C++程序  |  2188行  |  132.73 KB

/*-------------------------------------------------------------------------
 * drawElements Quality Program OpenGL ES 3.0 Module
 * -------------------------------------------------
 *
 * Copyright 2014 The Android Open Source Project
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *      http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 *
 *//*!
 * \file
 * \brief Shader operators tests.
 *//*--------------------------------------------------------------------*/

#include "es3fShaderOperatorTests.hpp"
#include "glsShaderRenderCase.hpp"
#include "gluShaderUtil.hpp"
#include "tcuStringTemplate.hpp"
#include "tcuVectorUtil.hpp"
#include "glwFunctions.hpp"
#include "glwEnums.hpp"

#include "deStringUtil.hpp"
#include "deInt32.h"
#include "deMemory.h"

#include <map>
#include <limits>

using namespace tcu;
using namespace glu;
using namespace deqp::gls;

using std::map;
using std::pair;
using std::vector;
using std::string;
using std::ostringstream;

namespace deqp
{
namespace gles3
{
namespace Functional
{

#if defined(abs)
#	undef abs
#endif

using de::min;
using de::max;
using de::clamp;

// \note VS2013 gets confused without these
using tcu::asinh;
using tcu::acosh;
using tcu::atanh;
using tcu::exp2;
using tcu::log2;
using tcu::trunc;

inline float abs		(float v)			{ return deFloatAbs(v); }

inline bool logicalAnd	(bool a, bool b)	{ return (a && b); }
inline bool logicalOr	(bool a, bool b)	{ return (a || b); }
inline bool logicalXor	(bool a, bool b)	{ return (a != b); }

// \note stdlib.h defines div() that is not compatible with the macros.
template<typename T> inline T div (T a, T b) { return a / b; }

template<typename T> inline T leftShift (T value, int amount) { return value << amount; }

inline deUint32	rightShift (deUint32 value, int amount)		{ return value >> amount; }
inline int		rightShift (int value, int amount)			{ return (value >> amount) | (value >= 0 ? 0 : ~(~0U >> amount)); } // \note Arithmetic shift.

template<typename T, int Size> Vector<T, Size> leftShift (const Vector<T, Size>& value, const Vector<int, Size>& amount)
{
	Vector<T, Size> result;
	for (int i = 0; i < Size; i++)
		result[i] = leftShift(value[i], amount[i]);
	return result;
}

template<typename T, int Size> Vector<T, Size> rightShift (const Vector<T, Size>& value, const Vector<int, Size>& amount)
{
	Vector<T, Size> result;
	for (int i = 0; i < Size; i++)
		result[i] = rightShift(value[i], amount[i]);
	return result;
}

template<typename T, int Size> Vector<T, Size> leftShiftVecScalar	(const Vector<T, Size>& value, int amount) { return leftShift(value, Vector<int, Size>(amount)); }
template<typename T, int Size> Vector<T, Size> rightShiftVecScalar	(const Vector<T, Size>& value, int amount) { return rightShift(value, Vector<int, Size>(amount)); }

template<typename T, int Size>
inline Vector<T, Size> minVecScalar (const Vector<T, Size>& v, T s)
{
	Vector<T, Size> res;
	for (int i = 0; i < Size; i++)
		res[i] = min(v[i], s);
	return res;
}

template<typename T, int Size>
inline Vector<T, Size> maxVecScalar (const Vector<T, Size>& v, T s)
{
	Vector<T, Size> res;
	for (int i = 0; i < Size; i++)
		res[i] = max(v[i], s);
	return res;
}

template<typename T, int Size>
inline Vector<T, Size> clampVecScalarScalar (const Vector<T, Size>& v, T s0, T s1)
{
	Vector<T, Size> res;
	for (int i = 0; i < Size; i++)
		res[i] = clamp(v[i], s0, s1);
	return res;
}

template<typename T, int Size>
inline Vector<T, Size> mixVecVecScalar (const Vector<T, Size>& v0, const Vector<T, Size>& v1, T s)
{
	Vector<T, Size> res;
	for (int i = 0; i < Size; i++)
		res[i] = mix(v0[i], v1[i], s);
	return res;
}

template<typename T, int Size>
inline Vector<T, Size> stepScalarVec (T s, const Vector<T, Size>& v)
{
	Vector<T, Size> res;
	for (int i = 0; i < Size; i++)
		res[i] = step(s, v[i]);
	return res;
}

template<typename T, int Size>
inline Vector<T, Size> smoothStepScalarScalarVec (T s0, T s1, const Vector<T, Size>& v)
{
	Vector<T, Size> res;
	for (int i = 0; i < Size; i++)
		res[i] = smoothStep(s0, s1, v[i]);
	return res;
}

inline float	addOne (float v)	{ return v + 1.0f; };
inline float	subOne (float v)	{ return v - 1.0f; };
inline int		addOne (int v)		{ return v + 1; };
inline int		subOne (int v)		{ return v - 1; };
inline deUint32	addOne (deUint32 v)	{ return v + 1; };
inline deUint32	subOne (deUint32 v)	{ return v - 1; };

template<int Size> inline Vector<float, Size>		addOne (const Vector<float, Size>& v)		{ return v + 1.0f; };
template<int Size> inline Vector<float, Size>		subOne (const Vector<float, Size>& v)		{ return v - 1.0f; };
template<int Size> inline Vector<int, Size>			addOne (const Vector<int, Size>& v)			{ return v + 1; };
template<int Size> inline Vector<int, Size>			subOne (const Vector<int, Size>& v)			{ return v - 1; };
template<int Size> inline Vector<deUint32, Size>	addOne (const Vector<deUint32, Size>& v)	{ return v + 1U; };
template<int Size> inline Vector<deUint32, Size>	subOne (const Vector<deUint32, Size>& v)	{ return (v.asInt() - 1).asUint(); };

template<typename T> inline T selection	(bool cond, T a, T b)	{ return cond ? a : b; };

// Vec-scalar and scalar-vec binary operators.

// \note This one is done separately due to how the overloaded minus operator is implemented for vector-scalar operands.
template<int Size>				inline Vector<deUint32, Size>	subVecScalar			(const Vector<deUint32, Size>& v, deUint32 s)	{ return (v.asInt() - (int)s).asUint(); };

template<typename T, int Size>	inline Vector<T, Size>			addVecScalar			(const Vector<T, Size>& v, T s)					{ return v + s; };
template<typename T, int Size>	inline Vector<T, Size>			subVecScalar			(const Vector<T, Size>& v, T s)					{ return v - s; };
template<typename T, int Size>	inline Vector<T, Size>			mulVecScalar			(const Vector<T, Size>& v, T s)					{ return v * s; };
template<typename T, int Size>	inline Vector<T, Size>			divVecScalar			(const Vector<T, Size>& v, T s)					{ return v / s; };
template<typename T, int Size>	inline Vector<T, Size>			modVecScalar			(const Vector<T, Size>& v, T s)					{ return mod(v, Vector<T, Size>(s)); };
template<typename T, int Size>	inline Vector<T, Size>			bitwiseAndVecScalar		(const Vector<T, Size>& v, T s)					{ return bitwiseAnd(v, Vector<T, Size>(s)); };
template<typename T, int Size>	inline Vector<T, Size>			bitwiseOrVecScalar		(const Vector<T, Size>& v, T s)					{ return bitwiseOr(v, Vector<T, Size>(s)); };
template<typename T, int Size>	inline Vector<T, Size>			bitwiseXorVecScalar		(const Vector<T, Size>& v, T s)					{ return bitwiseXor(v, Vector<T, Size>(s)); };

template<typename T, int Size> inline Vector<T, Size>			addScalarVec			(T s, const Vector<T, Size>& v) 				{ return s + v; };
template<typename T, int Size> inline Vector<T, Size>			subScalarVec			(T s, const Vector<T, Size>& v) 				{ return s - v; };
template<typename T, int Size> inline Vector<T, Size>			mulScalarVec			(T s, const Vector<T, Size>& v) 				{ return s * v; };
template<typename T, int Size> inline Vector<T, Size>			divScalarVec			(T s, const Vector<T, Size>& v) 				{ return s / v; };
template<typename T, int Size> inline Vector<T, Size>			modScalarVec			(T s, const Vector<T, Size>& v)					{ return mod(Vector<T, Size>(s), v); };
template<typename T, int Size> inline Vector<T, Size>			bitwiseAndScalarVec		(T s, const Vector<T, Size>& v)					{ return bitwiseAnd(Vector<T, Size>(s), v); };
template<typename T, int Size> inline Vector<T, Size>			bitwiseOrScalarVec		(T s, const Vector<T, Size>& v)					{ return bitwiseOr(Vector<T, Size>(s), v); };
template<typename T, int Size> inline Vector<T, Size>			bitwiseXorScalarVec		(T s, const Vector<T, Size>& v)					{ return bitwiseXor(Vector<T, Size>(s), v); };

// Reference functions for specific sequence operations for the sequence operator tests.

// Reference for expression "in0, in2 + in1, in1 + in0"
inline Vec4		sequenceNoSideEffCase0 (const Vec4& in0, const Vec4& in1, const Vec4& in2)		{ DE_UNREF(in2); return in1 + in0; }
// Reference for expression "in0, in2 + in1, in1 + in0"
inline deUint32	sequenceNoSideEffCase1 (float in0, deUint32 in1, float in2)						{ DE_UNREF(in0); DE_UNREF(in2); return in1 + in1; }
// Reference for expression "in0 && in1, in0, ivec2(vec2(in0) + in2)"
inline IVec2	sequenceNoSideEffCase2 (bool in0, bool in1, const Vec2& in2)					{ DE_UNREF(in1); return IVec2((int)((float)in0 + in2.x()), (int)((float)in0 + in2.y())); }
// Reference for expression "in0 + vec4(in1), in2, in1"
inline IVec4	sequenceNoSideEffCase3 (const Vec4& in0, const IVec4& in1, const BVec4& in2)	{ DE_UNREF(in0); DE_UNREF(in2); return in1; }
// Reference for expression "in0++, in1 = in0 + in2, in2 = in1"
inline Vec4		sequenceSideEffCase0 (const Vec4& in0, const Vec4& in1, const Vec4& in2)		{ DE_UNREF(in1); return in0 + 1.0f + in2; }
// Reference for expression "in1++, in0 = float(in1), in1 = uint(in0 + in2)"
inline deUint32	sequenceSideEffCase1 (float in0, deUint32 in1, float in2)						{ DE_UNREF(in0); return (deUint32)(float(in1) + 1.0f + in2); }
// Reference for expression "in1 = in0, in2++, in2 = in2 + vec2(in1), ivec2(in2)"
inline IVec2	sequenceSideEffCase2 (bool in0, bool in1, const Vec2& in2)						{ DE_UNREF(in1); return (in2 + Vec2(1.0f) + Vec2((float)in0)).asInt(); }
// Reference for expression "in0 = in0 + vec4(in2), in1 = in1 + ivec4(in0), in1++"
inline IVec4	sequenceSideEffCase3 (const Vec4& in0, const IVec4& in1, const BVec4& in2)		{ return in1 + (in0 + Vec4((float)in2.x(), (float)in2.y(), (float)in2.z(), (float)in2.w())).asInt(); }

// ShaderEvalFunc-type wrappers for the above functions.
void evalSequenceNoSideEffCase0	(ShaderEvalContext& ctx) { ctx.color		= sequenceNoSideEffCase0		(ctx.in[0].swizzle(1, 2, 3, 0),	ctx.in[1].swizzle(3, 2, 1, 0),			ctx.in[2].swizzle(0, 3, 2, 1)); }
void evalSequenceNoSideEffCase1	(ShaderEvalContext& ctx) { ctx.color.x()	= (float)sequenceNoSideEffCase1	(ctx.in[0].z(),					(deUint32)ctx.in[1].x(),				ctx.in[2].y()); }
void evalSequenceNoSideEffCase2	(ShaderEvalContext& ctx) { ctx.color.yz()	= sequenceNoSideEffCase2		(ctx.in[0].z() > 0.0f,			ctx.in[1].x() > 0.0f,					ctx.in[2].swizzle(2, 1)).asFloat(); }
void evalSequenceNoSideEffCase3	(ShaderEvalContext& ctx) { ctx.color		= sequenceNoSideEffCase3		(ctx.in[0].swizzle(1, 2, 3, 0),	ctx.in[1].swizzle(3, 2, 1, 0).asInt(),	greaterThan(ctx.in[2].swizzle(0, 3, 2, 1), Vec4(0.0f, 0.0f, 0.0f, 0.0f))).asFloat(); }
void evalSequenceSideEffCase0	(ShaderEvalContext& ctx) { ctx.color		= sequenceSideEffCase0			(ctx.in[0].swizzle(1, 2, 3, 0),	ctx.in[1].swizzle(3, 2, 1, 0),			ctx.in[2].swizzle(0, 3, 2, 1)); }
void evalSequenceSideEffCase1	(ShaderEvalContext& ctx) { ctx.color.x()	= (float)sequenceSideEffCase1	(ctx.in[0].z(),					(deUint32)ctx.in[1].x(),				ctx.in[2].y()); }
void evalSequenceSideEffCase2	(ShaderEvalContext& ctx) { ctx.color.yz()	= sequenceSideEffCase2			(ctx.in[0].z() > 0.0f,			ctx.in[1].x() > 0.0f,					ctx.in[2].swizzle(2, 1)).asFloat(); }
void evalSequenceSideEffCase3	(ShaderEvalContext& ctx) { ctx.color		= sequenceSideEffCase3			(ctx.in[0].swizzle(1, 2, 3, 0),	ctx.in[1].swizzle(3, 2, 1, 0).asInt(),	greaterThan(ctx.in[2].swizzle(0, 3, 2, 1), Vec4(0.0f, 0.0f, 0.0f, 0.0f))).asFloat(); }

static string stringJoin (const vector<string>& elems, const string& delim)
{
	string result;
	for (int i = 0; i < (int)elems.size(); i++)
		result += (i > 0 ? delim : "") + elems[i];
	return result;
}

static string twoValuedVec4 (const string& first, const string& second, const BVec4& firstMask)
{
	vector<string> elems(4);
	for (int i = 0; i < 4; i++)
		elems[i] = firstMask[i] ? first : second;

	return "vec4(" + stringJoin(elems, ", ") + ")";
}

enum
{
	MAX_INPUTS = 3
};

enum PrecisionMask
{
	PRECMASK_NA				= 0,						//!< Precision not applicable (booleans)
	PRECMASK_LOWP			= (1<<PRECISION_LOWP),
	PRECMASK_MEDIUMP		= (1<<PRECISION_MEDIUMP),
	PRECMASK_HIGHP			= (1<<PRECISION_HIGHP),

	PRECMASK_LOWP_MEDIUMP	= PRECMASK_LOWP | PRECMASK_MEDIUMP,
	PRECMASK_MEDIUMP_HIGHP	= PRECMASK_MEDIUMP | PRECMASK_HIGHP,
	PRECMASK_ALL			= PRECMASK_LOWP | PRECMASK_MEDIUMP | PRECMASK_HIGHP
};

enum ValueType
{
	VALUE_NONE			= 0,
	VALUE_FLOAT			= (1<<0),	// float scalar
	VALUE_FLOAT_VEC		= (1<<1),	// float vector
	VALUE_FLOAT_GENTYPE	= (1<<2),	// float scalar/vector
	VALUE_VEC3			= (1<<3),	// vec3 only
	VALUE_MATRIX		= (1<<4),	// matrix
	VALUE_BOOL			= (1<<5),	// boolean scalar
	VALUE_BOOL_VEC		= (1<<6),	// boolean vector
	VALUE_BOOL_GENTYPE	= (1<<7),	// boolean scalar/vector
	VALUE_INT			= (1<<8),	// int scalar
	VALUE_INT_VEC		= (1<<9),	// int vector
	VALUE_INT_GENTYPE	= (1<<10),	// int scalar/vector
	VALUE_UINT			= (1<<11),	// uint scalar
	VALUE_UINT_VEC		= (1<<12),	// uint vector
	VALUE_UINT_GENTYPE	= (1<<13),	// uint scalar/vector

	// Shorthands.
	F				= VALUE_FLOAT,
	FV				= VALUE_FLOAT_VEC,
	GT				= VALUE_FLOAT_GENTYPE,
	V3				= VALUE_VEC3,
	M				= VALUE_MATRIX,
	B				= VALUE_BOOL,
	BV				= VALUE_BOOL_VEC,
	BGT				= VALUE_BOOL_GENTYPE,
	I				= VALUE_INT,
	IV				= VALUE_INT_VEC,
	IGT				= VALUE_INT_GENTYPE,
	U				= VALUE_UINT,
	UV				= VALUE_UINT_VEC,
	UGT				= VALUE_UINT_GENTYPE
};

static inline bool isScalarType (ValueType type)
{
	return type == VALUE_FLOAT || type == VALUE_BOOL || type == VALUE_INT || type == VALUE_UINT;
}

static inline bool isFloatType (ValueType type)
{
	return (type & (VALUE_FLOAT | VALUE_FLOAT_VEC | VALUE_FLOAT_GENTYPE)) != 0;
}

static inline bool isIntType (ValueType type)
{
	return (type & (VALUE_INT | VALUE_INT_VEC | VALUE_INT_GENTYPE)) != 0;
}

static inline bool isUintType (ValueType type)
{
	return (type & (VALUE_UINT | VALUE_UINT_VEC | VALUE_UINT_GENTYPE)) != 0;
}

static inline bool isBoolType (ValueType type)
{
	return (type & (VALUE_BOOL | VALUE_BOOL_VEC | VALUE_BOOL_GENTYPE)) != 0;
}

static inline float getGLSLUintMaxAsFloat (const glw::Functions& gl, ShaderType shaderType, Precision uintPrecision)
{
	deUint32 intPrecisionGL;
	deUint32 shaderTypeGL;

	switch (uintPrecision)
	{
		case PRECISION_LOWP:		intPrecisionGL = GL_LOW_INT;		break;
		case PRECISION_MEDIUMP:		intPrecisionGL = GL_MEDIUM_INT;		break;
		case PRECISION_HIGHP:		intPrecisionGL = GL_HIGH_INT;		break;
		default:
			DE_ASSERT(false);
			intPrecisionGL = 0;
	}

	switch (shaderType)
	{
		case SHADERTYPE_VERTEX:		shaderTypeGL = GL_VERTEX_SHADER;	break;
		case SHADERTYPE_FRAGMENT:	shaderTypeGL = GL_FRAGMENT_SHADER;	break;
		default:
			DE_ASSERT(false);
			shaderTypeGL = 0;
	}

	glw::GLint range[2]		= { -1, -1 };
	glw::GLint precision	= -1;

	gl.getShaderPrecisionFormat(shaderTypeGL, intPrecisionGL, &range[0], &precision);
	GLU_EXPECT_NO_ERROR(gl.getError(), "glGetShaderPrecisionFormat failed");

	TCU_CHECK(de::inBounds(range[0], 8, 32));

	const int numBitsInType = range[0] + 1;
	return (float)((1ull << numBitsInType) - 1);
}

// Float scalar that can be either constant or a symbol that can be evaluated later.
class FloatScalar
{
public:
	enum Symbol
	{
		SYMBOL_LOWP_UINT_MAX = 0,
		SYMBOL_MEDIUMP_UINT_MAX,

		SYMBOL_LOWP_UINT_MAX_RECIPROCAL,
		SYMBOL_MEDIUMP_UINT_MAX_RECIPROCAL,

		SYMBOL_ONE_MINUS_UINT32MAX_DIV_LOWP_UINT_MAX,
		SYMBOL_ONE_MINUS_UINT32MAX_DIV_MEDIUMP_UINT_MAX,

		SYMBOL_LAST
	};

	FloatScalar (float c)	: m_isConstant(true),	m_value(c) {}
	FloatScalar (Symbol s)	: m_isConstant(false),	m_value(s) {}

	float getValue (const glw::Functions& gl, ShaderType shaderType) const
	{
		if (m_isConstant)
			return m_value.constant;
		else
		{
			switch (m_value.symbol)
			{
				case SYMBOL_LOWP_UINT_MAX:								return getGLSLUintMaxAsFloat(gl, shaderType, PRECISION_LOWP);
				case SYMBOL_MEDIUMP_UINT_MAX:							return getGLSLUintMaxAsFloat(gl, shaderType, PRECISION_MEDIUMP);

				case SYMBOL_LOWP_UINT_MAX_RECIPROCAL:					return 1.0f / getGLSLUintMaxAsFloat(gl, shaderType, PRECISION_LOWP);
				case SYMBOL_MEDIUMP_UINT_MAX_RECIPROCAL:				return 1.0f / getGLSLUintMaxAsFloat(gl, shaderType, PRECISION_MEDIUMP);

				case SYMBOL_ONE_MINUS_UINT32MAX_DIV_LOWP_UINT_MAX:		return 1.0f - (float)std::numeric_limits<deUint32>::max() / getGLSLUintMaxAsFloat(gl, shaderType, PRECISION_LOWP);
				case SYMBOL_ONE_MINUS_UINT32MAX_DIV_MEDIUMP_UINT_MAX:	return 1.0f - (float)std::numeric_limits<deUint32>::max() / getGLSLUintMaxAsFloat(gl, shaderType, PRECISION_MEDIUMP);

				default:
					DE_ASSERT(false);
					return 0.0f;
			}
		}
	}

private:
	bool m_isConstant;

	union ConstantOrSymbol
	{
		float	constant;
		Symbol	symbol;

		ConstantOrSymbol (float c)	: constant	(c) {}
		ConstantOrSymbol (Symbol s)	: symbol	(s) {}
	} m_value;
};

struct Value
{
	Value (ValueType valueType_, const FloatScalar& rangeMin_, const FloatScalar& rangeMax_)
		: valueType	(valueType_)
		, rangeMin	(rangeMin_)
		, rangeMax	(rangeMax_)
	{
	}

	ValueType		valueType;
	FloatScalar		rangeMin;
	FloatScalar		rangeMax;
};

enum OperationType
{
	FUNCTION = 0,
	OPERATOR,
	SIDE_EFFECT_OPERATOR // Test the side-effect (as opposed to the result) of a side-effect operator.
};

struct BuiltinFuncInfo
{
	BuiltinFuncInfo (const char* caseName_, const char* shaderFuncName_, ValueType outValue_,
					 Value input0_, Value input1_, Value input2_,
					 const FloatScalar& resultScale_, const FloatScalar& resultBias_,
					 deUint32 precisionMask_,
					 ShaderEvalFunc evalFuncScalar_, ShaderEvalFunc evalFuncVec2_, ShaderEvalFunc evalFuncVec3_, ShaderEvalFunc evalFuncVec4_,
					 OperationType type_=FUNCTION, bool isUnaryPrefix_=true)
		: caseName			(caseName_)
		, shaderFuncName	(shaderFuncName_)
		, outValue			(outValue_)
		, input0			(input0_)
		, input1			(input1_)
		, input2			(input2_)
		, resultScale		(resultScale_)
		, resultBias		(resultBias_)
		, referenceScale	(resultScale_)
		, referenceBias		(resultBias_)
		, precisionMask		(precisionMask_)
		, evalFuncScalar	(evalFuncScalar_)
		, evalFuncVec2		(evalFuncVec2_)
		, evalFuncVec3		(evalFuncVec3_)
		, evalFuncVec4		(evalFuncVec4_)
		, type				(type_)
		, isUnaryPrefix		(isUnaryPrefix_)
	{
	}

	BuiltinFuncInfo (const char* caseName_, const char* shaderFuncName_, ValueType outValue_,
					 Value input0_, Value input1_, Value input2_,
					 const FloatScalar& resultScale_, const FloatScalar& resultBias_, const FloatScalar& referenceScale_, const FloatScalar& referenceBias_,
					 deUint32 precisionMask_,
					 ShaderEvalFunc evalFuncScalar_, ShaderEvalFunc evalFuncVec2_, ShaderEvalFunc evalFuncVec3_, ShaderEvalFunc evalFuncVec4_,
					 OperationType type_=FUNCTION, bool isUnaryPrefix_=true)
		: caseName			(caseName_)
		, shaderFuncName	(shaderFuncName_)
		, outValue			(outValue_)
		, input0			(input0_)
		, input1			(input1_)
		, input2			(input2_)
		, resultScale		(resultScale_)
		, resultBias		(resultBias_)
		, referenceScale	(referenceScale_)
		, referenceBias		(referenceBias_)
		, precisionMask		(precisionMask_)
		, evalFuncScalar	(evalFuncScalar_)
		, evalFuncVec2		(evalFuncVec2_)
		, evalFuncVec3		(evalFuncVec3_)
		, evalFuncVec4		(evalFuncVec4_)
		, type				(type_)
		, isUnaryPrefix		(isUnaryPrefix_)
	{
	}

	const char*		caseName;			//!< Name of case.
	const char*		shaderFuncName;		//!< Name in shading language.
	ValueType		outValue;
	Value			input0;
	Value			input1;
	Value			input2;
	FloatScalar		resultScale;
	FloatScalar		resultBias;
	FloatScalar		referenceScale;
	FloatScalar		referenceBias;
	deUint32		precisionMask;
	ShaderEvalFunc	evalFuncScalar;
	ShaderEvalFunc	evalFuncVec2;
	ShaderEvalFunc	evalFuncVec3;
	ShaderEvalFunc	evalFuncVec4;
	OperationType	type;
	bool			isUnaryPrefix;		//!< Whether a unary operator is a prefix operator; redundant unless unary.
};

static inline BuiltinFuncInfo BuiltinOperInfo (const char* caseName_, const char* shaderFuncName_, ValueType outValue_, Value input0_, Value input1_, Value input2_, const FloatScalar& resultScale_, const FloatScalar& resultBias_, deUint32 precisionMask_, ShaderEvalFunc evalFuncScalar_, ShaderEvalFunc evalFuncVec2_, ShaderEvalFunc evalFuncVec3_, ShaderEvalFunc evalFuncVec4_)
{
	return BuiltinFuncInfo(caseName_, shaderFuncName_, outValue_, input0_, input1_, input2_, resultScale_, resultBias_, resultScale_, resultBias_, precisionMask_, evalFuncScalar_, evalFuncVec2_, evalFuncVec3_, evalFuncVec4_, OPERATOR);
}

static inline BuiltinFuncInfo BuiltinOperInfoSeparateRefScaleBias (const char* caseName_, const char* shaderFuncName_, ValueType outValue_, Value input0_, Value input1_, Value input2_, const FloatScalar& resultScale_, const FloatScalar& resultBias_, deUint32 precisionMask_, ShaderEvalFunc evalFuncScalar_, ShaderEvalFunc evalFuncVec2_, ShaderEvalFunc evalFuncVec3_, ShaderEvalFunc evalFuncVec4_, const FloatScalar& referenceScale_, const FloatScalar& referenceBias_)
{
	return BuiltinFuncInfo(caseName_, shaderFuncName_, outValue_, input0_, input1_, input2_, resultScale_, resultBias_, referenceScale_, referenceBias_, precisionMask_, evalFuncScalar_, evalFuncVec2_, evalFuncVec3_, evalFuncVec4_, OPERATOR);
}

