Add texture gather functions (and extension check) for GLSL 400 and GL_ARB_texture_gather.

git-svn-id: https://cvs.khronos.org/svn/repos/ogl/trunk/ecosystem/public/sdk/tools/glslang@23631 e7fa87d3-cd2b-0410-9028-fcbf551c1848
2024-11-09 20:10:06 +00:00 · 2013-10-21 19:25:59 +00:00 · 2013-10-21 19:25:59 +00:00 · 21a8770f92
commit 21a8770f92
parent a65dc63d46
11 changed files with 301 additions and 36 deletions
--- a/Test/130.frag
+++ b/Test/130.frag
@ -17,3 +17,17 @@ void main()
 {
    float clip = gl_ClipDistance[3];
 }
+
+uniform samplerCube sampC;
+
+void foo()
+{
+    vec4 s = textureGather(sampC, vec3(0.2));
+}
+
+#extension GL_ARB_texture_gather : enable
+
+void bar()
+{
+    vec4 s = textureGather(sampC, vec3(0.2));
+}
--- a/Test/400.frag
+++ b/Test/400.frag
@ -1,13 +1,24 @@
-#version 330 core
+#version 400 core

 in vec2 c2D;
 flat in int i;
 out vec4 outp;
 uniform sampler2D arrayedSampler[5];
+uniform usampler2DRect samp2dr;
+uniform isampler2DArray isamp2DA;

 void main()
 {
    vec4 v;
    v = texture(arrayedSampler[i], c2D);
    outp.x = gl_ClipDistance[1];
+
+    ivec2 offsets[4];
+    uvec4 uv4 = textureGatherOffsets(samp2dr, c2D, offsets, 2);
+    vec4 v4 = textureGather(arrayedSampler[0], c2D);
+    ivec4 iv4 = textureGatherOffset(isamp2DA, vec3(0.1), ivec2(1), 3);
+    iv4 = textureGatherOffset(isamp2DA, vec3(0.1), ivec2(1), i);  // ERROR, last argument not const
+    iv4 = textureGatherOffset(isamp2DA, vec3(0.1), ivec2(1), 4);  // ERROR, last argument out of range
+    iv4 = textureGatherOffset(isamp2DA, vec3(0.1), ivec2(1), 1+2);
+    iv4 = textureGatherOffset(isamp2DA, vec3(0.1), ivec2(i));
 }
--- a/Test/baseResults/130.frag.out
+++ b/Test/baseResults/130.frag.out
@ -1,5 +1,8 @@
 Warning, version 130 is not yet complete; most features are present, but a few are missing.
-0:? Sequence
+ERROR: 0:25: 'texture gather function' : not supported for this version or the enabled extensions 
+ERROR: 1 compilation errors.  No code generated.
+
+ERROR: node is still EOpNull!
 0:16  Function Definition: main( (void)
 0:16    Function Parameters: 
 0:18    Sequence
@ -10,6 +13,30 @@ Warning, version 130 is not yet complete; most features are present, but a few a
 0:18            'gl_ClipDistance' (smooth in unsized array of float)
 0:18            Constant:
 0:18              3 (const int)
+0:23  Function Definition: foo( (void)
+0:23    Function Parameters: 
+0:25    Sequence
+0:25      Sequence
+0:25        move second child to first child (4-component vector of float)
+0:25          's' (4-component vector of float)
+0:25          Function Call: textureGather(sC1;vf3; (4-component vector of float)
+0:25            'sampC' (uniform samplerCube)
+0:25            Constant:
+0:25              0.200000
+0:25              0.200000
+0:25              0.200000
+0:30  Function Definition: bar( (void)
+0:30    Function Parameters: 
+0:32    Sequence
+0:32      Sequence
+0:32        move second child to first child (4-component vector of float)
+0:32          's' (4-component vector of float)
+0:32          Function Call: textureGather(sC1;vf3; (4-component vector of float)
+0:32            'sampC' (uniform samplerCube)
+0:32            Constant:
+0:32              0.200000
+0:32              0.200000
+0:32              0.200000
 0:?   Linker Objects
 0:?     'a' (3-component vector of float)
 0:?     'b' (float)
@ -19,4 +46,5 @@ Warning, version 130 is not yet complete; most features are present, but a few a
 0:?     'fflat' (flat in float)
 0:?     'fsmooth' (smooth in float)
 0:?     'fnop' (noperspective in float)
+0:?     'sampC' (uniform samplerCube)

--- a/Test/baseResults/400.frag.out
+++ b/Test/baseResults/400.frag.out
@ -1,30 +1,113 @@
-Warning, version 330 is not yet complete; some version-specific features are present, but many are missing.
-ERROR: 0:11: 'variable indexing sampler array' : not supported for this version or the enabled extensions 
-ERROR: 1 compilation errors.  No code generated.
+Warning, version 400 is not yet complete; some version-specific features are present, but many are missing.
+ERROR: 0:20: 'texture gather component' : must be a constant 
+ERROR: 0:21: 'texture gather component' : must be 0, 1, 2, or 3 
+ERROR: 2 compilation errors.  No code generated.

