+                                                                  TIntermTyped::CreateIndexNode(i));
+                for (unsigned int j = 0; j < block.fields.size(); ++j)
+                {
+                    TIntermBinary *element =
+                        new TIntermBinary(EOpIndexDirectInterfaceBlock, instanceSymbol,
+                                          TIntermTyped::CreateIndexNode(j));
+                    sequence->insert(sequence->begin(), element);
+                }
+            }
+        }
+        else
+        {
+            TType type                 = GetInterfaceBlockType(block);
+            TString name               = TString(block.instanceName.c_str());
+            TIntermSymbol *blockSymbol = new TIntermSymbol(0, name, type);
+            for (unsigned int i = 0; i < block.fields.size(); ++i)
+            {
+                TIntermBinary *element = new TIntermBinary(
+                    EOpIndexDirectInterfaceBlock, blockSymbol, TIntermTyped::CreateIndexNode(i));
+
+                sequence->insert(sequence->begin(), element);
+            }
+        }
+    }
+}
+
+}  // namespace anonymous
+
+void UseInterfaceBlockFields(TIntermNode *root, const InterfaceBlockList &blocks)
+{
+    UseUniformBlockMembers useUniformBlock(blocks);
+    root->traverse(&useUniformBlock);
+}
+
+}  // namespace sh

src/compiler/translator/UseInterfaceBlockFields.h

+//
+// Copyright 2016 The ANGLE Project Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+//
+
+// UseInterfaceBlockFields.h: insert statements to reference all members in InterfaceBlock list at
+// the beginning of main. This is to work around a Mac driver that treats unused standard/shared
+// uniform blocks as inactive.
+
+#ifndef COMPILER_TRANSLATOR_USEINTERFACEBLOCKFIELDS_H_
+#define COMPILER_TRANSLATOR_USEINTERFACEBLOCKFIELDS_H_
+
+#include <GLSLANG/ShaderLang.h>
+
+class TIntermNode;
+namespace sh
+{
+
+using InterfaceBlockList = std::vector<sh::InterfaceBlock>;
+
+void UseInterfaceBlockFields(TIntermNode *root, const InterfaceBlockList &blocks);
+}  // namespace sh
+
+#endif  // COMPILER_TRANSLATOR_USEINTERFACEBLOCKFIELDS_H_

src/compiler/translator/util.cpp

@@ -278,10 +278,39 @@ InterpolationType GetInterpolationType(TQualifier qualifier)
     }
 }
 
+TType GetInterfaceBlockType(const sh::InterfaceBlock &block)
+{
+    TType type;
+    TFieldList *fields = new TFieldList;
+    TSourceLoc loc;
+    for (const auto &field : block.fields)
+    {
+        TType *fieldType = new TType(GetShaderVariableType(field));
+        fields->push_back(new TField(fieldType, new TString(field.name.c_str()), loc));
+    }
+
+    TInterfaceBlock *interfaceBlock = new TInterfaceBlock(
+        new TString(block.name.c_str()), fields, new TString(block.instanceName.c_str()),
+        block.arraySize, TLayoutQualifier::create());
+
+    type.setBasicType(EbtInterfaceBlock);
+    type.setInterfaceBlock(interfaceBlock);
+    type.setArraySize(block.arraySize);
+    return type;
+}
+
 TType GetShaderVariableBasicType(const sh::ShaderVariable &var)
 {
     switch (var.type)
     {
+        case GL_BOOL:
+            return TType(EbtBool);
+        case GL_BOOL_VEC2:
+            return TType(EbtBool, 2);
+        case GL_BOOL_VEC3:
+            return TType(EbtBool, 3);
+        case GL_BOOL_VEC4:
+            return TType(EbtBool, 4);
         case GL_FLOAT:
             return TType(EbtFloat);
         case GL_FLOAT_VEC2:

src/compiler/translator/util.h

@@ -38,6 +38,7 @@ bool IsVarying(TQualifier qualifier);
 InterpolationType GetInterpolationType(TQualifier qualifier);
 TString ArrayString(const TType &type);
 
+TType GetInterfaceBlockType(const sh::InterfaceBlock &block);
 TType GetShaderVariableBasicType(const sh::ShaderVariable &var);
 TType GetShaderVariableType(const sh::ShaderVariable &var);
 

src/libANGLE/renderer/gl/ShaderGL.cpp

@@ -70,6 +70,11 @@ ShCompileOptions ShaderGL::prepareSourceAndReturnOptions(std::stringstream *sour
         options |= SH_EMULATE_ISNAN_FLOAT_FUNCTION;
     }
 