// For postfix (unary) operators.
static inline BuiltinFuncInfo BuiltinPostOperInfo (const char* caseName_, const char* shaderFuncName_, ValueType outValue_, Value input0_, Value input1_, Value input2_, const FloatScalar& resultScale_, const FloatScalar& resultBias_, deUint32 precisionMask_, ShaderEvalFunc evalFuncScalar_, ShaderEvalFunc evalFuncVec2_, ShaderEvalFunc evalFuncVec3_, ShaderEvalFunc evalFuncVec4_)
{
	return BuiltinFuncInfo(caseName_, shaderFuncName_, outValue_, input0_, input1_, input2_, resultScale_, resultBias_, resultScale_, resultBias_, precisionMask_, evalFuncScalar_, evalFuncVec2_, evalFuncVec3_, evalFuncVec4_, OPERATOR, false);
}

static inline BuiltinFuncInfo BuiltinSideEffOperInfo (const char* caseName_, const char* shaderFuncName_, ValueType outValue_, Value input0_, Value input1_, Value input2_, const FloatScalar& resultScale_, const FloatScalar& resultBias_, deUint32 precisionMask_, ShaderEvalFunc evalFuncScalar_, ShaderEvalFunc evalFuncVec2_, ShaderEvalFunc evalFuncVec3_, ShaderEvalFunc evalFuncVec4_)
{
	return BuiltinFuncInfo(caseName_, shaderFuncName_, outValue_, input0_, input1_, input2_, resultScale_, resultBias_, resultScale_, resultBias_, precisionMask_, evalFuncScalar_, evalFuncVec2_, evalFuncVec3_, evalFuncVec4_, SIDE_EFFECT_OPERATOR);
}

// For postfix (unary) operators, testing side-effect.
static inline BuiltinFuncInfo BuiltinPostSideEffOperInfo (const char* caseName_, const char* shaderFuncName_, ValueType outValue_, Value input0_, Value input1_, Value input2_, const FloatScalar& resultScale_, const FloatScalar& resultBias_, deUint32 precisionMask_, ShaderEvalFunc evalFuncScalar_, ShaderEvalFunc evalFuncVec2_, ShaderEvalFunc evalFuncVec3_, ShaderEvalFunc evalFuncVec4_)
{
	return BuiltinFuncInfo(caseName_, shaderFuncName_, outValue_, input0_, input1_, input2_, resultScale_, resultBias_, resultScale_, resultBias_, precisionMask_, evalFuncScalar_, evalFuncVec2_, evalFuncVec3_, evalFuncVec4_, SIDE_EFFECT_OPERATOR, false);
}

// BuiltinFuncGroup

struct BuiltinFuncGroup
{
						BuiltinFuncGroup	(const char* name_, const char* description_) : name(name_), description(description_) {}
	BuiltinFuncGroup&	operator<<			(const BuiltinFuncInfo& info) { funcInfos.push_back(info); return *this; }

	const char*						name;
	const char*						description;
	std::vector<BuiltinFuncInfo>	funcInfos;
};

static const char* s_inSwizzles[MAX_INPUTS][4] =
{
	{ "z", "wy", "zxy", "yzwx" },
	{ "x", "yx", "yzx", "wzyx" },
	{ "y", "zy", "wyz", "xwzy" }
};

static const char* s_outSwizzles[]	= { "x", "yz", "xyz", "xyzw" };

static const BVec4 s_outSwizzleChannelMasks[] =
{
	BVec4(true,  false, false, false),
	BVec4(false, true,  true,  false),
	BVec4(true,  true,  true,  false),
	BVec4(true,  true,  true,  true )
};

// OperatorShaderEvaluator

class OperatorShaderEvaluator : public ShaderEvaluator
{
public:
	OperatorShaderEvaluator (const glw::Functions& gl, ShaderType shaderType, ShaderEvalFunc evalFunc, const FloatScalar& scale, const FloatScalar& bias, int resultScalarSize)
		: m_gl							(gl)
		, m_shaderType					(shaderType)
		, m_evalFunc					(evalFunc)
		, m_scale						(scale)
		, m_bias						(bias)
		, m_resultScalarSize			(resultScalarSize)
		, m_areScaleAndBiasEvaluated	(false)
		, m_evaluatedScale				(-1.0f)
		, m_evaluatedBias				(-1.0f)
	{
		DE_ASSERT(de::inRange(resultScalarSize, 1, 4));
	}

	virtual ~OperatorShaderEvaluator (void)
	{
	}

	virtual void evaluate (ShaderEvalContext& ctx)
	{
		m_evalFunc(ctx);

		if (!m_areScaleAndBiasEvaluated)
		{
			m_evaluatedScale	= m_scale.getValue(m_gl, m_shaderType);
			m_evaluatedBias		= m_bias.getValue(m_gl, m_shaderType);
			m_areScaleAndBiasEvaluated = true;
		}

		for (int i = 0; i < 4; i++)
			if (s_outSwizzleChannelMasks[m_resultScalarSize-1][i])
				ctx.color[i] = ctx.color[i] * m_evaluatedScale + m_evaluatedBias;
	}

private:
	const glw::Functions&	m_gl;
	ShaderType				m_shaderType;
	ShaderEvalFunc			m_evalFunc;
	FloatScalar				m_scale;
	FloatScalar				m_bias;
	int						m_resultScalarSize;

	bool					m_areScaleAndBiasEvaluated;
	float					m_evaluatedScale;
	float					m_evaluatedBias;
};

// Concrete value.

struct ShaderValue
{
	ShaderValue (DataType type_, const FloatScalar& rangeMin_, const FloatScalar& rangeMax_)
		: type		(type_)
		, rangeMin	(rangeMin_)
		, rangeMax	(rangeMax_)
	{
	}

	ShaderValue (void)
		: type		(TYPE_LAST)
		, rangeMin	(0.0f)
		, rangeMax	(0.0f)
	{
	}

	DataType		type;
	FloatScalar		rangeMin;
	FloatScalar		rangeMax;
};

struct ShaderDataSpec
{
	ShaderDataSpec (void)
		: resultScale		(1.0f)
		, resultBias		(0.0f)
		, referenceScale	(1.0f)
		, referenceBias		(0.0f)
		, precision			(PRECISION_LAST)
		, output			(TYPE_LAST)
		, numInputs			(0)
	{
	}

	FloatScalar		resultScale;
	FloatScalar		resultBias;
	FloatScalar		referenceScale;
	FloatScalar		referenceBias;
	Precision		precision;
	DataType		output;
	int				numInputs;
	ShaderValue		inputs[MAX_INPUTS];
};

// ShaderOperatorCase

class ShaderOperatorCase : public ShaderRenderCase
{
public:
								ShaderOperatorCase		(Context& context, const char* caseName, const char* description, bool isVertexCase, ShaderEvalFunc evalFunc, const string& shaderOp, const ShaderDataSpec& spec);
	virtual						~ShaderOperatorCase		(void);

protected:
	void						setupShaderData			(void);

private:
								ShaderOperatorCase		(const ShaderOperatorCase&);	// not allowed!
	ShaderOperatorCase&			operator=				(const ShaderOperatorCase&);	// not allowed!

	ShaderDataSpec				m_spec;
	string						m_shaderOp;
	OperatorShaderEvaluator		m_evaluator;
};

ShaderOperatorCase::ShaderOperatorCase (Context& context, const char* caseName, const char* description, bool isVertexCase, ShaderEvalFunc evalFunc, const string& shaderOp, const ShaderDataSpec& spec)
	: ShaderRenderCase	(context.getTestContext(), context.getRenderContext(), context.getContextInfo(), caseName, description, isVertexCase, m_evaluator)
	, m_spec			(spec)
	, m_shaderOp		(shaderOp)
	, m_evaluator		(m_renderCtx.getFunctions(), isVertexCase ? SHADERTYPE_VERTEX : SHADERTYPE_FRAGMENT, evalFunc, spec.referenceScale, spec.referenceBias, getDataTypeScalarSize(spec.output))
{
}

void ShaderOperatorCase::setupShaderData (void)
{
	ShaderType		shaderType	= m_isVertexCase ? SHADERTYPE_VERTEX : SHADERTYPE_FRAGMENT;
	const char*		precision	= m_spec.precision != PRECISION_LAST ? getPrecisionName(m_spec.precision) : DE_NULL;
	const char*		inputPrecision[MAX_INPUTS];

	ostringstream	vtx;
	ostringstream	frag;
	ostringstream&	op			= m_isVertexCase ? vtx : frag;

	vtx << "#version 300 es\n";
	frag << "#version 300 es\n";

	// Compute precision for inputs.
	for (int i = 0; i < m_spec.numInputs; i++)
	{
		bool		isBoolVal	= de::inRange<int>(m_spec.inputs[i].type, TYPE_BOOL, TYPE_BOOL_VEC4);
		bool		isIntVal	= de::inRange<int>(m_spec.inputs[i].type, TYPE_INT, TYPE_INT_VEC4);
		bool		isUintVal	= de::inRange<int>(m_spec.inputs[i].type, TYPE_UINT, TYPE_UINT_VEC4);
		// \note Mediump interpolators are used for booleans, and highp for integers.
		Precision	prec		= isBoolVal	? PRECISION_MEDIUMP
								: isIntVal || isUintVal ? PRECISION_HIGHP
								: m_spec.precision;
		inputPrecision[i] = getPrecisionName(prec);
	}

	// Attributes.
	vtx << "in highp vec4 a_position;\n";
	for (int i = 0; i < m_spec.numInputs; i++)
		vtx << "in " << inputPrecision[i] << " vec4 a_in" << i << ";\n";

	// Color output.
	frag << "layout(location = 0) out mediump vec4 o_color;\n";

	if (m_isVertexCase)
	{
		vtx << "out mediump vec4 v_color;\n";
		frag << "in mediump vec4 v_color;\n";
	}
	else
	{
		for (int i = 0; i < m_spec.numInputs; i++)
		{
			vtx << "out " << inputPrecision[i] << " vec4 v_in" << i << ";\n";
			frag << "in " << inputPrecision[i] << " vec4 v_in" << i << ";\n";
		}
	}

	vtx << "\n";
	vtx << "void main()\n";
	vtx << "{\n";
	vtx << "	gl_Position = a_position;\n";

	frag << "\n";
	frag << "void main()\n";
	frag << "{\n";

	// Expression inputs.
	string prefix = m_isVertexCase ? "a_" : "v_";
	for (int i = 0; i < m_spec.numInputs; i++)
	{
		DataType		inType		= m_spec.inputs[i].type;
		int				inSize		= getDataTypeScalarSize(inType);
		bool			isInt		= de::inRange<int>(inType, TYPE_INT, TYPE_INT_VEC4);
		bool			isUint		= de::inRange<int>(inType, TYPE_UINT, TYPE_UINT_VEC4);
		bool			isBool		= de::inRange<int>(inType, TYPE_BOOL, TYPE_BOOL_VEC4);
		const char*		typeName	= getDataTypeName(inType);
		const char*		swizzle		= s_inSwizzles[i][inSize-1];

		op << "\t";
		if (precision && !isBool) op << precision << " ";

		op << typeName << " in" << i << " = ";

		if (isBool)
		{
			if (inSize == 1)	op << "(";
			else				op << "greaterThan(";
		}
		else if (isInt || isUint)
			op << typeName << "(";

		op << prefix << "in" << i << "." << swizzle;

		if (isBool)
		{
			if (inSize == 1)	op << " > 0.0)";
			else				op << ", vec" << inSize << "(0.0))";
		}
		else if (isInt || isUint)
			op << ")";

		op << ";\n";
	}

	// Result variable.
	{
		const char* outTypeName = getDataTypeName(m_spec.output);
		bool		isBoolOut	= de::inRange<int>(m_spec.output, TYPE_BOOL, TYPE_BOOL_VEC4);

		op << "\t";
		if (precision && !isBoolOut) op << precision << " ";
		op << outTypeName << " res = " << outTypeName << "(0.0);\n\n";
	}

	// Expression.
	op << "\t" << m_shaderOp << "\n\n";

	// Convert to color.
	bool	isResFloatVec	= de::inRange<int>(m_spec.output, TYPE_FLOAT, TYPE_FLOAT_VEC4);
	int		outScalarSize	= getDataTypeScalarSize(m_spec.output);

	op << "\thighp vec4 color = vec4(0.0, 0.0, 0.0, 1.0);\n";
	op << "\tcolor." << s_outSwizzles[outScalarSize-1] << " = ";

	if (!isResFloatVec && outScalarSize == 1)
		op << "float(res)";
	else if (!isResFloatVec)
		op << "vec" << outScalarSize << "(res)";
	else
		op << "res";

	op << ";\n";

	// Scale & bias.
	float	resultScale		= m_spec.resultScale.getValue(m_renderCtx.getFunctions(), shaderType);
	float	resultBias		= m_spec.resultBias.getValue(m_renderCtx.getFunctions(), shaderType);
	if ((resultScale != 1.0f) || (resultBias != 0.0f))
	{
		op << "\tcolor = color";
		if (resultScale != 1.0f) op << " * " << twoValuedVec4(de::toString(resultScale),		"1.0", s_outSwizzleChannelMasks[outScalarSize-1]);
		if (resultBias != 0.0f)  op << " + " << twoValuedVec4(de::floatToString(resultBias, 2),	"0.0", s_outSwizzleChannelMasks[outScalarSize-1]);
		op << ";\n";
	}

	// ..
	if (m_isVertexCase)
	{
		vtx << "	v_color = color;\n";
		frag << "	o_color = v_color;\n";
	}
	else
	{
		for (int i = 0; i < m_spec.numInputs; i++)
		vtx << "	v_in" << i << " = a_in" << i << ";\n";
		frag << "	o_color = color;\n";
	}

	vtx << "}\n";
	frag << "}\n";

	m_vertShaderSource = vtx.str();
	m_fragShaderSource = frag.str();

	// Setup the user attributes.
	m_userAttribTransforms.resize(m_spec.numInputs);
	for (int inputNdx = 0; inputNdx < m_spec.numInputs; inputNdx++)
	{
		const ShaderValue& v = m_spec.inputs[inputNdx];
		DE_ASSERT(v.type != TYPE_LAST);

		float rangeMin	= v.rangeMin.getValue(m_renderCtx.getFunctions(), shaderType);
		float rangeMax	= v.rangeMax.getValue(m_renderCtx.getFunctions(), shaderType);
		float scale		= rangeMax - rangeMin;
		float minBias	= rangeMin;
		float maxBias	= rangeMax;
		Mat4  attribMatrix;

		for (int rowNdx = 0; rowNdx < 4; rowNdx++)
		{
			Vec4 row;

			switch ((rowNdx + inputNdx) % 4)
			{
				case 0:	row = Vec4(scale, 0.0f, 0.0f, minBias);		break;
				case 1:	row = Vec4(0.0f, scale, 0.0f, minBias);		break;
				case 2:	row = Vec4(-scale, 0.0f, 0.0f, maxBias);	break;
				case 3:	row = Vec4(0.0f, -scale, 0.0f, maxBias);	break;
				default: DE_ASSERT(false);
			}

			attribMatrix.setRow(rowNdx, row);
		}

		m_userAttribTransforms[inputNdx] = attribMatrix;
	}
}

ShaderOperatorCase::~ShaderOperatorCase (void)
{
}

// ShaderOperatorTests.

ShaderOperatorTests::ShaderOperatorTests(Context& context)
	: TestCaseGroup(context, "operator", "Operator tests.")
{
}

ShaderOperatorTests::~ShaderOperatorTests (void)
{
}

// Vector math functions.
template<typename T> inline T nop (T f) { return f; }

template <typename T, int Size>
Vector<T, Size> nop (const Vector<T, Size>& v) { return v; }

#define DECLARE_UNARY_GENTYPE_FUNCS(FUNC_NAME)																			\
	void eval_##FUNC_NAME##_float	(ShaderEvalContext& c) { c.color.x()	= FUNC_NAME(c.in[0].swizzle(2)).x(); }		\
	void eval_##FUNC_NAME##_vec2	(ShaderEvalContext& c) { c.color.yz()	= FUNC_NAME(c.in[0].swizzle(3, 1)); }		\
	void eval_##FUNC_NAME##_vec3	(ShaderEvalContext& c) { c.color.xyz()	= FUNC_NAME(c.in[0].swizzle(2, 0, 1)); }	\
	void eval_##FUNC_NAME##_vec4	(ShaderEvalContext& c) { c.color		= FUNC_NAME(c.in[0].swizzle(1, 2, 3, 0)); }

#define DECLARE_BINARY_GENTYPE_FUNCS(FUNC_NAME)																											\
	void eval_##FUNC_NAME##_float	(ShaderEvalContext& c) { c.color.x()	= FUNC_NAME(c.in[0].swizzle(2),          c.in[1].swizzle(0)).x(); }			\
	void eval_##FUNC_NAME##_vec2	(ShaderEvalContext& c) { c.color.yz()	= FUNC_NAME(c.in[0].swizzle(3, 1),       c.in[1].swizzle(1, 0)); }			\
	void eval_##FUNC_NAME##_vec3	(ShaderEvalContext& c) { c.color.xyz()	= FUNC_NAME(c.in[0].swizzle(2, 0, 1),    c.in[1].swizzle(1, 2, 0)); }		\
	void eval_##FUNC_NAME##_vec4	(ShaderEvalContext& c) { c.color		= FUNC_NAME(c.in[0].swizzle(1, 2, 3, 0), c.in[1].swizzle(3, 2, 1, 0)); }

#define DECLARE_TERNARY_GENTYPE_FUNCS(FUNC_NAME)																																	\
	void eval_##FUNC_NAME##_float	(ShaderEvalContext& c) { c.color.x()	= FUNC_NAME(c.in[0].swizzle(2),          c.in[1].swizzle(0),          c.in[2].swizzle(1)).x(); }		\
	void eval_##FUNC_NAME##_vec2	(ShaderEvalContext& c) { c.color.yz()	= FUNC_NAME(c.in[0].swizzle(3, 1),       c.in[1].swizzle(1, 0),       c.in[2].swizzle(2, 1)); }			\
	void eval_##FUNC_NAME##_vec3	(ShaderEvalContext& c) { c.color.xyz()	= FUNC_NAME(c.in[0].swizzle(2, 0, 1),    c.in[1].swizzle(1, 2, 0),    c.in[2].swizzle(3, 1, 2)); }		\
	void eval_##FUNC_NAME##_vec4	(ShaderEvalContext& c) { c.color		= FUNC_NAME(c.in[0].swizzle(1, 2, 3, 0), c.in[1].swizzle(3, 2, 1, 0), c.in[2].swizzle(0, 3, 2, 1)); }

#define DECLARE_UNARY_SCALAR_GENTYPE_FUNCS(FUNC_NAME)																	\
	void eval_##FUNC_NAME##_float	(ShaderEvalContext& c) { c.color.x()	= FUNC_NAME(c.in[0].swizzle(2)); }			\
	void eval_##FUNC_NAME##_vec2	(ShaderEvalContext& c) { c.color.x()	= FUNC_NAME(c.in[0].swizzle(3, 1)); }		\
	void eval_##FUNC_NAME##_vec3	(ShaderEvalContext& c) { c.color.x()	= FUNC_NAME(c.in[0].swizzle(2, 0, 1)); }	\
	void eval_##FUNC_NAME##_vec4	(ShaderEvalContext& c) { c.color.x()	= FUNC_NAME(c.in[0].swizzle(1, 2, 3, 0)); }

#define DECLARE_BINARY_SCALAR_GENTYPE_FUNCS(FUNC_NAME)																									\
	void eval_##FUNC_NAME##_float	(ShaderEvalContext& c) { c.color.x()	= FUNC_NAME(c.in[0].swizzle(2),          c.in[1].swizzle(0)); }				\
	void eval_##FUNC_NAME##_vec2	(ShaderEvalContext& c) { c.color.x()	= FUNC_NAME(c.in[0].swizzle(3, 1),       c.in[1].swizzle(1, 0)); }			\
	void eval_##FUNC_NAME##_vec3	(ShaderEvalContext& c) { c.color.x()	= FUNC_NAME(c.in[0].swizzle(2, 0, 1),    c.in[1].swizzle(1, 2, 0)); }		\
	void eval_##FUNC_NAME##_vec4	(ShaderEvalContext& c) { c.color.x()	= FUNC_NAME(c.in[0].swizzle(1, 2, 3, 0), c.in[1].swizzle(3, 2, 1, 0)); }

#define DECLARE_BINARY_BOOL_FUNCS(FUNC_NAME)																		\
	void eval_##FUNC_NAME##_bool	(ShaderEvalContext& c) { c.color.x()	= (float)FUNC_NAME(c.in[0].z() > 0.0f, c.in[1].x() > 0.0f); }

#define DECLARE_UNARY_BOOL_GENTYPE_FUNCS(FUNC_NAME)																											\
	void eval_##FUNC_NAME##_bool	(ShaderEvalContext& c) { c.color.x()	= (float)FUNC_NAME(c.in[0].z() > 0.0f); }										\
	void eval_##FUNC_NAME##_bvec2	(ShaderEvalContext& c) { c.color.yz()	= FUNC_NAME(greaterThan(c.in[0].swizzle(3, 1), Vec2(0.0f))).asFloat(); }		\
	void eval_##FUNC_NAME##_bvec3	(ShaderEvalContext& c) { c.color.xyz()	= FUNC_NAME(greaterThan(c.in[0].swizzle(2, 0, 1), Vec3(0.0f))).asFloat(); }		\
	void eval_##FUNC_NAME##_bvec4	(ShaderEvalContext& c) { c.color		= FUNC_NAME(greaterThan(c.in[0].swizzle(1, 2, 3, 0), Vec4(0.0f))).asFloat(); }

#define DECLARE_TERNARY_BOOL_GENTYPE_FUNCS(FUNC_NAME)																																																					\
	void eval_##FUNC_NAME##_bool	(ShaderEvalContext& c) { c.color.x()	= (float)FUNC_NAME(c.in[0].z() > 0.0f,                            c.in[1].x() > 0.0f,                                   c.in[2].y() > 0.0f); }												\
	void eval_##FUNC_NAME##_bvec2	(ShaderEvalContext& c) { c.color.yz()	= FUNC_NAME(greaterThan(c.in[0].swizzle(3, 1), Vec2(0.0f)),       greaterThan(c.in[1].swizzle(1, 0), Vec2(0.0f)),       greaterThan(c.in[2].swizzle(2, 1), Vec2(0.0f))).asFloat(); }		\
	void eval_##FUNC_NAME##_bvec3	(ShaderEvalContext& c) { c.color.xyz()	= FUNC_NAME(greaterThan(c.in[0].swizzle(2, 0, 1), Vec3(0.0f)),    greaterThan(c.in[1].swizzle(1, 2, 0), Vec3(0.0f)),    greaterThan(c.in[2].swizzle(3, 1, 2), Vec3(0.0f))).asFloat(); }		\
	void eval_##FUNC_NAME##_bvec4	(ShaderEvalContext& c) { c.color		= FUNC_NAME(greaterThan(c.in[0].swizzle(1, 2, 3, 0), Vec4(0.0f)), greaterThan(c.in[1].swizzle(3, 2, 1, 0), Vec4(0.0f)), greaterThan(c.in[2].swizzle(0, 3, 2, 1), Vec4(0.0f))).asFloat(); }

#define DECLARE_UNARY_INT_GENTYPE_FUNCS(FUNC_NAME)																						\
	void eval_##FUNC_NAME##_int		(ShaderEvalContext& c) { c.color.x()	= (float)FUNC_NAME((int)c.in[0].z()); }						\
	void eval_##FUNC_NAME##_ivec2	(ShaderEvalContext& c) { c.color.yz()	= FUNC_NAME(c.in[0].swizzle(3, 1).asInt()).asFloat(); }		\
	void eval_##FUNC_NAME##_ivec3	(ShaderEvalContext& c) { c.color.xyz()	= FUNC_NAME(c.in[0].swizzle(2, 0, 1).asInt()).asFloat(); }	\
	void eval_##FUNC_NAME##_ivec4	(ShaderEvalContext& c) { c.color		= FUNC_NAME(c.in[0].swizzle(1, 2, 3, 0).asInt()).asFloat(); }

#define DECLARE_BINARY_INT_GENTYPE_FUNCS(FUNC_NAME)																																\
	void eval_##FUNC_NAME##_int		(ShaderEvalContext& c) { c.color.x()	= (float)FUNC_NAME((int)c.in[0].z(),				(int)c.in[1].x()); }							\
	void eval_##FUNC_NAME##_ivec2	(ShaderEvalContext& c) { c.color.yz()	= FUNC_NAME(c.in[0].swizzle(3, 1).asInt(),			c.in[1].swizzle(1, 0).asInt()).asFloat(); }		\
	void eval_##FUNC_NAME##_ivec3	(ShaderEvalContext& c) { c.color.xyz()	= FUNC_NAME(c.in[0].swizzle(2, 0, 1).asInt(),		c.in[1].swizzle(1, 2, 0).asInt()).asFloat(); }	\
	void eval_##FUNC_NAME##_ivec4	(ShaderEvalContext& c) { c.color		= FUNC_NAME(c.in[0].swizzle(1, 2, 3, 0).asInt(),	c.in[1].swizzle(3, 2, 1, 0).asInt()).asFloat(); }

