finally commit Rohit's work as the start of a new (currently

unsupported) module, MoreVectorization.

CCMAIL:rpg.314@gmail.com

5 files changed, 127 insertions, 1 deletions

diff --git a/unsupported/Eigen/CMakeLists.txt b/unsupported/Eigen/CMakeLists.txt
index eb54f88ac..49c6817fb 100644
--- a/unsupported/Eigen/CMakeLists.txt
+++ b/unsupported/Eigen/CMakeLists.txt
@@ -1,4 +1,4 @@
-set(Eigen_HEADERS AdolcForward BVH IterativeSolvers)
+set(Eigen_HEADERS AdolcForward BVH IterativeSolvers MoreVectorization)
 
 install(FILES
   ${Eigen_HEADERS}
diff --git a/unsupported/Eigen/MoreVectorization b/unsupported/Eigen/MoreVectorization
new file mode 100644
index 000000000..e60526a3d
--- /dev/null
+++ b/unsupported/Eigen/MoreVectorization
@@ -0,0 +1,16 @@
+#ifndef EIGEN_MOREVECTORIZATION_MODULE_H
+#define EIGEN_MOREVECTORIZATION_MODULE_H
+
+#include <Eigen/Core>
+
+namespace Eigen {
+
+/** \ingroup Unsupported_modules
+  * \defgroup MoreVectorization additional vectorization module
+  */
+
+#include "src/MoreVectorization/MathFunctions.h"
+
+}
+
+#endif // EIGEN_MOREVECTORIZATION_MODULE_H
diff --git a/unsupported/Eigen/src/CMakeLists.txt b/unsupported/Eigen/src/CMakeLists.txt
index c2f63db20..3a688afd8 100644
--- a/unsupported/Eigen/src/CMakeLists.txt
+++ b/unsupported/Eigen/src/CMakeLists.txt
@@ -1,3 +1,4 @@
 ADD_SUBDIRECTORY(IterativeSolvers)
 ADD_SUBDIRECTORY(BVH)
 ADD_SUBDIRECTORY(AutoDiff)
+ADD_SUBDIRECTORY(MoreVectorization)
diff --git a/unsupported/Eigen/src/MoreVectorization/CMakeLists.txt b/unsupported/Eigen/src/MoreVectorization/CMakeLists.txt
new file mode 100644
index 000000000..1b887cc8e
--- /dev/null
+++ b/unsupported/Eigen/src/MoreVectorization/CMakeLists.txt
@@ -0,0 +1,6 @@
+FILE(GLOB Eigen_MoreVectorization_SRCS "*.h")
+
+INSTALL(FILES
+  ${Eigen_MoreVectorization_SRCS}
+  DESTINATION ${INCLUDE_INSTALL_DIR}/unsupported/Eigen/src/MoreVectorization COMPONENT Devel
+  )
diff --git a/unsupported/Eigen/src/MoreVectorization/MathFunctions.h b/unsupported/Eigen/src/MoreVectorization/MathFunctions.h
new file mode 100644
index 000000000..351131472
--- /dev/null
+++ b/unsupported/Eigen/src/MoreVectorization/MathFunctions.h
@@ -0,0 +1,103 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra. Eigen itself is part of the KDE project.
+//
+// Copyright (C) 2009 Rohit Garg <rpg.314@gmail.com>
+// Copyright (C) 2009 Benoit Jacob <jacob.benoit.1@gmail.com>
+//
+// Eigen is free software; you can redistribute it and/or
+// modify it under the terms of the GNU Lesser General Public
+// License as published by the Free Software Foundation; either
+// version 3 of the License, or (at your option) any later version.
+//
+// Alternatively, you can redistribute it and/or
+// modify it under the terms of the GNU General Public License as
+// published by the Free Software Foundation; either version 2 of
+// the License, or (at your option) any later version.
+//
+// Eigen is distributed in the hope that it will be useful, but WITHOUT ANY
+// WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+// FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License or the
+// GNU General Public License for more details.
+//
+// You should have received a copy of the GNU Lesser General Public
+// License and a copy of the GNU General Public License along with
+// Eigen. If not, see <http://www.gnu.org/licenses/>.
+
+#ifndef EIGEN_MOREVECTORIZATION_MATHFUNCTIONS_H
+#define EIGEN_MOREVECTORIZATION_MATHFUNCTIONS_H
+
+/** \internal \returns the arcsin of \a a (coeff-wise) */
+template<typename Packet> inline static Packet ei_pasin(Packet a) { return std::asin(a); }
+
+#ifdef EIGEN_VECTORIZE_SSE
+
+template<> EIGEN_DONT_INLINE Packet4f ei_pasin(Packet4f x)
+{
+  _EIGEN_DECLARE_CONST_Packet4f(half, 0.5);
+  _EIGEN_DECLARE_CONST_Packet4f(minus_half, -0.5);
+  _EIGEN_DECLARE_CONST_Packet4f(3half, 1.5);
+
+  _EIGEN_DECLARE_CONST_Packet4f_FROM_INT(sign_mask, 0x80000000);
+
+  _EIGEN_DECLARE_CONST_Packet4f(pi, 3.141592654);
+  _EIGEN_DECLARE_CONST_Packet4f(pi_over_2, 3.141592654*0.5);
+
+  _EIGEN_DECLARE_CONST_Packet4f(asin1, 4.2163199048E-2);
+  _EIGEN_DECLARE_CONST_Packet4f(asin2, 2.4181311049E-2);
+  _EIGEN_DECLARE_CONST_Packet4f(asin3, 4.5470025998E-2);
+  _EIGEN_DECLARE_CONST_Packet4f(asin4, 7.4953002686E-2);
+  _EIGEN_DECLARE_CONST_Packet4f(asin5, 1.6666752422E-1);
+
+  Packet4f a = ei_pabs(x);//got the absolute value
+
+  Packet4f sign_bit= _mm_and_ps(x, ei_p4f_sign_mask);//extracted the sign bit
+
+  Packet4f z1,z2;//will need them during computation    
+
+
+//will compute the two branches for asin
+//so first compare with half
+
+  Packet4f branch_mask= _mm_cmpgt_ps(a, ei_p4f_half);//this is to select which branch to take
+//both will be taken, and finally results will be merged
+//the branch for values >0.5
+
+    {
+//the core series expansion 
+    z1=ei_pmadd(ei_p4f_minus_half,a,ei_p4f_half);
+    Packet4f x1=ei_psqrt(z1);
+    Packet4f s1=ei_pmadd(ei_p4f_asin1, z1, ei_p4f_asin2);
+    Packet4f s2=ei_pmadd(s1, z1, ei_p4f_asin3);
+    Packet4f s3=ei_pmadd(s2,z1, ei_p4f_asin4);
+    Packet4f s4=ei_pmadd(s3,z1, ei_p4f_asin5);
+    Packet4f temp=ei_pmul(s4,z1);//not really a madd but a mul by z so that the next term can be a madd
+    z1=ei_pmadd(temp,x1,x1);
+    z1=ei_padd(z1,z1);
+    z1=ei_psub(ei_p4f_pi_over_2,z1);
+    }
+
+    {
+//the core series expansion 
+    Packet4f x2=a;
+    z2=ei_pmul(x2,x2);
+    Packet4f s1=ei_pmadd(ei_p4f_asin1, z2, ei_p4f_asin2);
+    Packet4f s2=ei_pmadd(s1, z2, ei_p4f_asin3);
+    Packet4f s3=ei_pmadd(s2,z2, ei_p4f_asin4);
+    Packet4f s4=ei_pmadd(s3,z2, ei_p4f_asin5);
+    Packet4f temp=ei_pmul(s4,z2);//not really a madd but a mul by z so that the next term can be a madd
+    z2=ei_pmadd(temp,x2,x2);
+    }
+
+/* select the correct result from the two branch evaluations */
+  z1  = _mm_and_ps(branch_mask, z1);
+  z2  = _mm_andnot_ps(branch_mask, z2);
+  Packet4f z  = _mm_or_ps(z1,z2);
+
+/* update the sign */
+  return _mm_xor_ps(z, sign_bit);
+}
+
+
+#endif
+
+#endif // EIGEN_MOREVECTORIZATION_MATHFUNCTIONS_H