+    if (mWorkarounds.useUnusedBlocksWithStandardOrSharedLayout)
+    {
+        options |= SH_USE_UNUSED_STANDARD_SHARED_BLOCKS;
+    }
+
     return options;
 }
 

src/libANGLE/renderer/gl/WorkaroundsGL.h

@@ -24,7 +24,8 @@ struct WorkaroundsGL
           unpackOverlappingRowsSeparatelyUnpackBuffer(false),
           emulateAbsIntFunction(false),
           addAndTrueToLoopCondition(false),
-          emulateIsnanFloat(false)
+          emulateIsnanFloat(false),
+          useUnusedBlocksWithStandardOrSharedLayout(false)
     {
     }
 
@@ -105,6 +106,11 @@ struct WorkaroundsGL
     // this bug, we use an expression to emulate function isnan().
     // Tracking bug: http://crbug.com/650547
     bool emulateIsnanFloat;
+
+    // On Mac with OpenGL version 4.1, unused std140 or shared uniform blocks will be
+    // treated as inactive which is not consistent with WebGL2.0 spec. Reference all members in a
+    // unused std140 or shared uniform block at the beginning of main to work around it.
+    bool useUnusedBlocksWithStandardOrSharedLayout;
 };
 }
 

src/libANGLE/renderer/gl/renderergl_utils.cpp

@@ -897,6 +897,7 @@ void GenerateWorkarounds(const FunctionsGL *functions, WorkaroundsGL *workaround
 
 #if defined(ANGLE_PLATFORM_APPLE)
     workarounds->doWhileGLSLCausesGPUHang = true;
+    workarounds->useUnusedBlocksWithStandardOrSharedLayout = true;
 #endif
 
     workarounds->finishDoesNotCauseQueriesToBeAvailable =

src/tests/gl_tests/UniformBufferTest.cpp

@@ -5,6 +5,7 @@
 //
 
 #include "test_utils/ANGLETest.h"
+#include "test_utils/gl_raii.h"
 
 using namespace angle;
 
@@ -396,6 +397,159 @@ TEST_P(UniformBufferTest, ActiveUniformNames)
     EXPECT_EQ("blockName.f", std::string(&strBuffer[0]));
 }
 