#define DECLARE_UNARY_UINT_GENTYPE_FUNCS(FUNC_NAME)																						\
	void eval_##FUNC_NAME##_uint	(ShaderEvalContext& c) { c.color.x()	= (float)FUNC_NAME((deUint32)c.in[0].z()); }				\
	void eval_##FUNC_NAME##_uvec2	(ShaderEvalContext& c) { c.color.yz()	= FUNC_NAME(c.in[0].swizzle(3, 1).asUint()).asFloat(); }	\
	void eval_##FUNC_NAME##_uvec3	(ShaderEvalContext& c) { c.color.xyz()	= FUNC_NAME(c.in[0].swizzle(2, 0, 1).asUint()).asFloat(); }	\
	void eval_##FUNC_NAME##_uvec4	(ShaderEvalContext& c) { c.color		= FUNC_NAME(c.in[0].swizzle(1, 2, 3, 0).asUint()).asFloat(); }

#define DECLARE_BINARY_UINT_GENTYPE_FUNCS(FUNC_NAME)																															\
	void eval_##FUNC_NAME##_uint	(ShaderEvalContext& c) { c.color.x()	= (float)FUNC_NAME((deUint32)c.in[0].z(),			(deUint32)c.in[1].x()); }						\
	void eval_##FUNC_NAME##_uvec2	(ShaderEvalContext& c) { c.color.yz()	= FUNC_NAME(c.in[0].swizzle(3, 1).asUint(),			c.in[1].swizzle(1, 0).asUint()).asFloat(); }	\
	void eval_##FUNC_NAME##_uvec3	(ShaderEvalContext& c) { c.color.xyz()	= FUNC_NAME(c.in[0].swizzle(2, 0, 1).asUint(),		c.in[1].swizzle(1, 2, 0).asUint()).asFloat(); }	\
	void eval_##FUNC_NAME##_uvec4	(ShaderEvalContext& c) { c.color		= FUNC_NAME(c.in[0].swizzle(1, 2, 3, 0).asUint(),	c.in[1].swizzle(3, 2, 1, 0).asUint()).asFloat(); }

#define DECLARE_TERNARY_INT_GENTYPE_FUNCS(FUNC_NAME)																																								\
	void eval_##FUNC_NAME##_int		(ShaderEvalContext& c) { c.color.x()	= (float)FUNC_NAME((int)c.in[0].z(),				(int)c.in[1].x(),					(int)c.in[2].y()); }							\
	void eval_##FUNC_NAME##_ivec2	(ShaderEvalContext& c) { c.color.yz()	= FUNC_NAME(c.in[0].swizzle(3, 1).asInt(),			c.in[1].swizzle(1, 0).asInt(),       c.in[2].swizzle(2, 1).asInt()).asFloat(); }	\
	void eval_##FUNC_NAME##_ivec3	(ShaderEvalContext& c) { c.color.xyz()	= FUNC_NAME(c.in[0].swizzle(2, 0, 1).asInt(),		c.in[1].swizzle(1, 2, 0).asInt(),    c.in[2].swizzle(3, 1, 2).asInt()).asFloat(); }	\
	void eval_##FUNC_NAME##_ivec4	(ShaderEvalContext& c) { c.color		= FUNC_NAME(c.in[0].swizzle(1, 2, 3, 0).asInt(),	c.in[1].swizzle(3, 2, 1, 0).asInt(), c.in[2].swizzle(0, 3, 2, 1).asInt()).asFloat(); }

#define DECLARE_TERNARY_UINT_GENTYPE_FUNCS(FUNC_NAME)																																									\
	void eval_##FUNC_NAME##_uint	(ShaderEvalContext& c) { c.color.x()	= (float)FUNC_NAME((deUint32)c.in[0].z(),			(deUint32)c.in[1].x(),					(deUint32)c.in[2].y()); }						\
	void eval_##FUNC_NAME##_uvec2	(ShaderEvalContext& c) { c.color.yz()	= FUNC_NAME(c.in[0].swizzle(3, 1).asUint(),			c.in[1].swizzle(1, 0).asUint(),			c.in[2].swizzle(2, 1).asUint()).asFloat(); }	\
	void eval_##FUNC_NAME##_uvec3	(ShaderEvalContext& c) { c.color.xyz()	= FUNC_NAME(c.in[0].swizzle(2, 0, 1).asUint(),		c.in[1].swizzle(1, 2, 0).asUint(),		c.in[2].swizzle(3, 1, 2).asUint()).asFloat(); }	\
	void eval_##FUNC_NAME##_uvec4	(ShaderEvalContext& c) { c.color		= FUNC_NAME(c.in[0].swizzle(1, 2, 3, 0).asUint(),	c.in[1].swizzle(3, 2, 1, 0).asUint(),	c.in[2].swizzle(0, 3, 2, 1).asUint()).asFloat(); }

#define DECLARE_VEC_FLOAT_FUNCS(FUNC_NAME)																								\
	void eval_##FUNC_NAME##_vec2	(ShaderEvalContext& c) { c.color.yz()	= FUNC_NAME(c.in[0].swizzle(3, 1),			c.in[1].x()); } \
	void eval_##FUNC_NAME##_vec3	(ShaderEvalContext& c) { c.color.xyz()	= FUNC_NAME(c.in[0].swizzle(2, 0, 1),		c.in[1].x()); } \
	void eval_##FUNC_NAME##_vec4	(ShaderEvalContext& c) { c.color		= FUNC_NAME(c.in[0].swizzle(1, 2, 3, 0),	c.in[1].x()); }

#define DECLARE_VEC_FLOAT_FLOAT_FUNCS(FUNC_NAME) \
	void eval_##FUNC_NAME##_vec2	(ShaderEvalContext& c) { c.color.yz()	= FUNC_NAME(c.in[0].swizzle(3, 1),			c.in[1].x(), c.in[2].y()); } \
	void eval_##FUNC_NAME##_vec3	(ShaderEvalContext& c) { c.color.xyz()	= FUNC_NAME(c.in[0].swizzle(2, 0, 1),		c.in[1].x(), c.in[2].y()); } \
	void eval_##FUNC_NAME##_vec4	(ShaderEvalContext& c) { c.color		= FUNC_NAME(c.in[0].swizzle(1, 2, 3, 0),	c.in[1].x(), c.in[2].y()); }

#define DECLARE_VEC_VEC_FLOAT_FUNCS(FUNC_NAME) \
	void eval_##FUNC_NAME##_vec2	(ShaderEvalContext& c) { c.color.yz()	= FUNC_NAME(c.in[0].swizzle(3, 1),			c.in[1].swizzle(1, 0),			c.in[2].y()); } \
	void eval_##FUNC_NAME##_vec3	(ShaderEvalContext& c) { c.color.xyz()	= FUNC_NAME(c.in[0].swizzle(2, 0, 1),		c.in[1].swizzle(1, 2, 0),		c.in[2].y()); } \
	void eval_##FUNC_NAME##_vec4	(ShaderEvalContext& c) { c.color		= FUNC_NAME(c.in[0].swizzle(1, 2, 3, 0),	c.in[1].swizzle(3, 2, 1, 0),	c.in[2].y()); }

#define DECLARE_FLOAT_FLOAT_VEC_FUNCS(FUNC_NAME) \
	void eval_##FUNC_NAME##_vec2	(ShaderEvalContext& c) { c.color.yz()	= FUNC_NAME(c.in[0].z(), c.in[1].x(), c.in[2].swizzle(2, 1)); }			\
	void eval_##FUNC_NAME##_vec3	(ShaderEvalContext& c) { c.color.xyz()	= FUNC_NAME(c.in[0].z(), c.in[1].x(), c.in[2].swizzle(3, 1, 2)); }		\
	void eval_##FUNC_NAME##_vec4	(ShaderEvalContext& c) { c.color		= FUNC_NAME(c.in[0].z(), c.in[1].x(), c.in[2].swizzle(0, 3, 2, 1)); }

#define DECLARE_FLOAT_VEC_FUNCS(FUNC_NAME)																												\
	void eval_##FUNC_NAME##_vec2	(ShaderEvalContext& c) { c.color.yz()	= FUNC_NAME(c.in[0].z(),					c.in[1].swizzle(1, 0)); }		\
	void eval_##FUNC_NAME##_vec3	(ShaderEvalContext& c) { c.color.xyz()	= FUNC_NAME(c.in[0].z(),					c.in[1].swizzle(1, 2, 0)); }	\
	void eval_##FUNC_NAME##_vec4	(ShaderEvalContext& c) { c.color		= FUNC_NAME(c.in[0].z(),					c.in[1].swizzle(3, 2, 1, 0)); }

#define DECLARE_IVEC_INT_FUNCS(FUNC_NAME)																														\
	void eval_##FUNC_NAME##_ivec2	(ShaderEvalContext& c) { c.color.yz()	= FUNC_NAME(c.in[0].swizzle(3, 1).asInt(),			(int)c.in[1].x()).asFloat(); }	\
	void eval_##FUNC_NAME##_ivec3	(ShaderEvalContext& c) { c.color.xyz()	= FUNC_NAME(c.in[0].swizzle(2, 0, 1).asInt(),		(int)c.in[1].x()).asFloat(); }	\
	void eval_##FUNC_NAME##_ivec4	(ShaderEvalContext& c) { c.color		= FUNC_NAME(c.in[0].swizzle(1, 2, 3, 0).asInt(),	(int)c.in[1].x()).asFloat(); }

#define DECLARE_IVEC_INT_INT_FUNCS(FUNC_NAME) \
	void eval_##FUNC_NAME##_ivec2	(ShaderEvalContext& c) { c.color.yz()	= FUNC_NAME(c.in[0].swizzle(3, 1).asInt(),			(int)c.in[1].x(), (int)c.in[2].y()).asFloat(); } \
	void eval_##FUNC_NAME##_ivec3	(ShaderEvalContext& c) { c.color.xyz()	= FUNC_NAME(c.in[0].swizzle(2, 0, 1).asInt(),		(int)c.in[1].x(), (int)c.in[2].y()).asFloat(); } \
	void eval_##FUNC_NAME##_ivec4	(ShaderEvalContext& c) { c.color		= FUNC_NAME(c.in[0].swizzle(1, 2, 3, 0).asInt(),	(int)c.in[1].x(), (int)c.in[2].y()).asFloat(); }

#define DECLARE_INT_IVEC_FUNCS(FUNC_NAME)																																	\
	void eval_##FUNC_NAME##_ivec2	(ShaderEvalContext& c) { c.color.yz()	= FUNC_NAME((int)c.in[0].z(),					c.in[1].swizzle(1, 0).asInt()).asFloat(); }		\
	void eval_##FUNC_NAME##_ivec3	(ShaderEvalContext& c) { c.color.xyz()	= FUNC_NAME((int)c.in[0].z(),					c.in[1].swizzle(1, 2, 0).asInt()).asFloat(); }	\
	void eval_##FUNC_NAME##_ivec4	(ShaderEvalContext& c) { c.color		= FUNC_NAME((int)c.in[0].z(),					c.in[1].swizzle(3, 2, 1, 0).asInt()).asFloat(); }

#define DECLARE_UVEC_UINT_FUNCS(FUNC_NAME)																															\
	void eval_##FUNC_NAME##_uvec2	(ShaderEvalContext& c) { c.color.yz()	= FUNC_NAME(c.in[0].swizzle(3, 1).asUint(),			(deUint32)c.in[1].x()).asFloat(); }	\
	void eval_##FUNC_NAME##_uvec3	(ShaderEvalContext& c) { c.color.xyz()	= FUNC_NAME(c.in[0].swizzle(2, 0, 1).asUint(),		(deUint32)c.in[1].x()).asFloat(); }	\
	void eval_##FUNC_NAME##_uvec4	(ShaderEvalContext& c) { c.color		= FUNC_NAME(c.in[0].swizzle(1, 2, 3, 0).asUint(),	(deUint32)c.in[1].x()).asFloat(); }

#define DECLARE_UVEC_UINT_UINT_FUNCS(FUNC_NAME) \
	void eval_##FUNC_NAME##_uvec2	(ShaderEvalContext& c) { c.color.yz()	= FUNC_NAME(c.in[0].swizzle(3, 1).asUint(),			(deUint32)c.in[1].x(), (deUint32)c.in[2].y()).asFloat(); } \
	void eval_##FUNC_NAME##_uvec3	(ShaderEvalContext& c) { c.color.xyz()	= FUNC_NAME(c.in[0].swizzle(2, 0, 1).asUint(),		(deUint32)c.in[1].x(), (deUint32)c.in[2].y()).asFloat(); } \
	void eval_##FUNC_NAME##_uvec4	(ShaderEvalContext& c) { c.color		= FUNC_NAME(c.in[0].swizzle(1, 2, 3, 0).asUint(),	(deUint32)c.in[1].x(), (deUint32)c.in[2].y()).asFloat(); }

#define DECLARE_UINT_UVEC_FUNCS(FUNC_NAME)																																		\
	void eval_##FUNC_NAME##_uvec2	(ShaderEvalContext& c) { c.color.yz()	= FUNC_NAME((deUint32)c.in[0].z(),					c.in[1].swizzle(1, 0).asUint()).asFloat(); }	\
	void eval_##FUNC_NAME##_uvec3	(ShaderEvalContext& c) { c.color.xyz()	= FUNC_NAME((deUint32)c.in[0].z(),					c.in[1].swizzle(1, 2, 0).asUint()).asFloat(); }	\
	void eval_##FUNC_NAME##_uvec4	(ShaderEvalContext& c) { c.color		= FUNC_NAME((deUint32)c.in[0].z(),					c.in[1].swizzle(3, 2, 1, 0).asUint()).asFloat(); }

#define DECLARE_BINARY_INT_VEC_FUNCS(FUNC_NAME)																																	\
	void eval_##FUNC_NAME##_ivec2	(ShaderEvalContext& c) { c.color.yz()	= FUNC_NAME(c.in[0].swizzle(3, 1).asInt(),			c.in[1].swizzle(1, 0).asInt()).asFloat(); }		\
	void eval_##FUNC_NAME##_ivec3	(ShaderEvalContext& c) { c.color.xyz()	= FUNC_NAME(c.in[0].swizzle(2, 0, 1).asInt(),		c.in[1].swizzle(1, 2, 0).asInt()).asFloat(); }	\
	void eval_##FUNC_NAME##_ivec4	(ShaderEvalContext& c) { c.color		= FUNC_NAME(c.in[0].swizzle(1, 2, 3, 0).asInt(),	c.in[1].swizzle(3, 2, 1, 0).asInt()).asFloat(); }

#define DECLARE_BINARY_UINT_VEC_FUNCS(FUNC_NAME)																																\
	void eval_##FUNC_NAME##_uvec2	(ShaderEvalContext& c) { c.color.yz()	= FUNC_NAME(c.in[0].swizzle(3, 1).asUint(),			c.in[1].swizzle(1, 0).asUint()).asFloat(); }	\
	void eval_##FUNC_NAME##_uvec3	(ShaderEvalContext& c) { c.color.xyz()	= FUNC_NAME(c.in[0].swizzle(2, 0, 1).asUint(),		c.in[1].swizzle(1, 2, 0).asUint()).asFloat(); }	\
	void eval_##FUNC_NAME##_uvec4	(ShaderEvalContext& c) { c.color		= FUNC_NAME(c.in[0].swizzle(1, 2, 3, 0).asUint(),	c.in[1].swizzle(3, 2, 1, 0).asUint()).asFloat(); }

#define DECLARE_UINT_INT_GENTYPE_FUNCS(FUNC_NAME)																																\
	void eval_##FUNC_NAME##_uint	(ShaderEvalContext& c) { c.color.x()	= (float)FUNC_NAME((deUint32)c.in[0].z(),			(int)c.in[1].x()); }							\
	void eval_##FUNC_NAME##_uvec2	(ShaderEvalContext& c) { c.color.yz()	= FUNC_NAME(c.in[0].swizzle(3, 1).asUint(),			c.in[1].swizzle(1, 0).asInt()).asFloat(); }		\
	void eval_##FUNC_NAME##_uvec3	(ShaderEvalContext& c) { c.color.xyz()	= FUNC_NAME(c.in[0].swizzle(2, 0, 1).asUint(),		c.in[1].swizzle(1, 2, 0).asInt()).asFloat(); }	\
	void eval_##FUNC_NAME##_uvec4	(ShaderEvalContext& c) { c.color		= FUNC_NAME(c.in[0].swizzle(1, 2, 3, 0).asUint(),	c.in[1].swizzle(3, 2, 1, 0).asInt()).asFloat(); }

#define DECLARE_UVEC_INT_FUNCS(FUNC_NAME)																														\
	void eval_##FUNC_NAME##_uvec2	(ShaderEvalContext& c) { c.color.yz()	= FUNC_NAME(c.in[0].swizzle(3, 1).asUint(),			(int)c.in[1].x()).asFloat(); }	\
	void eval_##FUNC_NAME##_uvec3	(ShaderEvalContext& c) { c.color.xyz()	= FUNC_NAME(c.in[0].swizzle(2, 0, 1).asUint(),		(int)c.in[1].x()).asFloat(); }	\
	void eval_##FUNC_NAME##_uvec4	(ShaderEvalContext& c) { c.color		= FUNC_NAME(c.in[0].swizzle(1, 2, 3, 0).asUint(),	(int)c.in[1].x()).asFloat(); }


// Operators.

DECLARE_UNARY_GENTYPE_FUNCS(nop)
DECLARE_UNARY_GENTYPE_FUNCS(negate)
DECLARE_UNARY_GENTYPE_FUNCS(addOne)
DECLARE_UNARY_GENTYPE_FUNCS(subOne)
DECLARE_BINARY_GENTYPE_FUNCS(add)
DECLARE_BINARY_GENTYPE_FUNCS(sub)
DECLARE_BINARY_GENTYPE_FUNCS(mul)
DECLARE_BINARY_GENTYPE_FUNCS(div)

void eval_selection_float	(ShaderEvalContext& c) { c.color.x()	= selection(c.in[0].z() > 0.0f,		c.in[1].x(),					c.in[2].y()); }
void eval_selection_vec2	(ShaderEvalContext& c) { c.color.yz()	= selection(c.in[0].z() > 0.0f,		c.in[1].swizzle(1, 0),			c.in[2].swizzle(2, 1)); }
void eval_selection_vec3	(ShaderEvalContext& c) { c.color.xyz()	= selection(c.in[0].z() > 0.0f,		c.in[1].swizzle(1, 2, 0),		c.in[2].swizzle(3, 1, 2)); }
void eval_selection_vec4	(ShaderEvalContext& c) { c.color		= selection(c.in[0].z() > 0.0f,		c.in[1].swizzle(3, 2, 1, 0),	c.in[2].swizzle(0, 3, 2, 1)); }

DECLARE_UNARY_INT_GENTYPE_FUNCS(nop)
DECLARE_UNARY_INT_GENTYPE_FUNCS(negate)
DECLARE_UNARY_INT_GENTYPE_FUNCS(addOne)
DECLARE_UNARY_INT_GENTYPE_FUNCS(subOne)
DECLARE_UNARY_INT_GENTYPE_FUNCS(bitwiseNot)
DECLARE_BINARY_INT_GENTYPE_FUNCS(add)
DECLARE_BINARY_INT_GENTYPE_FUNCS(sub)
DECLARE_BINARY_INT_GENTYPE_FUNCS(mul)
DECLARE_BINARY_INT_GENTYPE_FUNCS(div)
DECLARE_BINARY_INT_GENTYPE_FUNCS(mod)
DECLARE_BINARY_INT_GENTYPE_FUNCS(bitwiseAnd)
DECLARE_BINARY_INT_GENTYPE_FUNCS(bitwiseOr)
DECLARE_BINARY_INT_GENTYPE_FUNCS(bitwiseXor)

void eval_leftShift_int		(ShaderEvalContext& c) { c.color.x()	= (float)leftShift((int)c.in[0].z(),				(int)c.in[1].x()); }
DECLARE_BINARY_INT_VEC_FUNCS(leftShift)
void eval_rightShift_int	(ShaderEvalContext& c) { c.color.x()	= (float)rightShift((int)c.in[0].z(),				(int)c.in[1].x()); }
DECLARE_BINARY_INT_VEC_FUNCS(rightShift)
DECLARE_IVEC_INT_FUNCS(leftShiftVecScalar)
DECLARE_IVEC_INT_FUNCS(rightShiftVecScalar)

void eval_selection_int		(ShaderEvalContext& c) { c.color.x()	= (float)selection(c.in[0].z() > 0.0f,	(int)c.in[1].x(),						(int)c.in[2].y()); }
void eval_selection_ivec2	(ShaderEvalContext& c) { c.color.yz()	= selection(c.in[0].z() > 0.0f,			c.in[1].swizzle(1, 0).asInt(),			c.in[2].swizzle(2, 1).asInt()).asFloat(); }
void eval_selection_ivec3	(ShaderEvalContext& c) { c.color.xyz()	= selection(c.in[0].z() > 0.0f,			c.in[1].swizzle(1, 2, 0).asInt(),		c.in[2].swizzle(3, 1, 2).asInt()).asFloat(); }
void eval_selection_ivec4	(ShaderEvalContext& c) { c.color		= selection(c.in[0].z() > 0.0f,			c.in[1].swizzle(3, 2, 1, 0).asInt(),	c.in[2].swizzle(0, 3, 2, 1).asInt()).asFloat(); }

DECLARE_UNARY_UINT_GENTYPE_FUNCS(nop)
DECLARE_UNARY_UINT_GENTYPE_FUNCS(negate)
DECLARE_UNARY_UINT_GENTYPE_FUNCS(bitwiseNot)
DECLARE_UNARY_UINT_GENTYPE_FUNCS(addOne)
DECLARE_UNARY_UINT_GENTYPE_FUNCS(subOne)
DECLARE_BINARY_UINT_GENTYPE_FUNCS(add)
DECLARE_BINARY_UINT_GENTYPE_FUNCS(sub)
DECLARE_BINARY_UINT_GENTYPE_FUNCS(mul)
DECLARE_BINARY_UINT_GENTYPE_FUNCS(div)
DECLARE_BINARY_UINT_GENTYPE_FUNCS(mod)
DECLARE_BINARY_UINT_GENTYPE_FUNCS(bitwiseAnd)
DECLARE_BINARY_UINT_GENTYPE_FUNCS(bitwiseOr)
DECLARE_BINARY_UINT_GENTYPE_FUNCS(bitwiseXor)

DECLARE_UINT_INT_GENTYPE_FUNCS(leftShift)
DECLARE_UINT_INT_GENTYPE_FUNCS(rightShift)
DECLARE_UVEC_INT_FUNCS(leftShiftVecScalar)
DECLARE_UVEC_INT_FUNCS(rightShiftVecScalar)

void eval_selection_uint	(ShaderEvalContext& c) { c.color.x()	= (float)selection(c.in[0].z() > 0.0f,	(deUint32)c.in[1].x(),					(deUint32)c.in[2].y()); }
void eval_selection_uvec2	(ShaderEvalContext& c) { c.color.yz()	= selection(c.in[0].z() > 0.0f,			c.in[1].swizzle(1, 0).asUint(),			c.in[2].swizzle(2, 1).asUint()).asFloat(); }
void eval_selection_uvec3	(ShaderEvalContext& c) { c.color.xyz()	= selection(c.in[0].z() > 0.0f,			c.in[1].swizzle(1, 2, 0).asUint(),		c.in[2].swizzle(3, 1, 2).asUint()).asFloat(); }
void eval_selection_uvec4	(ShaderEvalContext& c) { c.color		= selection(c.in[0].z() > 0.0f,			c.in[1].swizzle(3, 2, 1, 0).asUint(),	c.in[2].swizzle(0, 3, 2, 1).asUint()).asFloat(); }

DECLARE_UNARY_BOOL_GENTYPE_FUNCS(boolNot)
DECLARE_BINARY_BOOL_FUNCS(logicalAnd)
DECLARE_BINARY_BOOL_FUNCS(logicalOr)
DECLARE_BINARY_BOOL_FUNCS(logicalXor)

void eval_selection_bool	(ShaderEvalContext& c) { c.color.x()	= (float)selection(c.in[0].z() > 0.0f,	c.in[1].x() > 0.0f,														c.in[2].y() > 0.0f); }
void eval_selection_bvec2	(ShaderEvalContext& c) { c.color.yz()	= selection(c.in[0].z() > 0.0f,			greaterThan(c.in[1].swizzle(1, 0), Vec2(0.0f, 0.0f)),					greaterThan(c.in[2].swizzle(2, 1), Vec2(0.0f, 0.0f))).asFloat(); }
void eval_selection_bvec3	(ShaderEvalContext& c) { c.color.xyz()	= selection(c.in[0].z() > 0.0f,			greaterThan(c.in[1].swizzle(1, 2, 0), Vec3(0.0f, 0.0f, 0.0f)),			greaterThan(c.in[2].swizzle(3, 1, 2), Vec3(0.0f, 0.0f, 0.0f))).asFloat(); }
void eval_selection_bvec4	(ShaderEvalContext& c) { c.color		= selection(c.in[0].z() > 0.0f,			greaterThan(c.in[1].swizzle(3, 2, 1, 0), Vec4(0.0f, 0.0f, 0.0f, 0.0f)),	greaterThan(c.in[2].swizzle(0, 3, 2, 1), Vec4(0.0f, 0.0f, 0.0f, 0.0f))).asFloat(); }

DECLARE_VEC_FLOAT_FUNCS(addVecScalar)
DECLARE_VEC_FLOAT_FUNCS(subVecScalar)
DECLARE_VEC_FLOAT_FUNCS(mulVecScalar)
DECLARE_VEC_FLOAT_FUNCS(divVecScalar)

DECLARE_FLOAT_VEC_FUNCS(addScalarVec)
DECLARE_FLOAT_VEC_FUNCS(subScalarVec)
DECLARE_FLOAT_VEC_FUNCS(mulScalarVec)
DECLARE_FLOAT_VEC_FUNCS(divScalarVec)

DECLARE_IVEC_INT_FUNCS(addVecScalar)
DECLARE_IVEC_INT_FUNCS(subVecScalar)
DECLARE_IVEC_INT_FUNCS(mulVecScalar)
DECLARE_IVEC_INT_FUNCS(divVecScalar)
DECLARE_IVEC_INT_FUNCS(modVecScalar)
DECLARE_IVEC_INT_FUNCS(bitwiseAndVecScalar)
DECLARE_IVEC_INT_FUNCS(bitwiseOrVecScalar)
DECLARE_IVEC_INT_FUNCS(bitwiseXorVecScalar)

DECLARE_INT_IVEC_FUNCS(addScalarVec)
DECLARE_INT_IVEC_FUNCS(subScalarVec)
DECLARE_INT_IVEC_FUNCS(mulScalarVec)
DECLARE_INT_IVEC_FUNCS(divScalarVec)
DECLARE_INT_IVEC_FUNCS(modScalarVec)
DECLARE_INT_IVEC_FUNCS(bitwiseAndScalarVec)
DECLARE_INT_IVEC_FUNCS(bitwiseOrScalarVec)
DECLARE_INT_IVEC_FUNCS(bitwiseXorScalarVec)

DECLARE_UVEC_UINT_FUNCS(addVecScalar)
DECLARE_UVEC_UINT_FUNCS(subVecScalar)
DECLARE_UVEC_UINT_FUNCS(mulVecScalar)
DECLARE_UVEC_UINT_FUNCS(divVecScalar)
DECLARE_UVEC_UINT_FUNCS(modVecScalar)
DECLARE_UVEC_UINT_FUNCS(bitwiseAndVecScalar)
DECLARE_UVEC_UINT_FUNCS(bitwiseOrVecScalar)
DECLARE_UVEC_UINT_FUNCS(bitwiseXorVecScalar)

DECLARE_UINT_UVEC_FUNCS(addScalarVec)
DECLARE_UINT_UVEC_FUNCS(subScalarVec)
DECLARE_UINT_UVEC_FUNCS(mulScalarVec)
DECLARE_UINT_UVEC_FUNCS(divScalarVec)
DECLARE_UINT_UVEC_FUNCS(modScalarVec)
DECLARE_UINT_UVEC_FUNCS(bitwiseAndScalarVec)
DECLARE_UINT_UVEC_FUNCS(bitwiseOrScalarVec)
DECLARE_UINT_UVEC_FUNCS(bitwiseXorScalarVec)

// Built-in functions.

