isnan and isinf implementation

src/Reactor/LLVMReactor.cpp

@@ -5791,6 +5791,16 @@ namespace sw
 		return RValue<Int4>(Nucleus::createSExt(Nucleus::createFCmpOGT(x.value, y.value), Int4::getType()));
 	}
 
+	RValue<Int4> IsInf(RValue<Float4> x)
+	{
+		return CmpEQ(As<Int4>(x) & Int4(0x7FFFFFFF), Int4(0x7F800000));
+	}
+
+	RValue<Int4> IsNan(RValue<Float4> x)
+	{
+		return ~CmpEQ(x, x);
+	}
+
 	RValue<Float4> Round(RValue<Float4> x)
 	{
 		if(CPUID::supportsSSE4_1())

src/Reactor/Reactor.hpp

@@ -2110,6 +2110,8 @@ namespace sw
 	RValue<Int4> CmpNEQ(RValue<Float4> x, RValue<Float4> y);
 	RValue<Int4> CmpNLT(RValue<Float4> x, RValue<Float4> y);
 	RValue<Int4> CmpNLE(RValue<Float4> x, RValue<Float4> y);
+	RValue<Int4> IsInf(RValue<Float4> x);
+	RValue<Int4> IsNan(RValue<Float4> x);
 	RValue<Float4> Round(RValue<Float4> x);
 	RValue<Float4> Trunc(RValue<Float4> x);
 	RValue<Float4> Frac(RValue<Float4> x);

src/Reactor/SubzeroReactor.cpp

@@ -6967,6 +6967,16 @@ namespace sw
 		return RValue<Int4>(Nucleus::createFCmpOGT(x.value, y.value));
 	}
 
+	RValue<Int4> IsInf(RValue<Float4> x)
+	{
+		return CmpEQ(As<Int4>(x) & Int4(0x7FFFFFFF), Int4(0x7F800000));
+	}
+
+	RValue<Int4> IsNan(RValue<Float4> x)
+	{
+		return ~CmpEQ(x, x);
+	}
+
 	RValue<Float4> Round(RValue<Float4> x)
 	{
 		if(emulateIntrinsics || CPUID::ARM)

src/Shader/PixelProgram.cpp

@@ -230,6 +230,8 @@ namespace sw
 			case Shader::OPCODE_LRP:        lrp(d, s0, s1, s2);                            break;
 			case Shader::OPCODE_STEP:       step(d, s0, s1);                               break;
 			case Shader::OPCODE_SMOOTH:     smooth(d, s0, s1, s2);                         break;
+			case Shader::OPCODE_ISINF:      isinf(d, s0);                                  break;
+			case Shader::OPCODE_ISNAN:      isnan(d, s0);                                  break;
 			case Shader::OPCODE_FLOATBITSTOINT:
 			case Shader::OPCODE_FLOATBITSTOUINT:
 			case Shader::OPCODE_INTBITSTOFLOAT:

src/Shader/ShaderCore.cpp

@@ -1030,6 +1030,22 @@ namespace sw
 		dst.w = src0.w * (src1.w - src2.w) + src2.w;
 	}
 
+	void ShaderCore::isinf(Vector4f &dst, const Vector4f &src)
+	{
+		dst.x = As<Float4>(IsInf(src.x));
+		dst.y = As<Float4>(IsInf(src.y));
+		dst.z = As<Float4>(IsInf(src.z));
+		dst.w = As<Float4>(IsInf(src.w));
+	}
+
+	void ShaderCore::isnan(Vector4f &dst, const Vector4f &src)
+	{
+		dst.x = As<Float4>(IsNan(src.x));
+		dst.y = As<Float4>(IsNan(src.y));
+		dst.z = As<Float4>(IsNan(src.z));
+		dst.w = As<Float4>(IsNan(src.w));
+	}
+
 	void ShaderCore::smooth(Vector4f &dst, const Vector4f &edge0, const Vector4f &edge1, const Vector4f &x)
 	{
 		Float4 tx = Min(Max((x.x - edge0.x) / (edge1.x - edge0.x), Float4(0.0f)), Float4(1.0f)); dst.x = tx * tx * (Float4(3.0f) - Float4(2.0f) * tx);

src/Shader/ShaderCore.hpp

@@ -282,6 +282,8 @@ namespace sw
 		void lit(Vector4f &dst, const Vector4f &src);
 		void att(Vector4f &dst, const Vector4f &src0, const Vector4f &src1);
 		void lrp(Vector4f &dst, const Vector4f &src0, const Vector4f &src1, const Vector4f &src2);
+		void isinf(Vector4f &dst, const Vector4f &src);
+		void isnan(Vector4f &dst, const Vector4f &src);
 		void smooth(Vector4f &dst, const Vector4f &src0, const Vector4f &src1, const Vector4f &src2);
 		void packHalf2x16(Vector4f &dst, const Vector4f &src);
 		void unpackHalf2x16(Vector4f &dst, const Vector4f &src);

src/Shader/VertexProgram.cpp

@@ -208,6 +208,8 @@ namespace sw
 			case Shader::OPCODE_LRP:        lrp(d, s0, s1, s2);             break;
 			case Shader::OPCODE_STEP:       step(d, s0, s1);                break;
 			case Shader::OPCODE_SMOOTH:     smooth(d, s0, s1, s2);          break;
+			case Shader::OPCODE_ISINF:      isinf(d, s0);                   break;
+			case Shader::OPCODE_ISNAN:      isnan(d, s0);                   break;
 			case Shader::OPCODE_FLOATBITSTOINT:
 			case Shader::OPCODE_FLOATBITSTOUINT:
 			case Shader::OPCODE_INTBITSTOFLOAT: