Replace tuple-style cast with pointer/stride access

It was noted in #1 (closed) that the get operator will not fully vectorize due to the tuple-style cast hidden in getStructMember. The pointers upfront are now cached in attempts to alleviate this. This changes the manner of data access in the AoSoA for arrays:

return getStructMember<I>(_raw_data[idx.s()]);

to:

return static_cast<struct_member_array_type<I> >( data<I>() + idx.s() * _strides[I] );

If this allows vectorization, then it may be well worth the extra arithmetic needed to index into the data block. Furthermore, if loop-invariant hoisting were done well by the compiler, idx.s() * _strides[I] would be hoisted above the inner loop over arrays as a further optimization.

Cacheing pointers and strides also had a nice side effect for future C interfaces - template-free pointer and stride access! A user can now get strides and (un-typed void*) pointers directly through the following interface:

size_t PosX = 0;
double* pos_x = (double*) aosoa.pointer(PosX);
size_t pos_x_stride = aosoa.stride(PosX);

core/src/Cabana_AoSoA.hpp

@@ -97,7 +97,7 @@ class AoSoA<MemberDataTypes<Types...>,Device,ArraySize>
     // Struct member array element pointer type at a given index I.
     template<std::size_t I>
     using struct_member_pointer_type =
-        typename std::add_pointer<struct_member_data_type<I> >::type;
+        typename std::add_pointer<struct_member_value_type<I> >::type;
 
     // Struct member array element const pointer type at a given index I.
     template<std::size_t I>
@@ -263,7 +263,7 @@ class AoSoA<MemberDataTypes<Types...>,Device,ArraySize>
                             struct_member_reference_type<I> >::type
     get( const Index& idx )
     {
-        return data<I>()[idx.s() * _strides[I] + idx.i()];
+        return array<I>(idx.s())[idx.i()];
     }
 
     template<std::size_t I>
@@ -272,7 +272,7 @@ class AoSoA<MemberDataTypes<Types...>,Device,ArraySize>
                             struct_member_const_reference_type<I> >::type
     get( const Index& idx ) const
     {
-        return data<I>()[idx.s() * _strides[I] + idx.i()];
+        return array<I>(idx.s())[idx.i()];
     }
 
     // Rank 1
@@ -283,7 +283,7 @@ class AoSoA<MemberDataTypes<Types...>,Device,ArraySize>
     get( const Index& idx,
          const int d0 )
     {
-        return data<I>()[idx.s() * _strides[I] + idx.i()][d0];
+        return array<I>(idx.s())[idx.i()][d0];
     }
 
     template<std::size_t I>
@@ -293,7 +293,7 @@ class AoSoA<MemberDataTypes<Types...>,Device,ArraySize>
     get( const Index& idx,
          const int d0 ) const
     {
-        return data<I>()[idx.s() * _strides[I] + idx.i()][d0];
+        return array<I>(idx.s())[idx.i()][d0];
     }
 
     // Rank 2
@@ -305,7 +305,7 @@ class AoSoA<MemberDataTypes<Types...>,Device,ArraySize>
          const int d0,
          const int d1 )
     {
-        return data<I>()[idx.s() * _strides[I] + idx.i()][d0][d1];
+        return array<I>(idx.s())[idx.i()][d0][d1];
     }
 
     template<std::size_t I>
@@ -316,7 +316,7 @@ class AoSoA<MemberDataTypes<Types...>,Device,ArraySize>
          const int d0,
          const int d1 ) const
     {
-        return data<I>()[idx.s() * _strides[I] + idx.i()][d0][d1];
+        return array<I>(idx.s())[idx.i()][d0][d1];
     }
 
     // Rank 3
@@ -329,7 +329,7 @@ class AoSoA<MemberDataTypes<Types...>,Device,ArraySize>
          const int d1,
          const int d2 )
     {
-        return data<I>()[idx.s() * _strides[I] + idx.i()][d0][d1][d2];
+        return array<I>(idx.s())[idx.i()][d0][d1][d2];
     }
 
     template<std::size_t I>
@@ -341,7 +341,7 @@ class AoSoA<MemberDataTypes<Types...>,Device,ArraySize>
          const int d1,
          const int d2 ) const
     {
-        return data<I>()[idx.s() * _strides[I] + idx.i()][d0][d1][d2];
+        return array<I>(idx.s())[idx.i()][d0][d1][d2];
     }
 
