Finish implementing compute shaders, within #version 430, partly based on a submission.

git-svn-id: https://cvs.khronos.org/svn/repos/ogl/trunk/ecosystem/public/sdk/tools/glslang@27674 e7fa87d3-cd2b-0410-9028-fcbf551c1848

Test/430.comp

@@ -1,7 +1,21 @@
 #version 430 core
 
+layout(local_size_x = 2) in;
+layout(local_size_x = 16) in;     // ERROR, changing
+layout(local_size_z = 4096) in;   // ERROR, too large
+layout(local_size_x = 2) in;
+
+const int total = gl_MaxComputeWorkGroupCount.y 
+                + gl_MaxComputeUniformComponents
+                + gl_MaxComputeTextureImageUnits
+                + gl_MaxComputeImageUniforms
+                + gl_MaxComputeAtomicCounters
+                + gl_MaxComputeAtomicCounterBuffers;
+
 void main()
 {
+    barrier();
+    memoryBarrier();
     memoryBarrierAtomicCounter();
     memoryBarrierBuffer();
     memoryBarrierShared();
@@ -9,4 +23,16 @@ void main()
     groupMemoryBarrier();
 }
 
-layout(location = 2) in vec3 v3;
+layout(location = 2) in vec3 v3;      // ERROR
+in float f;                           // ERROR
+out float fo;                         // ERROR
+
+shared vec4 s;
+layout(location = 2) shared vec4 sl;  // ERROR
+shared float fs = 4.2;                // ERROR
+
+layout(local_size_x = 2, local_size_y = 3, local_size_z = 4) out;  // ERROR
+
+int arrX[gl_WorkGroupSize.x];
+int arrY[gl_WorkGroupSize.y];
+int arrZ[gl_WorkGroupSize.z];

Test/430.vert

@@ -142,3 +142,16 @@ out bblck5 {
     layout(xfb_offset=0) vec4 bbv1;
     layout(xfb_stride=80, xfb_buffer=1, xfb_offset=64) vec4 bbv2;
 } bbinst5;
+
+shared vec4 sharedv;
+
+void fooBarrier()
+{
+    barrier();                       // ERROR
+    memoryBarrier();
+    memoryBarrierAtomicCounter();
+    memoryBarrierBuffer();
+    memoryBarrierShared();           // ERROR
+    memoryBarrierImage();
+    groupMemoryBarrier();            // ERROR
+}

Test/baseResults/300layout.vert.out

@@ -15,10 +15,9 @@ ERROR: 0:38: 'output block' : not supported with this profile: es
 ERROR: 0:42: 'location qualifier on output' : not supported in this stage: vertex
 ERROR: 0:50: 'shared' : not supported with this profile: es
 ERROR: 0:50: 'shared' : not supported in this stage: vertex
-ERROR: 0:50: '' : memory qualifiers can only be used on image types 
 ERROR: 0:54: 'layout' : cannot specify packing on a variable declaration 
 ERROR: 0:57: 'location' : overlapping use of location 40
-ERROR: 19 compilation errors.  No code generated.
+ERROR: 18 compilation errors.  No code generated.
 
 
 Shader version: 300

Test/baseResults/430.comp.out

@@ -1,36 +1,81 @@
 430.comp
 Warning, version 430 is not yet complete; most version-specific features are present, but some are missing.
-ERROR: 0:12: 'location qualifier on input' : not supported in this stage: compute
+ERROR: 0:4: 'local_size' : cannot change previously set size 
-ERROR: 1 compilation errors.  No code generated.
+ERROR: 0:5: 'local_size' : too large; see gl_MaxComputeWorkGroupSize 
+ERROR: 0:26: 'in' : global storage input qualifier cannot be used in a compute shader 
+ERROR: 0:26: 'location qualifier on input' : not supported in this stage: compute
+ERROR: 0:27: 'in' : global storage input qualifier cannot be used in a compute shader 
+ERROR: 0:28: 'out' : global storage output qualifier cannot be used in a compute shader 
+ERROR: 0:31: 'shared' : cannot apply layout qualifiers to a shared variable 
+ERROR: 0:31: 'location' : can only appy to uniform, buffer, in, or out storage qualifiers 
+ERROR: 0:32: 'shared' :  cannot initialize this type of qualifier  
+ERROR: 0:34: 'local_size' : can only apply to 'in' 
+ERROR: 0:34: 'local_size' : can only apply to 'in' 
+ERROR: 0:34: 'local_size' : can only apply to 'in' 
+ERROR: 12 compilation errors.  No code generated.
 
 
 Shader version: 430
+local_size = (2, 1, 4096)
 ERROR: node is still EOpNull!
-0:3  Function Definition: main( (void)
+0:15  Function Definition: main( (void)
-0:3    Function Parameters: 
+0:15    Function Parameters: 
-0:5    Sequence
+0:17    Sequence
-0:5      MemoryBarrierAtomicCounter (void)
+0:17      Barrier (void)
-0:6      MemoryBarrierBuffer (void)
+0:18      MemoryBarrier (void)
-0:7      MemoryBarrierShared (void)
+0:19      MemoryBarrierAtomicCounter (void)
-0:8      MemoryBarrierImage (void)
+0:20      MemoryBarrierBuffer (void)
-0:9      GroupMemoryBarrier (void)
+0:21      MemoryBarrierShared (void)
+0:22      MemoryBarrierImage (void)
+0:23      GroupMemoryBarrier (void)
 0:?   Linker Objects
+0:?     'gl_WorkGroupSize' (const 3-component vector of uint)
+0:?       2 (const uint)
+0:?       1 (const uint)
+0:?       4096 (const uint)
+0:?     'total' (const int)
+0:?       66592 (const int)
 0:?     'v3' (layout(location=2 ) in 3-component vector of float)
+0:?     'f' (in float)
+0:?     'fo' (out float)
+0:?     's' (shared 4-component vector of float)
+0:?     'sl' (layout(location=2 ) shared 4-component vector of float)
+0:?     'fs' (shared float)
+0:?     'arrX' (2-element array of int)
+0:?     'arrY' (1-element array of int)
+0:?     'arrZ' (4096-element array of int)
 
 
 Linked compute stage:
 
 
 Shader version: 430
+local_size = (2, 1, 4096)
 ERROR: node is still EOpNull!
-0:3  Function Definition: main( (void)
+0:15  Function Definition: main( (void)
-0:3    Function Parameters: 
+0:15    Function Parameters: 
-0:5    Sequence
+0:17    Sequence
-0:5      MemoryBarrierAtomicCounter (void)
+0:17      Barrier (void)
-0:6      MemoryBarrierBuffer (void)
+0:18      MemoryBarrier (void)
-0:7      MemoryBarrierShared (void)
+0:19      MemoryBarrierAtomicCounter (void)
-0:8      MemoryBarrierImage (void)
+0:20      MemoryBarrierBuffer (void)
-0:9      GroupMemoryBarrier (void)
+0:21      MemoryBarrierShared (void)
+0:22      MemoryBarrierImage (void)
+0:23      GroupMemoryBarrier (void)
 0:?   Linker Objects
+0:?     'gl_WorkGroupSize' (const 3-component vector of uint)
+0:?       2 (const uint)
+0:?       1 (const uint)
+0:?       4096 (const uint)
+0:?     'total' (const int)
+0:?       66592 (const int)
 0:?     'v3' (layout(location=2 ) in 3-component vector of float)
+0:?     'f' (in float)
+0:?     'fo' (out float)
+0:?     's' (shared 4-component vector of float)
+0:?     'sl' (layout(location=2 ) shared 4-component vector of float)
+0:?     'fs' (shared float)
+0:?     'arrX' (2-element array of int)
+0:?     'arrY' (1-element array of int)
+0:?     'arrZ' (4096-element array of int)
 

Test/baseResults/430.vert.out

@@ -44,7 +44,11 @@ ERROR: 0:92: 'transform feedback qualifier' : not supported for this version or
 ERROR: 0:93: 'transform feedback qualifier' : not supported for this version or the enabled extensions 
 ERROR: 0:93: 'transform feedback qualifier' : not supported for this version or the enabled extensions 
 ERROR: 0:93: 'transform feedback qualifier' : not supported for this version or the enabled extensions 
-ERROR: 44 compilation errors.  No code generated.
+ERROR: 0:146: 'shared' : not supported in this stage: vertex
+ERROR: 0:150: 'barrier' : no matching overloaded function found 
+ERROR: 0:154: 'memoryBarrierShared' : no matching overloaded function found 
+ERROR: 0:156: 'groupMemoryBarrier' : no matching overloaded function found 
+ERROR: 48 compilation errors.  No code generated.
 
