mewin/glm
1
0
Fork 0
Code Releases Activity

Added SIMD sqrt functions support, ticket #2

Browse Source
This commit is contained in:
Christophe Riccio 2011-02-08 10:37:47 +00:00
parent 9fbfb8c208
commit d77694f8ac
3 changed files with 51 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
glm/core/intrinsic_common.inl
View File

@ -258,7 +258,7 @@ inline __m128 sse_inf_ps(__m128 x)
} }
// SSE scalar reciprocal sqrt using rsqrt op, plus one Newton-Rhaphson iteration // SSE scalar reciprocal sqrt using rsqrt op, plus one Newton-Rhaphson iteration
// By Elan Ruskin, // By Elan Ruskin, http://assemblyrequired.crashworks.org/
inline __m128 sse_sqrt_wip_ss(__m128 const & x) inline __m128 sse_sqrt_wip_ss(__m128 const & x)
{ {
__m128 recip = _mm_rsqrt_ss(x); // "estimate" opcode __m128 recip = _mm_rsqrt_ss(x); // "estimate" opcode

22
glm/gtx/simd_vec4.hpp
View File

@ -401,6 +401,28 @@ namespace glm
detail::fvec4SIMD const & I, detail::fvec4SIMD const & I,
detail::fvec4SIMD const & N, detail::fvec4SIMD const & N,
float const & eta); float const & eta);
//! Returns the positive square root of x.
//! (From GLM_GTX_simd_vec4 extension, exponential function)
detail::fvec4SIMD simdSqrt(
detail::fvec4SIMD const & x);
//! Returns the positive square root of x with an accuracy slight lower or equal than simdSqrt but much faster.
//! (From GLM_GTX_simd_vec4 extension, exponential function)
detail::fvec4SIMD simdFastSqrt(
detail::fvec4SIMD const & x);
//! Returns the reciprocal of the positive square root of x.
//! (From GLM_GTX_simd_vec4 extension, exponential function)
detail::fvec4SIMD simdInversesqrt(
detail::fvec4SIMD const & x);
//! Returns the reciprocal of the positive square root of x,
//! faster than simdInversesqrt but less accurate.
//! (From GLM_GTX_simd_vec4 extension, exponential function)
detail::fvec4SIMD simdFastInversesqrt(
detail::fvec4SIMD const & x);
}//namespace simd_vec4 }//namespace simd_vec4
}//namespace gtx }//namespace gtx
}//namespace glm }//namespace glm

28
glm/gtx/simd_vec4.inl
View File

@ -634,6 +634,34 @@ namespace glm
return detail::sse_rfa_ps(I.Data, N.Data, _mm_set1_ps(eta)); return detail::sse_rfa_ps(I.Data, N.Data, _mm_set1_ps(eta));
} }
inline detail::fvec4SIMD simdSqrt(detail::fvec4SIMD const & x)
{
return _mm_sqrt_ps(x.Data);
}
inline detail::fvec4SIMD simdFastSqrt(detail::fvec4SIMD const & x)
{
}
// SSE scalar reciprocal sqrt using rsqrt op, plus one Newton-Rhaphson iteration
// By Elan Ruskin, http://assemblyrequired.crashworks.org/
inline detail::fvec4SIMD simdInversesqrt(detail::fvec4SIMD const & x)
{
GLM_ALIGN(4) static const __m128 three = {3, 3, 3, 3}; // aligned consts for fast load
GLM_ALIGN(4) static const __m128 half = {0.5,0.5,0.5,0.5};
__m128 recip = _mm_rsqrt_ps(x.Data); // "estimate" opcode
__m128 halfrecip = _mm_mul_ps(half, recip);
__m128 threeminus_xrr = _mm_sub_ps(three, _mm_mul_ps(x.Data, _mm_mul_ps(recip, recip)));
return _mm_mul_ps(halfrecip, threeminus_xrr);
}
inline detail::fvec4SIMD simdFastInversesqrt(detail::fvec4SIMD const & x)
{
return _mm_rsqrt_ps(x.Data);
}
}//namespace simd_vec4 }//namespace simd_vec4
}//namespace gtx }//namespace gtx
}//namespace glm }//namespace glm