Front-end: Implement GL_OES_gpu_shader5.

Test/310.frag

@@ -30,11 +30,11 @@ void main()
     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));
+    iv4 = textureGatherOffset(isamp2DA, vec3(0.1), ivec2(0.5));
+    iv4 = textureGatherOffset(isamp2DA, vec3(0.1), ivec2(i));     // ERROR, offset not constant
 }
 
 out vec4 outp;
-
 void foo23()
 {
     const ivec2[3] offsets = ivec2[3](ivec2(1,2), ivec2(3,4), ivec2(15,16));
@@ -193,3 +193,32 @@ void foo_GS()
     highp int l = gl_Layer;
     highp int p = gl_PrimitiveID;
 }
+
+in vec2 inf, ing;
+uniform ivec2 offsets[4];
+uniform sampler2D sArray[4];
+uniform int sIndex;
+layout(binding = 0) uniform atomic_uint auArray[2];
+uniform ubName { int i; } ubInst[4];
+buffer bbName { int i; } bbInst[4];
+uniform writeonly image2D iArray[5];
+const ivec2 constOffsets[4] = ivec2[4](ivec2(0.1), ivec2(0.2), ivec2(0.3), ivec2(0.4));
+
+void pfooBad()
+{
+    precise vec2 h;                                            // ERROR reserved
+    h = fma(inf, ing, h);                                      // ERROR, not available
+    textureGatherOffset(sArray[0], vec2(0.1), ivec2(inf));     // ERROR, offset not constant
+    textureGatherOffsets(sArray[0], vec2(0.1), constOffsets);  // ERROR, not available
+}
+
+#extension GL_OES_gpu_shader5 : enable
+
+void pfoo()
+{
+    precise vec2 h;
+    h = fma(inf, ing, h);
+    textureGatherOffset(sArray[0], vec2(0.1), ivec2(inf));
+    textureGatherOffsets(sArray[0], vec2(0.1), constOffsets);
+    textureGatherOffsets(sArray[0], vec2(0.1), offsets);       // ERROR, offset not constant
+}

Test/310.vert

@@ -139,3 +139,45 @@ centroid out cmo {              // ERROR, no centroid on a block
 invariant out imo {             // ERROR, no invariant on a block
     int i;
 } imon;
+
+in vec2 inf, ing;
+uniform ivec2 offsets[4];
+uniform sampler2D sArray[4];
+uniform int sIndex;
+layout(binding = 0) uniform atomic_uint auArray[2];
+uniform ubName { int i; } ubInst[4];
+buffer bbName { int i; } bbInst[4];
+uniform writeonly image2D iArray[5];
+const ivec2 constOffsets[4] = ivec2[4](ivec2(0.1), ivec2(0.2), ivec2(0.3), ivec2(0.4));
+
+void pfooBad()
+{
+    precise vec2 h;          // ERROR reserved
+    h = fma(inf, ing, h);    // ERROR, not available
+    sArray[sIndex + 1];      // ERRRO, not supported
+    auArray[sIndex + 1];
+    ubInst[1];
+    bbInst[2];
+    ubInst[sIndex + 1];      // ERRRO, not supported
+    bbInst[sIndex];          // ERRRO, not supported
+    iArray[2];
+    iArray[sIndex * 2];      // ERRRO, not supported
+    textureGatherOffset(sArray[0], vec2(0.1), ivec2(inf));     // ERROR, offset not constant
+    textureGatherOffsets(sArray[0], vec2(0.1), constOffsets);  // ERROR, not available
+}
+
+#extension GL_OES_gpu_shader5 : enable
+
+void pfoo()
+{
+    precise vec2 h;
+    h = fma(inf, ing, h);
+    sArray[sIndex + 1];
+    ubInst[sIndex + 1];
+    bbInst[sIndex - 2];      // ERROR, still not supported
+    iArray[2];
+    iArray[sIndex - 2];
+    textureGatherOffset(sArray[0], vec2(0.1), ivec2(inf));
+    textureGatherOffsets(sArray[0], vec2(0.1), constOffsets);
+    textureGatherOffsets(sArray[0], vec2(0.1), offsets);   // ERROR, offset not constant
+}

Test/baseResults/300.frag.out

@@ -10,7 +10,7 @@ ERROR: 0:63: 'texel offset' : argument must be compile-time constant
 ERROR: 0:64: 'texel offset' : argument must be compile-time constant 
 ERROR: 0:66: 'texel offset' : argument must be compile-time constant 
 ERROR: 0:68: 'texel offset' : argument must be compile-time constant 
-ERROR: 0:69: 'variable indexing sampler array' : not supported with this profile: es
+ERROR: 0:69: 'variable indexing sampler array' : not supported for this version or the enabled extensions 
 ERROR: 0:73: 'texel offset' : argument must be compile-time constant 
 ERROR: 0:83: 'double' : Reserved word. 
 ERROR: 0:83: 'double' : not supported with this profile: es

