Merge pull request #1990 from ianromanick/INTEL_shader_integer_functions2

SPIRV/GlslangToSpv.cpp

@@ -2608,6 +2608,17 @@ bool TGlslangToSpvTraverser::visitAggregate(glslang::TVisit visit, glslang::TInt
         atomic = true;
         break;
 
+    case glslang::EOpAbsDifference:
+    case glslang::EOpAddSaturate:
+    case glslang::EOpSubSaturate:
+    case glslang::EOpAverage:
+    case glslang::EOpAverageRounded:
+    case glslang::EOpMul32x16:
+        builder.addCapability(spv::CapabilityIntegerFunctions2INTEL);
+        builder.addExtension("SPV_INTEL_shader_integer_functions2");
+        binOp = node->getOp();
+        break;
+
     case glslang::EOpIgnoreIntersectionNV:
     case glslang::EOpTerminateRayNV:
     case glslang::EOpTraceNV:
@@ -5231,6 +5242,30 @@ spv::Id TGlslangToSpvTraverser::createBinaryOperation(glslang::TOperator op, OpD
         binOp = spv::OpLogicalNotEqual;
         break;
 
+    case glslang::EOpAbsDifference:
+        binOp = isUnsigned ? spv::OpAbsUSubINTEL : spv::OpAbsISubINTEL;
+        break;
+
+    case glslang::EOpAddSaturate:
+        binOp = isUnsigned ? spv::OpUAddSatINTEL : spv::OpIAddSatINTEL;
+        break;
+
+    case glslang::EOpSubSaturate:
+        binOp = isUnsigned ? spv::OpUSubSatINTEL : spv::OpISubSatINTEL;
+        break;
+
+    case glslang::EOpAverage:
+        binOp = isUnsigned ? spv::OpUAverageINTEL : spv::OpIAverageINTEL;
+        break;
+
+    case glslang::EOpAverageRounded:
+        binOp = isUnsigned ? spv::OpUAverageRoundedINTEL : spv::OpIAverageRoundedINTEL;
+        break;
+
+    case glslang::EOpMul32x16:
+        binOp = isUnsigned ? spv::OpUMul32x16INTEL : spv::OpIMul32x16INTEL;
+        break;
+
     case glslang::EOpLessThan:
     case glslang::EOpGreaterThan:
     case glslang::EOpLessThanEqual:
@@ -5745,6 +5780,18 @@ spv::Id TGlslangToSpvTraverser::createUnaryOperation(glslang::TOperator op, OpDe
             libCall = spv::GLSLstd450FindSMsb;
         break;
 
+    case glslang::EOpCountLeadingZeros:
+        builder.addCapability(spv::CapabilityIntegerFunctions2INTEL);
+        builder.addExtension("SPV_INTEL_shader_integer_functions2");
+        unaryOp = spv::OpUCountLeadingZerosINTEL;
+        break;
+
+    case glslang::EOpCountTrailingZeros:
+        builder.addCapability(spv::CapabilityIntegerFunctions2INTEL);
+        builder.addExtension("SPV_INTEL_shader_integer_functions2");
+        unaryOp = spv::OpUCountTrailingZerosINTEL;
+        break;
+
     case glslang::EOpBallot:
     case glslang::EOpReadFirstInvocation:
     case glslang::EOpAnyInvocation:

SPIRV/doc.cpp

@@ -931,6 +931,8 @@ const char* CapabilityString(int info)
     case CapabilityDemoteToHelperInvocationEXT:             return "DemoteToHelperInvocationEXT";
     case CapabilityShaderClockKHR:                          return "ShaderClockKHR";
 
+    case CapabilityIntegerFunctions2INTEL:              return "CapabilityIntegerFunctions2INTEL";
+
     default: return "Bad";
     }
 }