DECLARE_UNARY_GENTYPE_FUNCS(radians)
DECLARE_UNARY_GENTYPE_FUNCS(degrees)
DECLARE_UNARY_GENTYPE_FUNCS(sin)
DECLARE_UNARY_GENTYPE_FUNCS(cos)
DECLARE_UNARY_GENTYPE_FUNCS(tan)
DECLARE_UNARY_GENTYPE_FUNCS(asin)
DECLARE_UNARY_GENTYPE_FUNCS(acos)
DECLARE_UNARY_GENTYPE_FUNCS(atan)
DECLARE_BINARY_GENTYPE_FUNCS(atan2)
DECLARE_UNARY_GENTYPE_FUNCS(sinh)
DECLARE_UNARY_GENTYPE_FUNCS(cosh)
DECLARE_UNARY_GENTYPE_FUNCS(tanh)
DECLARE_UNARY_GENTYPE_FUNCS(asinh)
DECLARE_UNARY_GENTYPE_FUNCS(acosh)
DECLARE_UNARY_GENTYPE_FUNCS(atanh)

DECLARE_BINARY_GENTYPE_FUNCS(pow)
DECLARE_UNARY_GENTYPE_FUNCS(exp)
DECLARE_UNARY_GENTYPE_FUNCS(log)
DECLARE_UNARY_GENTYPE_FUNCS(exp2)
DECLARE_UNARY_GENTYPE_FUNCS(log2)
DECLARE_UNARY_GENTYPE_FUNCS(sqrt)
DECLARE_UNARY_GENTYPE_FUNCS(inverseSqrt)

DECLARE_UNARY_GENTYPE_FUNCS(abs)
DECLARE_UNARY_GENTYPE_FUNCS(sign)
DECLARE_UNARY_GENTYPE_FUNCS(floor)
DECLARE_UNARY_GENTYPE_FUNCS(trunc)
DECLARE_UNARY_GENTYPE_FUNCS(roundToEven)
DECLARE_UNARY_GENTYPE_FUNCS(ceil)
DECLARE_UNARY_GENTYPE_FUNCS(fract)
DECLARE_BINARY_GENTYPE_FUNCS(mod)
DECLARE_VEC_FLOAT_FUNCS(modVecScalar)
DECLARE_BINARY_GENTYPE_FUNCS(min)
DECLARE_VEC_FLOAT_FUNCS(minVecScalar)
DECLARE_BINARY_INT_GENTYPE_FUNCS(min)
DECLARE_IVEC_INT_FUNCS(minVecScalar)
DECLARE_BINARY_UINT_GENTYPE_FUNCS(min)
DECLARE_UVEC_UINT_FUNCS(minVecScalar)
DECLARE_BINARY_GENTYPE_FUNCS(max)
DECLARE_VEC_FLOAT_FUNCS(maxVecScalar)
DECLARE_BINARY_INT_GENTYPE_FUNCS(max)
DECLARE_IVEC_INT_FUNCS(maxVecScalar)
DECLARE_BINARY_UINT_GENTYPE_FUNCS(max)
DECLARE_UVEC_UINT_FUNCS(maxVecScalar)
DECLARE_TERNARY_GENTYPE_FUNCS(clamp)
DECLARE_VEC_FLOAT_FLOAT_FUNCS(clampVecScalarScalar)
DECLARE_TERNARY_INT_GENTYPE_FUNCS(clamp)
DECLARE_IVEC_INT_INT_FUNCS(clampVecScalarScalar)
DECLARE_TERNARY_UINT_GENTYPE_FUNCS(clamp)
DECLARE_UVEC_UINT_UINT_FUNCS(clampVecScalarScalar)
DECLARE_TERNARY_GENTYPE_FUNCS(mix)
DECLARE_VEC_VEC_FLOAT_FUNCS(mixVecVecScalar)
DECLARE_BINARY_GENTYPE_FUNCS(step)
DECLARE_FLOAT_VEC_FUNCS(stepScalarVec)
DECLARE_TERNARY_GENTYPE_FUNCS(smoothStep)
DECLARE_FLOAT_FLOAT_VEC_FUNCS(smoothStepScalarScalarVec)

DECLARE_UNARY_SCALAR_GENTYPE_FUNCS(length)
DECLARE_BINARY_SCALAR_GENTYPE_FUNCS(distance)
DECLARE_BINARY_SCALAR_GENTYPE_FUNCS(dot)
void eval_cross_vec3 (ShaderEvalContext& c) { c.color.xyz()	= cross(c.in[0].swizzle(2, 0, 1), c.in[1].swizzle(1, 2, 0)); }

DECLARE_UNARY_GENTYPE_FUNCS(normalize)
DECLARE_TERNARY_GENTYPE_FUNCS(faceForward)
DECLARE_BINARY_GENTYPE_FUNCS(reflect)

void eval_refract_float	(ShaderEvalContext& c) { c.color.x()	= refract(c.in[0].z(),                 c.in[1].x(),                 c.in[2].y()); }
void eval_refract_vec2	(ShaderEvalContext& c) { c.color.yz()	= refract(c.in[0].swizzle(3, 1),       c.in[1].swizzle(1, 0),       c.in[2].y()); }
void eval_refract_vec3	(ShaderEvalContext& c) { c.color.xyz()	= refract(c.in[0].swizzle(2, 0, 1),    c.in[1].swizzle(1, 2, 0),    c.in[2].y()); }
void eval_refract_vec4	(ShaderEvalContext& c) { c.color		= refract(c.in[0].swizzle(1, 2, 3, 0), c.in[1].swizzle(3, 2, 1, 0), c.in[2].y()); }

// Compare functions.

#define DECLARE_FLOAT_COMPARE_FUNCS(FUNC_NAME)																											\
	void eval_##FUNC_NAME##_float	(ShaderEvalContext& c) { c.color.x() = (float)FUNC_NAME(c.in[0].z(),          c.in[1].x()); }						\
	void eval_##FUNC_NAME##_vec2	(ShaderEvalContext& c) { c.color.x() = (float)FUNC_NAME(c.in[0].swizzle(3, 1),       c.in[1].swizzle(1, 0)); }		\
	void eval_##FUNC_NAME##_vec3	(ShaderEvalContext& c) { c.color.x() = (float)FUNC_NAME(c.in[0].swizzle(2, 0, 1),    c.in[1].swizzle(1, 2, 0)); }	\
	void eval_##FUNC_NAME##_vec4	(ShaderEvalContext& c) { c.color.x() = (float)FUNC_NAME(c.in[0].swizzle(1, 2, 3, 0), c.in[1].swizzle(3, 2, 1, 0)); }

#define DECLARE_FLOAT_CWISE_COMPARE_FUNCS(FUNC_NAME)																											\
	void eval_##FUNC_NAME##_float	(ShaderEvalContext& c) { c.color.x()	= (float)FUNC_NAME(c.in[0].z(),          c.in[1].x()); }							\
	void eval_##FUNC_NAME##_vec2	(ShaderEvalContext& c) { c.color.yz()	= FUNC_NAME(c.in[0].swizzle(3, 1),       c.in[1].swizzle(1, 0)).asFloat(); }		\
	void eval_##FUNC_NAME##_vec3	(ShaderEvalContext& c) { c.color.xyz()	= FUNC_NAME(c.in[0].swizzle(2, 0, 1),    c.in[1].swizzle(1, 2, 0)).asFloat(); }		\
	void eval_##FUNC_NAME##_vec4	(ShaderEvalContext& c) { c.color		= FUNC_NAME(c.in[0].swizzle(1, 2, 3, 0), c.in[1].swizzle(3, 2, 1, 0)).asFloat(); }

#define DECLARE_INT_COMPARE_FUNCS(FUNC_NAME)																																	\
	void eval_##FUNC_NAME##_int		(ShaderEvalContext& c) { c.color.x() = (float)FUNC_NAME(chopToInt(c.in[0].z()), chopToInt(c.in[1].x())); }									\
	void eval_##FUNC_NAME##_ivec2	(ShaderEvalContext& c) { c.color.x() = (float)FUNC_NAME(chopToInt(c.in[0].swizzle(3, 1)),       chopToInt(c.in[1].swizzle(1, 0))); }		\
	void eval_##FUNC_NAME##_ivec3	(ShaderEvalContext& c) { c.color.x() = (float)FUNC_NAME(chopToInt(c.in[0].swizzle(2, 0, 1)),    chopToInt(c.in[1].swizzle(1, 2, 0))); }		\
	void eval_##FUNC_NAME##_ivec4	(ShaderEvalContext& c) { c.color.x() = (float)FUNC_NAME(chopToInt(c.in[0].swizzle(1, 2, 3, 0)), chopToInt(c.in[1].swizzle(3, 2, 1, 0))); }

#define DECLARE_INT_CWISE_COMPARE_FUNCS(FUNC_NAME)																																	\
	void eval_##FUNC_NAME##_int		(ShaderEvalContext& c) { c.color.x()	= (float)FUNC_NAME(chopToInt(c.in[0].z()), chopToInt(c.in[1].x())); }									\
	void eval_##FUNC_NAME##_ivec2	(ShaderEvalContext& c) { c.color.yz()	= FUNC_NAME(chopToInt(c.in[0].swizzle(3, 1)),       chopToInt(c.in[1].swizzle(1, 0))).asFloat(); }		\
	void eval_##FUNC_NAME##_ivec3	(ShaderEvalContext& c) { c.color.xyz()	= FUNC_NAME(chopToInt(c.in[0].swizzle(2, 0, 1)),    chopToInt(c.in[1].swizzle(1, 2, 0))).asFloat(); }	\
	void eval_##FUNC_NAME##_ivec4	(ShaderEvalContext& c) { c.color		= FUNC_NAME(chopToInt(c.in[0].swizzle(1, 2, 3, 0)), chopToInt(c.in[1].swizzle(3, 2, 1, 0))).asFloat(); }

#define DECLARE_UINT_COMPARE_FUNCS(FUNC_NAME)																																\
	void eval_##FUNC_NAME##_uint	(ShaderEvalContext& c) { c.color.x() = (float)FUNC_NAME((deUint32)c.in[0].z(), (deUint32)c.in[1].x()); }								\
	void eval_##FUNC_NAME##_uvec2	(ShaderEvalContext& c) { c.color.x() = (float)FUNC_NAME(c.in[0].swizzle(3, 1).asUint(),       c.in[1].swizzle(1, 0).asUint()); }		\
	void eval_##FUNC_NAME##_uvec3	(ShaderEvalContext& c) { c.color.x() = (float)FUNC_NAME(c.in[0].swizzle(2, 0, 1).asUint(),    c.in[1].swizzle(1, 2, 0).asUint()); }		\
	void eval_##FUNC_NAME##_uvec4	(ShaderEvalContext& c) { c.color.x() = (float)FUNC_NAME(c.in[0].swizzle(1, 2, 3, 0).asUint(), c.in[1].swizzle(3, 2, 1, 0).asUint()); }

#define DECLARE_UINT_CWISE_COMPARE_FUNCS(FUNC_NAME)																																\
	void eval_##FUNC_NAME##_uint	(ShaderEvalContext& c) { c.color.x()	= (float)FUNC_NAME((deUint32)c.in[0].z(), (deUint32)c.in[1].x()); }									\
	void eval_##FUNC_NAME##_uvec2	(ShaderEvalContext& c) { c.color.yz()	= FUNC_NAME(c.in[0].swizzle(3, 1).asUint(),       c.in[1].swizzle(1, 0).asUint()).asFloat(); }		\
	void eval_##FUNC_NAME##_uvec3	(ShaderEvalContext& c) { c.color.xyz()	= FUNC_NAME(c.in[0].swizzle(2, 0, 1).asUint(),    c.in[1].swizzle(1, 2, 0).asUint()).asFloat(); }	\
	void eval_##FUNC_NAME##_uvec4	(ShaderEvalContext& c) { c.color		= FUNC_NAME(c.in[0].swizzle(1, 2, 3, 0).asUint(), c.in[1].swizzle(3, 2, 1, 0).asUint()).asFloat(); }

#define DECLARE_BOOL_COMPARE_FUNCS(FUNC_NAME)																																								\
	void eval_##FUNC_NAME##_bool	(ShaderEvalContext& c) { c.color.x() = (float)FUNC_NAME(c.in[0].z() > 0.0f, c.in[1].x() > 0.0f); }																		\
	void eval_##FUNC_NAME##_bvec2	(ShaderEvalContext& c) { c.color.x() = (float)FUNC_NAME(greaterThan(c.in[0].swizzle(3, 1), Vec2(0.0f)),       greaterThan(c.in[1].swizzle(1, 0), Vec2(0.0f))); }		\
	void eval_##FUNC_NAME##_bvec3	(ShaderEvalContext& c) { c.color.x() = (float)FUNC_NAME(greaterThan(c.in[0].swizzle(2, 0, 1), Vec3(0.0f)),    greaterThan(c.in[1].swizzle(1, 2, 0), Vec3(0.0f))); }		\
	void eval_##FUNC_NAME##_bvec4	(ShaderEvalContext& c) { c.color.x() = (float)FUNC_NAME(greaterThan(c.in[0].swizzle(1, 2, 3, 0), Vec4(0.0f)), greaterThan(c.in[1].swizzle(3, 2, 1, 0), Vec4(0.0f))); }

#define DECLARE_BOOL_CWISE_COMPARE_FUNCS(FUNC_NAME)																																								\
	void eval_##FUNC_NAME##_bool	(ShaderEvalContext& c) { c.color.x()	= (float)FUNC_NAME(c.in[0].z() > 0.0f, c.in[1].x() > 0.0f); }																		\
	void eval_##FUNC_NAME##_bvec2	(ShaderEvalContext& c) { c.color.yz()	= FUNC_NAME(greaterThan(c.in[0].swizzle(3, 1), Vec2(0.0f)),       greaterThan(c.in[1].swizzle(1, 0), Vec2(0.0f))).asFloat(); }		\
	void eval_##FUNC_NAME##_bvec3	(ShaderEvalContext& c) { c.color.xyz()	= FUNC_NAME(greaterThan(c.in[0].swizzle(2, 0, 1), Vec3(0.0f)),    greaterThan(c.in[1].swizzle(1, 2, 0), Vec3(0.0f))).asFloat(); }	\
	void eval_##FUNC_NAME##_bvec4	(ShaderEvalContext& c) { c.color		= FUNC_NAME(greaterThan(c.in[0].swizzle(1, 2, 3, 0), Vec4(0.0f)), greaterThan(c.in[1].swizzle(3, 2, 1, 0), Vec4(0.0f))).asFloat(); }

DECLARE_FLOAT_COMPARE_FUNCS(allEqual)
DECLARE_FLOAT_COMPARE_FUNCS(anyNotEqual)
DECLARE_FLOAT_CWISE_COMPARE_FUNCS(lessThan)
DECLARE_FLOAT_CWISE_COMPARE_FUNCS(lessThanEqual)
DECLARE_FLOAT_CWISE_COMPARE_FUNCS(greaterThan)
DECLARE_FLOAT_CWISE_COMPARE_FUNCS(greaterThanEqual)
DECLARE_FLOAT_CWISE_COMPARE_FUNCS(equal)
DECLARE_FLOAT_CWISE_COMPARE_FUNCS(notEqual)

DECLARE_INT_COMPARE_FUNCS(allEqual)
DECLARE_INT_COMPARE_FUNCS(anyNotEqual)
DECLARE_INT_CWISE_COMPARE_FUNCS(lessThan)
DECLARE_INT_CWISE_COMPARE_FUNCS(lessThanEqual)
DECLARE_INT_CWISE_COMPARE_FUNCS(greaterThan)
DECLARE_INT_CWISE_COMPARE_FUNCS(greaterThanEqual)
DECLARE_INT_CWISE_COMPARE_FUNCS(equal)
DECLARE_INT_CWISE_COMPARE_FUNCS(notEqual)

DECLARE_UINT_COMPARE_FUNCS(allEqual)
DECLARE_UINT_COMPARE_FUNCS(anyNotEqual)
DECLARE_UINT_CWISE_COMPARE_FUNCS(lessThan)
DECLARE_UINT_CWISE_COMPARE_FUNCS(lessThanEqual)
DECLARE_UINT_CWISE_COMPARE_FUNCS(greaterThan)
DECLARE_UINT_CWISE_COMPARE_FUNCS(greaterThanEqual)
DECLARE_UINT_CWISE_COMPARE_FUNCS(equal)
DECLARE_UINT_CWISE_COMPARE_FUNCS(notEqual)

DECLARE_BOOL_COMPARE_FUNCS(allEqual)
DECLARE_BOOL_COMPARE_FUNCS(anyNotEqual)
DECLARE_BOOL_CWISE_COMPARE_FUNCS(equal)
DECLARE_BOOL_CWISE_COMPARE_FUNCS(notEqual)

// Boolean functions.

#define DECLARE_UNARY_SCALAR_BVEC_FUNCS(GLSL_NAME, FUNC_NAME)																							\
	void eval_##GLSL_NAME##_bvec2	(ShaderEvalContext& c) { c.color.x()	= float(FUNC_NAME(greaterThan(c.in[0].swizzle(3, 1), Vec2(0.0f)))); }		\
	void eval_##GLSL_NAME##_bvec3	(ShaderEvalContext& c) { c.color.x()	= float(FUNC_NAME(greaterThan(c.in[0].swizzle(2, 0, 1), Vec3(0.0f)))); }	\
	void eval_##GLSL_NAME##_bvec4	(ShaderEvalContext& c) { c.color.x()	= float(FUNC_NAME(greaterThan(c.in[0].swizzle(1, 2, 3, 0), Vec4(0.0f)))); }

#define DECLARE_UNARY_BVEC_BVEC_FUNCS(GLSL_NAME, FUNC_NAME)																								\
	void eval_##GLSL_NAME##_bvec2	(ShaderEvalContext& c) { c.color.yz()	= FUNC_NAME(greaterThan(c.in[0].swizzle(3, 1), Vec2(0.0f))).asFloat(); }	\
	void eval_##GLSL_NAME##_bvec3	(ShaderEvalContext& c) { c.color.xyz()	= FUNC_NAME(greaterThan(c.in[0].swizzle(2, 0, 1), Vec3(0.0f))).asFloat(); }	\
	void eval_##GLSL_NAME##_bvec4	(ShaderEvalContext& c) { c.color.xyzw()	= FUNC_NAME(greaterThan(c.in[0].swizzle(1, 2, 3, 0), Vec4(0.0f))).asFloat(); }

DECLARE_UNARY_SCALAR_BVEC_FUNCS(any, boolAny);
DECLARE_UNARY_SCALAR_BVEC_FUNCS(all, boolAll);

void ShaderOperatorTests::init (void)
{
	#define BOOL_FUNCS(FUNC_NAME)			eval_##FUNC_NAME##_bool, DE_NULL, DE_NULL, DE_NULL

	#define FLOAT_VEC_FUNCS(FUNC_NAME)		DE_NULL, eval_##FUNC_NAME##_vec2, eval_##FUNC_NAME##_vec3, eval_##FUNC_NAME##_vec4
	#define INT_VEC_FUNCS(FUNC_NAME)		DE_NULL, eval_##FUNC_NAME##_ivec2, eval_##FUNC_NAME##_ivec3, eval_##FUNC_NAME##_ivec4
	#define UINT_VEC_FUNCS(FUNC_NAME)		DE_NULL, eval_##FUNC_NAME##_uvec2, eval_##FUNC_NAME##_uvec3, eval_##FUNC_NAME##_uvec4
	#define BOOL_VEC_FUNCS(FUNC_NAME)		DE_NULL, eval_##FUNC_NAME##_bvec2, eval_##FUNC_NAME##_bvec3, eval_##FUNC_NAME##_bvec4

	#define FLOAT_GENTYPE_FUNCS(FUNC_NAME)	eval_##FUNC_NAME##_float, eval_##FUNC_NAME##_vec2, eval_##FUNC_NAME##_vec3, eval_##FUNC_NAME##_vec4
	#define INT_GENTYPE_FUNCS(FUNC_NAME)	eval_##FUNC_NAME##_int, eval_##FUNC_NAME##_ivec2, eval_##FUNC_NAME##_ivec3, eval_##FUNC_NAME##_ivec4
	#define UINT_GENTYPE_FUNCS(FUNC_NAME)	eval_##FUNC_NAME##_uint, eval_##FUNC_NAME##_uvec2, eval_##FUNC_NAME##_uvec3, eval_##FUNC_NAME##_uvec4
	#define BOOL_GENTYPE_FUNCS(FUNC_NAME)	eval_##FUNC_NAME##_bool, eval_##FUNC_NAME##_bvec2, eval_##FUNC_NAME##_bvec3, eval_##FUNC_NAME##_bvec4

	// Shorthands.
	Value					notUsed		= Value(VALUE_NONE, 0.0f, 0.0f);
	FloatScalar::Symbol		lUMax		= FloatScalar::SYMBOL_LOWP_UINT_MAX;
	FloatScalar::Symbol		mUMax		= FloatScalar::SYMBOL_MEDIUMP_UINT_MAX;
	FloatScalar::Symbol		lUMaxR		= FloatScalar::SYMBOL_LOWP_UINT_MAX_RECIPROCAL;
	FloatScalar::Symbol		mUMaxR		= FloatScalar::SYMBOL_MEDIUMP_UINT_MAX_RECIPROCAL;

	std::vector<BuiltinFuncGroup> funcInfoGroups;

	// Unary operators.
	funcInfoGroups.push_back(
		BuiltinFuncGroup("unary_operator", "Unary operator tests")
		<< BuiltinOperInfo						("plus",			"+",	GT,		Value(GT,  -1.0f, 1.0f),	notUsed,	notUsed,	0.5f,	0.5f,	PRECMASK_ALL,		FLOAT_GENTYPE_FUNCS(nop))
		<< BuiltinOperInfo						("plus",			"+",	IGT,	Value(IGT, -5.0f, 5.0f),	notUsed,	notUsed,	0.1f,	0.5f,	PRECMASK_ALL,		INT_GENTYPE_FUNCS(nop))
		<< BuiltinOperInfo						("plus",			"+",	UGT,	Value(UGT,  0.0f, 2e2f),	notUsed,	notUsed,	5e-3f,	0.0f,	PRECMASK_ALL,		UINT_GENTYPE_FUNCS(nop))
		<< BuiltinOperInfo						("minus",			"-",	GT,		Value(GT,  -1.0f, 1.0f),	notUsed,	notUsed,	0.5f,	0.5f,	PRECMASK_ALL,		FLOAT_GENTYPE_FUNCS(negate))
		<< BuiltinOperInfo						("minus",			"-",	IGT,	Value(IGT, -5.0f, 5.0f),	notUsed,	notUsed,	0.1f,	0.5f,	PRECMASK_ALL,		INT_GENTYPE_FUNCS(negate))
		<< BuiltinOperInfoSeparateRefScaleBias	("minus",			"-",	UGT,	Value(UGT,  0.0f, lUMax),	notUsed,	notUsed,	lUMaxR,	0.0f,	PRECMASK_LOWP,		UINT_GENTYPE_FUNCS(negate), lUMaxR, FloatScalar::SYMBOL_ONE_MINUS_UINT32MAX_DIV_LOWP_UINT_MAX)
		<< BuiltinOperInfoSeparateRefScaleBias	("minus",			"-",	UGT,	Value(UGT,  0.0f, mUMax),	notUsed,	notUsed,	mUMaxR,	0.0f,	PRECMASK_MEDIUMP,	UINT_GENTYPE_FUNCS(negate), mUMaxR, FloatScalar::SYMBOL_ONE_MINUS_UINT32MAX_DIV_MEDIUMP_UINT_MAX)
		<< BuiltinOperInfo						("minus",			"-",	UGT,	Value(UGT,  0.0f, 4e9f),	notUsed,	notUsed,	2e-10f,	0.0f,	PRECMASK_HIGHP,		UINT_GENTYPE_FUNCS(negate))
		<< BuiltinOperInfo						("not",				"!",	B,		Value(B,   -1.0f, 1.0f),	notUsed,	notUsed,	1.0f,	0.0f,	PRECMASK_NA,		eval_boolNot_bool, DE_NULL, DE_NULL, DE_NULL)
		<< BuiltinOperInfo						("bitwise_not",		"~",	IGT,	Value(IGT, -1e5f, 1e5f),	notUsed,	notUsed,	5e-5f,	0.5f,	PRECMASK_HIGHP,		INT_GENTYPE_FUNCS(bitwiseNot))
		<< BuiltinOperInfo						("bitwise_not",		"~",	UGT,	Value(UGT,  0.0f, 2e9f),	notUsed,	notUsed,	2e-10f,	0.0f,	PRECMASK_HIGHP,		UINT_GENTYPE_FUNCS(bitwiseNot))