Test/baseResults/300block.frag.out

@@ -2,7 +2,7 @@
 ERROR: 0:10: '' : cannot nest a structure definition inside a structure or block 
 ERROR: 0:21: '' : cannot nest a structure definition inside a structure or block 
 ERROR: 0:20: 'sampler' : member of block cannot be a sampler type 
-ERROR: 0:45: 'variable indexing block array' : not supported with this profile: es
+ERROR: 0:45: 'variable indexing uniform block array' : not supported for this version or the enabled extensions 
 ERROR: 4 compilation errors.  No code generated.
 
 

Test/baseResults/310.tesc.out

@@ -31,7 +31,9 @@ ERROR: 0:80: '' : array size required
 ERROR: 0:86: 'location' : overlapping use of location 4
 ERROR: 0:90: 'location' : overlapping use of location 4
 ERROR: 0:94: 'precise' : Reserved word. 
-ERROR: 0:95: 'fma' : no matching overloaded function found 
+ERROR: 0:95: 'fma' : required extension not requested: Possible extensions include:
+GL_EXT_gpu_shader5
+GL_OES_gpu_shader5
 ERROR: 0:104: 'sample' : Reserved word. 
 ERROR: 0:106: 'vertices' : can only apply to a standalone qualifier 
 ERROR: 0:107: 'vertices' : inconsistent output number of vertices for array size of misSized
@@ -188,8 +190,10 @@ ERROR: node is still EOpNull!
 0:?     Sequence
 0:95      move second child to first child (temp highp float)
 0:95        'd' (temp highp float)
-0:95        Constant:
-0:95          0.000000
+0:95        Function Call: fma(f1;f1;f1; (global highp float)
+0:95          'd' (temp highp float)
+0:95          'd' (temp highp float)
+0:95          'd' (temp highp float)
 0:112  Function Definition: pointSize2( (global void)
 0:112    Function Parameters: 
 0:114    Sequence
@@ -390,8 +394,10 @@ ERROR: node is still EOpNull!
 0:?     Sequence
 0:95      move second child to first child (temp highp float)
 0:95        'd' (temp highp float)
-0:95        Constant:
-0:95          0.000000
+0:95        Function Call: fma(f1;f1;f1; (global highp float)
+0:95          'd' (temp highp float)
+0:95          'd' (temp highp float)
+0:95          'd' (temp highp float)
 0:112  Function Definition: pointSize2( (global void)
 0:112    Function Parameters: 
 0:114    Sequence

Test/baseResults/preprocessor.extensions.vert.err

 Warning, version 310 is not yet complete; most version-specific features are present, but some are missing.
-WARNING: 0:5: '#extension' : extension is only partially supported: GL_EXT_gpu_shader5
 WARNING: 0:6: '#extension' : extension not supported: GL_EXT_shader_texture_image_samples
 

Test/baseResults/preprocessor.line.frag.err

-Warning, version 310 is not yet complete; most version-specific features are present, but some are missing.
-
+Warning, version 310 is not yet complete; most version-specific features are present, but some are missing.
+

Test/baseResults/preprocessor.line.frag.out

-#version 310 es
-#line 1 2
-#pragma something
-void main(){ }
-
+#version 310 es
+#line 1 2
+#pragma something
+void main(){ }
+

Test/baseResults/preprocessor.line.vert.out

-#line 300
-
-#line 2
-
-
-
-
-
-#line 10
-
-
-#line 2
-
-#line 0
-
-
-
-#line 4
-
-
-
-
-
-#line 8
-
-void main(){
-  gl_Position = vec4(10);
-}
-
-#line 8 4
-
-
-
-
-#line 12 3
-
-#line 1
-
-
+#line 300
+
+#line 2
+
+
+
+
+
+#line 10
+
+
+#line 2
+
+#line 0
+
+
+
+#line 4
+
+
+
+
+
+#line 8
+
+void main(){
+  gl_Position = vec4(10);
+}
+
+#line 8 4
+
+
+
+
+#line 12 3
+
+#line 1
+
+

glslang/MachineIndependent/Initialize.cpp

@@ -493,19 +493,24 @@ void TBuiltIns::initialize(int version, EProfile profile)
             "\n");
     }
 