Test/baseResults/spv.specConstant.vert.out

@@ -11,7 +11,7 @@ spv.specConstant.vert
                               Source GLSL 400
                               Name 4  "main"
                               Name 9  "arraySize"
-                              Name 14  "foo(vf4[s4393];"
+                              Name 14  "foo(vf4[s4529];"
                               Name 13  "p"
                               Name 17  "builtin_spec_constant("
                               Name 20  "color"
@@ -102,10 +102,10 @@ spv.specConstant.vert
                               Store 20(color) 46
               48:          10 Load 22(ucol)
                               Store 47(param) 48
-              49:           2 FunctionCall 14(foo(vf4[s4393];) 47(param)
+              49:           2 FunctionCall 14(foo(vf4[s4529];) 47(param)
                               Return
                               FunctionEnd
-14(foo(vf4[s4393];):           2 Function None 12
+14(foo(vf4[s4529];):           2 Function None 12
            13(p):     11(ptr) FunctionParameter
               15:             Label
               54:     24(ptr) AccessChain 53(dupUcol) 23

glslang/Include/intermediate.h

@@ -899,6 +899,15 @@ enum TOperator {
     EOpFindLSB,
     EOpFindMSB,
 
+    EOpCountLeadingZeros,
+    EOpCountTrailingZeros,
+    EOpAbsDifference,
+    EOpAddSaturate,
+    EOpSubSaturate,
+    EOpAverage,
+    EOpAverageRounded,
+    EOpMul32x16,
+
     EOpTraceNV,
     EOpReportIntersectionNV,
     EOpIgnoreIntersectionNV,

glslang/MachineIndependent/Initialize.cpp

@@ -2854,6 +2854,181 @@ void TBuiltIns::initialize(int version, EProfile profile, const SpvVersion& spvV
             "\n");
     }
 
