diff --git a/Eigen/src/Core/CoreEvaluators.h b/Eigen/src/Core/CoreEvaluators.h
index ce00566a5..850877079 100644
--- a/Eigen/src/Core/CoreEvaluators.h
+++ b/Eigen/src/Core/CoreEvaluators.h
@@ -934,6 +934,16 @@ struct unary_evaluator<Replicate<ArgType, RowFactor, ColFactor> >
     
     return m_argImpl.coeff(actual_row, actual_col);
   }
+  
+  EIGEN_DEVICE_FUNC CoeffReturnType coeff(Index index) const
+  {
+    // try to avoid using modulo; this is a pure optimization strategy
+    const Index actual_index = internal::traits<XprType>::RowsAtCompileTime==1
+                                  ? (ColFactor==1 ?  index : index%m_cols.value())
+                                  : (RowFactor==1 ?  index : index%m_rows.value());
+    
+    return m_argImpl.coeff(actual_index);
+  }
 
   template<int LoadMode>
   PacketReturnType packet(Index row, Index col) const
@@ -947,6 +957,16 @@ struct unary_evaluator<Replicate<ArgType, RowFactor, ColFactor> >
 
     return m_argImpl.template packet<LoadMode>(actual_row, actual_col);
   }
+  
+  template<int LoadMode>
+  PacketReturnType packet(Index index) const
+  {
+    const Index actual_index = internal::traits<XprType>::RowsAtCompileTime==1
+                                  ? (ColFactor==1 ?  index : index%m_cols.value())
+                                  : (RowFactor==1 ?  index : index%m_rows.value());
+
+    return m_argImpl.template packet<LoadMode>(actual_index);
+  }
  
 protected:
   const ArgTypeNested m_arg; // FIXME is it OK to store both the argument and its evaluator?? (we have the same situation in evaluator_product)
diff --git a/Eigen/src/Core/Replicate.h b/Eigen/src/Core/Replicate.h
index 3777049ee..b976cbd54 100644
--- a/Eigen/src/Core/Replicate.h
+++ b/Eigen/src/Core/Replicate.h
@@ -91,31 +91,6 @@ template<typename MatrixType,int RowFactor,int ColFactor> class Replicate
     inline Index rows() const { return m_matrix.rows() * m_rowFactor.value(); }
     inline Index cols() const { return m_matrix.cols() * m_colFactor.value(); }
 
-    inline Scalar coeff(Index rowId, Index colId) const
-    {
-      // try to avoid using modulo; this is a pure optimization strategy
-      const Index actual_row  = internal::traits<MatrixType>::RowsAtCompileTime==1 ? 0
-                            : RowFactor==1 ? rowId
-                            : rowId%m_matrix.rows();
-      const Index actual_col  = internal::traits<MatrixType>::ColsAtCompileTime==1 ? 0
-                            : ColFactor==1 ? colId
-                            : colId%m_matrix.cols();
-
-      return m_matrix.coeff(actual_row, actual_col);
-    }
-    template<int LoadMode>
-    inline PacketScalar packet(Index rowId, Index colId) const
-    {
-      const Index actual_row  = internal::traits<MatrixType>::RowsAtCompileTime==1 ? 0
-                            : RowFactor==1 ? rowId
-                            : rowId%m_matrix.rows();
-      const Index actual_col  = internal::traits<MatrixType>::ColsAtCompileTime==1 ? 0
-                            : ColFactor==1 ? colId
-                            : colId%m_matrix.cols();
-
-      return m_matrix.template packet<LoadMode>(actual_row, actual_col);
-    }
-
     const _MatrixTypeNested& nestedExpression() const
     { 
       return m_matrix;