		// Pre/post incr/decr side effect cases.
		<< BuiltinSideEffOperInfo		("pre_increment_effect",	"++",	GT,		Value(GT,	-1.0f, 1.0f),	notUsed,	notUsed,	0.5f, 0.0f,		PRECMASK_ALL,	FLOAT_GENTYPE_FUNCS(addOne))
		<< BuiltinSideEffOperInfo		("pre_increment_effect",	"++",	IGT,	Value(IGT,	-6.0f, 4.0f),	notUsed,	notUsed,	0.1f, 0.5f,		PRECMASK_ALL,	INT_GENTYPE_FUNCS(addOne))
		<< BuiltinSideEffOperInfo		("pre_increment_effect",	"++",	UGT,	Value(UGT,	 0.0f, 9.0f),	notUsed,	notUsed,	0.1f, 0.0f,		PRECMASK_ALL,	UINT_GENTYPE_FUNCS(addOne))
		<< BuiltinSideEffOperInfo		("pre_decrement_effect",	"--",	GT,		Value(GT,	-1.0f, 1.0f),	notUsed,	notUsed,	0.5f, 1.0f,		PRECMASK_ALL,	FLOAT_GENTYPE_FUNCS(subOne))
		<< BuiltinSideEffOperInfo		("pre_decrement_effect",	"--",	IGT,	Value(IGT,	-4.0f, 6.0f),	notUsed,	notUsed,	0.1f, 0.5f,		PRECMASK_ALL,	INT_GENTYPE_FUNCS(subOne))
		<< BuiltinSideEffOperInfo		("pre_decrement_effect",	"--",	UGT,	Value(UGT,	 1.0f, 10.0f),	notUsed,	notUsed,	0.1f, 0.0f,		PRECMASK_ALL,	UINT_GENTYPE_FUNCS(subOne))
		<< BuiltinPostSideEffOperInfo	("post_increment_effect",	"++",	GT,		Value(GT,	-1.0f, 1.0f),	notUsed,	notUsed,	0.5f, 0.0f,		PRECMASK_ALL,	FLOAT_GENTYPE_FUNCS(addOne))
		<< BuiltinPostSideEffOperInfo	("post_increment_effect",	"++",	IGT,	Value(IGT,	-6.0f, 4.0f),	notUsed,	notUsed,	0.1f, 0.5f,		PRECMASK_ALL,	INT_GENTYPE_FUNCS(addOne))
		<< BuiltinPostSideEffOperInfo	("post_increment_effect",	"++",	UGT,	Value(UGT,	 0.0f, 9.0f),	notUsed,	notUsed,	0.1f, 0.0f,		PRECMASK_ALL,	UINT_GENTYPE_FUNCS(addOne))
		<< BuiltinPostSideEffOperInfo	("post_decrement_effect",	"--",	GT,		Value(GT,	-1.0f, 1.0f),	notUsed,	notUsed,	0.5f, 1.0f,		PRECMASK_ALL,	FLOAT_GENTYPE_FUNCS(subOne))
		<< BuiltinPostSideEffOperInfo	("post_decrement_effect",	"--",	IGT,	Value(IGT,	-4.0f, 6.0f),	notUsed,	notUsed,	0.1f, 0.5f,		PRECMASK_ALL,	INT_GENTYPE_FUNCS(subOne))
		<< BuiltinPostSideEffOperInfo	("post_decrement_effect",	"--",	UGT,	Value(UGT,	 1.0f, 10.0f),	notUsed,	notUsed,	0.1f, 0.0f,		PRECMASK_ALL,	UINT_GENTYPE_FUNCS(subOne))

		// Pre/post incr/decr result cases.
		<< BuiltinOperInfo				("pre_increment_result",	"++",	GT,		Value(GT,	-1.0f, 1.0f),	notUsed,	notUsed,	0.5f, 0.0f,		PRECMASK_ALL,	FLOAT_GENTYPE_FUNCS(addOne))
		<< BuiltinOperInfo				("pre_increment_result",	"++",	IGT,	Value(IGT,	-6.0f, 4.0f),	notUsed,	notUsed,	0.1f, 0.5f,		PRECMASK_ALL,	INT_GENTYPE_FUNCS(addOne))
		<< BuiltinOperInfo				("pre_increment_result",	"++",	UGT,	Value(UGT,	 0.0f, 9.0f),	notUsed,	notUsed,	0.1f, 0.0f,		PRECMASK_ALL,	UINT_GENTYPE_FUNCS(addOne))
		<< BuiltinOperInfo				("pre_decrement_result",	"--",	GT,		Value(GT,	-1.0f, 1.0f),	notUsed,	notUsed,	0.5f, 1.0f,		PRECMASK_ALL,	FLOAT_GENTYPE_FUNCS(subOne))
		<< BuiltinOperInfo				("pre_decrement_result",	"--",	IGT,	Value(IGT,	-4.0f, 6.0f),	notUsed,	notUsed,	0.1f, 0.5f,		PRECMASK_ALL,	INT_GENTYPE_FUNCS(subOne))
		<< BuiltinOperInfo				("pre_decrement_result",	"--",	UGT,	Value(UGT,	 1.0f, 10.0f),	notUsed,	notUsed,	0.1f, 0.0f,		PRECMASK_ALL,	UINT_GENTYPE_FUNCS(subOne))
		<< BuiltinPostOperInfo			("post_increment_result",	"++",	GT,		Value(GT,	-1.0f, 1.0f),	notUsed,	notUsed,	0.5f, 0.5f,		PRECMASK_ALL,	FLOAT_GENTYPE_FUNCS(nop))
		<< BuiltinPostOperInfo			("post_increment_result",	"++",	IGT,	Value(IGT,	-5.0f, 5.0f),	notUsed,	notUsed,	0.1f, 0.5f,		PRECMASK_ALL,	INT_GENTYPE_FUNCS(nop))
		<< BuiltinPostOperInfo			("post_increment_result",	"++",	UGT,	Value(UGT,	 0.0f, 9.0f),	notUsed,	notUsed,	0.1f, 0.0f,		PRECMASK_ALL,	UINT_GENTYPE_FUNCS(nop))
		<< BuiltinPostOperInfo			("post_decrement_result",	"--",	GT,		Value(GT,	-1.0f, 1.0f),	notUsed,	notUsed,	0.5f, 0.5f,		PRECMASK_ALL,	FLOAT_GENTYPE_FUNCS(nop))
		<< BuiltinPostOperInfo			("post_decrement_result",	"--",	IGT,	Value(IGT,	-5.0f, 5.0f),	notUsed,	notUsed,	0.1f, 0.5f,		PRECMASK_ALL,	INT_GENTYPE_FUNCS(nop))
		<< BuiltinPostOperInfo			("post_decrement_result",	"--",	UGT,	Value(UGT,	 1.0f, 10.0f),	notUsed,	notUsed,	0.1f, 0.0f,		PRECMASK_ALL,	UINT_GENTYPE_FUNCS(nop))
	);

	BuiltinFuncGroup binaryOpGroup("binary_operator", "Binary operator tests");

	// Normal binary operations and their corresponding assignment operations have lots in common; generate both in the following loop.

	for (int binaryOperatorType = 0; binaryOperatorType <= 2; binaryOperatorType++) // 0: normal op test, 1: assignment op side-effect test, 2: assignment op result test
	{
		bool		isNormalOp		= binaryOperatorType == 0;
		bool		isAssignEff		= binaryOperatorType == 1;
		bool		isAssignRes		= binaryOperatorType == 2;

		DE_ASSERT(isNormalOp || isAssignEff || isAssignRes);
		DE_UNREF(isAssignRes);

		const char*	addName			= isNormalOp ? "add"			: isAssignEff ? "add_assign_effect"			: "add_assign_result";
		const char*	subName			= isNormalOp ? "sub"			: isAssignEff ? "sub_assign_effect"			: "sub_assign_result";
		const char*	mulName			= isNormalOp ? "mul"			: isAssignEff ? "mul_assign_effect"			: "mul_assign_result";
		const char*	divName			= isNormalOp ? "div"			: isAssignEff ? "div_assign_effect"			: "div_assign_result";
		const char* modName			= isNormalOp ? "mod"			: isAssignEff ? "mod_assign_effect"			: "mod_assign_result";
		const char* andName			= isNormalOp ? "bitwise_and"	: isAssignEff ? "bitwise_and_assign_effect"	: "bitwise_and_assign_result";
		const char* orName			= isNormalOp ? "bitwise_or"		: isAssignEff ? "bitwise_or_assign_effect"	: "bitwise_or_assign_result";
		const char* xorName			= isNormalOp ? "bitwise_xor"	: isAssignEff ? "bitwise_xor_assign_effect"	: "bitwise_xor_assign_result";
		const char* leftShiftName	= isNormalOp ? "left_shift"		: isAssignEff ? "left_shift_assign_effect"	: "left_shift_assign_result";
		const char* rightShiftName	= isNormalOp ? "right_shift"	: isAssignEff ? "right_shift_assign_effect"	: "right_shift_assign_result";
		const char*	addOp			= isNormalOp ? "+" : "+=";
		const char*	subOp			= isNormalOp ? "-" : "-=";
		const char*	mulOp			= isNormalOp ? "*" : "*=";
		const char*	divOp			= isNormalOp ? "/" : "/=";
		const char*	modOp			= isNormalOp ? "%" : "%=";
		const char*	andOp			= isNormalOp ? "&" : "&=";
		const char*	orOp			= isNormalOp ? "|" : "|=";
		const char*	xorOp			= isNormalOp ? "^" : "^=";
		const char*	leftShiftOp		= isNormalOp ? "<<" : "<<=";
		const char*	rightShiftOp	= isNormalOp ? ">>" : ">>=";

		// Pointer to appropriate OperInfo function.
		BuiltinFuncInfo (*operInfoFunc)(const char*, const char*, ValueType, Value, Value, Value, const FloatScalar&, const FloatScalar&, deUint32, ShaderEvalFunc, ShaderEvalFunc, ShaderEvalFunc, ShaderEvalFunc) =
			isAssignEff ? BuiltinSideEffOperInfo : BuiltinOperInfo;

		DE_ASSERT(operInfoFunc != DE_NULL);

		// The following cases will be added for each operator, precision and fundamental type (float, int, uint) combination, where applicable:
		// gentype <op> gentype
		// vector <op> scalar
		// For normal (non-assigning) operators only:
		//   scalar <op> vector

		// The add operator.

		binaryOpGroup
			<< operInfoFunc(addName,	addOp,	GT,		Value(GT,  -1.0f, 1.0f),	Value(GT,  -1.0f, 1.0f),	notUsed,	1.0f,	0.0f,	PRECMASK_ALL,			FLOAT_GENTYPE_FUNCS(add))
			<< operInfoFunc(addName,	addOp,	IGT,	Value(IGT, -4.0f, 6.0f),	Value(IGT, -6.0f, 5.0f),	notUsed,	0.1f,	0.5f,	PRECMASK_LOWP_MEDIUMP,	INT_GENTYPE_FUNCS(add))
			<< operInfoFunc(addName,	addOp,	IGT,	Value(IGT, -2e9f, 2e9f),	Value(IGT, -2e9f, 2e9f),	notUsed,	4e-10f,	0.5f,	PRECMASK_HIGHP,			INT_GENTYPE_FUNCS(add))
			<< operInfoFunc(addName,	addOp,	UGT,	Value(UGT,  0.0f, 1e2f),	Value(UGT,  0.0f, 1e2f),	notUsed,	5e-3f,	0.0f,	PRECMASK_LOWP_MEDIUMP,	UINT_GENTYPE_FUNCS(add))
			<< operInfoFunc(addName,	addOp,	UGT,	Value(UGT,  0.0f, 4e9f),	Value(UGT,  0.0f, 4e9f),	notUsed,	2e-10f,	0.0f,	PRECMASK_HIGHP,			UINT_GENTYPE_FUNCS(add))
			<< operInfoFunc(addName,	addOp,	FV,		Value(FV,  -1.0f, 1.0f),	Value(F,   -1.0f, 1.0f),	notUsed,	1.0f,	0.0f,	PRECMASK_ALL,			FLOAT_VEC_FUNCS(addVecScalar))
			<< operInfoFunc(addName,	addOp,	IV,		Value(IV,  -4.0f, 6.0f),	Value(I,   -6.0f, 5.0f),	notUsed,	0.1f,	0.5f,	PRECMASK_LOWP_MEDIUMP,	INT_VEC_FUNCS(addVecScalar))
			<< operInfoFunc(addName,	addOp,	IV,		Value(IV,  -2e9f, 2e9f),	Value(I,   -2e9f, 2e9f),	notUsed,	4e-10f,	0.5f,	PRECMASK_HIGHP,			INT_VEC_FUNCS(addVecScalar))
			<< operInfoFunc(addName,	addOp,	UV,		Value(UV,   0.0f, 1e2f),	Value(U,    0.0f, 1e2f),	notUsed,	5e-3f,	0.0f,	PRECMASK_LOWP_MEDIUMP,	UINT_VEC_FUNCS(addVecScalar))
			<< operInfoFunc(addName,	addOp,	UV,		Value(UV,   0.0f, 4e9f),	Value(U,    0.0f, 4e9f),	notUsed,	2e-10f,	0.0f,	PRECMASK_HIGHP,			UINT_VEC_FUNCS(addVecScalar));

		if (isNormalOp)
			binaryOpGroup
				<< operInfoFunc(addName,	addOp,	FV,		Value(F,   -1.0f, 1.0f),	Value(FV,  -1.0f, 1.0f),	notUsed,	1.0f,	0.0f,	PRECMASK_ALL,			FLOAT_VEC_FUNCS(addScalarVec))
				<< operInfoFunc(addName,	addOp,	IV,		Value(I,   -4.0f, 6.0f),	Value(IV,  -6.0f, 5.0f),	notUsed,	0.1f,	0.5f,	PRECMASK_LOWP_MEDIUMP,	INT_VEC_FUNCS(addScalarVec))
				<< operInfoFunc(addName,	addOp,	IV,		Value(I,   -2e9f, 2e9f),	Value(IV,  -2e9f, 2e9f),	notUsed,	4e-10f,	0.5f,	PRECMASK_HIGHP,			INT_VEC_FUNCS(addScalarVec))
				<< operInfoFunc(addName,	addOp,	UV,		Value(U,    0.0f, 1e2f),	Value(UV,   0.0f, 1e2f),	notUsed,	5e-3f,	0.0f,	PRECMASK_LOWP_MEDIUMP,	UINT_VEC_FUNCS(addScalarVec))
				<< operInfoFunc(addName,	addOp,	UV,		Value(U,    0.0f, 4e9f),	Value(UV,   0.0f, 4e9f),	notUsed,	2e-10f,	0.0f,	PRECMASK_HIGHP,			UINT_VEC_FUNCS(addScalarVec));

		// The subtract operator.

		binaryOpGroup
			<< operInfoFunc(subName,	subOp,	GT,		Value(GT,  -1.0f, 1.0f),	Value(GT,  -1.0f, 1.0f),	notUsed,	1.0f,	0.0f,	PRECMASK_ALL,			FLOAT_GENTYPE_FUNCS(sub))
			<< operInfoFunc(subName,	subOp,	IGT,	Value(IGT, -4.0f, 6.0f),	Value(IGT, -6.0f, 5.0f),	notUsed,	0.1f,	0.5f,	PRECMASK_LOWP_MEDIUMP,	INT_GENTYPE_FUNCS(sub))
			<< operInfoFunc(subName,	subOp,	IGT,	Value(IGT, -2e9f, 2e9f),	Value(IGT, -2e9f, 2e9f),	notUsed,	4e-10f,	0.5f,	PRECMASK_HIGHP,			INT_GENTYPE_FUNCS(sub))
			<< operInfoFunc(subName,	subOp,	UGT,	Value(UGT,  1e2f, 2e2f),	Value(UGT,  0.0f, 1e2f),	notUsed,	5e-3f,	0.0f,	PRECMASK_LOWP_MEDIUMP,	UINT_GENTYPE_FUNCS(sub))
			<< operInfoFunc(subName,	subOp,	UGT,	Value(UGT,  .5e9f, 3.7e9f),	Value(UGT,  0.0f, 3.9e9f),	notUsed,	2e-10f,	0.0f,	PRECMASK_HIGHP,			UINT_GENTYPE_FUNCS(sub))
			<< operInfoFunc(subName,	subOp,	FV,		Value(FV,  -1.0f, 1.0f),	Value(F,   -1.0f, 1.0f),	notUsed,	1.0f,	0.0f,	PRECMASK_ALL,			FLOAT_VEC_FUNCS(subVecScalar))
			<< operInfoFunc(subName,	subOp,	IV,		Value(IV,  -4.0f, 6.0f),	Value(I,   -6.0f, 5.0f),	notUsed,	0.1f,	0.5f,	PRECMASK_LOWP_MEDIUMP,	INT_VEC_FUNCS(subVecScalar))
			<< operInfoFunc(subName,	subOp,	IV,		Value(IV,  -2e9f, 2e9f),	Value(I,   -2e9f, 2e9f),	notUsed,	4e-10f,	0.5f,	PRECMASK_HIGHP,			INT_VEC_FUNCS(subVecScalar))
			<< operInfoFunc(subName,	subOp,	UV,		Value(UV,   1e2f, 2e2f),	Value(U,    0.0f, 1e2f),	notUsed,	5e-3f,	0.0f,	PRECMASK_LOWP_MEDIUMP,	UINT_VEC_FUNCS(subVecScalar))
			<< operInfoFunc(subName,	subOp,	UV,		Value(UV,   0.0f, 4e9f),	Value(U,    0.0f, 4e9f),	notUsed,	2e-10f,	0.0f,	PRECMASK_HIGHP,			UINT_VEC_FUNCS(subVecScalar));

		if (isNormalOp)
			binaryOpGroup
				<< operInfoFunc(subName,	subOp,	FV,		Value(F,   -1.0f, 1.0f),	Value(FV,  -1.0f, 1.0f),	notUsed,	1.0f,	0.0f,	PRECMASK_ALL,			FLOAT_VEC_FUNCS(subScalarVec))
				<< operInfoFunc(subName,	subOp,	IV,		Value(I,   -4.0f, 6.0f),	Value(IV,  -6.0f, 5.0f),	notUsed,	0.1f,	0.5f,	PRECMASK_LOWP_MEDIUMP,	INT_VEC_FUNCS(subScalarVec))
				<< operInfoFunc(subName,	subOp,	IV,		Value(I,   -2e9f, 2e9f),	Value(IV,  -2e9f, 2e9f),	notUsed,	4e-10f,	0.5f,	PRECMASK_HIGHP,			INT_VEC_FUNCS(subScalarVec))
				<< operInfoFunc(subName,	subOp,	UV,		Value(U,    1e2f, 2e2f),	Value(UV,    0.0f, 1e2f),	notUsed,	5e-3f,	0.0f,	PRECMASK_LOWP_MEDIUMP,	UINT_VEC_FUNCS(subScalarVec))
				<< operInfoFunc(subName,	subOp,	UV,		Value(U,    0.0f, 4e9f),	Value(UV,    0.0f, 4e9f),	notUsed,	2e-10f,	0.0f,	PRECMASK_HIGHP,			UINT_VEC_FUNCS(subScalarVec));

		// The multiply operator.

		binaryOpGroup
			<< operInfoFunc(mulName,	mulOp,	GT,		Value(GT,  -1.0f, 1.0f),	Value(GT,  -1.0f, 1.0f),	notUsed,	1.0f,	0.0f,	PRECMASK_ALL,			FLOAT_GENTYPE_FUNCS(mul))
			<< operInfoFunc(mulName,	mulOp,	IGT,	Value(IGT, -4.0f, 6.0f),	Value(IGT, -6.0f, 5.0f),	notUsed,	0.1f,	0.5f,	PRECMASK_LOWP_MEDIUMP,	INT_GENTYPE_FUNCS(mul))
			<< operInfoFunc(mulName,	mulOp,	IGT,	Value(IGT, -3e5f, 3e5f),	Value(IGT, -3e4f, 3e4f),	notUsed,	4e-10f,	0.5f,	PRECMASK_HIGHP,			INT_GENTYPE_FUNCS(mul))
			<< operInfoFunc(mulName,	mulOp,	UGT,	Value(UGT,  0.0f, 16.0f),	Value(UGT,  0.0f, 16.0f),	notUsed,	4e-3f,	0.0f,	PRECMASK_LOWP_MEDIUMP,	UINT_GENTYPE_FUNCS(mul))
			<< operInfoFunc(mulName,	mulOp,	UGT,	Value(UGT,  0.0f, 6e5f),	Value(UGT,  0.0f, 6e4f),	notUsed,	2e-10f,	0.0f,	PRECMASK_HIGHP,			UINT_GENTYPE_FUNCS(mul))
			<< operInfoFunc(mulName,	mulOp,	FV,		Value(FV,  -1.0f, 1.0f),	Value(F,   -1.0f,  1.0f),	notUsed,	1.0f,	0.0f,	PRECMASK_ALL,			FLOAT_VEC_FUNCS(mulVecScalar))
			<< operInfoFunc(mulName,	mulOp,	IV,		Value(IV,  -4.0f, 6.0f),	Value(I,   -6.0f,  5.0f),	notUsed,	0.1f,	0.5f,	PRECMASK_LOWP_MEDIUMP,	INT_VEC_FUNCS(mulVecScalar))
			<< operInfoFunc(mulName,	mulOp,	IV,		Value(IV,  -3e5f, 3e5f),	Value(I,   -3e4f,  3e4f),	notUsed,	4e-10f,	0.5f,	PRECMASK_HIGHP,			INT_VEC_FUNCS(mulVecScalar))
			<< operInfoFunc(mulName,	mulOp,	UV,		Value(UV,   0.0f, 16.0f),	Value(U,    0.0f, 16.0f),	notUsed,	4e-3f,	0.0f,	PRECMASK_LOWP_MEDIUMP,	UINT_VEC_FUNCS(mulVecScalar))
			<< operInfoFunc(mulName,	mulOp,	UV,		Value(UV,   0.0f, 6e5f),	Value(U,    0.0f, 6e4f),	notUsed,	2e-10f,	0.0f,	PRECMASK_HIGHP,			UINT_VEC_FUNCS(mulVecScalar));

		if (isNormalOp)
			binaryOpGroup
				<< operInfoFunc(mulName,	mulOp,	FV,		Value(F,   -1.0f, 1.0f),	Value(FV,  -1.0f,  1.0f),	notUsed,	1.0f,	0.0f,	PRECMASK_ALL,			FLOAT_VEC_FUNCS(mulScalarVec))
				<< operInfoFunc(mulName,	mulOp,	IV,		Value(I,   -4.0f, 6.0f),	Value(IV,  -6.0f,  5.0f),	notUsed,	0.1f,	0.5f,	PRECMASK_LOWP_MEDIUMP,	INT_VEC_FUNCS(mulScalarVec))
				<< operInfoFunc(mulName,	mulOp,	IV,		Value(I,   -3e5f, 3e5f),	Value(IV,  -3e4f,  3e4f),	notUsed,	4e-10f,	0.5f,	PRECMASK_HIGHP,			INT_VEC_FUNCS(mulScalarVec))
				<< operInfoFunc(mulName,	mulOp,	UV,		Value(U,    0.0f, 16.0f),	Value(UV,   0.0f, 16.0f),	notUsed,	4e-3f,	0.0f,	PRECMASK_LOWP_MEDIUMP,	UINT_VEC_FUNCS(mulScalarVec))
				<< operInfoFunc(mulName,	mulOp,	UV,		Value(U,    0.0f, 6e5f),	Value(UV,   0.0f, 6e4f),	notUsed,	2e-10f,	0.0f,	PRECMASK_HIGHP,			UINT_VEC_FUNCS(mulScalarVec));

		// The divide operator.