+    if ((profile != EEsProfile && version >= 130) ||
+        (profile == EEsProfile && version >= 300)) {
+        commonBuiltins.append(
+            "uint countLeadingZeros(uint);"
+            "uvec2 countLeadingZeros(uvec2);"
+            "uvec3 countLeadingZeros(uvec3);"
+            "uvec4 countLeadingZeros(uvec4);"
+
+            "uint countTrailingZeros(uint);"
+            "uvec2 countTrailingZeros(uvec2);"
+            "uvec3 countTrailingZeros(uvec3);"
+            "uvec4 countTrailingZeros(uvec4);"
+
+            "uint absoluteDifference(int, int);"
+            "uvec2 absoluteDifference(ivec2, ivec2);"
+            "uvec3 absoluteDifference(ivec3, ivec3);"
+            "uvec4 absoluteDifference(ivec4, ivec4);"
+
+            "uint16_t absoluteDifference(int16_t, int16_t);"
+            "u16vec2 absoluteDifference(i16vec2, i16vec2);"
+            "u16vec3 absoluteDifference(i16vec3, i16vec3);"
+            "u16vec4 absoluteDifference(i16vec4, i16vec4);"
+
+            "uint64_t absoluteDifference(int64_t, int64_t);"
+            "u64vec2 absoluteDifference(i64vec2, i64vec2);"
+            "u64vec3 absoluteDifference(i64vec3, i64vec3);"
+            "u64vec4 absoluteDifference(i64vec4, i64vec4);"
+
+            "uint absoluteDifference(uint, uint);"
+            "uvec2 absoluteDifference(uvec2, uvec2);"
+            "uvec3 absoluteDifference(uvec3, uvec3);"
+            "uvec4 absoluteDifference(uvec4, uvec4);"
+
+            "uint16_t absoluteDifference(uint16_t, uint16_t);"
+            "u16vec2 absoluteDifference(u16vec2, u16vec2);"
+            "u16vec3 absoluteDifference(u16vec3, u16vec3);"
+            "u16vec4 absoluteDifference(u16vec4, u16vec4);"
+
+            "uint64_t absoluteDifference(uint64_t, uint64_t);"
+            "u64vec2 absoluteDifference(u64vec2, u64vec2);"
+            "u64vec3 absoluteDifference(u64vec3, u64vec3);"
+            "u64vec4 absoluteDifference(u64vec4, u64vec4);"
+
+            "int addSaturate(int, int);"
+            "ivec2 addSaturate(ivec2, ivec2);"
+            "ivec3 addSaturate(ivec3, ivec3);"
+            "ivec4 addSaturate(ivec4, ivec4);"
+
+            "int16_t addSaturate(int16_t, int16_t);"
+            "i16vec2 addSaturate(i16vec2, i16vec2);"
+            "i16vec3 addSaturate(i16vec3, i16vec3);"
+            "i16vec4 addSaturate(i16vec4, i16vec4);"
+
+            "int64_t addSaturate(int64_t, int64_t);"
+            "i64vec2 addSaturate(i64vec2, i64vec2);"
+            "i64vec3 addSaturate(i64vec3, i64vec3);"
+            "i64vec4 addSaturate(i64vec4, i64vec4);"
+
+            "uint addSaturate(uint, uint);"
+            "uvec2 addSaturate(uvec2, uvec2);"
+            "uvec3 addSaturate(uvec3, uvec3);"
+            "uvec4 addSaturate(uvec4, uvec4);"
+
+            "uint16_t addSaturate(uint16_t, uint16_t);"
+            "u16vec2 addSaturate(u16vec2, u16vec2);"
+            "u16vec3 addSaturate(u16vec3, u16vec3);"
+            "u16vec4 addSaturate(u16vec4, u16vec4);"
+
+            "uint64_t addSaturate(uint64_t, uint64_t);"
+            "u64vec2 addSaturate(u64vec2, u64vec2);"
+            "u64vec3 addSaturate(u64vec3, u64vec3);"
+            "u64vec4 addSaturate(u64vec4, u64vec4);"
+
+            "int subtractSaturate(int, int);"
+            "ivec2 subtractSaturate(ivec2, ivec2);"
+            "ivec3 subtractSaturate(ivec3, ivec3);"
+            "ivec4 subtractSaturate(ivec4, ivec4);"
+
+            "int16_t subtractSaturate(int16_t, int16_t);"
+            "i16vec2 subtractSaturate(i16vec2, i16vec2);"
+            "i16vec3 subtractSaturate(i16vec3, i16vec3);"
+            "i16vec4 subtractSaturate(i16vec4, i16vec4);"
+
+            "int64_t subtractSaturate(int64_t, int64_t);"
+            "i64vec2 subtractSaturate(i64vec2, i64vec2);"
+            "i64vec3 subtractSaturate(i64vec3, i64vec3);"
+            "i64vec4 subtractSaturate(i64vec4, i64vec4);"
+
+            "uint subtractSaturate(uint, uint);"
+            "uvec2 subtractSaturate(uvec2, uvec2);"
+            "uvec3 subtractSaturate(uvec3, uvec3);"
+            "uvec4 subtractSaturate(uvec4, uvec4);"
+
+            "uint16_t subtractSaturate(uint16_t, uint16_t);"
+            "u16vec2 subtractSaturate(u16vec2, u16vec2);"
+            "u16vec3 subtractSaturate(u16vec3, u16vec3);"
+            "u16vec4 subtractSaturate(u16vec4, u16vec4);"
+
+            "uint64_t subtractSaturate(uint64_t, uint64_t);"
+            "u64vec2 subtractSaturate(u64vec2, u64vec2);"
+            "u64vec3 subtractSaturate(u64vec3, u64vec3);"
+            "u64vec4 subtractSaturate(u64vec4, u64vec4);"
+
+            "int average(int, int);"
+            "ivec2 average(ivec2, ivec2);"
+            "ivec3 average(ivec3, ivec3);"
+            "ivec4 average(ivec4, ivec4);"
+
+            "int16_t average(int16_t, int16_t);"
+            "i16vec2 average(i16vec2, i16vec2);"
+            "i16vec3 average(i16vec3, i16vec3);"
+            "i16vec4 average(i16vec4, i16vec4);"
+
+            "int64_t average(int64_t, int64_t);"
+            "i64vec2 average(i64vec2, i64vec2);"
+            "i64vec3 average(i64vec3, i64vec3);"
+            "i64vec4 average(i64vec4, i64vec4);"
+
+            "uint average(uint, uint);"
+            "uvec2 average(uvec2, uvec2);"
+            "uvec3 average(uvec3, uvec3);"
+            "uvec4 average(uvec4, uvec4);"
+
+            "uint16_t average(uint16_t, uint16_t);"
+            "u16vec2 average(u16vec2, u16vec2);"
+            "u16vec3 average(u16vec3, u16vec3);"
+            "u16vec4 average(u16vec4, u16vec4);"
+
+            "uint64_t average(uint64_t, uint64_t);"
+            "u64vec2 average(u64vec2, u64vec2);"
+            "u64vec3 average(u64vec3, u64vec3);"
+            "u64vec4 average(u64vec4, u64vec4);"
+
+            "int averageRounded(int, int);"
+            "ivec2 averageRounded(ivec2, ivec2);"
+            "ivec3 averageRounded(ivec3, ivec3);"
+            "ivec4 averageRounded(ivec4, ivec4);"
+
+            "int16_t averageRounded(int16_t, int16_t);"
+            "i16vec2 averageRounded(i16vec2, i16vec2);"
+            "i16vec3 averageRounded(i16vec3, i16vec3);"
+            "i16vec4 averageRounded(i16vec4, i16vec4);"
+
+            "int64_t averageRounded(int64_t, int64_t);"
+            "i64vec2 averageRounded(i64vec2, i64vec2);"
+            "i64vec3 averageRounded(i64vec3, i64vec3);"
+            "i64vec4 averageRounded(i64vec4, i64vec4);"
+
+            "uint averageRounded(uint, uint);"
+            "uvec2 averageRounded(uvec2, uvec2);"
+            "uvec3 averageRounded(uvec3, uvec3);"
+            "uvec4 averageRounded(uvec4, uvec4);"
+
+            "uint16_t averageRounded(uint16_t, uint16_t);"
+            "u16vec2 averageRounded(u16vec2, u16vec2);"
+            "u16vec3 averageRounded(u16vec3, u16vec3);"
+            "u16vec4 averageRounded(u16vec4, u16vec4);"
+
+            "uint64_t averageRounded(uint64_t, uint64_t);"
+            "u64vec2 averageRounded(u64vec2, u64vec2);"
+            "u64vec3 averageRounded(u64vec3, u64vec3);"
+            "u64vec4 averageRounded(u64vec4, u64vec4);"
+
+            "int multiply32x16(int, int);"
+            "ivec2 multiply32x16(ivec2, ivec2);"
+            "ivec3 multiply32x16(ivec3, ivec3);"
+            "ivec4 multiply32x16(ivec4, ivec4);"
+
+            "uint multiply32x16(uint, uint);"
+            "uvec2 multiply32x16(uvec2, uvec2);"
+            "uvec3 multiply32x16(uvec3, uvec3);"
+            "uvec4 multiply32x16(uvec4, uvec4);"
+            "\n");
+    }
+
     if ((profile != EEsProfile && version >= 450) ||
         (profile == EEsProfile && version >= 320)) {
         commonBuiltins.append(
@@ -6974,6 +7149,15 @@ void TBuiltIns::identifyBuiltIns(int version, EProfile profile, const SpvVersion
             symbolTable.setFunctionExtensions("fragmentFetchAMD",     1, &E_GL_AMD_shader_fragment_mask);
         }
 