 ERROR: node is still EOpNull!
-0:8  Function Definition: main( (void)
-0:8    Function Parameters: 
+0:10  Function Definition: main( (void)
+0:10    Function Parameters: 
 0:?     Sequence
-0:11      move second child to first child (4-component vector of float)
-0:11        'v' (4-component vector of float)
-0:11        Function Call: texture(s21;vf2; (4-component vector of float)
-0:11          indirect index (uniform sampler2D)
-0:11            'arrayedSampler' (uniform 5-element array of sampler2D)
-0:11            'i' (flat in int)
-0:11          'c2D' (smooth in 2-component vector of float)
-0:12      move second child to first child (float)
-0:12        direct index (float)
-0:12          'outp' (out 4-component vector of float)
-0:12          Constant:
-0:12            0 (const int)
-0:12        direct index (smooth in float)
-0:12          'gl_ClipDistance' (smooth in unsized array of float)
-0:12          Constant:
-0:12            1 (const int)
+0:13      move second child to first child (4-component vector of float)
+0:13        'v' (4-component vector of float)
+0:13        Function Call: texture(s21;vf2; (4-component vector of float)
+0:13          indirect index (uniform sampler2D)
+0:13            'arrayedSampler' (uniform 5-element array of sampler2D)
+0:13            'i' (flat in int)
+0:13          'c2D' (smooth in 2-component vector of float)
+0:14      move second child to first child (float)
+0:14        direct index (float)
+0:14          'outp' (out 4-component vector of float)
+0:14          Constant:
+0:14            0 (const int)
+0:14        direct index (smooth in float)
+0:14          'gl_ClipDistance' (smooth in unsized array of float)
+0:14          Constant:
+0:14            1 (const int)
+0:17      Sequence
+0:17        move second child to first child (4-component vector of uint)
+0:17          'uv4' (4-component vector of uint)
+0:17          Function Call: textureGatherOffsets(usR21;vf2;vi2[4];i1; (4-component vector of uint)
+0:17            'samp2dr' (uniform usampler2DRect)
+0:17            'c2D' (smooth in 2-component vector of float)
+0:17            'offsets' (4-element array of 2-component vector of int)
+0:17            Constant:
+0:17              2 (const int)
+0:18      Sequence
+0:18        move second child to first child (4-component vector of float)
+0:18          'v4' (4-component vector of float)
+0:18          Function Call: textureGather(s21;vf2; (4-component vector of float)
+0:18            direct index (uniform sampler2D)
+0:18              'arrayedSampler' (uniform 5-element array of sampler2D)
+0:18              Constant:
+0:18                0 (const int)
+0:18            'c2D' (smooth in 2-component vector of float)
+0:19      Sequence
+0:19        move second child to first child (4-component vector of int)
+0:19          'iv4' (4-component vector of int)
+0:19          Function Call: textureGatherOffset(isA21;vf3;vi2;i1; (4-component vector of int)
+0:19            'isamp2DA' (uniform isampler2DArray)
+0:19            Constant:
+0:19              0.100000
+0:19              0.100000
+0:19              0.100000
+0:19            Constant:
+0:19              1 (const int)
+0:19              1 (const int)
+0:19            Constant:
+0:19              3 (const int)
+0:20      move second child to first child (4-component vector of int)
+0:20        'iv4' (4-component vector of int)
+0:20        Function Call: textureGatherOffset(isA21;vf3;vi2;i1; (4-component vector of int)
+0:20          'isamp2DA' (uniform isampler2DArray)
+0:20          Constant:
+0:20            0.100000
+0:20            0.100000
+0:20            0.100000
+0:20          Constant:
+0:20            1 (const int)
+0:20            1 (const int)
+0:20          'i' (flat in int)
+0:21      move second child to first child (4-component vector of int)
+0:21        'iv4' (4-component vector of int)
+0:21        Function Call: textureGatherOffset(isA21;vf3;vi2;i1; (4-component vector of int)
+0:21          'isamp2DA' (uniform isampler2DArray)
+0:21          Constant:
+0:21            0.100000
+0:21            0.100000
+0:21            0.100000
+0:21          Constant:
+0:21            1 (const int)
+0:21            1 (const int)
+0:21          Constant:
+0:21            4 (const int)
+0:22      move second child to first child (4-component vector of int)
+0:22        'iv4' (4-component vector of int)
+0:22        Function Call: textureGatherOffset(isA21;vf3;vi2;i1; (4-component vector of int)
+0:22          'isamp2DA' (uniform isampler2DArray)
+0:22          Constant:
+0:22            0.100000
+0:22            0.100000
+0:22            0.100000
+0:22          Constant:
+0:22            1 (const int)
+0:22            1 (const int)
+0:22          Constant:
+0:22            3 (const int)
+0:23      move second child to first child (4-component vector of int)
+0:23        'iv4' (4-component vector of int)
+0:23        Function Call: textureGatherOffset(isA21;vf3;vi2; (4-component vector of int)
+0:23          'isamp2DA' (uniform isampler2DArray)
+0:23          Constant:
+0:23            0.100000
+0:23            0.100000
+0:23            0.100000
+0:23          Construct ivec2 (2-component vector of int)
+0:23            'i' (flat in int)
 0:?   Linker Objects
 0:?     'c2D' (smooth in 2-component vector of float)
 0:?     'i' (flat in int)
 0:?     'outp' (out 4-component vector of float)
 0:?     'arrayedSampler' (uniform 5-element array of sampler2D)
+0:?     'samp2dr' (uniform usampler2DRect)
+0:?     'isamp2DA' (uniform isampler2DArray)

--- a/Todo.txt
+++ b/Todo.txt
@ -115,10 +115,10 @@ Shader Functionality to Implement/Finish
         - bitCount(),findLSB(), andfindMSB()
      - New built-in to query LOD, textureQueryLod().
      - New overloaded function matching algorithm, handling selection from many valid multiple choices.
-      - Texture gather functions that return four texels with a single call.
-         - textureGather()
-         - textureGatherOffset()
-         - textureGatherOffsets()
+      + Texture gather functions that return four texels with a single call.
+         + textureGather()
+         + textureGatherOffset()
+         + textureGatherOffsets()
      + Add streams out from geometry shader. Output can be directed to streams through
         + EmitStreamVertex() and EndStreamPrimitive().
    GLSL 4.1
@ -168,8 +168,8 @@ Shader Functionality to Implement/Finish
      - Make GLSL consistent with the API regarding user clipping, by no longer referring to gl_Position when gl_ClipVertex is not written. Rather, user clipping becomes undefined.
      - Clarified that a comma sequence-operator expression cannot be a constant expression. E.g., “(2,3)” is not allowed, semantically, 
            as a valid constant expression 3, even though it is an expression that will evaluate to 3.
-      - Use vec2 instead of vec3 for coordinate in textureGather*(sampler2DRect,...).
-      - Clarify that textureGatherOffset() can take non-constants for the offsets.
+      + Use vec2 instead of vec3 for coordinate in textureGather*(sampler2DRect,...).
+      + Clarify that textureGatherOffset() can take non-constants for the offsets.
    GLSL 4.3
      - Add shader storage buffer objects, as per the ARB_shader_storage_buffer_object extension. This includes 
            - allowing the last member of a storage buffer block to be an array that does not know its size until render time
--- a/glslang/MachineIndependent/Initialize.cpp
+++ b/glslang/MachineIndependent/Initialize.cpp
@ -1397,6 +1397,11 @@ void TBuiltIns::initialize(int version, EProfile profile)
 //
 void TBuiltIns::add2ndGenerationSamplingImaging(int version, EProfile profile)
 {
+    //
+    // In this function proper, enumerate the types, then calls the next set of functions
+    // to enumerate all the uses for that type.
+    //
+
    TBasicType bTypes[3] = { EbtFloat, EbtInt, EbtUint };

    // enumerate all the types
@ -1455,8 +1460,11 @@ void TBuiltIns::add2ndGenerationSamplingImaging(int version, EProfile profile)