+// Tests active uniforms and blocks when the layout is std140, shared and packed.
+TEST_P(UniformBufferTest, ActiveUniformNumberAndName)
+{
+    // TODO(Jiajia): Figure out why this fails on Intel on Mac.
+    // This case can pass on Intel Mac-10.11/10.12. But it fails on Intel Mac-10.10.
+    if (IsIntel() && IsOSX())
+    {
+        std::cout << "Test skipped on Intel on Mac." << std::endl;
+        return;
+    }
+
+    // This case fails on all AMD platforms (Mac, Linux, Win).
+    if (IsAMD())
+    {
+        std::cout << "Test skipped on AMD." << std::endl;
+        return;
+    }
+
+    const std::string &vertexShaderSource =
+        "#version 300 es\n"
+        "in vec2 position;\n"
+        "out float v;\n"
+        "struct S {\n"
+        "  highp ivec3 a;\n"
+        "  mediump ivec2 b[4];\n"
+        "};\n"
+        "struct T {\n"
+        "  S c[2];\n"
+        "};\n"
+        "layout(std140) uniform blockName0 {\n"
+        "  S s0;\n"
+        "  lowp vec2 v0;\n"
+        "  T t0[2];\n"
+        "  highp uint u0;\n"
+        "};\n"
+        "layout(std140) uniform blockName1 {\n"
+        "  float f1;\n"
+        "  bool b1;\n"
+        "} instanceName1;\n"
+        "layout(shared) uniform blockName2 {\n"
+        "  float f2;\n"
+        "};\n"
+        "layout(packed) uniform blockName3 {\n"
+        "  float f3;\n"
+        "};\n"
+        "void main() {\n"
+        "  v = instanceName1.f1;\n"
+        "  gl_Position = vec4(position, 0, 1);\n"
+        "}";
+
+    const std::string &fragmentShaderSource =
+        "#version 300 es\n"
+        "precision highp float;\n"
+        "in float v;\n"
+        "out vec4 color;\n"
+        "void main() {\n"
+        "  color = vec4(v, 0, 0, 1);\n"
+        "}";
+
+    ANGLE_GL_PROGRAM(program, vertexShaderSource, fragmentShaderSource);
+
+    GLint activeUniforms;
+    glGetProgramiv(program.get(), GL_ACTIVE_UNIFORMS, &activeUniforms);
+
+    ASSERT_EQ(15, activeUniforms);
+
+    GLint activeUniformBlocks;
+    glGetProgramiv(program.get(), GL_ACTIVE_UNIFORM_BLOCKS, &activeUniformBlocks);
+
+    ASSERT_EQ(3, activeUniformBlocks);
+
+    GLint maxLength, size;
+    GLenum type;
+    GLsizei length;
+    glGetProgramiv(program.get(), GL_ACTIVE_UNIFORM_MAX_LENGTH, &maxLength);
+    std::vector<GLchar> strBuffer(maxLength + 1, 0);
+
+    glGetActiveUniform(program.get(), 0, maxLength, &length, &size, &type, &strBuffer[0]);
+    ASSERT_GL_NO_ERROR();
+    EXPECT_EQ(1, size);
+    EXPECT_EQ("s0.a", std::string(&strBuffer[0]));
+
+    glGetActiveUniform(program.get(), 1, maxLength, &length, &size, &type, &strBuffer[0]);
+    ASSERT_GL_NO_ERROR();
+    EXPECT_EQ(4, size);
+    EXPECT_EQ("s0.b[0]", std::string(&strBuffer[0]));
+
+    glGetActiveUniform(program.get(), 2, maxLength, &length, &size, &type, &strBuffer[0]);
+    ASSERT_GL_NO_ERROR();
+    EXPECT_EQ(1, size);
+    EXPECT_EQ("v0", std::string(&strBuffer[0]));
+
+    glGetActiveUniform(program.get(), 3, maxLength, &length, &size, &type, &strBuffer[0]);
+    ASSERT_GL_NO_ERROR();
+    EXPECT_EQ(1, size);
+    EXPECT_EQ("t0[0].c[0].a", std::string(&strBuffer[0]));
+
+    glGetActiveUniform(program.get(), 4, maxLength, &length, &size, &type, &strBuffer[0]);
+    ASSERT_GL_NO_ERROR();
+    EXPECT_EQ(4, size);
+    EXPECT_EQ("t0[0].c[0].b[0]", std::string(&strBuffer[0]));
+
+    glGetActiveUniform(program.get(), 5, maxLength, &length, &size, &type, &strBuffer[0]);
+    ASSERT_GL_NO_ERROR();
+    EXPECT_EQ(1, size);
+    EXPECT_EQ("t0[0].c[1].a", std::string(&strBuffer[0]));
+
+    glGetActiveUniform(program.get(), 6, maxLength, &length, &size, &type, &strBuffer[0]);
+    ASSERT_GL_NO_ERROR();
+    EXPECT_EQ(4, size);
+    EXPECT_EQ("t0[0].c[1].b[0]", std::string(&strBuffer[0]));
+
+    glGetActiveUniform(program.get(), 7, maxLength, &length, &size, &type, &strBuffer[0]);
+    ASSERT_GL_NO_ERROR();
+    EXPECT_EQ(1, size);
+    EXPECT_EQ("t0[1].c[0].a", std::string(&strBuffer[0]));
+
+    glGetActiveUniform(program.get(), 8, maxLength, &length, &size, &type, &strBuffer[0]);
+    ASSERT_GL_NO_ERROR();
+    EXPECT_EQ(4, size);
+    EXPECT_EQ("t0[1].c[0].b[0]", std::string(&strBuffer[0]));
+
+    glGetActiveUniform(program.get(), 9, maxLength, &length, &size, &type, &strBuffer[0]);
+    ASSERT_GL_NO_ERROR();
+    EXPECT_EQ(1, size);
+    EXPECT_EQ("t0[1].c[1].a", std::string(&strBuffer[0]));
+
+    glGetActiveUniform(program.get(), 10, maxLength, &length, &size, &type, &strBuffer[0]);
+    ASSERT_GL_NO_ERROR();
+    EXPECT_EQ(4, size);
+    EXPECT_EQ("t0[1].c[1].b[0]", std::string(&strBuffer[0]));
+
+    glGetActiveUniform(program.get(), 11, maxLength, &length, &size, &type, &strBuffer[0]);
+    ASSERT_GL_NO_ERROR();
+    EXPECT_EQ(1, size);
+    EXPECT_EQ("u0", std::string(&strBuffer[0]));
+
+    glGetActiveUniform(program.get(), 12, maxLength, &length, &size, &type, &strBuffer[0]);
+    ASSERT_GL_NO_ERROR();
+    EXPECT_EQ(1, size);
+    EXPECT_EQ("blockName1.f1", std::string(&strBuffer[0]));
+
+    glGetActiveUniform(program.get(), 13, maxLength, &length, &size, &type, &strBuffer[0]);
+    ASSERT_GL_NO_ERROR();
+    EXPECT_EQ(1, size);
+    EXPECT_EQ("blockName1.b1", std::string(&strBuffer[0]));
+
+    glGetActiveUniform(program.get(), 14, maxLength, &length, &size, &type, &strBuffer[0]);
+    ASSERT_GL_NO_ERROR();
+    EXPECT_EQ(1, size);
+    EXPECT_EQ("f2", std::string(&strBuffer[0]));
+}
+
 // Test that using a very large buffer to back a small uniform block works OK.
 TEST_P(UniformBufferTest, VeryLarge)
 {