+        symbolTable.setFunctionExtensions("countLeadingZeros",  1, &E_GL_INTEL_shader_integer_functions2);
+        symbolTable.setFunctionExtensions("countTrailingZeros", 1, &E_GL_INTEL_shader_integer_functions2);
+        symbolTable.setFunctionExtensions("absoluteDifference", 1, &E_GL_INTEL_shader_integer_functions2);
+        symbolTable.setFunctionExtensions("addSaturate",        1, &E_GL_INTEL_shader_integer_functions2);
+        symbolTable.setFunctionExtensions("subtractSaturate",   1, &E_GL_INTEL_shader_integer_functions2);
+        symbolTable.setFunctionExtensions("average",            1, &E_GL_INTEL_shader_integer_functions2);
+        symbolTable.setFunctionExtensions("averageRounded",     1, &E_GL_INTEL_shader_integer_functions2);
+        symbolTable.setFunctionExtensions("multiply32x16",      1, &E_GL_INTEL_shader_integer_functions2);
+
         symbolTable.setFunctionExtensions("textureFootprintNV",          1, &E_GL_NV_shader_texture_footprint);
         symbolTable.setFunctionExtensions("textureFootprintClampNV",     1, &E_GL_NV_shader_texture_footprint);
         symbolTable.setFunctionExtensions("textureFootprintLodNV",       1, &E_GL_NV_shader_texture_footprint);
@@ -8217,6 +8401,15 @@ void TBuiltIns::identifyBuiltIns(int version, EProfile profile, const SpvVersion
 
     symbolTable.relateToOperator("helperInvocationEXT",  EOpIsHelperInvocation);
 