-    if (profile != EEsProfile && version >= 400) {
+    if ((profile != EEsProfile && version >= 400) ||
+        (profile == EEsProfile && version >= 310)) {    // GL_OES_gpu_shader5
+
         commonBuiltins.append(
             "float  fma(float,  float,  float );"
             "vec2   fma(vec2,   vec2,   vec2  );"
             "vec3   fma(vec3,   vec3,   vec3  );"
             "vec4   fma(vec4,   vec4,   vec4  );"
-
-            "double fma(double, double, double);"
-            "dvec2  fma(dvec2,  dvec2,  dvec2 );"
-            "dvec3  fma(dvec3,  dvec3,  dvec3 );"
-            "dvec4  fma(dvec4,  dvec4,  dvec4 );"
-            
             "\n");
+
+        if (profile != EEsProfile) {
+            commonBuiltins.append(
+                "double fma(double, double, double);"
+                "dvec2  fma(dvec2,  dvec2,  dvec2 );"
+                "dvec3  fma(dvec3,  dvec3,  dvec3 );"
+                "dvec4  fma(dvec4,  dvec4,  dvec4 );"
+                "\n");
+        }
     }
 
     if ((profile == EEsProfile && version >= 310) ||
@@ -2221,9 +2226,6 @@ void TBuiltIns::addGatherFunctions(TSampler sampler, TString& typeName, int vers
 
     for (int offset = 0; offset < 3; ++offset) { // loop over three forms of offset in the call name:  none, Offset, and Offsets
 
-        if (profile == EEsProfile && offset == 2)
-            continue;
-
         for (int comp = 0; comp < 2; ++comp) { // loop over presence of comp argument
 
             if (comp > 0 && sampler.shadow)
@@ -2802,8 +2804,11 @@ void IdentifyBuiltIns(int version, EProfile profile, EShLanguage language, TSymb
     //
 
     // N.B.: a symbol should only be tagged once, and this function is called multiple times, once
-    // per stage that's used for this profile.  So, stick common ones in the fragment stage to
-    // ensure they are tagged exactly once.
+    // per stage that's used for this profile.  So
+    //  - generally, stick common ones in the fragment stage to ensure they are tagged exactly once
+    //  - for ES, which has different precisions for different stages, the coarsest-grained tagging 
+    //    for a built-in used in many stages needs to be once for the fragment stage and once for
+    //    the vertex stage
 
     switch(language) {
     case EShLangVertex:
@@ -2822,6 +2827,15 @@ void IdentifyBuiltIns(int version, EProfile profile, EShLanguage language, TSymb
         BuiltInVariable("gl_MultiTexCoord7", EbvMultiTexCoord7, symbolTable);
         BuiltInVariable("gl_FogCoord",       EbvFogFragCoord,   symbolTable);
 
+        if (profile == EEsProfile) {
+            symbolTable.setFunctionExtensions("texture2DGradEXT",     1, &E_GL_EXT_shader_texture_lod);
+            symbolTable.setFunctionExtensions("texture2DProjGradEXT", 1, &E_GL_EXT_shader_texture_lod);
+            symbolTable.setFunctionExtensions("textureCubeGradEXT",   1, &E_GL_EXT_shader_texture_lod);
+            symbolTable.setFunctionExtensions("textureGatherOffsets", Num_AEP_gpu_shader5, AEP_gpu_shader5);
+            if (version >= 310)
+                symbolTable.setFunctionExtensions("fma", Num_AEP_gpu_shader5, AEP_gpu_shader5);
+        }
+
         // Fall through
 
     case EShLangTessControl:
@@ -2931,15 +2945,21 @@ void IdentifyBuiltIns(int version, EProfile profile, EShLanguage language, TSymb
 
