Added vectorized implementation of fast inversesqrt

This commit is contained in:
Christophe Riccio 2014-01-04 13:27:06 +01:00
parent 9b57315681
commit 6ed8c3dbb3
3 changed files with 54 additions and 64 deletions

View File

@ -153,68 +153,53 @@ namespace detail
VECTORIZE_VEC(sqrt)
template <typename genType>
GLM_FUNC_QUALIFIER genType inversesqrt
namespace detail
{
template <template <class, precision> class vecType, typename T, precision P>
struct compute_inversesqrt
{
static vecType<T, P> call(vecType<T, P> const & x)
{
return static_cast<T>(1) / sqrt(x);
}
};
template <template <class, precision> class vecType>
struct compute_inversesqrt<vecType, float, lowp>
{
static vecType<float, lowp> call(vecType<float, lowp> const & x)
{
vecType<float, lowp> tmp(x);
vecType<float, lowp> xhalf(tmp * 0.5f);
vecType<uint, lowp> i = *reinterpret_cast<vecType<uint, lowp>*>(const_cast<vecType<float, lowp>*>(&x));
i = vecType<uint, lowp>(0x5f375a86) - (i >> vecType<uint, lowp>(1));
tmp = *reinterpret_cast<vecType<float, lowp>*>(&i);
tmp = tmp * (1.5f - xhalf * tmp * tmp);
return tmp;
}
};
}//namespace detail
// inversesqrt
GLM_FUNC_QUALIFIER float inversesqrt(float const & x)
{
return 1.0f / sqrt(x);
}
GLM_FUNC_QUALIFIER double inversesqrt(double const & x)
{
return 1.0 / sqrt(x);
}
template <template <class, precision> class vecType, typename T, precision P>
GLM_FUNC_QUALIFIER vecType<T, P> inversesqrt
(
genType const & x
vecType<T, P> const & x
)
{
GLM_STATIC_ASSERT(
std::numeric_limits<genType>::is_iec559,
"'inversesqrt' only accept floating-point inputs");
assert(x > genType(0));
return genType(1) / std::sqrt(x);
GLM_STATIC_ASSERT(std::numeric_limits<T>::is_iec559, "'inversesqrt' only accept floating-point inputs");
return detail::compute_inversesqrt<vecType, T, P>::call(x);
}
VECTORIZE_VEC(inversesqrt)
namespace detail
{
template <typename genType, typename genUType>
genType fastInversesqrt(genType const & v)
{
genType tmp(v);
genType xhalf(tmp * genType(0.5f));
genUType i = *reinterpret_cast<genUType*>(const_cast<genType*>(&v));
i = genUType(0x5f375a86) - (i >> genUType(1));
tmp = *reinterpret_cast<genType*>(&i);
tmp = tmp * (genType(1.5f) - xhalf * tmp * tmp);
return tmp;
}
}
template <>
GLM_FUNC_QUALIFIER lowp_vec1 inversesqrt(lowp_vec1 const & v)
{
assert(glm::all(glm::greaterThan(v, lowp_vec1(0))));
return detail::fastInversesqrt<lowp_vec1, uint>(v);
}
template <>
GLM_FUNC_QUALIFIER lowp_vec2 inversesqrt(lowp_vec2 const & v)
{
assert(glm::all(glm::greaterThan(v, lowp_vec2(0))));
return detail::fastInversesqrt<lowp_vec2, lowp_uvec2>(v);
}
template <>
GLM_FUNC_QUALIFIER lowp_vec3 inversesqrt(lowp_vec3 const & v)
{
assert(glm::all(glm::greaterThan(v, lowp_vec3(0))));
return detail::fastInversesqrt<lowp_vec3, lowp_uvec3>(v);
}
template <>
GLM_FUNC_QUALIFIER lowp_vec4 inversesqrt(lowp_vec4 const & v)
{
assert(glm::all(glm::greaterThan(v, lowp_vec4(0))));
return detail::fastInversesqrt<lowp_vec4, lowp_uvec4>(v);
}
}//namespace glm

View File

@ -74,7 +74,7 @@ namespace glm
GLM_FUNC_DECL T dot(
vecType<T, P> const & x,
vecType<T, P> const & y);
/*
/// Returns the dot product of x and y, i.e., result = x * y.
///
/// @tparam genType Floating-point vector types.
@ -85,7 +85,7 @@ namespace glm
GLM_FUNC_DECL genType dot(
genType const & x,
genType const & y);
*/
/// Returns the cross product of x and y.
///
/// @tparam valType Floating-point scalar types.

View File

@ -24,13 +24,18 @@ namespace glm
VECTORIZE_VEC(fastSqrt)
// fastInversesqrt
template <typename genType>
GLM_FUNC_QUALIFIER genType fastInverseSqrt
GLM_FUNC_QUALIFIER float fastInverseSqrt(float x)
{
return detail::compute_inversesqrt<detail::tvec1, float, lowp>::call(x).x;
}
template <template <class, precision> class vecType, typename T, precision P>
GLM_FUNC_QUALIFIER vecType<T, P> fastInverseSqrt
(
genType const & x
vecType<T, P> const & x
)
{
return detail::fastInversesqrt<genType, uint>(x);
return detail::compute_inversesqrt<vecType, T, P>::call(x);
}
VECTORIZE_VEC(fastInverseSqrt)