From d77694f8ac08c132afc24c9225af920dc5389c94 Mon Sep 17 00:00:00 2001
From: Christophe Riccio <g.truc.creation@gmail.com>
Date: Tue, 8 Feb 2011 10:37:47 +0000
Subject: [PATCH] Added SIMD sqrt functions support, ticket #2

---
 glm/core/intrinsic_common.inl |  2 +-
 glm/gtx/simd_vec4.hpp         | 22 ++++++++++++++++++++++
 glm/gtx/simd_vec4.inl         | 28 ++++++++++++++++++++++++++++
 3 files changed, 51 insertions(+), 1 deletion(-)

diff --git a/glm/core/intrinsic_common.inl b/glm/core/intrinsic_common.inl
index b11f000d..fd031ef8 100644
--- a/glm/core/intrinsic_common.inl
+++ b/glm/core/intrinsic_common.inl
@@ -258,7 +258,7 @@ inline __m128 sse_inf_ps(__m128 x)
 }
 
 // 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)
 {
 	__m128 recip = _mm_rsqrt_ss(x);  // "estimate" opcode
diff --git a/glm/gtx/simd_vec4.hpp b/glm/gtx/simd_vec4.hpp
index 186fd045..028db5e0 100644
--- a/glm/gtx/simd_vec4.hpp
+++ b/glm/gtx/simd_vec4.hpp
@@ -401,6 +401,28 @@ namespace glm
 			detail::fvec4SIMD const & I,
 			detail::fvec4SIMD const & N,
 			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 gtx
 }//namespace glm
diff --git a/glm/gtx/simd_vec4.inl b/glm/gtx/simd_vec4.inl
index 510a57e6..f74144b8 100644
--- a/glm/gtx/simd_vec4.inl
+++ b/glm/gtx/simd_vec4.inl
@@ -634,6 +634,34 @@ namespace glm
 			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 gtx
 }//namespace glm