         // built-in functions
 
-        if (version == 100) {
-            symbolTable.setFunctionExtensions("dFdx",   1, &E_GL_OES_standard_derivatives);
-            symbolTable.setFunctionExtensions("dFdy",   1, &E_GL_OES_standard_derivatives);
-            symbolTable.setFunctionExtensions("fwidth", 1, &E_GL_OES_standard_derivatives);
-        }
         if (profile == EEsProfile) {
-            symbolTable.setFunctionExtensions("texture2DLodEXT",     1, &E_GL_EXT_shader_texture_lod);
-            symbolTable.setFunctionExtensions("texture2DProjLodEXT", 1, &E_GL_EXT_shader_texture_lod);
-            symbolTable.setFunctionExtensions("textureCubeLodEXT",   1, &E_GL_EXT_shader_texture_lod);
+            symbolTable.setFunctionExtensions("texture2DLodEXT",      1, &E_GL_EXT_shader_texture_lod);
+            symbolTable.setFunctionExtensions("texture2DProjLodEXT",  1, &E_GL_EXT_shader_texture_lod);
+            symbolTable.setFunctionExtensions("textureCubeLodEXT",    1, &E_GL_EXT_shader_texture_lod);
+            symbolTable.setFunctionExtensions("texture2DGradEXT",     1, &E_GL_EXT_shader_texture_lod);
+            symbolTable.setFunctionExtensions("texture2DProjGradEXT", 1, &E_GL_EXT_shader_texture_lod);
+            symbolTable.setFunctionExtensions("textureCubeGradEXT",   1, &E_GL_EXT_shader_texture_lod);
+            symbolTable.setFunctionExtensions("textureGatherOffsets", Num_AEP_gpu_shader5, AEP_gpu_shader5);
+            if (version == 100) {
+                symbolTable.setFunctionExtensions("dFdx",   1, &E_GL_OES_standard_derivatives);
+                symbolTable.setFunctionExtensions("dFdy",   1, &E_GL_OES_standard_derivatives);
+                symbolTable.setFunctionExtensions("fwidth", 1, &E_GL_OES_standard_derivatives);
+            }
+            if (version >= 310)
+                symbolTable.setFunctionExtensions("fma", Num_AEP_gpu_shader5, AEP_gpu_shader5);
         } else if (version < 130) {
             symbolTable.setFunctionExtensions("texture1DLod",        1, &E_GL_ARB_shader_texture_lod);
             symbolTable.setFunctionExtensions("texture2DLod",        1, &E_GL_ARB_shader_texture_lod);
@@ -2973,12 +2993,6 @@ void IdentifyBuiltIns(int version, EProfile profile, EShLanguage language, TSymb
             symbolTable.setFunctionExtensions("shadow2DRectProjGradARB",  1, &E_GL_ARB_shader_texture_lod);
         }
 
-        if (profile == EEsProfile) {
-            symbolTable.setFunctionExtensions("texture2DGradEXT",     1, &E_GL_EXT_shader_texture_lod);
-            symbolTable.setFunctionExtensions("texture2DProjGradEXT", 1, &E_GL_EXT_shader_texture_lod);
-            symbolTable.setFunctionExtensions("textureCubeGradEXT",   1, &E_GL_EXT_shader_texture_lod);
-        }
-
         // E_GL_ARB_shader_image_load_store
         if (profile != EEsProfile && version < 420)
             symbolTable.setFunctionExtensions("memoryBarrier", 1, &E_GL_ARB_shader_image_load_store);

glslang/MachineIndependent/ParseHelper.cpp

@@ -505,13 +505,19 @@ TIntermTyped* TParseContext::handleBracketDereference(TSourceLoc loc, TIntermTyp
                 else
                     error(loc, "", "[", "array must be redeclared with a size before being indexed with a variable");
             }
-            if (base->getBasicType() == EbtBlock)
-                requireProfile(base->getLoc(), ~EEsProfile, "variable indexing block array");
-            else if (language == EShLangFragment && base->getQualifier().isPipeOutput())
+            if (base->getBasicType() == EbtBlock) {
+                if (base->getQualifier().storage == EvqBuffer)
+                    requireProfile(base->getLoc(), ~EEsProfile, "variable indexing buffer block array");
+                else if (base->getQualifier().storage == EvqUniform)
+                    profileRequires(base->getLoc(), EEsProfile, 0, Num_AEP_gpu_shader5, AEP_gpu_shader5, "variable indexing uniform block array");
+                else
+                    requireProfile(base->getLoc(), ~EEsProfile, "variable indexing in/out block array");
+            } else if (language == EShLangFragment && base->getQualifier().isPipeOutput())
                 requireProfile(base->getLoc(), ~EEsProfile, "variable indexing fragment shader ouput array");
             else if (base->getBasicType() == EbtSampler && version >= 130) {
                 const char* explanation = "variable indexing sampler array";
-                requireProfile(base->getLoc(), ECoreProfile | ECompatibilityProfile, explanation);
+                requireProfile(base->getLoc(), EEsProfile | ECoreProfile | ECompatibilityProfile, explanation);
+                profileRequires(base->getLoc(), EEsProfile, 0, Num_AEP_gpu_shader5, AEP_gpu_shader5, explanation);
                 profileRequires(base->getLoc(), ECoreProfile | ECompatibilityProfile, 400, nullptr, explanation);
             }
 