		binaryOpGroup
			<< operInfoFunc(divName,	divOp,	GT,		Value(GT,  -1.0f,    1.0f),		Value(GT,  -2.0f, -0.5f),	notUsed,	1.0f,	0.0f,	PRECMASK_ALL,			FLOAT_GENTYPE_FUNCS(div))
			<< operInfoFunc(divName,	divOp,	IGT,	Value(IGT, 24.0f,    24.0f),	Value(IGT, -4.0f, -1.0f),	notUsed,	0.04f,	1.0f,	PRECMASK_LOWP_MEDIUMP,	INT_GENTYPE_FUNCS(div))
			<< operInfoFunc(divName,	divOp,	IGT,	Value(IGT, 40320.0f, 40320.0f),	Value(IGT, -8.0f, -1.0f),	notUsed,	1e-5f,	0.5f,	PRECMASK_HIGHP,			INT_GENTYPE_FUNCS(div))
			<< operInfoFunc(divName,	divOp,	UGT,	Value(UGT,  0.0f,    24.0f),	Value(UGT,  1.0f,  4.0f),	notUsed,	0.04f,	0.0f,	PRECMASK_LOWP_MEDIUMP,	UINT_GENTYPE_FUNCS(div))
			<< operInfoFunc(divName,	divOp,	UGT,	Value(UGT,  0.0f,    40320.0f),	Value(UGT,  1.0f,  8.0f),	notUsed,	1e-5f,	0.0f,	PRECMASK_HIGHP,			UINT_GENTYPE_FUNCS(div))
			<< operInfoFunc(divName,	divOp,	FV,		Value(FV,  -1.0f,    1.0f),		Value(F,   -2.0f, -0.5f),	notUsed,	1.0f,	0.0f,	PRECMASK_ALL,			FLOAT_VEC_FUNCS(divVecScalar))
			<< operInfoFunc(divName,	divOp,	IV,		Value(IV,  24.0f,    24.0f),	Value(I,   -4.0f, -1.0f),	notUsed,	0.04f,	1.0f,	PRECMASK_LOWP_MEDIUMP,	INT_VEC_FUNCS(divVecScalar))
			<< operInfoFunc(divName,	divOp,	IV,		Value(IV,  40320.0f, 40320.0f),	Value(I,   -8.0f, -1.0f),	notUsed,	1e-5f,	0.5f,	PRECMASK_HIGHP,			INT_VEC_FUNCS(divVecScalar))
			<< operInfoFunc(divName,	divOp,	UV,		Value(UV,   0.0f,    24.0f),	Value(U,    1.0f,  4.0f),	notUsed,	0.04f,	0.0f,	PRECMASK_LOWP_MEDIUMP,	UINT_VEC_FUNCS(divVecScalar))
			<< operInfoFunc(divName,	divOp,	UV,		Value(UV,   0.0f,    40320.0f),	Value(U,    1.0f,  8.0f),	notUsed,	1e-5f,	0.0f,	PRECMASK_HIGHP,			UINT_VEC_FUNCS(divVecScalar));

		if (isNormalOp)
			binaryOpGroup
				<< operInfoFunc(divName,	divOp,	FV,		Value(F,   -1.0f,    1.0f),		Value(FV,  -2.0f, -0.5f),	notUsed,	1.0f,	0.0f,	PRECMASK_ALL,			FLOAT_VEC_FUNCS(divScalarVec))
				<< operInfoFunc(divName,	divOp,	IV,		Value(I,   24.0f,    24.0f),	Value(IV,  -4.0f, -1.0f),	notUsed,	0.04f,	1.0f,	PRECMASK_LOWP_MEDIUMP,	INT_VEC_FUNCS(divScalarVec))
				<< operInfoFunc(divName,	divOp,	IV,		Value(I,   40320.0f, 40320.0f),	Value(IV,  -8.0f, -1.0f),	notUsed,	1e-5f,	0.5f,	PRECMASK_HIGHP,			INT_VEC_FUNCS(divScalarVec))
				<< operInfoFunc(divName,	divOp,	UV,		Value(U,    0.0f,    24.0f),	Value(UV,   1.0f,  4.0f),	notUsed,	0.04f,	0.0f,	PRECMASK_LOWP_MEDIUMP,	UINT_VEC_FUNCS(divScalarVec))
				<< operInfoFunc(divName,	divOp,	UV,		Value(U,    0.0f,    40320.0f),	Value(UV,   1.0f,  8.0f),	notUsed,	1e-5f,	0.0f,	PRECMASK_HIGHP,			UINT_VEC_FUNCS(divScalarVec));

		// The modulus operator.

		binaryOpGroup
			<< operInfoFunc(modName,	modOp,	IGT,	Value(IGT,  0.0f, 6.0f),	Value(IGT,   1.1f,  6.1f),	notUsed,	0.25f,	0.5f,	PRECMASK_LOWP_MEDIUMP,	INT_GENTYPE_FUNCS(mod))
			<< operInfoFunc(modName,	modOp,	IGT,	Value(IGT,  0.0f, 14.0f),	Value(IGT,   1.1f, 11.1f),	notUsed,	0.1f,	0.5f,	PRECMASK_HIGHP,			INT_GENTYPE_FUNCS(mod))
			<< operInfoFunc(modName,	modOp,	UGT,	Value(UGT,  0.0f, 6.0f),	Value(UGT,   1.1f,  6.1f),	notUsed,	0.25f,	0.0f,	PRECMASK_LOWP_MEDIUMP,	UINT_GENTYPE_FUNCS(mod))
			<< operInfoFunc(modName,	modOp,	UGT,	Value(UGT,  0.0f, 24.0f),	Value(UGT,   1.1f, 11.1f),	notUsed,	0.1f,	0.0f,	PRECMASK_HIGHP,			UINT_GENTYPE_FUNCS(mod))
			<< operInfoFunc(modName,	modOp,	IV,		Value(IV,   0.0f, 6.0f),	Value(I,     1.1f,  6.1f),	notUsed,	0.25f,	0.5f,	PRECMASK_LOWP_MEDIUMP,	INT_VEC_FUNCS(modVecScalar))
			<< operInfoFunc(modName,	modOp,	IV,		Value(IV,   0.0f, 6.0f),	Value(I,     1.1f, 11.1f),	notUsed,	0.1f,	0.5f,	PRECMASK_HIGHP,			INT_VEC_FUNCS(modVecScalar))
			<< operInfoFunc(modName,	modOp,	UV,		Value(UV,   0.0f, 6.0f),	Value(U,     1.1f,  6.1f),	notUsed,	0.25f,	0.0f,	PRECMASK_LOWP_MEDIUMP,	UINT_VEC_FUNCS(modVecScalar))
			<< operInfoFunc(modName,	modOp,	UV,		Value(UV,   0.0f, 24.0f),	Value(U,     1.1f, 11.1f),	notUsed,	0.1f,	0.0f,	PRECMASK_HIGHP,			UINT_VEC_FUNCS(modVecScalar));

		if (isNormalOp)
			binaryOpGroup
				<< operInfoFunc(modName,	modOp,	IV,		Value(I,   0.0f, 6.0f),		Value(IV,     1.1f,  6.1f),	notUsed,	0.25f,	0.5f,	PRECMASK_LOWP_MEDIUMP,	INT_VEC_FUNCS(modScalarVec))
				<< operInfoFunc(modName,	modOp,	IV,		Value(I,   0.0f, 6.0f),		Value(IV,     1.1f, 11.1f),	notUsed,	0.1f,	0.5f,	PRECMASK_HIGHP,			INT_VEC_FUNCS(modScalarVec))
				<< operInfoFunc(modName,	modOp,	UV,		Value(U,   0.0f, 6.0f),		Value(UV,     1.1f,  6.1f),	notUsed,	0.25f,	0.0f,	PRECMASK_LOWP_MEDIUMP,	UINT_VEC_FUNCS(modScalarVec))
				<< operInfoFunc(modName,	modOp,	UV,		Value(U,   0.0f, 24.0f),	Value(UV,     1.1f, 11.1f),	notUsed,	0.1f,	0.0f,	PRECMASK_HIGHP,			UINT_VEC_FUNCS(modScalarVec));

		// The bitwise and operator.

		binaryOpGroup
			<< operInfoFunc(andName,	andOp,	IGT,	Value(IGT, -16.0f, 16.0f),	Value(IGT, -16.0f, 16.0f),	notUsed,	 0.03f,	0.5f,	PRECMASK_LOWP_MEDIUMP,	INT_GENTYPE_FUNCS(bitwiseAnd))
			<< operInfoFunc(andName,	andOp,	IGT,	Value(IGT,  -2e9f,  2e9f),	Value(IGT,  -2e9f,  2e9f),	notUsed,	4e-10f,	0.5f,	PRECMASK_HIGHP,			INT_GENTYPE_FUNCS(bitwiseAnd))
			<< operInfoFunc(andName,	andOp,	UGT,	Value(UGT,   0.0f, 32.0f),	Value(UGT,   0.0f, 32.0f),	notUsed,	 0.03f,	0.0f,	PRECMASK_LOWP_MEDIUMP,	UINT_GENTYPE_FUNCS(bitwiseAnd))
			<< operInfoFunc(andName,	andOp,	UGT,	Value(UGT,   0.0f,  4e9f),	Value(UGT,   0.0f,  4e9f),	notUsed,	2e-10f,	0.0f,	PRECMASK_HIGHP,			UINT_GENTYPE_FUNCS(bitwiseAnd))
			<< operInfoFunc(andName,	andOp,	IV,		Value(IV, -16.0f, 16.0f),	Value(I, -16.0f, 16.0f),	notUsed,	 0.03f,	0.5f,	PRECMASK_LOWP_MEDIUMP,	INT_VEC_FUNCS(bitwiseAndVecScalar))
			<< operInfoFunc(andName,	andOp,	IV,		Value(IV,  -2e9f,  2e9f),	Value(I,  -2e9f,  2e9f),	notUsed,	4e-10f,	0.5f,	PRECMASK_HIGHP,			INT_VEC_FUNCS(bitwiseAndVecScalar))
			<< operInfoFunc(andName,	andOp,	UV,		Value(UV,   0.0f, 32.0f),	Value(U,   0.0f, 32.0f),	notUsed,	 0.03f,	0.0f,	PRECMASK_LOWP_MEDIUMP,	UINT_VEC_FUNCS(bitwiseAndVecScalar))
			<< operInfoFunc(andName,	andOp,	UV,		Value(UV,   0.0f,  4e9f),	Value(U,   0.0f,  4e9f),	notUsed,	2e-10f,	0.0f,	PRECMASK_HIGHP,			UINT_VEC_FUNCS(bitwiseAndVecScalar));

		if (isNormalOp)
			binaryOpGroup
				<< operInfoFunc(andName,	andOp,	IV,		Value(I, -16.0f, 16.0f),	Value(IV, -16.0f, 16.0f),	notUsed,	 0.03f,	0.5f,	PRECMASK_LOWP_MEDIUMP,	INT_VEC_FUNCS(bitwiseAndScalarVec))
				<< operInfoFunc(andName,	andOp,	IV,		Value(I,  -2e9f,  2e9f),	Value(IV,  -2e9f,  2e9f),	notUsed,	4e-10f,	0.5f,	PRECMASK_HIGHP,			INT_VEC_FUNCS(bitwiseAndScalarVec))
				<< operInfoFunc(andName,	andOp,	UV,		Value(U,   0.0f, 32.0f),	Value(UV,   0.0f, 32.0f),	notUsed,	 0.03f,	0.0f,	PRECMASK_LOWP_MEDIUMP,	UINT_VEC_FUNCS(bitwiseAndScalarVec))
				<< operInfoFunc(andName,	andOp,	UV,		Value(U,   0.0f,  4e9f),	Value(UV,   0.0f,  4e9f),	notUsed,	2e-10f,	0.0f,	PRECMASK_HIGHP,			UINT_VEC_FUNCS(bitwiseAndScalarVec));

		// The bitwise or operator.

		binaryOpGroup
			<< operInfoFunc(orName,	orOp,	IGT,	Value(IGT, -16.0f, 16.0f),	Value(IGT, -16.0f, 16.0f),	notUsed,	 0.03f,	0.5f,	PRECMASK_LOWP_MEDIUMP,	INT_GENTYPE_FUNCS(bitwiseOr))
			<< operInfoFunc(orName,	orOp,	IGT,	Value(IGT,  -2e9f,  2e9f),	Value(IGT,  -2e9f,  2e9f),	notUsed,	4e-10f,	0.5f,	PRECMASK_HIGHP,			INT_GENTYPE_FUNCS(bitwiseOr))
			<< operInfoFunc(orName,	orOp,	UGT,	Value(UGT,   0.0f, 32.0f),	Value(UGT,   0.0f, 32.0f),	notUsed,	 0.03f,	0.0f,	PRECMASK_LOWP_MEDIUMP,	UINT_GENTYPE_FUNCS(bitwiseOr))
			<< operInfoFunc(orName,	orOp,	UGT,	Value(UGT,   0.0f,  4e9f),	Value(UGT,   0.0f,  4e9f),	notUsed,	2e-10f,	0.0f,	PRECMASK_HIGHP,			UINT_GENTYPE_FUNCS(bitwiseOr))
			<< operInfoFunc(orName,	orOp,	IV,		Value(IV, -16.0f, 16.0f),	Value(I, -16.0f, 16.0f),	notUsed,	 0.03f,	0.5f,	PRECMASK_LOWP_MEDIUMP,	INT_VEC_FUNCS(bitwiseOrVecScalar))
			<< operInfoFunc(orName,	orOp,	IV,		Value(IV,  -2e9f,  2e9f),	Value(I,  -2e9f,  2e9f),	notUsed,	4e-10f,	0.5f,	PRECMASK_HIGHP,			INT_VEC_FUNCS(bitwiseOrVecScalar))
			<< operInfoFunc(orName,	orOp,	UV,		Value(UV,   0.0f, 32.0f),	Value(U,   0.0f, 32.0f),	notUsed,	 0.03f,	0.0f,	PRECMASK_LOWP_MEDIUMP,	UINT_VEC_FUNCS(bitwiseOrVecScalar))
			<< operInfoFunc(orName,	orOp,	UV,		Value(UV,   0.0f,  4e9f),	Value(U,   0.0f,  4e9f),	notUsed,	2e-10f,	0.0f,	PRECMASK_HIGHP,			UINT_VEC_FUNCS(bitwiseOrVecScalar));

		if (isNormalOp)
			binaryOpGroup
				<< operInfoFunc(orName,	orOp,	IV,		Value(I, -16.0f, 16.0f),	Value(IV, -16.0f, 16.0f),	notUsed,	 0.03f,	0.5f,	PRECMASK_LOWP_MEDIUMP,	INT_VEC_FUNCS(bitwiseOrScalarVec))
				<< operInfoFunc(orName,	orOp,	IV,		Value(I,  -2e9f,  2e9f),	Value(IV,  -2e9f,  2e9f),	notUsed,	4e-10f,	0.5f,	PRECMASK_HIGHP,			INT_VEC_FUNCS(bitwiseOrScalarVec))
				<< operInfoFunc(orName,	orOp,	UV,		Value(U,   0.0f, 32.0f),	Value(UV,   0.0f, 32.0f),	notUsed,	 0.03f,	0.0f,	PRECMASK_LOWP_MEDIUMP,	UINT_VEC_FUNCS(bitwiseOrScalarVec))
				<< operInfoFunc(orName,	orOp,	UV,		Value(U,   0.0f,  4e9f),	Value(UV,   0.0f,  4e9f),	notUsed,	2e-10f,	0.0f,	PRECMASK_HIGHP,			UINT_VEC_FUNCS(bitwiseOrScalarVec));

		// The bitwise xor operator.

		binaryOpGroup
			<< operInfoFunc(xorName,	xorOp,	IGT,	Value(IGT, -16.0f, 16.0f),	Value(IGT, -16.0f, 16.0f),	notUsed,	 0.03f,	0.5f,	PRECMASK_LOWP_MEDIUMP,	INT_GENTYPE_FUNCS(bitwiseXor))
			<< operInfoFunc(xorName,	xorOp,	IGT,	Value(IGT,  -2e9f,  2e9f),	Value(IGT,  -2e9f,  2e9f),	notUsed,	4e-10f,	0.5f,	PRECMASK_HIGHP,			INT_GENTYPE_FUNCS(bitwiseXor))
			<< operInfoFunc(xorName,	xorOp,	UGT,	Value(UGT,   0.0f, 32.0f),	Value(UGT,   0.0f, 32.0f),	notUsed,	 0.03f,	0.0f,	PRECMASK_LOWP_MEDIUMP,	UINT_GENTYPE_FUNCS(bitwiseXor))
			<< operInfoFunc(xorName,	xorOp,	UGT,	Value(UGT,   0.0f,  4e9f),	Value(UGT,   0.0f,  4e9f),	notUsed,	2e-10f,	0.0f,	PRECMASK_HIGHP,			UINT_GENTYPE_FUNCS(bitwiseXor))
			<< operInfoFunc(xorName,	xorOp,	IV,		Value(IV, -16.0f, 16.0f),	Value(I, -16.0f, 16.0f),	notUsed,	 0.03f,	0.5f,	PRECMASK_LOWP_MEDIUMP,	INT_VEC_FUNCS(bitwiseXorVecScalar))
			<< operInfoFunc(xorName,	xorOp,	IV,		Value(IV,  -2e9f,  2e9f),	Value(I,  -2e9f,  2e9f),	notUsed,	4e-10f,	0.5f,	PRECMASK_HIGHP,			INT_VEC_FUNCS(bitwiseXorVecScalar))
			<< operInfoFunc(xorName,	xorOp,	UV,		Value(UV,   0.0f, 32.0f),	Value(U,   0.0f, 32.0f),	notUsed,	 0.03f,	0.0f,	PRECMASK_LOWP_MEDIUMP,	UINT_VEC_FUNCS(bitwiseXorVecScalar))
			<< operInfoFunc(xorName,	xorOp,	UV,		Value(UV,   0.0f,  4e9f),	Value(U,   0.0f,  4e9f),	notUsed,	2e-10f,	0.0f,	PRECMASK_HIGHP,			UINT_VEC_FUNCS(bitwiseXorVecScalar));

		if (isNormalOp)
			binaryOpGroup
				<< operInfoFunc(xorName,	xorOp,	IV,		Value(I, -16.0f, 16.0f),	Value(IV, -16.0f, 16.0f),	notUsed,	 0.03f,	0.5f,	PRECMASK_LOWP_MEDIUMP,	INT_VEC_FUNCS(bitwiseXorScalarVec))
				<< operInfoFunc(xorName,	xorOp,	IV,		Value(I,  -2e9f,  2e9f),	Value(IV,  -2e9f,  2e9f),	notUsed,	4e-10f,	0.5f,	PRECMASK_HIGHP,			INT_VEC_FUNCS(bitwiseXorScalarVec))
				<< operInfoFunc(xorName,	xorOp,	UV,		Value(U,   0.0f, 32.0f),	Value(UV,   0.0f, 32.0f),	notUsed,	 0.03f,	0.0f,	PRECMASK_LOWP_MEDIUMP,	UINT_VEC_FUNCS(bitwiseXorScalarVec))
				<< operInfoFunc(xorName,	xorOp,	UV,		Value(U,   0.0f,  4e9f),	Value(UV,   0.0f,  4e9f),	notUsed,	2e-10f,	0.0f,	PRECMASK_HIGHP,			UINT_VEC_FUNCS(bitwiseXorScalarVec));

		// The left shift operator. Second operand (shift amount) can be either int or uint, even for uint and int first operand, respectively.

		for (int isSignedAmount = 0; isSignedAmount <= 1; isSignedAmount++)
		{
			ValueType gType = isSignedAmount == 0 ? UGT	: IGT;
			ValueType sType = isSignedAmount == 0 ? U	: I;
			binaryOpGroup
				<< operInfoFunc(leftShiftName,	leftShiftOp,	IGT,	Value(IGT, -7.0f, 7.0f),	Value(gType, 0.0f, 4.0f),	notUsed,	4e-3f,  0.5f,	PRECMASK_LOWP_MEDIUMP,	INT_GENTYPE_FUNCS(leftShift))
				<< operInfoFunc(leftShiftName,	leftShiftOp,	IGT,	Value(IGT, -7.0f, 7.0f),	Value(gType, 0.0f, 27.0f),	notUsed,	5e-10f, 0.5f,	PRECMASK_HIGHP,			INT_GENTYPE_FUNCS(leftShift))
				<< operInfoFunc(leftShiftName,	leftShiftOp,	UGT,	Value(UGT,  0.0f, 7.0f),	Value(gType, 0.0f, 5.0f),	notUsed,	4e-3f,  0.0f,	PRECMASK_LOWP_MEDIUMP,	UINT_GENTYPE_FUNCS(leftShift))
				<< operInfoFunc(leftShiftName,	leftShiftOp,	UGT,	Value(UGT,  0.0f, 7.0f),	Value(gType, 0.0f, 28.0f),	notUsed,	5e-10f, 0.0f,	PRECMASK_HIGHP,			UINT_GENTYPE_FUNCS(leftShift))
				<< operInfoFunc(leftShiftName,	leftShiftOp,	IV,		Value(IV,  -7.0f, 7.0f),	Value(sType, 0.0f, 4.0f),	notUsed,	4e-3f,  0.5f,	PRECMASK_LOWP_MEDIUMP,	INT_VEC_FUNCS(leftShiftVecScalar))
				<< operInfoFunc(leftShiftName,	leftShiftOp,	IV,		Value(IV,  -7.0f, 7.0f),	Value(sType, 0.0f, 27.0f),	notUsed,	5e-10f, 0.5f,	PRECMASK_HIGHP,			INT_VEC_FUNCS(leftShiftVecScalar))
				<< operInfoFunc(leftShiftName,	leftShiftOp,	UV,		Value(UV,   0.0f, 7.0f),	Value(sType, 0.0f, 5.0f),	notUsed,	4e-3f,  0.0f,	PRECMASK_LOWP_MEDIUMP,	UINT_VEC_FUNCS(leftShiftVecScalar))
				<< operInfoFunc(leftShiftName,	leftShiftOp,	UV,		Value(UV,   0.0f, 7.0f),	Value(sType, 0.0f, 28.0f),	notUsed,	5e-10f, 0.0f,	PRECMASK_HIGHP,			UINT_VEC_FUNCS(leftShiftVecScalar));
		}

		// The right shift operator. Second operand (shift amount) can be either int or uint, even for uint and int first operand, respectively.

		for (int isSignedAmount = 0; isSignedAmount <= 1; isSignedAmount++)
		{
			ValueType gType = isSignedAmount == 0 ? UGT	: IGT;
			ValueType sType = isSignedAmount == 0 ? U	: I;
			binaryOpGroup
				<< operInfoFunc(rightShiftName,	rightShiftOp,	IGT,	Value(IGT, -127.0f, 127.0f),	Value(gType, 0.0f, 8.0f),	notUsed,	4e-3f,  0.5f,	PRECMASK_LOWP_MEDIUMP,	INT_GENTYPE_FUNCS(rightShift))
				<< operInfoFunc(rightShiftName,	rightShiftOp,	IGT,	Value(IGT, -2e9f, 2e9f),		Value(gType, 0.0f, 31.0f),	notUsed,	5e-10f, 0.5f,	PRECMASK_HIGHP,			INT_GENTYPE_FUNCS(rightShift))
				<< operInfoFunc(rightShiftName,	rightShiftOp,	UGT,	Value(UGT,  0.0f, 255.0f),		Value(gType, 0.0f, 8.0f),	notUsed,	4e-3f,  0.0f,	PRECMASK_LOWP_MEDIUMP,	UINT_GENTYPE_FUNCS(rightShift))
				<< operInfoFunc(rightShiftName,	rightShiftOp,	UGT,	Value(UGT,  0.0f, 4e9f),		Value(gType, 0.0f, 31.0f),	notUsed,	5e-10f, 0.0f,	PRECMASK_HIGHP,			UINT_GENTYPE_FUNCS(rightShift))
				<< operInfoFunc(rightShiftName,	rightShiftOp,	IV,		Value(IV,  -127.0f, 127.0f),	Value(sType, 0.0f, 8.0f),	notUsed,	4e-3f,  0.5f,	PRECMASK_LOWP_MEDIUMP,	INT_VEC_FUNCS(rightShiftVecScalar))
				<< operInfoFunc(rightShiftName,	rightShiftOp,	IV,		Value(IV,  -2e9f, 2e9f),		Value(sType, 0.0f, 31.0f),	notUsed,	5e-10f, 0.5f,	PRECMASK_HIGHP,			INT_VEC_FUNCS(rightShiftVecScalar))
				<< operInfoFunc(rightShiftName,	rightShiftOp,	UV,		Value(UV,   0.0f, 255.0f),		Value(sType, 0.0f, 8.0f),	notUsed,	4e-3f,  0.0f,	PRECMASK_LOWP_MEDIUMP,	UINT_VEC_FUNCS(rightShiftVecScalar))
				<< operInfoFunc(rightShiftName,	rightShiftOp,	UV,		Value(UV,   0.0f, 4e9f),		Value(sType, 0.0f, 31.0f),	notUsed,	5e-10f, 0.0f,	PRECMASK_HIGHP,			UINT_VEC_FUNCS(rightShiftVecScalar));
		}
	}

	// Rest of binary operators.