 
 Shader version: 430
@@ -70,6 +74,19 @@ ERROR: node is still EOpNull!
 0:31      'v3' (volatile in 3-component vector of float)
 0:31      'v2' (in 2-component vector of float)
 0:31      'cv3' (in 3-component vector of float)
+0:148  Function Definition: fooBarrier( (void)
+0:148    Function Parameters: 
+0:150    Sequence
+0:150      Constant:
+0:150        0.000000
+0:151      MemoryBarrier (void)
+0:152      MemoryBarrierAtomicCounter (void)
+0:153      MemoryBarrierBuffer (void)
+0:154      Constant:
+0:154        0.000000
+0:155      MemoryBarrierImage (void)
+0:156      Constant:
+0:156        0.000000
 0:?   Linker Objects
 0:?     'v4' (layout(location=3 ) 4-component vector of float)
 0:?     'uv4' (layout(location=4 ) uniform 4-component vector of float)
@@ -109,6 +126,7 @@ ERROR: node is still EOpNull!
 0:?     'bh' (layout(xfb_buffer=1 xfb_offset=32 xfb_stride=80 ) smooth out 4-component vector of float)
 0:?     'bbinst4' (layout(xfb_stride=80 ) out block{layout(xfb_buffer=1 xfb_offset=16 ) out 4-component vector of float bbv1})
 0:?     'bbinst5' (out block{layout(xfb_buffer=1 xfb_offset=0 ) out 4-component vector of float bbv1, layout(xfb_buffer=1 xfb_offset=64 xfb_stride=80 ) out 4-component vector of float bbv2})
+0:?     'sharedv' (shared 4-component vector of float)
 0:?     'gl_VertexID' (gl_VertexId int)
 0:?     'gl_InstanceID' (gl_InstanceId int)
 
@@ -142,6 +160,19 @@ ERROR: node is still EOpNull!
 0:31      'v3' (volatile in 3-component vector of float)
 0:31      'v2' (in 2-component vector of float)
 0:31      'cv3' (in 3-component vector of float)
+0:148  Function Definition: fooBarrier( (void)
+0:148    Function Parameters: 
+0:150    Sequence
+0:150      Constant:
+0:150        0.000000
+0:151      MemoryBarrier (void)
+0:152      MemoryBarrierAtomicCounter (void)
+0:153      MemoryBarrierBuffer (void)
+0:154      Constant:
+0:154        0.000000
+0:155      MemoryBarrierImage (void)
+0:156      Constant:
+0:156        0.000000
 0:?   Linker Objects
 0:?     'v4' (layout(location=3 ) 4-component vector of float)
 0:?     'uv4' (layout(location=4 ) uniform 4-component vector of float)
@@ -181,6 +212,7 @@ ERROR: node is still EOpNull!
 0:?     'bh' (layout(xfb_buffer=1 xfb_offset=32 xfb_stride=80 ) smooth out 4-component vector of float)
 0:?     'bbinst4' (layout(xfb_stride=80 ) out block{layout(xfb_buffer=1 xfb_offset=16 ) out 4-component vector of float bbv1})
 0:?     'bbinst5' (out block{layout(xfb_buffer=1 xfb_offset=0 ) out 4-component vector of float bbv1, layout(xfb_buffer=1 xfb_offset=64 xfb_stride=80 ) out 4-component vector of float bbv2})
+0:?     'sharedv' (shared 4-component vector of float)
 0:?     'gl_VertexID' (gl_VertexId int)
 0:?     'gl_InstanceID' (gl_InstanceId int)
 

Test/baseResults/specExamples.vert.out

@@ -40,7 +40,6 @@ ERROR: 0:122: '' : memory qualifiers can only be used on image types
 ERROR: 0:128: '' : memory qualifiers can only be used on image types 
 ERROR: 0:129: '' : memory qualifiers can only be used on image types 
 ERROR: 0:132: 'shared' : not supported in this stage: vertex
-ERROR: 0:132: '' : memory qualifiers can only be used on image types 
 ERROR: 0:134: '' : function does not return a value: funcA
 ERROR: 0:136: '' : function does not return a value: funcB
 ERROR: 0:153: '' : function does not return a value: func3
@@ -48,7 +47,7 @@ ERROR: 0:170: 'coherent' : argument cannot drop memory qualifier when passed to
 ERROR: 0:192: 'constructor' : constructing from a non-dereferenced array 
 ERROR: 0:193: 'constructor' : constructing from a non-dereferenced array 
 ERROR: 0:194: 'constructor' : constructing from a non-dereferenced array 
-ERROR: 46 compilation errors.  No code generated.
+ERROR: 45 compilation errors.  No code generated.
 
 
 Shader version: 430

Todo.txt

@@ -57,10 +57,9 @@ Link Validation
       + location match
       - block matching
       - component/binding/index/offset match check
-      - compute shader layout(local_size_*) matching
+      + compute shader layout(local_size_*) matching
       + mixed es/non-es profiles are an error
       - matching redeclarations of interface blocks
-      - 4.3: early_fragment_tests contradictions
       - 4.3: implicit array sizing is cross shader within a stage
       - 4.4: If gl_FragCoord is redeclared in any fragment shader in a program, it must be redeclared in all the fragment shaders in that program that have a static use gl_FragCoord
 
@@ -211,7 +210,7 @@ Shader Functionality to Implement/Finish
             - Arrays of other objects (uniform blocks) containing implicitly sized arrays will have the same implicit size for all 
                 elements of the array.
       - Arrays of arrays are now supported, as per the GL_ARB_arrays_of_arrays extension.
-      - Compute shaders are now supported, as per the GL_ARB_compute_shader extension.
+      + Compute shaders are now supported, as per the GL_ARB_compute_shader extension.
       - Added imageSize() built-ins to query the dimensions of an image.
       - All choice of depth or stencil texturing, for a packed depth-stencil texture, as per the 
         GL_ARB_stencil_texturing extension.

glslang/Include/BaseTypes.h

@@ -69,6 +69,7 @@ enum TStorageQualifier {
     EvqVaryingOut,    // pipeline ouput, read/write
     EvqUniform,       // read only, shader with app
     EvqBuffer,        // read only, shader with app
+    EvqShared,        // compute shader's read/write 'shared' qualifier
     
     // parameters
     EvqIn,            // also, for 'in' in the grammar before we know if it's a pipeline input or an 'in' parameter
@@ -109,6 +110,8 @@ __inline const char* GetStorageQualifierString(TStorageQualifier q)
     case EvqVaryingIn:      return "in";             break;
     case EvqVaryingOut:     return "out";            break;
     case EvqUniform:        return "uniform";        break;
+    case EvqBuffer:         return "buffer";         break;
+    case EvqShared:         return "shared";         break;
     case EvqIn:             return "in";             break;
     case EvqOut:            return "out";            break;
     case EvqInOut:          return "inout";          break;

glslang/Include/Types.h

@@ -342,7 +342,7 @@ public:
 
     bool isMemory() const
     {
-        return shared || coherent || volatil || restrict || readonly || writeonly;
+        return coherent || volatil || restrict || readonly || writeonly;
     }
     bool isInterpolation() const
     {
@@ -693,6 +693,7 @@ struct TShaderQualifiers {
     TVertexSpacing spacing;
     TVertexOrder order;
     bool pointMode;
+    int localSize[3];         // compute shader
     bool earlyFragmentTests;  // fragment input
 
     void init()
@@ -705,6 +706,9 @@ struct TShaderQualifiers {
         spacing = EvsNone;
         order = EvoNone;
         pointMode = false;
+        localSize[0] = 1;
+        localSize[1] = 1;
+        localSize[2] = 1;
         earlyFragmentTests = false;
     }
 
@@ -728,6 +732,10 @@ struct TShaderQualifiers {
             order = src.order;
         if (src.pointMode)
             pointMode = true;
+        for (int i = 0; i < 3; ++i) {
+            if (src.localSize[i] > 1)
+                localSize[i] = src.localSize[i];
+        }
         if (src.earlyFragmentTests)
             earlyFragmentTests = true;
     }

glslang/MachineIndependent/Initialize.cpp

@@ -1079,7 +1079,7 @@ void TBuiltIns::initialize(int version, EProfile profile)
     if (version >= 430) {
         stageBuiltins[EShLangCompute].append(
             "in uvec3 gl_NumWorkGroups;"
-            // TODO: 4.3 functionality: compute shader: constant with no initializer                "const uvec3 gl_WorkGroupSize;"
+            "const uvec3 gl_WorkGroupSize = uvec3(1,1,1);"
 