+    symbolTable.relateToOperator("countLeadingZeros",  EOpCountLeadingZeros);
+    symbolTable.relateToOperator("countTrailingZeros", EOpCountTrailingZeros);
+    symbolTable.relateToOperator("absoluteDifference", EOpAbsDifference);
+    symbolTable.relateToOperator("addSaturate",        EOpAddSaturate);
+    symbolTable.relateToOperator("subtractSaturate",   EOpSubSaturate);
+    symbolTable.relateToOperator("average",            EOpAverage);
+    symbolTable.relateToOperator("averageRounded",     EOpAverageRounded);
+    symbolTable.relateToOperator("multiply32x16",      EOpMul32x16);
+
     if (PureOperatorBuiltins) {
         symbolTable.relateToOperator("imageSize",               EOpImageQuerySize);
         symbolTable.relateToOperator("imageSamples",            EOpImageQuerySamples);

glslang/MachineIndependent/Versions.cpp

@@ -236,6 +236,8 @@ void TParseVersions::initializeExtensionBehavior()
     extensionBehavior[E_GL_AMD_shader_fragment_mask]                 = EBhDisable;
     extensionBehavior[E_GL_AMD_gpu_shader_half_float_fetch]          = EBhDisable;
 
+    extensionBehavior[E_GL_INTEL_shader_integer_functions2]          = EBhDisable;
+
     extensionBehavior[E_GL_NV_sample_mask_override_coverage]         = EBhDisable;
     extensionBehavior[E_SPV_NV_geometry_shader_passthrough]          = EBhDisable;
     extensionBehavior[E_GL_NV_viewport_array2]                       = EBhDisable;
@@ -434,6 +436,8 @@ void TParseVersions::getPreamble(std::string& preamble)
             "#define GL_AMD_shader_fragment_mask 1\n"
             "#define GL_AMD_gpu_shader_half_float_fetch 1\n"
 
+            "#define GL_INTEL_shader_integer_functions2 1\n"
+
             "#define GL_NV_sample_mask_override_coverage 1\n"
             "#define GL_NV_geometry_shader_passthrough 1\n"
             "#define GL_NV_viewport_array2 1\n"

glslang/MachineIndependent/Versions.h

@@ -206,6 +206,8 @@ const char* const E_GL_AMD_shader_image_load_store_lod          = "GL_AMD_shader
 const char* const E_GL_AMD_shader_fragment_mask                 = "GL_AMD_shader_fragment_mask";
 const char* const E_GL_AMD_gpu_shader_half_float_fetch          = "GL_AMD_gpu_shader_half_float_fetch";
 
+const char* const E_GL_INTEL_shader_integer_functions2          = "GL_INTEL_shader_integer_functions2";
+
 const char* const E_GL_NV_sample_mask_override_coverage         = "GL_NV_sample_mask_override_coverage";
 const char* const E_SPV_NV_geometry_shader_passthrough          = "GL_NV_geometry_shader_passthrough";
 const char* const E_GL_NV_viewport_array2                       = "GL_NV_viewport_array2";

glslang/MachineIndependent/intermOut.cpp

@@ -213,6 +213,13 @@ bool TOutputTraverser::visitBinary(TVisit /* visit */, TIntermBinary* node)
     case EOpLogicalXor: out.debug << "logical-xor"; break;
     case EOpLogicalAnd: out.debug << "logical-and"; break;
 
+    case EOpAbsDifference:          out.debug << "absoluteDifference";    break;
+    case EOpAddSaturate:            out.debug << "addSaturate";           break;
+    case EOpSubSaturate:            out.debug << "subtractSaturate";      break;
+    case EOpAverage:                out.debug << "average";               break;
+    case EOpAverageRounded:         out.debug << "averageRounded";        break;
+    case EOpMul32x16:               out.debug << "multiply32x16";         break;
+
     default: out.debug << "<unknown op>";
     }
 
@@ -557,6 +564,9 @@ bool TOutputTraverser::visitUnary(TVisit /* visit */, TIntermUnary* node)
     case EOpFindLSB:                out.debug << "findLSB";               break;
     case EOpFindMSB:                out.debug << "findMSB";               break;
 
+    case EOpCountLeadingZeros:      out.debug << "countLeadingZeros";     break;
+    case EOpCountTrailingZeros:     out.debug << "countTrailingZeros";    break;
+
     case EOpNoise:                  out.debug << "noise";                 break;
 
     case EOpBallot:                 out.debug << "ballot";                break;