	binaryOpGroup
		// Scalar relational operators.
		<< BuiltinOperInfo("less",				"<",	B,		Value(F,   -1.0f, 1.0f),	Value(F,   -1.0f, 1.0f),	notUsed,	1.0f, 0.0f,		PRECMASK_ALL,	eval_lessThan_float,			DE_NULL, DE_NULL, DE_NULL)
		<< BuiltinOperInfo("less",				"<",	B,		Value(I,   -5.0f, 5.0f),	Value(I,   -5.0f, 5.0f),	notUsed,	1.0f, 0.0f,		PRECMASK_ALL,	eval_lessThan_int,				DE_NULL, DE_NULL, DE_NULL)
		<< BuiltinOperInfo("less",				"<",	B,		Value(U,    0.0f, 16.0f),	Value(U,    0.0f, 16.0f),	notUsed,	1.0f, 0.0f,		PRECMASK_ALL,	eval_lessThan_uint,				DE_NULL, DE_NULL, DE_NULL)
		<< BuiltinOperInfo("less_or_equal",		"<=",	B,		Value(F,   -1.0f, 1.0f),	Value(F,   -1.0f, 1.0f),	notUsed,	1.0f, 0.0f,		PRECMASK_ALL,	eval_lessThanEqual_float,		DE_NULL, DE_NULL, DE_NULL)
		<< BuiltinOperInfo("less_or_equal",		"<=",	B,		Value(I,   -5.0f, 5.0f),	Value(I,   -5.0f, 5.0f),	notUsed,	1.0f, 0.0f,		PRECMASK_ALL,	eval_lessThanEqual_int,			DE_NULL, DE_NULL, DE_NULL)
		<< BuiltinOperInfo("less_or_equal",		"<=",	B,		Value(U,    0.0f, 16.0f),	Value(U,    0.0f, 16.0f),	notUsed,	1.0f, 0.0f,		PRECMASK_ALL,	eval_lessThanEqual_uint,		DE_NULL, DE_NULL, DE_NULL)
		<< BuiltinOperInfo("greater",			">",	B,		Value(F,   -1.0f, 1.0f),	Value(F,   -1.0f, 1.0f),	notUsed,	1.0f, 0.0f,		PRECMASK_ALL,	eval_greaterThan_float,			DE_NULL, DE_NULL, DE_NULL)
		<< BuiltinOperInfo("greater",			">",	B,		Value(I,   -5.0f, 5.0f),	Value(I,   -5.0f, 5.0f),	notUsed,	1.0f, 0.0f,		PRECMASK_ALL,	eval_greaterThan_int,			DE_NULL, DE_NULL, DE_NULL)
		<< BuiltinOperInfo("greater",			">",	B,		Value(U,    0.0f, 16.0f),	Value(U,    0.0f, 16.0f),	notUsed,	1.0f, 0.0f,		PRECMASK_ALL,	eval_greaterThan_uint,			DE_NULL, DE_NULL, DE_NULL)
		<< BuiltinOperInfo("greater_or_equal",	">=",	B,		Value(F,   -1.0f, 1.0f),	Value(F,   -1.0f, 1.0f),	notUsed,	1.0f, 0.0f,		PRECMASK_ALL,	eval_greaterThanEqual_float,	DE_NULL, DE_NULL, DE_NULL)
		<< BuiltinOperInfo("greater_or_equal",	">=",	B,		Value(I,   -5.0f, 5.0f),	Value(I,   -5.0f, 5.0f),	notUsed,	1.0f, 0.0f,		PRECMASK_ALL,	eval_greaterThanEqual_int,		DE_NULL, DE_NULL, DE_NULL)
		<< BuiltinOperInfo("greater_or_equal",	">=",	B,		Value(U,    0.0f, 16.0f),	Value(U,    0.0f, 16.0f),	notUsed,	1.0f, 0.0f,		PRECMASK_ALL,	eval_greaterThanEqual_uint,		DE_NULL, DE_NULL, DE_NULL)

		// Equality comparison operators.
		<< BuiltinOperInfo("equal",				"==",	B,		Value(GT,  -1.0f, 1.0f),	Value(GT,  -1.0f, 1.0f),	notUsed,	1.0f, 0.0f,		PRECMASK_ALL,	FLOAT_GENTYPE_FUNCS(allEqual))
		<< BuiltinOperInfo("equal",				"==",	B,		Value(IGT, -5.5f, 4.7f),	Value(IGT, -2.1f, 0.1f),	notUsed,	1.0f, 0.0f,		PRECMASK_ALL,	INT_GENTYPE_FUNCS(allEqual))
		<< BuiltinOperInfo("equal",				"==",	B,		Value(UGT,  0.0f, 8.0f),	Value(UGT,  3.5f, 4.5f),	notUsed,	1.0f, 0.0f,		PRECMASK_ALL,	UINT_GENTYPE_FUNCS(allEqual))
		<< BuiltinOperInfo("equal",				"==",	B,		Value(BGT, -2.1f, 2.1f),	Value(BGT, -1.1f, 3.0f),	notUsed,	1.0f, 0.0f,		PRECMASK_NA,	BOOL_GENTYPE_FUNCS(allEqual))
		<< BuiltinOperInfo("not_equal",			"!=",	B,		Value(GT,  -1.0f, 1.0f),	Value(GT,  -1.0f, 1.0f),	notUsed,	1.0f, 0.0f,		PRECMASK_ALL,	FLOAT_GENTYPE_FUNCS(anyNotEqual))
		<< BuiltinOperInfo("not_equal",			"!=",	B,		Value(IGT, -5.5f, 4.7f),	Value(IGT, -2.1f, 0.1f),	notUsed,	1.0f, 0.0f,		PRECMASK_ALL,	INT_GENTYPE_FUNCS(anyNotEqual))
		<< BuiltinOperInfo("not_equal",			"!=",	B,		Value(UGT,  0.0f, 8.0f),	Value(UGT,  3.5f, 4.5f),	notUsed,	1.0f, 0.0f,		PRECMASK_ALL,	UINT_GENTYPE_FUNCS(anyNotEqual))
		<< BuiltinOperInfo("not_equal",			"!=",	B,		Value(BGT, -2.1f, 2.1f),	Value(BGT, -1.1f, 3.0f),	notUsed,	1.0f, 0.0f,		PRECMASK_NA,	BOOL_GENTYPE_FUNCS(anyNotEqual))

		// Logical operators.
		<< BuiltinOperInfo("logical_and",	"&&",	B,	Value(B, -1.0f, 1.0f),	Value(B, -1.0f, 1.0f),	notUsed,	1.0f, 0.0f,		PRECMASK_NA,	BOOL_FUNCS(logicalAnd))
		<< BuiltinOperInfo("logical_or",	"||",	B,	Value(B, -1.0f, 1.0f),	Value(B, -1.0f, 1.0f),	notUsed,	1.0f, 0.0f,		PRECMASK_NA,	BOOL_FUNCS(logicalOr))
		<< BuiltinOperInfo("logical_xor",	"^^",	B,	Value(B, -1.0f, 1.0f),	Value(B, -1.0f, 1.0f),	notUsed,	1.0f, 0.0f,		PRECMASK_NA,	BOOL_FUNCS(logicalXor));

	funcInfoGroups.push_back(binaryOpGroup);

	// 8.1 Angle and Trigonometry Functions.
	funcInfoGroups.push_back(
		BuiltinFuncGroup("angle_and_trigonometry", "Angle and trigonometry function tests.")
		<< BuiltinFuncInfo("radians",		"radians",		GT,	Value(GT, -1.0f, 1.0f),		notUsed,					notUsed,					25.0f, 0.5f,	PRECMASK_MEDIUMP_HIGHP,		FLOAT_GENTYPE_FUNCS(radians) )
		<< BuiltinFuncInfo("degrees",		"degrees",		GT,	Value(GT, -1.0f, 1.0f),		notUsed,					notUsed,					0.04f, 0.5f,	PRECMASK_MEDIUMP_HIGHP,		FLOAT_GENTYPE_FUNCS(degrees) )
		<< BuiltinFuncInfo("sin",			"sin",			GT,	Value(GT, -5.0f, 5.0f),		notUsed,					notUsed,					0.5f, 0.5f,		PRECMASK_MEDIUMP_HIGHP,		FLOAT_GENTYPE_FUNCS(sin) )
		<< BuiltinFuncInfo("sin",			"sin",			GT,	Value(GT, -1.5f, 1.5f),		notUsed,					notUsed,					0.5f, 0.5f,		PRECMASK_LOWP,				FLOAT_GENTYPE_FUNCS(sin) )
		<< BuiltinFuncInfo("cos",			"cos",			GT,	Value(GT, -5.0f, 5.0f),		notUsed,					notUsed,					0.5f, 0.5f,		PRECMASK_MEDIUMP_HIGHP,		FLOAT_GENTYPE_FUNCS(cos) )
		<< BuiltinFuncInfo("cos",			"cos",			GT,	Value(GT, -1.5f, 1.5f),		notUsed,					notUsed,					0.5f, 0.5f,		PRECMASK_LOWP,				FLOAT_GENTYPE_FUNCS(cos) )
		<< BuiltinFuncInfo("tan",			"tan",			GT,	Value(GT, -5.0f, 5.0f),		notUsed,					notUsed,					0.5f, 0.5f,		PRECMASK_MEDIUMP_HIGHP,		FLOAT_GENTYPE_FUNCS(tan) )
		<< BuiltinFuncInfo("tan",			"tan",			GT,	Value(GT, -1.5f, 5.5f),		notUsed,					notUsed,					0.5f, 0.5f,		PRECMASK_LOWP,				FLOAT_GENTYPE_FUNCS(tan) )
		<< BuiltinFuncInfo("asin",			"asin",			GT,	Value(GT, -1.0f, 1.0f),		notUsed,					notUsed,					1.0f, 0.0f,		PRECMASK_MEDIUMP_HIGHP,		FLOAT_GENTYPE_FUNCS(asin) )
		<< BuiltinFuncInfo("acos",			"acos",			GT,	Value(GT, -1.0f, 1.0f),		notUsed,					notUsed,					1.0f, 0.0f,		PRECMASK_MEDIUMP_HIGHP,		FLOAT_GENTYPE_FUNCS(acos) )
		<< BuiltinFuncInfo("atan",			"atan",			GT,	Value(GT, -4.0f, 4.0f),		notUsed,					notUsed,					0.5f, 0.5f,		PRECMASK_MEDIUMP_HIGHP,		FLOAT_GENTYPE_FUNCS(atan) )
		<< BuiltinFuncInfo("atan2",			"atan",			GT,	Value(GT, -4.0f, 4.0f),		Value(GT, 0.5f, 2.0f),		notUsed,					0.5f, 0.5f,		PRECMASK_MEDIUMP_HIGHP,		FLOAT_GENTYPE_FUNCS(atan2) )
		<< BuiltinFuncInfo("sinh",			"sinh",			GT,	Value(GT, -5.0f, 5.0f),		notUsed,					notUsed,					0.5f, 0.5f,		PRECMASK_MEDIUMP_HIGHP,		FLOAT_GENTYPE_FUNCS(sinh) )
		<< BuiltinFuncInfo("sinh",			"sinh",			GT,	Value(GT, -1.5f, 1.5f),		notUsed,					notUsed,					0.5f, 0.5f,		PRECMASK_LOWP,				FLOAT_GENTYPE_FUNCS(sinh) )
		<< BuiltinFuncInfo("cosh",			"cosh",			GT,	Value(GT, -5.0f, 5.0f),		notUsed,					notUsed,					0.5f, 0.5f,		PRECMASK_MEDIUMP_HIGHP,		FLOAT_GENTYPE_FUNCS(cosh) )
		<< BuiltinFuncInfo("cosh",			"cosh",			GT,	Value(GT, -1.5f, 1.5f),		notUsed,					notUsed,					0.5f, 0.5f,		PRECMASK_LOWP,				FLOAT_GENTYPE_FUNCS(cosh) )
		<< BuiltinFuncInfo("tanh",			"tanh",			GT,	Value(GT, -5.0f, 5.0f),		notUsed,					notUsed,					0.5f, 0.5f,		PRECMASK_MEDIUMP_HIGHP,		FLOAT_GENTYPE_FUNCS(tanh) )
		<< BuiltinFuncInfo("tanh",			"tanh",			GT,	Value(GT, -1.5f, 5.5f),		notUsed,					notUsed,					0.5f, 0.5f,		PRECMASK_LOWP,				FLOAT_GENTYPE_FUNCS(tanh) )
		<< BuiltinFuncInfo("asinh",			"asinh",		GT,	Value(GT, -1.0f, 1.0f),		notUsed,					notUsed,					1.0f, 0.0f,		PRECMASK_MEDIUMP_HIGHP,		FLOAT_GENTYPE_FUNCS(asinh) )
		<< BuiltinFuncInfo("acosh",			"acosh",		GT,	Value(GT, 1.0f, 2.2f),		notUsed,					notUsed,					1.0f, 0.0f,		PRECMASK_MEDIUMP_HIGHP,		FLOAT_GENTYPE_FUNCS(acosh) )
		<< BuiltinFuncInfo("atanh",			"atanh",		GT,	Value(GT, -0.99f, 0.99f),	notUsed,					notUsed,					1.0f, 0.0f,		PRECMASK_MEDIUMP_HIGHP,		FLOAT_GENTYPE_FUNCS(atanh) )
	);

	// 8.2 Exponential Functions.
	funcInfoGroups.push_back(
		BuiltinFuncGroup("exponential", "Exponential function tests")
		<< BuiltinFuncInfo("pow",			"pow",			GT,	Value(GT, 0.1f, 8.0f),		Value(GT, -4.0f, 2.0f),		notUsed,					1.0f, 0.0f,		PRECMASK_MEDIUMP_HIGHP,		FLOAT_GENTYPE_FUNCS(pow) )
		<< BuiltinFuncInfo("exp",			"exp",			GT,	Value(GT, -6.0f, 3.0f),		notUsed,					notUsed,					0.5f, 0.0f,		PRECMASK_MEDIUMP_HIGHP,		FLOAT_GENTYPE_FUNCS(exp) )
		<< BuiltinFuncInfo("log",			"log",			GT,	Value(GT, 0.1f, 10.0f),		notUsed,					notUsed,					0.5f, 0.3f,		PRECMASK_MEDIUMP_HIGHP,		FLOAT_GENTYPE_FUNCS(log) )
		<< BuiltinFuncInfo("exp2",			"exp2",			GT,	Value(GT, -7.0f, 2.0f),		notUsed,					notUsed,					1.0f, 0.0f,		PRECMASK_MEDIUMP_HIGHP,		FLOAT_GENTYPE_FUNCS(exp2) )
		<< BuiltinFuncInfo("log2",			"log2",			GT,	Value(GT, 0.1f, 10.0f),		notUsed,					notUsed,					1.0f, 0.0f,		PRECMASK_MEDIUMP_HIGHP,		FLOAT_GENTYPE_FUNCS(log2) )
		<< BuiltinFuncInfo("sqrt",			"sqrt",			GT,	Value(GT, 0.0f, 10.0f),		notUsed,					notUsed,					0.3f, 0.0f,		PRECMASK_MEDIUMP_HIGHP,		FLOAT_GENTYPE_FUNCS(sqrt) )
		<< BuiltinFuncInfo("inversesqrt",	"inversesqrt",	GT,	Value(GT, 0.5f, 10.0f),		notUsed,					notUsed,					1.0f, 0.0f,		PRECMASK_MEDIUMP_HIGHP,		FLOAT_GENTYPE_FUNCS(inverseSqrt) )
	);

	// 8.3 Common Functions.
	funcInfoGroups.push_back(
		BuiltinFuncGroup("common_functions", "Common function tests.")
		<< BuiltinFuncInfo("abs",			"abs",			GT,	Value(GT, -2.0f, 2.0f),		notUsed,					notUsed,					0.5f, 0.5f,		PRECMASK_ALL,				FLOAT_GENTYPE_FUNCS(abs) )
		<< BuiltinFuncInfo("sign",			"sign",			GT,	Value(GT, -1.5f, 1.5f),		notUsed,					notUsed,					0.3f, 0.5f,		PRECMASK_ALL,				FLOAT_GENTYPE_FUNCS(sign) )
		<< BuiltinFuncInfo("floor",			"floor",		GT,	Value(GT, -2.5f, 2.5f),		notUsed,					notUsed,					0.2f, 0.7f,		PRECMASK_ALL,				FLOAT_GENTYPE_FUNCS(floor) )
		<< BuiltinFuncInfo("trunc",			"trunc",		GT,	Value(GT, -2.5f, 2.5f),		notUsed,					notUsed,					0.2f, 0.7f,		PRECMASK_ALL,				FLOAT_GENTYPE_FUNCS(trunc) )
		<< BuiltinFuncInfo("round",			"round",		GT,	Value(GT, -2.5f, 2.5f),		notUsed,					notUsed,					0.2f, 0.7f,		PRECMASK_ALL,				FLOAT_GENTYPE_FUNCS(roundToEven) )
		<< BuiltinFuncInfo("roundEven",		"roundEven",	GT,	Value(GT, -2.5f, 2.5f),		notUsed,					notUsed,					0.2f, 0.7f,		PRECMASK_ALL,				FLOAT_GENTYPE_FUNCS(roundToEven) )
		<< BuiltinFuncInfo("ceil",			"ceil",			GT,	Value(GT, -2.5f, 2.5f),		notUsed,					notUsed,					0.2f, 0.5f,		PRECMASK_ALL,				FLOAT_GENTYPE_FUNCS(ceil) )
		<< BuiltinFuncInfo("fract",			"fract",		GT,	Value(GT, -1.5f, 1.5f),		notUsed,					notUsed,					0.8f, 0.1f,		PRECMASK_ALL,				FLOAT_GENTYPE_FUNCS(fract) )
		<< BuiltinFuncInfo("mod",			"mod",			GT,	Value(GT, -2.0f, 2.0f),		Value(GT, 0.9f, 6.0f),		notUsed,					0.5f, 0.5f,		PRECMASK_MEDIUMP_HIGHP,		FLOAT_GENTYPE_FUNCS(mod) )
		<< BuiltinFuncInfo("mod",			"mod",			GT,	Value(FV, -2.0f, 2.0f),		Value(F, 0.9f, 6.0f),		notUsed,					0.5f, 0.5f,		PRECMASK_MEDIUMP_HIGHP,		FLOAT_VEC_FUNCS(modVecScalar) )
		<< BuiltinFuncInfo("min",			"min",			GT,	Value(GT, -1.0f, 1.0f),		Value(GT, -1.0f, 1.0f),		notUsed,					0.5f, 0.5f,		PRECMASK_ALL,				FLOAT_GENTYPE_FUNCS(min) )
		<< BuiltinFuncInfo("min",			"min",			GT,	Value(FV, -1.0f, 1.0f),		Value(F, -1.0f, 1.0f),		notUsed,					0.5f, 0.5f,		PRECMASK_ALL,				FLOAT_VEC_FUNCS(minVecScalar) )
		<< BuiltinFuncInfo("min",			"min",			IGT,Value(IGT, -4.0f, 4.0f),	Value(IGT, -4.0f, 4.0f),	notUsed,					0.125f, 0.5f,	PRECMASK_ALL,				INT_GENTYPE_FUNCS(min) )
		<< BuiltinFuncInfo("min",			"min",			IGT,Value(IV,  -4.0f, 4.0f),	Value(I, -4.0f, 4.0f),		notUsed,					0.125f, 0.5f,	PRECMASK_ALL,				INT_VEC_FUNCS(minVecScalar) )
		<< BuiltinFuncInfo("min",			"min",			UGT,Value(UGT, 0.0f, 8.0f),		Value(UGT, 0.0f, 8.0f),		notUsed,					0.125f, 0.0f,	PRECMASK_ALL,				UINT_GENTYPE_FUNCS(min) )
		<< BuiltinFuncInfo("min",			"min",			UGT,Value(UV,  0.0f, 8.0f),		Value(U, 0.0f, 8.0f),		notUsed,					0.125f, 0.0f,	PRECMASK_ALL,				UINT_VEC_FUNCS(minVecScalar) )
		<< BuiltinFuncInfo("max",			"max",			GT,	Value(GT, -1.0f, 1.0f),		Value(GT, -1.0f, 1.0f),		notUsed,					0.5f, 0.5f,		PRECMASK_ALL,				FLOAT_GENTYPE_FUNCS(max) )
		<< BuiltinFuncInfo("max",			"max",			GT,	Value(FV, -1.0f, 1.0f),		Value(F, -1.0f, 1.0f),		notUsed,					0.5f, 0.5f,		PRECMASK_ALL,				FLOAT_VEC_FUNCS(maxVecScalar) )
		<< BuiltinFuncInfo("max",			"max",			IGT,Value(IGT, -4.0f, 4.0f),	Value(IGT, -4.0f, 4.0f),	notUsed,					0.125f, 0.5f,	PRECMASK_ALL,				INT_GENTYPE_FUNCS(max) )
		<< BuiltinFuncInfo("max",			"max",			IGT,Value(IV,  -4.0f, 4.0f),	Value(I, -4.0f, 4.0f),		notUsed,					0.125f, 0.5f,	PRECMASK_ALL,				INT_VEC_FUNCS(maxVecScalar) )
		<< BuiltinFuncInfo("max",			"max",			UGT,Value(UGT, 0.0f, 8.0f),		Value(UGT, 0.0f, 8.0f),		notUsed,					0.125f, 0.0f,	PRECMASK_ALL,				UINT_GENTYPE_FUNCS(max) )
		<< BuiltinFuncInfo("max",			"max",			UGT,Value(UV,  0.0f, 8.0f),		Value(U, 0.0f, 8.0f),		notUsed,					0.125f, 0.0f,	PRECMASK_ALL,				UINT_VEC_FUNCS(maxVecScalar) )
		<< BuiltinFuncInfo("clamp",			"clamp",		GT,	Value(GT, -1.0f, 1.0f),		Value(GT, -0.5f, 0.5f),		Value(GT, 0.5f, 1.0f),		0.5f, 0.5f,		PRECMASK_ALL,				FLOAT_GENTYPE_FUNCS(clamp) )
		<< BuiltinFuncInfo("clamp",			"clamp",		GT,	Value(FV, -1.0f, 1.0f),		Value(F, -0.5f, 0.5f),		Value(F, 0.5f, 1.0f),		0.5f, 0.5f,		PRECMASK_ALL,				FLOAT_VEC_FUNCS(clampVecScalarScalar) )
		<< BuiltinFuncInfo("clamp",			"clamp",		IGT,Value(IGT, -4.0f, 4.0f),	Value(IGT, -2.0f, 2.0f),	Value(IGT, 2.0f, 4.0f),		0.125f, 0.5f,	PRECMASK_ALL,				INT_GENTYPE_FUNCS(clamp) )
		<< BuiltinFuncInfo("clamp",			"clamp",		IGT,Value(IV,  -4.0f, 4.0f),	Value(I, -2.0f, 2.0f),		Value(I, 2.0f, 4.0f),		0.125f, 0.5f,	PRECMASK_ALL,				INT_VEC_FUNCS(clampVecScalarScalar) )
		<< BuiltinFuncInfo("clamp",			"clamp",		UGT,Value(UGT, 0.0f, 8.0f),		Value(UGT, 2.0f, 6.0f),		Value(UGT, 6.0f, 8.0f),		0.125f, 0.0f,	PRECMASK_ALL,				UINT_GENTYPE_FUNCS(clamp) )
		<< BuiltinFuncInfo("clamp",			"clamp",		UGT,Value(UV,  0.0f, 8.0f),		Value(U,   2.0f, 6.0f),		Value(U, 6.0f, 8.0f),		0.125f, 0.0f,	PRECMASK_ALL,				UINT_VEC_FUNCS(clampVecScalarScalar) )
		<< BuiltinFuncInfo("mix",			"mix",			GT,	Value(GT, -1.0f, 1.0f),		Value(GT, -1.0f, 1.0f),		Value(GT, 0.0f, 1.0f),		0.5f, 0.5f,		PRECMASK_ALL,				FLOAT_GENTYPE_FUNCS(mix) )
		<< BuiltinFuncInfo("mix",			"mix",			GT,	Value(FV, -1.0f, 1.0f),		Value(FV, -1.0f, 1.0f),		Value(F, 0.0f, 1.0f),		0.5f, 0.5f,		PRECMASK_ALL,				FLOAT_VEC_FUNCS(mixVecVecScalar) )
		<< BuiltinFuncInfo("step",			"step",			GT,	Value(GT, -1.0f, 1.0f),		Value(GT, -1.0f, 0.0f),		notUsed,					0.5f, 0.25f,	PRECMASK_ALL,				FLOAT_GENTYPE_FUNCS(step) )
		<< BuiltinFuncInfo("step",			"step",			GT,	Value(F, -1.0f, 1.0f),		Value(FV, -1.0f, 0.0f),		notUsed,					0.5f, 0.25f,	PRECMASK_ALL,				FLOAT_VEC_FUNCS(stepScalarVec) )
		<< BuiltinFuncInfo("smoothstep",	"smoothstep",	GT,	Value(GT, -0.5f, 0.0f),		Value(GT, 0.1f, 1.0f),		Value(GT, -1.0f, 1.0f),		0.5f, 0.5f,		PRECMASK_ALL,				FLOAT_GENTYPE_FUNCS(smoothStep) )
		<< BuiltinFuncInfo("smoothstep",	"smoothstep",	GT,	Value(F, -0.5f, 0.0f),		Value(F, 0.1f, 1.0f),		Value(FV, -1.0f, 1.0f),		0.5f, 0.5f,		PRECMASK_ALL,				FLOAT_VEC_FUNCS(smoothStepScalarScalarVec) )
	);

	// 8.4 Geometric Functions.
	funcInfoGroups.push_back(
		BuiltinFuncGroup("geometric", "Geometric function tests.")
		<< BuiltinFuncInfo("length",		"length",		F,	Value(GT, -5.0f, 5.0f),		notUsed,					notUsed,					0.1f, 0.5f,		PRECMASK_MEDIUMP_HIGHP,		FLOAT_GENTYPE_FUNCS(length) )
		<< BuiltinFuncInfo("distance",		"distance",		F,	Value(GT, -5.0f, 5.0f),		Value(GT, -5.0f, 5.0f),		notUsed,					0.1f, 0.5f,		PRECMASK_MEDIUMP_HIGHP,		FLOAT_GENTYPE_FUNCS(distance) )
		<< BuiltinFuncInfo("dot",			"dot",			F,	Value(GT, -5.0f, 5.0f),		Value(GT, -5.0f, 5.0f),		notUsed,					0.1f, 0.5f,		PRECMASK_MEDIUMP_HIGHP,		FLOAT_GENTYPE_FUNCS(dot) )
		<< BuiltinFuncInfo("cross",			"cross",		V3,	Value(GT, -5.0f, 5.0f),		Value(GT, -5.0f, 5.0f),		notUsed,					0.1f, 0.5f,		PRECMASK_MEDIUMP_HIGHP,		DE_NULL, DE_NULL, eval_cross_vec3, DE_NULL )
		<< BuiltinFuncInfo("normalize",		"normalize",	GT,	Value(GT, 0.1f, 4.0f),		notUsed,					notUsed,					0.5f, 0.5f,		PRECMASK_MEDIUMP_HIGHP,		FLOAT_GENTYPE_FUNCS(normalize) )
		<< BuiltinFuncInfo("faceforward",	"faceforward",	GT,	Value(GT, -5.0f, 5.0f),		Value(GT, -5.0f, 5.0f),		Value(GT, -1.0f, 1.0f),		0.5f, 0.5f,		PRECMASK_MEDIUMP_HIGHP,		FLOAT_GENTYPE_FUNCS(faceForward) )
		<< BuiltinFuncInfo("reflect",		"reflect",		GT,	Value(GT, -0.8f, -0.5f),	Value(GT, 0.5f, 0.8f),		notUsed,					0.5f, 0.5f,		PRECMASK_MEDIUMP_HIGHP,		FLOAT_GENTYPE_FUNCS(reflect) )
		<< BuiltinFuncInfo("refract",		"refract",		GT,	Value(GT, -0.8f, 1.2f),		Value(GT, -1.1f, 0.5f),		Value(F, 0.2f, 1.5f),		0.5f, 0.5f,		PRECMASK_MEDIUMP_HIGHP,		FLOAT_GENTYPE_FUNCS(refract) )
	);