             "in uvec3 gl_WorkGroupID;"
             "in uvec3 gl_LocalInvocationID;"

glslang/MachineIndependent/ParseHelper.cpp

@@ -1969,6 +1969,8 @@ void TParseContext::globalQualifierCheck(TSourceLoc loc, const TQualifier& quali
             break;
 
         case EShLangCompute:
+            if (! symbolTable.atBuiltInLevel())
+                error(loc, "global storage input qualifier cannot be used in a compute shader", "in", "");
             break;
 
         default:
@@ -2005,6 +2007,7 @@ void TParseContext::globalQualifierCheck(TSourceLoc loc, const TQualifier& quali
         break;
 
         case EShLangCompute:
+            error(loc, "global storage output qualifier cannot be used in a compute shader", "out", "");
             break;
 
         default:
@@ -3178,6 +3181,18 @@ void TParseContext::setLayoutQualifier(TSourceLoc loc, TPublicType& publicType,
         break;
 
     case EShLangCompute:
+        if (id == "local_size_x") {
+            publicType.shaderQualifiers.localSize[0] = value;
+            return;
+        }
+        if (id == "local_size_y") {
+            publicType.shaderQualifiers.localSize[1] = value;
+            return;
+        }
+        if (id == "local_size_z") {
+            publicType.shaderQualifiers.localSize[2] = value;
+            return;
+        }
         break;
 
     default:
@@ -3401,6 +3416,9 @@ void TParseContext::layoutTypeCheck(TSourceLoc loc, const TType& type)
 // if there are blocks, atomic counters, variables, etc.
 void TParseContext::layoutQualifierCheck(TSourceLoc loc, const TQualifier& qualifier)
 {
+    if (qualifier.storage == EvqShared && qualifier.hasLayout())
+        error(loc, "cannot apply layout qualifiers to a shared variable", "shared", "");
+
     // "It is a compile-time error to use *component* without also specifying the location qualifier (order does not matter)."
     if (qualifier.layoutComponent != TQualifier::layoutComponentEnd && qualifier.layoutLocation == TQualifier::layoutLocationEnd)
         error(loc, "must specify 'location' to use 'component'", "component", "");
@@ -3490,6 +3508,10 @@ void TParseContext::checkNoShaderLayouts(TSourceLoc loc, const TShaderQualifiers
         else
             assert(0);
     }
+    for (int i = 0; i < 3; ++i) {
+        if (shaderQualifiers.localSize[i] > 1)
+            error(loc, message, "local_size", "");
+    }
 }
 
 //
@@ -4561,6 +4583,34 @@ void TParseContext::updateStandaloneQualifierDefaults(TSourceLoc loc, const TPub
         else
             error(loc, "can only apply to 'in'", "point_mode", "");
     }
+    for (int i = 0; i < 3; ++i) {
+        if (publicType.shaderQualifiers.localSize[i] > 1) {
+            if (publicType.qualifier.storage == EvqVaryingIn) {
+                if (! intermediate.setLocalSize(i, publicType.shaderQualifiers.localSize[i]))
+                    error(loc, "cannot change previously set size", "local_size", "");
+                else {
+                    int max;
+                    switch (i) {
+                    case 0: max = resources.maxComputeWorkGroupSizeX; break;
+                    case 1: max = resources.maxComputeWorkGroupSizeY; break;
+                    case 2: max = resources.maxComputeWorkGroupSizeZ; break;
+                    default: break;
+                    }
+                    if (intermediate.getLocalSize(i) > (unsigned int)max)
+                        error(loc, "too large; see gl_MaxComputeWorkGroupSize", "local_size", "");
+
+                    // Fix the existing constant gl_WorkGroupSize with this new information.
+                    bool builtIn;
+                    TSymbol* symbol = symbolTable.find("gl_WorkGroupSize", &builtIn);
+                    if (builtIn)
+                        makeEditable(symbol);
+                    TVariable* workGroupSize = symbol->getAsVariable();
+                    workGroupSize->getWritableConstArray()[i].setUConst(intermediate.getLocalSize(i));
+                }
+            } else
+                error(loc, "can only apply to 'in'", "local_size", "");
+        }
+    }
     if (publicType.shaderQualifiers.earlyFragmentTests) {
         if (publicType.qualifier.storage == EvqVaryingIn)
             intermediate.setEarlyFragmentTests();

