From 684d76eba3244d5ae75602b8ec06e81ef74da79b Mon Sep 17 00:00:00 2001
From: Benoit Jacob <jacob.benoit.1@gmail.com>
Date: Tue, 24 Nov 2009 15:12:43 -0500
Subject: add SSE4 support, start with integer multiplication

---
 Eigen/Core                           | 6 ++++++
 Eigen/src/Core/arch/SSE/PacketMath.h | 4 ++++
 2 files changed, 10 insertions(+)

(limited to 'Eigen')

diff --git a/Eigen/Core b/Eigen/Core
index 2968e36c6..e095aee11 100644
--- a/Eigen/Core
+++ b/Eigen/Core
@@ -67,6 +67,12 @@
     #ifdef __SSSE3__
       #include <tmmintrin.h>
     #endif
+    #ifdef __SSE4_1__
+      #include <smmintrin.h>
+    #endif
+    #ifdef __SSE4_2__
+      #include <nmmintrin.h>
+    #endif
   #elif defined __ALTIVEC__
     #define EIGEN_VECTORIZE
     #define EIGEN_VECTORIZE_ALTIVEC
diff --git a/Eigen/src/Core/arch/SSE/PacketMath.h b/Eigen/src/Core/arch/SSE/PacketMath.h
index 60ccadc21..69f6979bd 100644
--- a/Eigen/src/Core/arch/SSE/PacketMath.h
+++ b/Eigen/src/Core/arch/SSE/PacketMath.h
@@ -118,6 +118,9 @@ template<> EIGEN_STRONG_INLINE Packet4f ei_pmul<Packet4f>(const Packet4f& a, con
 template<> EIGEN_STRONG_INLINE Packet2d ei_pmul<Packet2d>(const Packet2d& a, const Packet2d& b) { return _mm_mul_pd(a,b); }
 template<> EIGEN_STRONG_INLINE Packet4i ei_pmul<Packet4i>(const Packet4i& a, const Packet4i& b)
 {
+#ifdef __SSE4_1__
+  return _mm_mullo_epi32(a,b);
+#else
   // this version is slightly faster than 4 scalar products
   return ei_vec4i_swizzle1(
             ei_vec4i_swizzle2(
@@ -126,6 +129,7 @@ template<> EIGEN_STRONG_INLINE Packet4i ei_pmul<Packet4i>(const Packet4i& a, con
                             ei_vec4i_swizzle1(b,1,0,3,2)),
               0,2,0,2),
             0,2,1,3);
+#endif
 }
 
 template<> EIGEN_STRONG_INLINE Packet4f ei_pdiv<Packet4f>(const Packet4f& a, const Packet4f& b) { return _mm_div_ps(a,b); }
-- 
cgit v1.2.3