From c94fde118a2b6b2b415a7860ba066ba240bc8ac3 Mon Sep 17 00:00:00 2001
From: Gael Guennebaud <g.gael@free.fr>
Date: Fri, 28 Mar 2014 09:11:06 +0100
Subject: [PATCH] Enable vectorization of gemv for PacketSize>4 through
 unaligned loads (still better than no vectorization)

---
 Eigen/src/Core/products/GeneralMatrixVector.h | 18 +++++++++++++-----
 1 file changed, 13 insertions(+), 5 deletions(-)

diff --git a/Eigen/src/Core/products/GeneralMatrixVector.h b/Eigen/src/Core/products/GeneralMatrixVector.h
index 59fc54c4e..340c51394 100644
--- a/Eigen/src/Core/products/GeneralMatrixVector.h
+++ b/Eigen/src/Core/products/GeneralMatrixVector.h
@@ -136,12 +136,17 @@ EIGEN_DONT_INLINE void general_matrix_vector_product<Index,LhsScalar,ColMajor,Co
   // find how many columns do we have to skip to be aligned with the result (if possible)
   Index skipColumns = 0;
   // if the data cannot be aligned (TODO add some compile time tests when possible, e.g. for floats)
-  // TODO: extend the code to support aligned loads whenever possible when LhsPacketSize > 4.
-  if( (size_t(lhs)%sizeof(LhsScalar)) || (size_t(res)%sizeof(ResScalar)) || LhsPacketSize > 4)
+  if( (size_t(lhs)%sizeof(LhsScalar)) || (size_t(res)%sizeof(ResScalar)) )
   {
     alignedSize = 0;
     alignedStart = 0;
   }
+  else if(LhsPacketSize > 4)
+  {
+    // TODO: extend the code to support aligned loads whenever possible when LhsPacketSize > 4.
+    // Currently, it seems to be better to perform unaligned loads anyway
+    alignmentPattern = NoneAligned;
+  }
   else if (LhsPacketSize>1)
   {
     eigen_internal_assert(size_t(lhs+lhsAlignmentOffset)%sizeof(LhsPacket)==0 || size<LhsPacketSize);
@@ -401,13 +406,16 @@ EIGEN_DONT_INLINE void general_matrix_vector_product<Index,LhsScalar,RowMajor,Co
   // find how many rows do we have to skip to be aligned with rhs (if possible)
   Index skipRows = 0;
   // if the data cannot be aligned (TODO add some compile time tests when possible, e.g. for floats)
-  // TODO: extend the code to support aligned loads whenever possible when LhsPacketSize > 4.
-  if( (sizeof(LhsScalar)!=sizeof(RhsScalar)) || (size_t(lhs)%sizeof(LhsScalar)) || (size_t(rhs)%sizeof(RhsScalar)) ||
-      (LhsPacketSize > 4))
+  if( (sizeof(LhsScalar)!=sizeof(RhsScalar)) || (size_t(lhs)%sizeof(LhsScalar)) || (size_t(rhs)%sizeof(RhsScalar)) )
   {
     alignedSize = 0;
     alignedStart = 0;
   }
+  else if(LhsPacketSize > 4)
+  {
+    // TODO: extend the code to support aligned loads whenever possible when LhsPacketSize > 4.
+    alignmentPattern = NoneAligned;
+  }
   else if (LhsPacketSize>1)
   {
     eigen_internal_assert(size_t(lhs+lhsAlignmentOffset)%sizeof(LhsPacket)==0  || depth<LhsPacketSize);