     // Rank 4
@@ -355,7 +355,7 @@ class AoSoA<MemberDataTypes<Types...>,Device,ArraySize>
          const int d2,
          const int d3 )
     {
-        return data<I>()[idx.s() * _strides[I] + idx.i()][d0][d1][d2][d3];
+        return array<I>(idx.s())[idx.i()][d0][d1][d2][d3];
     }
 
     template<std::size_t I>
@@ -368,51 +368,70 @@ class AoSoA<MemberDataTypes<Types...>,Device,ArraySize>
          const int d2,
          const int d3 ) const
     {
-        return data<I>()[idx.s() * _strides[I] + idx.i()][d0][d1][d2][d3];
+        return array<I>(idx.s())[idx.i()][d0][d1][d2][d3];
     }
 
     // -------------------------------
     // Raw data access.
 
-    // Get the stride between SoA data for a given member at index I.
+    // Get the stride between SoA data for a given member at index I. Note
+    // that this strides are computed in the context of the *value_type* for
+    // each member.
     CABANA_INLINE_FUNCTION
     std::size_t stride( const std::size_t I ) const
     {
         return _strides[I];
     }
 
+    // Get an un-typed raw pointer to the data for a given member at index
+    // I. Users will need to cast this pointer to the appropriate type for the
+    // stride associated with this member to mean anything.
+    CABANA_INLINE_FUNCTION
+    void* data( const std::size_t I )
+    {
+        return _pointers[I];
+    }
+
+    CABANA_INLINE_FUNCTION
+    const void* data( const std::size_t I ) const
+    {
+        return _pointers[I];
+    }
+
+  private:
+
     // Get a typed pointer to the data for a given member at index I.
     template<std::size_t I>
     CABANA_INLINE_FUNCTION
-    struct_member_pointer_type<I> data()
+    struct_member_pointer_type<I> typedPointer()
     {
         return static_cast<struct_member_pointer_type<I> >( _pointers[I] );
     }
 
     template<std::size_t I>
     CABANA_INLINE_FUNCTION
-    struct_member_const_pointer_type<I> data() const
+    struct_member_const_pointer_type<I> typedPointer() const
     {
         return static_cast<struct_member_pointer_type<I> >( _pointers[I] );
     }
 
-    // Get an un-typed raw pointer to the data for a given member at index
-    // I. Users will need to cast this pointer to the appropriate type for the
-    // stride associated with this member to mean anything.
+    // Get the array at the given struct index.
+    template<std::size_t I>
     CABANA_INLINE_FUNCTION
-    void* pointer( const std::size_t I )
+    struct_member_array_type<I> array( const std::size_t s )
     {
-        return _pointers[I];
+        return reinterpret_cast<struct_member_array_type<I> >(
+            typedPointer<I>() + s * _strides[I] );
     }
 
+    template<std::size_t I>
     CABANA_INLINE_FUNCTION
-    const void* pointer( const std::size_t I ) const
+    struct_member_const_array_type<I> array( const std::size_t s ) const
     {
-        return _pointers[I];
+        return reinterpret_cast<struct_member_array_type<I> >(
+            typedPointer<I>() + s * _strides[I] );
     }
 
-  private:
-
     // Store the pointers and strides for each member element.
     template<std::size_t N>
     void assignPointersAndStrides()
@@ -460,7 +479,8 @@ class AoSoA<MemberDataTypes<Types...>,Device,ArraySize>
     // Pointers to the first element of each member.
     void* _pointers[number_of_members];
 
-    // Strides for each member.
+    // Strides for each member. Note that these strides are computed in the
+    // context of the *value_type* of each member.
     std::size_t _strides[number_of_members];
 };