                            if (image)
                                addImageFunctions(sampler, typeName, version, profile);
-                            else
+                            else {
                                addSamplingFunctions(sampler, typeName, version, profile);
+                                if (version >= 130)
+                                    addGatherFunctions(sampler, typeName, version, profile);
+                            }
                        }
                    }
                }
@ -1469,6 +1477,8 @@ void TBuiltIns::add2ndGenerationSamplingImaging(int version, EProfile profile)
 // Helper function for add2ndGenerationSamplingImaging(), 
 // when adding context-independent built-in functions.
 //
+// Add all the query functions for the given type.
+//
 void TBuiltIns::addQueryFunctions(TSampler sampler, TString& typeName, int version, EProfile profile)
 {
    //
@ -1502,6 +1512,8 @@ void TBuiltIns::addQueryFunctions(TSampler sampler, TString& typeName, int versi
 // Helper function for add2ndGenerationSamplingImaging(), 
 // when adding context-independent built-in functions.
 //
+// Add all the image access functions for the given type.
+//
 void TBuiltIns::addImageFunctions(TSampler sampler, TString& typeName, int version, EProfile profile)
 {
    // TODO: 4.2 Functionality: imaging functions
@ -1511,6 +1523,8 @@ void TBuiltIns::addImageFunctions(TSampler sampler, TString& typeName, int versi
 // Helper function for add2ndGenerationSamplingImaging(), 
 // when adding context-independent built-in functions.
 //
+// Add all the texture lookup functions for the given type.
+//
 void TBuiltIns::addSamplingFunctions(TSampler sampler, TString& typeName, int version, EProfile profile)
 {
    // make one string per stage to contain all functions of the passed-in type for that stage
@ -1690,6 +1704,89 @@ void TBuiltIns::addSamplingFunctions(TSampler sampler, TString& typeName, int ve
    }
 }

+
+//
+// Helper function for add2ndGenerationSamplingImaging(), 
+// when adding context-independent built-in functions.
+//
+// Add all the texture gather functions for the given type.
+//
+void TBuiltIns::addGatherFunctions(TSampler sampler, TString& typeName, int version, EProfile profile)
+{
+    switch (sampler.dim) {
+    case Esd2D:
+    case EsdRect:
+    case EsdCube:
+        break;
+    default:
+        return;
+    }
+
+    if (sampler.ms)
+        return;
+
+    // make one string per stage to contain all functions of the passed-in type for that stage
+    TString functions[EShLangCount];
+
+    for (int offset = 0; offset < 3; ++offset) { // loop over three forms of offset in the call name:  none, Offset, and Offsets
+
+        for (int comp = 0; comp < 2; ++comp) { // loop over presence of comp argument
+
+            if (comp > 0 && sampler.shadow)
+                continue;
+
+            if (offset > 0 && sampler.dim == EsdCube)
+                continue;
+
+            TString s;
+
+            // return type
+            s.append(prefixes[sampler.type]);
+            s.append("vec4 ");
+
+            // name
+            s.append("textureGather");
+            switch (offset) {
+            case 1:
+                s.append("Offset");
+                break;
+            case 2:
+                s.append("Offsets");
+            default:
+                break;
+            }
+            s.append("(");
+
+            // sampler type argument
+            s.append(typeName);
+
+            // P coordinate argument
+            s.append(",vec");
+            int totalDims = dimMap[sampler.dim] + (sampler.arrayed ? 1 : 0);
+            s.append(postfixes[totalDims]);
+
+            // refZ argument
+            if (sampler.shadow)
+                s.append(",float");
+
+            // offset argument
+            if (offset > 0) {
+                s.append(",ivec2");
+                if (offset == 2)
+                    s.append("[4]");
+            }
+
+            // comp argument
+            if (comp)
+                s.append(",int");
+
+            s.append(");\n");
+            commonBuiltins.append(s);
+            //printf("%s", s.c_str());
+        }
+    }
+}
+
 //
 // Add context-dependent built-in functions and variables that are present
 // for the given version and profile.  Share common ones across stages, otherwise
--- a/glslang/MachineIndependent/Initialize.h
+++ b/glslang/MachineIndependent/Initialize.h
@ -69,6 +69,7 @@ protected:
    void addQueryFunctions(TSampler, TString& typeName, int version, EProfile profile);
    void addImageFunctions(TSampler, TString& typeName, int version, EProfile profile);
    void addSamplingFunctions(TSampler, TString& typeName, int version, EProfile profile);
+    void addGatherFunctions(TSampler, TString& typeName, int version, EProfile profile);

    TString commonBuiltins;
    TString stageBuiltins[EShLangCount];
--- a/glslang/MachineIndependent/ParseHelper.cpp
+++ b/glslang/MachineIndependent/ParseHelper.cpp
@ -867,10 +867,8 @@ TIntermTyped* TParseContext::handleFunctionCall(TSourceLoc loc, TFunction* fnCal
                    qualifierList.push_back(qual);
                }

