Implement experimental SIMD 4x4 matrix products with AVX2

- The new functions are enabled only when compiling with MSVC and "/arch:AVX2".

cml/common/size_tags.h

@@ -78,6 +78,10 @@ template<class T> struct is_fixed_size
     std::is_same<size_tag_of_t<T>, fixed_size_tag>::value;
 };
 
+/** Helper to detect fixed-size types. */
+template<class T> constexpr auto is_fixed_size_v
+  = is_fixed_size<T>::value;
+
 /** Wrapper for enable_if to detect types tagged with fixed_size_tag. */
 template<class Sub, class T = void>
 struct enable_if_fixed_size : std::enable_if<is_fixed_size<Sub>::value, T>

cml/matrix/detail/get.h

@@ -4,14 +4,16 @@
 
 #pragma once
 
+#include <cml/matrix/readable_matrix.h>
 #include <cml/matrix/writable_matrix.h>
 
 namespace cml::detail {
 
 /** Helper to return the passed-in value in response to a matrix index @c
  * (i,j).
  */
-template<class Other>
+template<class Other,
+  typename = std::enable_if_t<!cml::is_matrix<Other>::value>>
 inline auto
 get(const Other& v, int, int) -> const Other&
 {
@@ -20,17 +22,16 @@ get(const Other& v, int, int) -> const Other&
 
 /** Helper to return element @c (i,j) of @c array. */
 template<class Other, int Rows, int Cols>
-inline const Other&
-get(Other const (&array)[Rows][Cols], int i, int j)
--> const Other&
+inline auto
+get(Other const (&array)[Rows][Cols], int i, int j) -> const Other&
 {
   return array[i][j];
 }
 
 /** Helper to return element @c (i,j) of @c array. */
-template<class Other, int Rows, int Cols> inline auto
-get(Other (&array)[Rows][Cols], int i, int j)
--> Other&
+template<class Other, int Rows, int Cols>
+inline auto
+get(Other (&array)[Rows][Cols], int i, int j) -> Other&
 {
   return array[i][j];
 }
@@ -44,11 +45,12 @@ get(const readable_matrix<Sub>& sub, int i, int j) ->
   return sub.get(i, j);
 }
 
-template<class Sub> inline auto
-get(writable_matrix<Sub>& sub, int i, int j)
--> typename matrix_traits<Sub>::mutable_value
+template<class Sub>
+inline auto
+get(writable_matrix<Sub>& sub, int i, int j) ->
+  typename matrix_traits<Sub>::mutable_value
 {
   return sub.get(i, j);
 }
 
-}
+}  // namespace cml::detail

cml/matrix/matrix_product.h

@@ -1,23 +1,23 @@
-/*-------------------------------------------------------------------------
- @@COPYRIGHT@@
- *-----------------------------------------------------------------------*/
-
-#pragma once
-
-#include <cml/matrix/readable_matrix.h>
-#include <cml/matrix/promotion.h>
-
-namespace cml {
-
-/** Multiply two matrices, and return the result as a temporary. */
-template<class Sub1, class Sub2, enable_if_matrix_t<Sub1>* = nullptr,
-  enable_if_matrix_t<Sub2>* = nullptr>
-auto operator*(Sub1&& sub1, Sub2&& sub2)
-  -> matrix_inner_product_promote_t<actual_operand_type_of_t<decltype(sub1)>,
-    actual_operand_type_of_t<decltype(sub2)>>;
-
-}  // namespace cml
-
-#define __CML_MATRIX_MATRIX_PRODUCT_TPP
-#include <cml/matrix/matrix_product.tpp>
-#undef __CML_MATRIX_MATRIX_PRODUCT_TPP
+/*-------------------------------------------------------------------------
+ @@COPYRIGHT@@
+ *-----------------------------------------------------------------------*/
+
+#pragma once
+
+#include <cml/matrix/readable_matrix.h>
+#include <cml/matrix/promotion.h>
+#include <cml/matrix/detail/resize.h>
+
+namespace cml {
+
+/** Multiply two matrices and return the result as a temporary. */
+template<class Sub1, class Sub2, typename = enable_if_matrix_t<Sub1>,
+  typename = enable_if_matrix_t<Sub2>>
+auto operator*(Sub1&& sub1, Sub2&& sub2) -> matrix_inner_product_promote_t<
+  actual_type_of_t<Sub1>, actual_type_of_t<Sub2>>;
+
+}  // namespace cml
+
+#define __CML_MATRIX_MATRIX_PRODUCT_TPP
+#include <cml/matrix/matrix_product.tpp>
+#undef __CML_MATRIX_MATRIX_PRODUCT_TPP