fixes for cuda 11.4

correaa · Feb 18, 2025 · dd63476 · dd63476
1 parent 19f7ac7
commit dd63476
Show file tree

Hide file tree

Showing 5 changed files with 20 additions and 15 deletions.
diff --git a/include/boost/multi/adaptors/blas/numeric.hpp b/include/boost/multi/adaptors/blas/numeric.hpp
@@ -295,15 +295,16 @@ auto conj(A&& array) -> A&& {
 }
 
 template<
-	class A, class D = std::decay_t<A>, typename Elem = typename D::element_type,
+	class A, class D = std::decay_t<A>, typename Elem = typename D::element,
 	typename Ptr = std::decay_t<decltype(std::declval<A&&>().base())>,
+	// typename = decltype(std::declval<A>().size()),
 	// typename = decltype(std::declval<A&&>().template static_array_cast<Elem, conjugater<Ptr>>()),
 	std::enable_if_t<!is_conjugated<A>{} && is_complex_array<A>{}, int> =0>  // NOLINT(modernize-use-constraints) TODO(correaa) for C++20
 auto conj(A&& array) -> decltype(auto) {
 	return std::forward<A>(array).template static_array_cast<Elem, conjugater<Ptr>>();
 }
 
-template<class A, class D = std::decay_t<A>, typename Elem = typename D::element_type,
+template<class A, class D = std::decay_t<A>, typename Elem = typename D::element,
          typename Ptr = typename decltype(std::declval<A&&>().base())::underlying_type,
 		 std::enable_if_t<is_conjugated<A>{}, int> =0>  // NOLINT(modernize-use-constraints) TODO(correaa) for C++20
 auto conj(A&& array)

diff --git a/include/boost/multi/array.hpp b/include/boost/multi/array.hpp
@@ -252,12 +252,13 @@ struct static_array  // NOLINT(fuchsia-multiple-inheritance) : multiple inherita
 
 	template<
 		class Range, class = std::enable_if_t<!std::is_base_of<static_array, std::decay_t<Range>>{}>,
-		class = decltype(/*static_array*/ (std::declval<Range const&>().begin() - std::declval<Range const&>().end())),  // instantiation of static_array here gives a compiler error in 11.0, partially defined type?
+		class = typename Range::iterator,  // this solves a problem in cuda 11.x when element_iterator is made to be pointer-like
+		// class = decltype(/*static_array*/ (std::declval<Range const&>().begin() - std::declval<Range const&>().end())),  // instantiation of static_array here gives a compiler error in 11.0, partially defined type?
 		class = std::enable_if_t<! is_subarray<Range const&>::value>  // NOLINT(modernize-use-constraints) TODO(correaa) in C++20
 	>
 	// cppcheck-suppress noExplicitConstructor ; because I want to use equal for lazy assigments form range-expressions // NOLINTNEXTLINE(runtime/explicit)
 	static_array(Range const& rng)  // NOLINT(google-explicit-constructor,hicpp-explicit-conversions) : to allow terse syntax  // NOSONAR
-	: static_array{std::begin(rng), std::end(rng)} {}  // Sonar: Prefer free functions over member functions when handling objects of generic type "Range".
+	: static_array(std::begin(rng), std::end(rng)) {}  // Sonar: Prefer free functions over member functions when handling objects of generic type "Range".
 
 	template<class TT>
 	auto uninitialized_fill_elements(TT const& value) {

diff --git a/include/boost/multi/array_ref.hpp b/include/boost/multi/array_ref.hpp
@@ -488,7 +488,7 @@ struct array_iterator  // NOLINT(fuchsia-multiple-inheritance) for facades
 	using element_ptr = ElementPtr;
 	using element_const_ptr = typename std::pointer_traits<ElementPtr>::template rebind<element const>;
 	using value_type = typename subarray<element, D-1, element_ptr>::decay_type;
-	// using element_type = value_type;
+	using element_type = value_type;
 
 	using pointer         = subarray<element, D - 1, element_ptr>*;
 	using reference       = std::conditional_t<

diff --git a/include/boost/multi/detail/tuple_zip.hpp b/include/boost/multi/detail/tuple_zip.hpp
@@ -303,15 +303,6 @@ constexpr auto get(tuple<T0, Ts...> const& t) -> auto const& {  // NOLINT(readab
 		return get<N - 1>(t.tail());
 	}
 }
-#if defined __NVCC__
-#ifdef __NVCC_DIAG_PRAGMA_SUPPORT__
-#pragma nv_diagnostic pop
-#else
-#pragma diagnostic pop
-#endif
-#elif defined      __NVCOMPILER
-#pragma diagnostic pop
-#endif
 
 template<std::size_t N, class T0, class... Ts>
 constexpr auto get(tuple<T0, Ts...>& tup) -> auto& {
@@ -330,6 +321,17 @@ constexpr auto get(tuple<T0, Ts...>&& tup) -> auto&& {
 		return get<N - 1>(std::move(std::move(tup).tail()));
 	}
 }
+
+#if defined __NVCC__
+#ifdef __NVCC_DIAG_PRAGMA_SUPPORT__
+#pragma nv_diagnostic pop
+#else
+#pragma diagnostic pop
+#endif
+#elif defined      __NVCOMPILER
+#pragma diagnostic pop
+#endif
+
 #if ! defined(_MSC_VER)
 #pragma GCC diagnostic pop
 #endif

diff --git a/test/array_cref.cpp b/test/array_cref.cpp
@@ -4,7 +4,8 @@
 // https://www.boost.org/LICENSE_1_0.txt
 
 #ifdef __NVCC__  // affects cuda 11.4.3
-#pragma diag_suppress 20014  // error #20014-D: calling a __host__ function from a __host__ __device__ function is not allowed
+// #pragma diag_suppress 20014  // error #20014-D: calling a __host__ function from a __host__ __device__ function is not allowed
+#pragma nv_diag_suppress 20014  // error #20014-D: calling a __host__ function from a __host__ __device__ function is not allowed
 #endif
 
 #include <boost/multi/array.hpp>  // for array, array_ref, subarray, arra...