-                // built-in texturing functions get their return value precision from the precision of the sampler
-                if (builtIn && fnCandidate->getType().getQualifier().precision == EpqNone &&
-                    fnCandidate->getParamCount() > 0 && (*fnCandidate)[0].type->getBasicType() == EbtSampler)
-                    result->getQualifier().precision = result->getAsAggregate()->getSequence()[0]->getAsTyped()->getQualifier().precision;
+                if (builtIn)
+                    nonOpBuiltInCheck(loc, *fnCandidate, result->getAsAggregate());
            }
        } else {
            // error message was put out by PaFindFunction()
@ -892,6 +890,36 @@ TIntermTyped* TParseContext::handleFunctionCall(TSourceLoc loc, TFunction* fnCal
    return result;
 }

+//
+// Do additional checking of built-in function calls that were not mapped
+// to built-in operations (e.g., texturing functions).
+//
+// Assumes there has been a semantically correct match to a built-in function.
+//
+void TParseContext::nonOpBuiltInCheck(TSourceLoc loc, const TFunction& fnCandidate, TIntermAggregate* callNode)
+{
+    // built-in texturing functions get their return value precision from the precision of the sampler
+    if (fnCandidate.getType().getQualifier().precision == EpqNone &&
+        fnCandidate.getParamCount() > 0 && fnCandidate[0].type->getBasicType() == EbtSampler)
+        callNode->getQualifier().precision = callNode->getAsAggregate()->getSequence()[0]->getAsTyped()->getQualifier().precision;
+
+    if (fnCandidate.getName().compare(0, 13, "textureGather") == 0) {
+        const char* feature = "texture gather function";
+        requireProfile(loc, ~EEsProfile, feature);
+        profileRequires(loc, ~EEsProfile, 400, GL_ARB_texture_gather, feature);
+        int lastArgIndex = fnCandidate.getParamCount() - 1;
+        if (fnCandidate[lastArgIndex].type->getBasicType() == EbtInt && fnCandidate[lastArgIndex].type->isScalar()) {
+            // the last integral argument to a texture gather must be a constant int between 0 and 3
+            if (callNode->getSequence()[lastArgIndex]->getAsConstantUnion()) {
+                int value = callNode->getSequence()[lastArgIndex]->getAsConstantUnion()->getConstArray()[0].getIConst();
+                if (value < 0 || value > 3)
+                    error(loc, "must be 0, 1, 2, or 3", "texture gather component", "");
+            } else
+                error(loc, "must be a constant", "texture gather component", "");
+        }
+    }
+}
+
 //
 // Handle seeing a built-in-type constructor call in the grammar.
 //
--- a/glslang/MachineIndependent/ParseHelper.h
+++ b/glslang/MachineIndependent/ParseHelper.h
@ -82,6 +82,7 @@ public:
    TFunction* handleFunctionDeclarator(TSourceLoc loc, TFunction& function);
    TIntermAggregate* handleFunctionPrototype(TSourceLoc, TFunction&);
    TIntermTyped* handleFunctionCall(TSourceLoc, TFunction*, TIntermNode*, TIntermAggregate*);
+    void nonOpBuiltInCheck(TSourceLoc, const TFunction&, TIntermAggregate*);
    TFunction* handleConstructorCall(TSourceLoc, TPublicType&);

    bool parseVectorFields(TSourceLoc, const TString&, int vecSize, TVectorFields&);
--- a/glslang/MachineIndependent/Versions.cpp
+++ b/glslang/MachineIndependent/Versions.cpp
@ -138,9 +138,10 @@ namespace glslang {
 //
 void TParseContext::initializeExtensionBehavior()
 {
-    extensionBehavior[GL_ARB_texture_rectangle] = EBhDisable;
-    extensionBehavior[GL_3DL_array_objects] = EBhDisable;
+    extensionBehavior[GL_ARB_texture_rectangle]        = EBhDisable;
+    extensionBehavior[GL_3DL_array_objects]            = EBhDisable;
    extensionBehavior[GL_ARB_shading_language_420pack] = EBhDisable;
+    extensionBehavior[GL_ARB_texture_gather]           = EBhDisable;
 }

 //
--- a/glslang/MachineIndependent/Versions.h
+++ b/glslang/MachineIndependent/Versions.h
@ -74,6 +74,7 @@ typedef enum {
 const char* const GL_ARB_texture_rectangle        = "GL_ARB_texture_rectangle";
 const char* const GL_3DL_array_objects            = "GL_3DL_array_objects";
 const char* const GL_ARB_shading_language_420pack = "GL_ARB_shading_language_420pack";
+const char* const GL_ARB_texture_gather           = "GL_ARB_texture_gather";

 } // end namespace glslang