From: Martin Kronbichler <martin.kronbichler@it.uu.se>
Date: Tue, 30 Nov 2021 08:14:35 +0000 (+0100)
Subject: Use narrowest possible SIMD vector for MappingQ::InternalData
X-Git-Tag: v9.4.0-rc1~784^2~1
X-Git-Url: https://gitweb.dealii.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=c955cb5c33e73221dbc90345476184a356cebca2;p=dealii.git

Use narrowest possible SIMD vector for MappingQ::InternalData
---

diff --git a/include/deal.II/fe/mapping_q.h b/include/deal.II/fe/mapping_q.h
index 7cf2303251..15be913847 100644
--- a/include/deal.II/fe/mapping_q.h
+++ b/include/deal.II/fe/mapping_q.h
@@ -474,43 +474,51 @@ public:
      */
     QGaussLobatto<1> line_support_points;
 
+    /**
+     * A vectorized array type to reflect the necessary number of components
+     * for all interpolations to be done by this class.
+     */
+    using VectorizedArrayType =
+      VectorizedArray<double,
+                      std::min<std::size_t>(VectorizedArray<double>::size(),
+                                            (dim <= 2 ? 2 : 4))>;
+
     /**
      * In case the quadrature rule given represents a tensor product
      * we need to store the evaluations of the 1d polynomials at
      * the 1d quadrature points. That is what this variable is for.
      */
-    internal::MatrixFreeFunctions::ShapeInfo<VectorizedArray<double>>
-      shape_info;
+    internal::MatrixFreeFunctions::ShapeInfo<VectorizedArrayType> shape_info;
 
     /**
      * In case the quadrature rule given represents a tensor product
      * we need to store temporary data in this object.
      */
-    mutable AlignedVector<VectorizedArray<double>> scratch;
+    mutable AlignedVector<VectorizedArrayType> scratch;
 
     /**
      * In case the quadrature rule given represents a tensor product
      * the values at the mapped support points are stored in this object.
      */
-    mutable AlignedVector<VectorizedArray<double>> values_dofs;
+    mutable AlignedVector<VectorizedArrayType> values_dofs;
 
     /**
      * In case the quadrature rule given represents a tensor product
      * the values at the quadrature points are stored in this object.
      */
-    mutable AlignedVector<VectorizedArray<double>> values_quad;
+    mutable AlignedVector<VectorizedArrayType> values_quad;
 
     /**
      * In case the quadrature rule given represents a tensor product
      * the gradients at the quadrature points are stored in this object.
      */
-    mutable AlignedVector<VectorizedArray<double>> gradients_quad;
+    mutable AlignedVector<VectorizedArrayType> gradients_quad;
 
     /**
      * In case the quadrature rule given represents a tensor product
      * the hessians at the quadrature points are stored in this object.
      */
-    mutable AlignedVector<VectorizedArray<double>> hessians_quad;
+    mutable AlignedVector<VectorizedArrayType> hessians_quad;
 
     /**
      * Indicates whether the given Quadrature object is a tensor product.
diff --git a/include/deal.II/fe/mapping_q_internal.h b/include/deal.II/fe/mapping_q_internal.h
index 3f432f3bcd..9df99a0404 100644
--- a/include/deal.II/fe/mapping_q_internal.h
+++ b/include/deal.II/fe/mapping_q_internal.h
@@ -1079,9 +1079,12 @@ namespace internal
     {
       const UpdateFlags update_flags = data.update_each;
 
+      using VectorizedArrayType =
+        typename dealii::MappingQ<dim,
+                                  spacedim>::InternalData::VectorizedArrayType;
       const unsigned int     n_shape_values = data.n_shape_functions;
       const unsigned int     n_q_points     = data.shape_info.n_q_points;
-      constexpr unsigned int n_lanes        = VectorizedArray<double>::size();
+      constexpr unsigned int n_lanes        = VectorizedArrayType::size();
       constexpr unsigned int n_comp         = 1 + (spacedim - 1) / n_lanes;
       constexpr unsigned int n_hessians     = (dim * (dim + 1)) / 2;
 
@@ -1148,7 +1151,7 @@ namespace internal
               }
 
           // do the actual tensorized evaluation
-          internal::FEEvaluationFactory<dim, double, VectorizedArray<double>>::
+          internal::FEEvaluationFactory<dim, double, VectorizedArrayType>::
             evaluate(n_comp,
                      evaluation_flag,
                      data.shape_info,