	// 8.5 Matrix Functions.
	// separate matrix tests?
//	funcInfoGroups.push_back(
//		BuiltinFuncGroup("matrix", "Matrix function tests.")
//		<< BuiltinFuncInfo("matrixCompMult",	"matrixCompMult",	M, ... )
//	);

	// 8.6 Vector Relational Functions.
	funcInfoGroups.push_back(
		BuiltinFuncGroup("float_compare", "Floating point comparison tests.")
		<< BuiltinFuncInfo("lessThan",			"lessThan",			BV,	Value(FV, -1.0f, 1.0f),		Value(FV, -1.0f, 1.0f),	notUsed, 1.0f, 0.0f, PRECMASK_ALL,	FLOAT_VEC_FUNCS(lessThan) )
		<< BuiltinFuncInfo("lessThanEqual",		"lessThanEqual",	BV,	Value(FV, -1.0f, 1.0f),		Value(FV, -1.0f, 1.0f),	notUsed, 1.0f, 0.0f, PRECMASK_ALL,	FLOAT_VEC_FUNCS(lessThanEqual) )
		<< BuiltinFuncInfo("greaterThan",		"greaterThan",		BV,	Value(FV, -1.0f, 1.0f),		Value(FV, -1.0f, 1.0f),	notUsed, 1.0f, 0.0f, PRECMASK_ALL,	FLOAT_VEC_FUNCS(greaterThan) )
		<< BuiltinFuncInfo("greaterThanEqual",	"greaterThanEqual",	BV,	Value(FV, -1.0f, 1.0f),		Value(FV, -1.0f, 1.0f),	notUsed, 1.0f, 0.0f, PRECMASK_ALL,	FLOAT_VEC_FUNCS(greaterThanEqual) )
		<< BuiltinFuncInfo("equal",				"equal",			BV,	Value(FV, -1.0f, 1.0f),		Value(FV, -1.0f, 1.0f),	notUsed, 1.0f, 0.0f, PRECMASK_ALL,	FLOAT_VEC_FUNCS(equal) )
		<< BuiltinFuncInfo("notEqual",			"notEqual",			BV,	Value(FV, -1.0f, 1.0f),		Value(FV, -1.0f, 1.0f),	notUsed, 1.0f, 0.0f, PRECMASK_ALL,	FLOAT_VEC_FUNCS(notEqual) )
	);

	funcInfoGroups.push_back(
		BuiltinFuncGroup("int_compare", "Integer comparison tests.")
		<< BuiltinFuncInfo("lessThan",			"lessThan",			BV,	Value(IV, -5.2f, 4.9f),		Value(IV, -5.0f, 5.0f),	notUsed, 1.0f, 0.0f, PRECMASK_ALL,	INT_VEC_FUNCS(lessThan) )
		<< BuiltinFuncInfo("lessThanEqual",		"lessThanEqual",	BV,	Value(IV, -5.2f, 4.9f),		Value(IV, -5.0f, 5.0f),	notUsed, 1.0f, 0.0f, PRECMASK_ALL,	INT_VEC_FUNCS(lessThanEqual) )
		<< BuiltinFuncInfo("greaterThan",		"greaterThan",		BV,	Value(IV, -5.2f, 4.9f),		Value(IV, -5.0f, 5.0f),	notUsed, 1.0f, 0.0f, PRECMASK_ALL,	INT_VEC_FUNCS(greaterThan) )
		<< BuiltinFuncInfo("greaterThanEqual",	"greaterThanEqual",	BV,	Value(IV, -5.2f, 4.9f),		Value(IV, -5.0f, 5.0f),	notUsed, 1.0f, 0.0f, PRECMASK_ALL,	INT_VEC_FUNCS(greaterThanEqual) )
		<< BuiltinFuncInfo("equal",				"equal",			BV,	Value(IV, -5.2f, 4.9f),		Value(IV, -5.0f, 5.0f),	notUsed, 1.0f, 0.0f, PRECMASK_ALL,	INT_VEC_FUNCS(equal) )
		<< BuiltinFuncInfo("notEqual",			"notEqual",			BV,	Value(IV, -5.2f, 4.9f),		Value(IV, -5.0f, 5.0f),	notUsed, 1.0f, 0.0f, PRECMASK_ALL,	INT_VEC_FUNCS(notEqual) )
	);

	funcInfoGroups.push_back(
		BuiltinFuncGroup("bool_compare", "Boolean comparison tests.")
		<< BuiltinFuncInfo("equal",				"equal",			BV,	Value(BV, -5.2f, 4.9f),		Value(BV, -5.0f, 5.0f),	notUsed, 1.0f, 0.0f, PRECMASK_NA,	BOOL_VEC_FUNCS(equal) )
		<< BuiltinFuncInfo("notEqual",			"notEqual",			BV,	Value(BV, -5.2f, 4.9f),		Value(BV, -5.0f, 5.0f),	notUsed, 1.0f, 0.0f, PRECMASK_NA,	BOOL_VEC_FUNCS(notEqual) )
		<< BuiltinFuncInfo("any",				"any",				B,	Value(BV, -1.0f, 0.3f),		notUsed,				notUsed, 1.0f, 0.0f, PRECMASK_NA,	BOOL_VEC_FUNCS(any) )
		<< BuiltinFuncInfo("all",				"all",				B,	Value(BV, -0.3f, 1.0f),		notUsed,				notUsed, 1.0f, 0.0f, PRECMASK_NA,	BOOL_VEC_FUNCS(all) )
		<< BuiltinFuncInfo("not",				"not",				BV,	Value(BV, -1.0f, 1.0f),		notUsed,				notUsed, 1.0f, 0.0f, PRECMASK_NA,	BOOL_VEC_FUNCS(boolNot) )
	);

	static const ShaderType s_shaderTypes[] =
	{
		SHADERTYPE_VERTEX,
		SHADERTYPE_FRAGMENT
	};

	static const DataType s_floatTypes[] =
	{
		TYPE_FLOAT,
		TYPE_FLOAT_VEC2,
		TYPE_FLOAT_VEC3,
		TYPE_FLOAT_VEC4
	};

	static const DataType s_intTypes[] =
	{
		TYPE_INT,
		TYPE_INT_VEC2,
		TYPE_INT_VEC3,
		TYPE_INT_VEC4
	};

	static const DataType s_uintTypes[] =
	{
		TYPE_UINT,
		TYPE_UINT_VEC2,
		TYPE_UINT_VEC3,
		TYPE_UINT_VEC4
	};

	static const DataType s_boolTypes[] =
	{
		TYPE_BOOL,
		TYPE_BOOL_VEC2,
		TYPE_BOOL_VEC3,
		TYPE_BOOL_VEC4
	};

	for (int outerGroupNdx = 0; outerGroupNdx < (int)funcInfoGroups.size(); outerGroupNdx++)
	{
		// Create outer group.
		const BuiltinFuncGroup& outerGroupInfo = funcInfoGroups[outerGroupNdx];
		TestCaseGroup* outerGroup = new TestCaseGroup(m_context, outerGroupInfo.name, outerGroupInfo.description);
		addChild(outerGroup);

		// Only create new group if name differs from previous one.
		TestCaseGroup* innerGroup = DE_NULL;

		for (int funcInfoNdx = 0; funcInfoNdx < (int)outerGroupInfo.funcInfos.size(); funcInfoNdx++)
		{
			const BuiltinFuncInfo&	funcInfo 		= outerGroupInfo.funcInfos[funcInfoNdx];
			const char*				shaderFuncName	= funcInfo.shaderFuncName;
			bool					isBoolCase		= (funcInfo.precisionMask == PRECMASK_NA);
			bool					isBoolOut		= (funcInfo.outValue & (VALUE_BOOL | VALUE_BOOL_VEC | VALUE_BOOL_GENTYPE)) != 0;
			bool					isIntOut		= (funcInfo.outValue & (VALUE_INT | VALUE_INT_VEC | VALUE_INT_GENTYPE)) != 0;
			bool					isUintOut		= (funcInfo.outValue & (VALUE_UINT | VALUE_UINT_VEC | VALUE_UINT_GENTYPE)) != 0;
			bool					isFloatOut		= !isBoolOut && !isIntOut && !isUintOut;

			if (!innerGroup || (string(innerGroup->getName()) != funcInfo.caseName))
			{
				string groupDesc = string("Built-in function ") + shaderFuncName + "() tests.";
				innerGroup = new TestCaseGroup(m_context, funcInfo.caseName, groupDesc.c_str());
				outerGroup->addChild(innerGroup);
			}

			for (int inScalarSize = 1; inScalarSize <= 4; inScalarSize++)
			{
				int			outScalarSize	= ((funcInfo.outValue == VALUE_FLOAT) || (funcInfo.outValue == VALUE_BOOL)) ? 1 : inScalarSize; // \todo [petri] Int.
				DataType	outDataType		= isFloatOut	? s_floatTypes[outScalarSize - 1]
											: isIntOut		? s_intTypes[outScalarSize - 1]
											: isUintOut		? s_uintTypes[outScalarSize - 1]
											: isBoolOut		? s_boolTypes[outScalarSize - 1]
											: TYPE_LAST;

				ShaderEvalFunc evalFunc = DE_NULL;
				if      (inScalarSize == 1)	evalFunc = funcInfo.evalFuncScalar;
				else if (inScalarSize == 2)	evalFunc = funcInfo.evalFuncVec2;
				else if (inScalarSize == 3)	evalFunc = funcInfo.evalFuncVec3;
				else if (inScalarSize == 4)	evalFunc = funcInfo.evalFuncVec4;
				else DE_ASSERT(false);

				// Skip if no valid eval func.
				// \todo [petri] Better check for V3 only etc. cases?
				if (evalFunc == DE_NULL)
					continue;

				for (int precision = 0; precision < PRECISION_LAST; precision++)
				{
					if ((funcInfo.precisionMask & (1<<precision)) ||
						(funcInfo.precisionMask == PRECMASK_NA && precision == PRECISION_MEDIUMP)) // use mediump interpolators for booleans
					{
						const char*	precisionStr	= getPrecisionName((Precision)precision);
						string		precisionPrefix	= isBoolCase ? "" : (string(precisionStr) + "_");

						for (int shaderTypeNdx = 0; shaderTypeNdx < DE_LENGTH_OF_ARRAY(s_shaderTypes); shaderTypeNdx++)
						{
							ShaderType		shaderType		= s_shaderTypes[shaderTypeNdx];
							ShaderDataSpec	shaderSpec;
							const char*		shaderTypeName	= getShaderTypeName(shaderType);
							bool			isVertexCase	= (ShaderType)shaderType == SHADERTYPE_VERTEX;
							bool			isUnaryOp		= (funcInfo.input1.valueType == VALUE_NONE);

							// \note Data type names will be added to description and name in a following loop.
							string desc	= string("Built-in function ") + shaderFuncName + "(";
							string name = precisionPrefix;

							// Generate shader op.
							string shaderOp = string("res = ");

							// Setup shader data info.
							shaderSpec.numInputs		= 0;
							shaderSpec.precision		= isBoolCase ? PRECISION_LAST : (Precision)precision;
							shaderSpec.output			= outDataType;
							shaderSpec.resultScale		= funcInfo.resultScale;
							shaderSpec.resultBias		= funcInfo.resultBias;
							shaderSpec.referenceScale	= funcInfo.referenceScale;
							shaderSpec.referenceBias	= funcInfo.referenceBias;

							if (funcInfo.type == OPERATOR)
							{
								if (isUnaryOp && funcInfo.isUnaryPrefix)
									shaderOp += shaderFuncName;
							}
							else if (funcInfo.type == FUNCTION)
								shaderOp += string(shaderFuncName) + "(";
							else // SIDE_EFFECT_OPERATOR
								shaderOp += "in0;\n\t";

							for (int inputNdx = 0; inputNdx < MAX_INPUTS; inputNdx++)
							{
								const Value&	prevV			= (inputNdx == 1) ? funcInfo.input0 : (inputNdx == 2) ? funcInfo.input1 : funcInfo.input2;
								const Value&	v				= (inputNdx == 0) ? funcInfo.input0 : (inputNdx == 1) ? funcInfo.input1 : funcInfo.input2;

								if (v.valueType == VALUE_NONE)
									continue; // Skip unused input.

								int				prevInScalarSize	= isScalarType(prevV.valueType) ? 1 : inScalarSize;
								DataType		prevInDataType		= isFloatType(prevV.valueType)	? s_floatTypes[prevInScalarSize - 1]
																	: isIntType(prevV.valueType)	? s_intTypes[prevInScalarSize - 1]
																	: isUintType(prevV.valueType)	? s_uintTypes[prevInScalarSize - 1]
																	: isBoolType(prevV.valueType)	? s_boolTypes[prevInScalarSize - 1]
																	: TYPE_LAST;

								int				curInScalarSize		= isScalarType(v.valueType) ? 1 : inScalarSize;
								DataType		curInDataType		= isFloatType(v.valueType)	? s_floatTypes[curInScalarSize - 1]
																	: isIntType(v.valueType)	? s_intTypes[curInScalarSize - 1]
																	: isUintType(v.valueType)	? s_uintTypes[curInScalarSize - 1]
																	: isBoolType(v.valueType)	? s_boolTypes[curInScalarSize - 1]
																	: TYPE_LAST;

								// Write input type(s) to case description and name.

								if (inputNdx > 0)
									desc += ", ";

								desc += getDataTypeName(curInDataType);

								if (inputNdx == 0 || prevInDataType != curInDataType) // \note Only write input type to case name if different from previous input type (avoid overly long names).
									name += string("") + getDataTypeName(curInDataType) + "_";

								// Generate op input source.

								if (funcInfo.type == OPERATOR || funcInfo.type == FUNCTION)
								{
									if (inputNdx != 0)
									{
										if (funcInfo.type == OPERATOR && !isUnaryOp)
											shaderOp += " " + string(shaderFuncName) + " ";
										else
											shaderOp += ", ";
									}

									shaderOp += "in" + de::toString(inputNdx);

									if (funcInfo.type == OPERATOR && isUnaryOp && !funcInfo.isUnaryPrefix)
										shaderOp += string(shaderFuncName);
								}
								else
								{
									DE_ASSERT(funcInfo.type == SIDE_EFFECT_OPERATOR);

									if (inputNdx != 0 || (isUnaryOp && funcInfo.isUnaryPrefix))
										shaderOp += string("") + (isUnaryOp ? "" : " ") + shaderFuncName + (isUnaryOp ? "" : " ");

									shaderOp += inputNdx == 0 ? "res" : "in" + de::toString(inputNdx); // \note in0 has already been assigned to res, so start from in1.

									if (isUnaryOp && !funcInfo.isUnaryPrefix)
										shaderOp += shaderFuncName;
								}

								// Fill in shader info.
								shaderSpec.inputs[shaderSpec.numInputs++] = ShaderValue(curInDataType, v.rangeMin, v.rangeMax);
							}

							if (funcInfo.type == FUNCTION)
								shaderOp += ")";

							shaderOp += ";";

							desc += ").";
							name += shaderTypeName;

							// Create the test case.
							innerGroup->addChild(new ShaderOperatorCase(m_context, name.c_str(), desc.c_str(), isVertexCase, evalFunc, shaderOp, shaderSpec));
						}
					}
				}
			}
		}
	}

	// The ?: selection operator.

	static const struct
	{
		DataType		type; // The type of "Y" and "Z" operands in "X ? Y : Z" (X is always bool).
		ShaderEvalFunc	evalFunc;
	} s_selectionInfo[] =
	{
		{ TYPE_FLOAT,		eval_selection_float	},
		{ TYPE_FLOAT_VEC2,	eval_selection_vec2		},
		{ TYPE_FLOAT_VEC3,	eval_selection_vec3		},
		{ TYPE_FLOAT_VEC4,	eval_selection_vec4		},
		{ TYPE_INT,			eval_selection_int		},
		{ TYPE_INT_VEC2,	eval_selection_ivec2	},
		{ TYPE_INT_VEC3,	eval_selection_ivec3	},
		{ TYPE_INT_VEC4,	eval_selection_ivec4	},
		{ TYPE_UINT,		eval_selection_uint		},
		{ TYPE_UINT_VEC2,	eval_selection_uvec2	},
		{ TYPE_UINT_VEC3,	eval_selection_uvec3	},
		{ TYPE_UINT_VEC4,	eval_selection_uvec4	},
		{ TYPE_BOOL,		eval_selection_bool		},
		{ TYPE_BOOL_VEC2,	eval_selection_bvec2	},
		{ TYPE_BOOL_VEC3,	eval_selection_bvec3	},
		{ TYPE_BOOL_VEC4,	eval_selection_bvec4	}
	};

	TestCaseGroup* selectionGroup = new TestCaseGroup(m_context, "selection", "Selection operator tests");
	addChild(selectionGroup);

	for (int typeNdx = 0; typeNdx < DE_LENGTH_OF_ARRAY(s_selectionInfo); typeNdx++)
	{
		DataType		curType			= s_selectionInfo[typeNdx].type;
		ShaderEvalFunc	evalFunc		= s_selectionInfo[typeNdx].evalFunc;
		bool			isBoolCase		= isDataTypeBoolOrBVec(curType);
		bool			isFloatCase		= isDataTypeFloatOrVec(curType);
		bool			isIntCase		= isDataTypeIntOrIVec(curType);
		bool			isUintCase		= isDataTypeUintOrUVec(curType);
		const char*		dataTypeStr		= getDataTypeName(curType);

		DE_ASSERT(isBoolCase || isFloatCase || isIntCase || isUintCase);
		DE_UNREF(isIntCase);

		for (int precision = 0; precision < (int)PRECISION_LAST; precision++)
		{
			if (isBoolCase && precision != PRECISION_MEDIUMP) // Use mediump interpolators for booleans.
				continue;

			const char*	precisionStr	= getPrecisionName((Precision)precision);
			string		precisionPrefix	= isBoolCase ? "" : (string(precisionStr) + "_");

			for (int shaderTypeNdx = 0; shaderTypeNdx < DE_LENGTH_OF_ARRAY(s_shaderTypes); shaderTypeNdx++)
			{
				ShaderType		shaderType		= s_shaderTypes[shaderTypeNdx];
				ShaderDataSpec	shaderSpec;
				const char*		shaderTypeName	= getShaderTypeName(shaderType);
				bool			isVertexCase	= (ShaderType)shaderType == SHADERTYPE_VERTEX;

				string name	= precisionPrefix + dataTypeStr + "_" + shaderTypeName;

				shaderSpec.numInputs		= 3;
				shaderSpec.precision		= isBoolCase ? PRECISION_LAST : (Precision)precision;
				shaderSpec.output			= curType;
				shaderSpec.resultScale		= isBoolCase ? 1.0f : isFloatCase ? 0.5f : isUintCase ? 0.5f : 0.1f;
				shaderSpec.resultBias		= isBoolCase ? 0.0f : isFloatCase ? 0.5f : isUintCase ? 0.0f : 0.5f;
				shaderSpec.referenceScale	= shaderSpec.resultScale;
				shaderSpec.referenceBias	= shaderSpec.resultBias;

				float rangeMin = isBoolCase ? -1.0f : isFloatCase ? -1.0f : isUintCase ? 0.0f : -5.0f;
				float rangeMax = isBoolCase ?  1.0f : isFloatCase ?  1.0f : isUintCase ? 2.0f :  5.0f;

				shaderSpec.inputs[0] = ShaderValue(TYPE_BOOL, -1.0f, 1.0f);
				shaderSpec.inputs[1] = ShaderValue(curType, rangeMin, rangeMax);
				shaderSpec.inputs[2] = ShaderValue(curType, rangeMin, rangeMax);

				selectionGroup->addChild(new ShaderOperatorCase(m_context, name.c_str(), "", isVertexCase, evalFunc, "res = in0 ? in1 : in2;", shaderSpec));
			}
		}
	}

	// The sequence operator (comma).

	TestCaseGroup* sequenceGroup = new TestCaseGroup(m_context, "sequence", "Sequence operator tests");
	addChild(sequenceGroup);

	TestCaseGroup* sequenceNoSideEffGroup = new TestCaseGroup(m_context, "no_side_effects", "Sequence tests without side-effects");
	TestCaseGroup* sequenceSideEffGroup = new TestCaseGroup(m_context, "side_effects", "Sequence tests with side-effects");
	sequenceGroup->addChild(sequenceNoSideEffGroup);
	sequenceGroup->addChild(sequenceSideEffGroup);

	static const struct
	{
		bool			containsSideEffects;
		const char*		caseName;
		const char*		expressionStr;
		int				numInputs;
		DataType		inputTypes[MAX_INPUTS];
		DataType		resultType;
		ShaderEvalFunc	evalFunc;
	} s_sequenceCases[] =
	{
		{ false,	"vec4",					"in0, in2 + in1, in1 + in0",							3,	{ TYPE_FLOAT_VEC4,	TYPE_FLOAT_VEC4,	TYPE_FLOAT_VEC4	},	TYPE_FLOAT_VEC4,	evalSequenceNoSideEffCase0 },
		{ false,	"float_uint",			"in0 + in2, in1 + in1",									3,	{ TYPE_FLOAT,		TYPE_UINT,			TYPE_FLOAT		},	TYPE_UINT,			evalSequenceNoSideEffCase1 },
		{ false,	"bool_vec2",			"in0 && in1, in0, ivec2(vec2(in0) + in2)",				3,	{ TYPE_BOOL,		TYPE_BOOL,			TYPE_FLOAT_VEC2	},	TYPE_INT_VEC2,		evalSequenceNoSideEffCase2 },
		{ false,	"vec4_ivec4_bvec4",		"in0 + vec4(in1), in2, in1",							3,	{ TYPE_FLOAT_VEC4,	TYPE_INT_VEC4,		TYPE_BOOL_VEC4	},	TYPE_INT_VEC4,		evalSequenceNoSideEffCase3 },

		{ true,		"vec4",					"in0++, in1 = in0 + in2, in2 = in1",					3,	{ TYPE_FLOAT_VEC4,	TYPE_FLOAT_VEC4,	TYPE_FLOAT_VEC4	},	TYPE_FLOAT_VEC4,	evalSequenceSideEffCase0 },
		{ true,		"float_uint",			"in1++, in0 = float(in1), in1 = uint(in0 + in2)",		3,	{ TYPE_FLOAT,		TYPE_UINT,			TYPE_FLOAT		},	TYPE_UINT,			evalSequenceSideEffCase1 },
		{ true,		"bool_vec2",			"in1 = in0, in2++, in2 = in2 + vec2(in1), ivec2(in2)",	3,	{ TYPE_BOOL,		TYPE_BOOL,			TYPE_FLOAT_VEC2	},	TYPE_INT_VEC2,		evalSequenceSideEffCase2 },
		{ true,		"vec4_ivec4_bvec4",		"in0 = in0 + vec4(in2), in1 = in1 + ivec4(in0), in1++",	3,	{ TYPE_FLOAT_VEC4,	TYPE_INT_VEC4,		TYPE_BOOL_VEC4	},	TYPE_INT_VEC4,		evalSequenceSideEffCase3 }
	};

	for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(s_sequenceCases); caseNdx++)
	{
		for (int precision = 0; precision < (int)PRECISION_LAST; precision++)
		{
			for (int shaderTypeNdx = 0; shaderTypeNdx < DE_LENGTH_OF_ARRAY(s_shaderTypes); shaderTypeNdx++)
			{
				ShaderType		shaderType		= s_shaderTypes[shaderTypeNdx];
				ShaderDataSpec	shaderSpec;
				const char*		shaderTypeName	= getShaderTypeName(shaderType);
				bool			isVertexCase	= (ShaderType)shaderType == SHADERTYPE_VERTEX;

				string name	= string("") + getPrecisionName((Precision)precision) + "_" + s_sequenceCases[caseNdx].caseName + "_" + shaderTypeName;

				shaderSpec.numInputs		= s_sequenceCases[caseNdx].numInputs;
				shaderSpec.precision		= (Precision)precision;
				shaderSpec.output			= s_sequenceCases[caseNdx].resultType;
				shaderSpec.resultScale		= 0.5f;
				shaderSpec.resultBias		= 0.0f;
				shaderSpec.referenceScale	= shaderSpec.resultScale;
				shaderSpec.referenceBias	= shaderSpec.resultBias;

				for (int inputNdx = 0; inputNdx < s_sequenceCases[caseNdx].numInputs; inputNdx++)
				{
					DataType	type		= s_sequenceCases[caseNdx].inputTypes[inputNdx];
					float		rangeMin	= isDataTypeFloatOrVec(type) ? -0.5f : isDataTypeIntOrIVec(type) ? -2.0f : isDataTypeUintOrUVec(type) ? 0.0f : -1.0f;
					float		rangeMax	= isDataTypeFloatOrVec(type) ?  0.5f : isDataTypeIntOrIVec(type) ?  2.0f : isDataTypeUintOrUVec(type) ? 2.0f :  1.0f;

					shaderSpec.inputs[inputNdx] = ShaderValue(type, rangeMin, rangeMax);
				}

				string expression = string("") + "res = (" + s_sequenceCases[caseNdx].expressionStr + ");";

				TestCaseGroup* group = s_sequenceCases[caseNdx].containsSideEffects ? sequenceSideEffGroup : sequenceNoSideEffGroup;
				group->addChild(new ShaderOperatorCase(m_context, name.c_str(), "", isVertexCase, s_sequenceCases[caseNdx].evalFunc, expression.c_str(), shaderSpec));
			}
		}
	}
}

} // Functional
} // gles3
} // deqp