@@ -1332,6 +1338,9 @@ void TParseContext::nonOpBuiltInCheck(TSourceLoc loc, const TFunction& fnCandida
                     profileRequires(loc, ~EEsProfile, 400, E_GL_ARB_texture_gather, feature);
                 else
                     profileRequires(loc, ~EEsProfile, 400, E_GL_ARB_gpu_shader5, feature);
+                int offsetArg = fnCandidate[0].type->getSampler().shadow ? 3 : 2;
+                if (! callNode.getSequence()[offsetArg]->getAsConstantUnion())
+                    profileRequires(loc, EEsProfile, 0, Num_AEP_gpu_shader5, AEP_gpu_shader5, "non-constant offset argument");
                 if (! fnCandidate[0].type->getSampler().shadow)
                     compArg = 3;
             } else if (fnCandidate.getName().compare("textureGatherOffsets") == 0) {

glslang/MachineIndependent/Scan.cpp

@@ -941,12 +941,14 @@ int TScanContext::tokenizeIdentifier()
         return keyword;
 
     case PRECISE:
-        if (parseContext.profile == EEsProfile && parseContext.version >= 310)
+        if ((parseContext.profile == EEsProfile && parseContext.extensionsTurnedOn(Num_AEP_gpu_shader5, AEP_gpu_shader5)) || 
+            (parseContext.profile != EEsProfile && parseContext.version >= 400))
+            return keyword;
+        if (parseContext.profile == EEsProfile && parseContext.version == 310) {
             reservedWord();
-        else if (parseContext.profile == EEsProfile ||
-            (parseContext.profile != EEsProfile && parseContext.version < 400))
-            return identifierOrType();
-        return keyword;
+            return keyword;
+        }
+        return identifierOrType();
 
     case INVARIANT:
         if (parseContext.profile != EEsProfile && parseContext.version < 120)

glslang/MachineIndependent/SymbolTable.h

@@ -200,7 +200,7 @@ public:
         TSymbol(name),
         mangledName(*name + '('),
         op(tOp),
-        defined(false), prototyped(false) { returnType.shallowCopy(retType); }    
+        defined(false), prototyped(false) { returnType.shallowCopy(retType); }
     virtual TFunction* clone() const;
     virtual ~TFunction();
 
@@ -486,8 +486,8 @@ public:
     //
 protected:
     static const int globalLevel = 3;
-    bool isSharedLevel(int level)  { return level <= 1; }    // exclude all per-compile levels
-    bool isBuiltInLevel(int level) { return level <= 2; }    // exclude user globals
+    bool isSharedLevel(int level)  { return level <= 1; }              // exclude all per-compile levels
+    bool isBuiltInLevel(int level) { return level <= 2; }              // exclude user globals
     bool isGlobalLevel(int level)  { return level <= globalLevel; }    // include user globals
 public:
     bool isEmpty() { return table.size() == 0; }

glslang/MachineIndependent/Versions.cpp

@@ -183,7 +183,7 @@ void TParseContext::initializeExtensionBehavior()
     extensionBehavior[E_GL_OES_texture_storage_multisample_2d_array] = EBhDisablePartial;
     extensionBehavior[E_GL_EXT_geometry_shader]                      = EBhDisable;
     extensionBehavior[E_GL_EXT_geometry_point_size]                  = EBhDisable;
-    extensionBehavior[E_GL_EXT_gpu_shader5]                          = EBhDisablePartial;
+    extensionBehavior[E_GL_EXT_gpu_shader5]                          = EBhDisable;
     extensionBehavior[E_GL_EXT_primitive_bounding_box]               = EBhDisablePartial;
     extensionBehavior[E_GL_EXT_shader_io_blocks]                     = EBhDisable;
     extensionBehavior[E_GL_EXT_tessellation_shader]                  = EBhDisable;
@@ -194,7 +194,7 @@ void TParseContext::initializeExtensionBehavior()
     // OES matching AEP
     extensionBehavior[E_GL_OES_geometry_shader]          = EBhDisable;
     extensionBehavior[E_GL_OES_geometry_point_size]      = EBhDisable;
-    extensionBehavior[E_GL_OES_gpu_shader5]              = EBhDisablePartial;
+    extensionBehavior[E_GL_OES_gpu_shader5]              = EBhDisable;
     extensionBehavior[E_GL_OES_primitive_bounding_box]   = EBhDisablePartial;
     extensionBehavior[E_GL_OES_shader_io_blocks]         = EBhDisable;
     extensionBehavior[E_GL_OES_tessellation_shader]      = EBhDisable;