glslang/MachineIndependent/SymbolTable.cpp

@@ -242,8 +242,7 @@ TVariable::TVariable(const TVariable& copyOf) : TSymbol(copyOf)
 
     if (! copyOf.unionArray.empty()) {
         assert(! copyOf.type.isStruct());
-        TConstUnionArray newArray(1);
+        TConstUnionArray newArray(copyOf.unionArray, 0, copyOf.unionArray.size());
-        newArray[0] = copyOf.unionArray[0];
         unionArray = newArray;
     }
 }

glslang/MachineIndependent/SymbolTable.h

@@ -154,6 +154,7 @@ public:
     virtual TType& getWritableType() { assert(writable); return type; }
     virtual bool isUserType() const { return userType; }
     virtual const TConstUnionArray& getConstArray() const { return unionArray; }
+    virtual TConstUnionArray& getWritableConstArray() { assert(writable); return unionArray; }
     virtual void setConstArray(const TConstUnionArray& constArray) { unionArray = constArray; }
 
     virtual void dump(TInfoSink &infoSink) const;

glslang/MachineIndependent/glslang.y

@@ -1220,7 +1220,7 @@ storage_qualifier
         parseContext.profileRequires($1.loc, ECoreProfile, 430, 0, "shared");
         parseContext.requireStage($1.loc, EShLangCompute, "shared");
         $$.init($1.loc);
-        $$.qualifier.shared = true;
+        $$.qualifier.storage = EvqShared;
     }
     | COHERENT {
         $$.init($1.loc);

glslang/MachineIndependent/intermOut.cpp

@@ -616,6 +616,7 @@ void TIntermediate::output(TInfoSink& infoSink, bool tree)
         break;
 
     case EShLangCompute:
+        infoSink.debug << "local_size = (" << localSize[0] << ", " << localSize[1] << ", " << localSize[2] << ")\n";
         break;
 
     default:

glslang/MachineIndependent/linkValidate.cpp

@@ -117,6 +117,13 @@ void TIntermediate::merge(TInfoSink& infoSink, TIntermediate& unit)
     if (unit.pointMode)
         pointMode = true;
 
+    for (int i = 0; i < 3; ++i) {
+        if (localSize[i] > 1)
+            localSize[i] = unit.localSize[i];
+        else if (localSize[i] != unit.localSize[i])
+            error(infoSink, "Contradictory local size");
+    }
+
     if (unit.xfbMode)
         xfbMode = true;
     for (size_t b = 0; b < xfbBuffers.size(); ++b) {

glslang/MachineIndependent/localintermediate.h

@@ -114,6 +114,9 @@ public:
         invocations(0), vertices(0), inputPrimitive(ElgNone), outputPrimitive(ElgNone), pixelCenterInteger(false), originUpperLeft(false),
         vertexSpacing(EvsNone), vertexOrder(EvoNone), pointMode(false), earlyFragmentTests(false), xfbMode(false)
     {
+        localSize[0] = 1;
+        localSize[1] = 1;
+        localSize[2] = 1;
         xfbBuffers.resize(TQualifier::layoutXfbBufferEnd);
     }
     void setLimits(const TBuiltInResource& r) { resources = r; }
@@ -221,6 +224,16 @@ public:
     TVertexOrder getVertexOrder() const { return vertexOrder; }
     void setPointMode() { pointMode = true; }
     bool getPointMode() const { return pointMode; }
+    
+    bool setLocalSize(int dim, int size)
+    {
+        if (localSize[dim] > 1)
+            return size == localSize[dim];
+        localSize[dim] = size;
+        return true;
+    }
+    unsigned int getLocalSize(int dim) const { return localSize[dim]; }
+
     void setXfbMode() { xfbMode = true; }
     bool getXfbMode() const { return xfbMode; }
     bool setOutputPrimitive(TLayoutGeometry p)
@@ -289,6 +302,7 @@ protected:
     TVertexSpacing vertexSpacing;
     TVertexOrder vertexOrder;
     bool pointMode;
+    int localSize[3];
     bool earlyFragmentTests;
     bool xfbMode;