Add more dialects to host only TU tests (#346)

* Expand host TU testing up to c++20, clear test files during compose to save space
* Fix missing H/D annotations in cmath
* Fix Span using deduction guides in gcc-6+c++17.
* Block gcc-6 on deduction guides test in <span>
* Remove compiler exclusion test from host_only tests, add a try_compile for dialect instead
* Fix usage of long double builtins on newer MSVC versions

.upstream-tests/test/host_only/CMakeLists.txt

@@ -1,4 +1,13 @@
-function(libcudacxx_add_standalone_header_test test_file_output_var header_under_test)
+add_custom_target(libcudacxx_tu_tests)
+
+set(cpp_std_versions 11 14 17 20)
+
+set(cpp_11_exclusions)
+set(cpp_14_exclusions)
+set(cpp_17_exclusions)
+set(cpp_20_exclusions)
+
+function(libcudacxx_add_standalone_header_test test_file_output_var header_under_test version)
     # ex: build/.../header_tests/cuda/std/version.cpp
     set(test_cpp "${CMAKE_CURRENT_BINARY_DIR}/header_tests/${header_under_test}.cpp")
 
@@ -9,22 +18,46 @@ function(libcudacxx_add_standalone_header_test test_file_output_var header_under
 
     set(${test_file_output_var} ${test_cpp} PARENT_SCOPE)
     # ex: cuda/std/version -> cuda_std_version
-    string(REPLACE "/" "_" executable_name ${header_under_test})
-    add_executable(${executable_name} ${test_cpp})
-    target_include_directories(${executable_name} PRIVATE ${CMAKE_SOURCE_DIR}/include)
-    target_compile_options(${executable_name} PRIVATE
+
+    string(REPLACE "/" "_" executable_name "${header_under_test}-${version}")
+    list(FIND cpp_${version}_exclusions ${header_under_test} header_excluded)
+
+    if ("${header_excluded}" EQUAL "-1")
+        add_library(${executable_name} OBJECT ${test_cpp})
+        target_include_directories(${executable_name} PRIVATE ${CMAKE_SOURCE_DIR}/include)
+        target_compile_options(${executable_name} PRIVATE
         $<$<OR:$<CXX_COMPILER_ID:Clang>,$<CXX_COMPILER_ID:AppleClang>,$<CXX_COMPILER_ID:GNU>>:
-            -Wall -Werror>
+        -Wall -Werror>
         $<$<CXX_COMPILER_ID:MSVC>:
-            /W4 /WX>)
-    set_target_properties(${executable_name} PROPERTIES CXX_STANDARD 11)
+        /W4 /WX>)
+        set_target_properties(${executable_name} PROPERTIES CXX_STANDARD ${version})
+        set_target_properties(${executable_name} PROPERTIES CXX_STANDARD_REQUIRED ON)
+        set_target_properties(${executable_name} PROPERTIES CXX_EXTENSIONS OFF)
+        add_dependencies(libcudacxx_tu_tests ${executable_name})
+    endif()
+
 endfunction()
 
 # Don't generate CUDA targets, they fail currently.
 # file(GLOB cuda_headers LIST_DIRECTORIES false RELATIVE ${CMAKE_SOURCE_DIR}/include ${CMAKE_SOURCE_DIR}/include/cuda/*)
 file(GLOB cuda_std_headers LIST_DIRECTORIES false RELATIVE ${CMAKE_SOURCE_DIR}/include ${CMAKE_SOURCE_DIR}/include/cuda/std/*)
 
-foreach(header IN LISTS cuda_headers cuda_std_headers)
-    libcudacxx_add_standalone_header_test(test_file ${header})
-    message(STATUS "Detected ${header} ... Writing ${test_file}")
+foreach(version IN LISTS cpp_std_versions)
+    if("${CMAKE_CXX_COMPILER_ID}" STREQUAL "MSVC")
+        set(cxx_version "/std:c++${version}")
+    else()
+        set(cxx_version "-std=c++${version}")
+    endif()
+
+    try_compile(
+        dialect_supported
+        ${CMAKE_BINARY_DIR}/dialect_check_${version} "${CMAKE_CURRENT_SOURCE_DIR}/detect_dialect.cpp"
+        COMPILE_DEFINITIONS ${cxx_version}
+    )
+
+    if (dialect_supported)
+        foreach(header IN LISTS cuda_headers cuda_std_headers)
+            libcudacxx_add_standalone_header_test(test_file ${header} ${version})
+        endforeach()
+    endif()
 endforeach()

.upstream-tests/test/host_only/detect_dialect.cpp

@@ -0,0 +1,7 @@
+#include <type_traits>
+#include <atomic>
+#include <thread>
+
+int main() {
+    return 0;
+}

.upstream-tests/test/std/views/views.span/span.cons/deduct.pass.cpp

@@ -8,6 +8,9 @@
 //===----------------------------------------------------------------------===//
 // UNSUPPORTED: c++03, c++11, c++14
 
+// gcc-6 does not support deduction guides until 7
+// UNSUPPORTED: gcc-6
+
 // <span>
 
 //   template<class It, class EndOrSize>
@@ -94,6 +97,6 @@ int main(int, char**)
   test_iterator_sentinel();
   test_c_array();
   test_std_array();
- 
+
   return 0;
 }

environments/linux/docker/compose.yml

@@ -14,7 +14,7 @@ services:
       args: &ubuntu22-args
         BUILDKIT_INLINE_CACHE: "1"
         USE_LLVM_INSTALLER: "1"
-        ROOT_IMAGE: "nvcr.io/nvidia/cuda:11.8.0-devel-ubuntu22.04"
+        ROOT_IMAGE: "nvcr.io/nvidia/cuda:12.0.0-devel-ubuntu22.04"
         COMPILERS: "g++-11 g++-12 clang-11 clang-12 clang-13 clang-14 clang-15"
     deploy: &deploy-template
       resources:
@@ -38,7 +38,7 @@ services:
       args: &ubuntu20-args
         BUILDKIT_INLINE_CACHE: "1"
         USE_LLVM_INSTALLER: 0
-        ROOT_IMAGE: "nvcr.io/nvidia/cuda:11.8.0-devel-ubuntu20.04"
+        ROOT_IMAGE: "nvcr.io/nvidia/cuda:12.0.0-devel-ubuntu20.04"
         COMPILERS: "g++-8 g++-9 g++-10 clang-8 clang-9 clang-10"
     deploy: *deploy-template
 
@@ -56,7 +56,7 @@ services:
       args: &ubuntu18-args
         BUILDKIT_INLINE_CACHE: "1"
         USE_LLVM_INSTALLER: 0
-        ROOT_IMAGE: "nvcr.io/nvidia/cuda:11.8.0-devel-ubuntu18.04"
+        ROOT_IMAGE: "nvcr.io/nvidia/cuda:12.0.0-devel-ubuntu18.04"
         COMPILERS: "g++-6 g++-7 clang-7"
     deploy: *deploy-template
 

environments/linux/docker/ubuntu.base.Dockerfile

@@ -77,6 +77,7 @@ ADD ./.upstream-tests/utils                   /libcudacxx/utils
 
 # Install compiler and configure project
 RUN cmake -S /libcudacxx -B /build \
+        -G Ninja \
         -DLIBCUDACXX_ENABLE_STATIC_LIBRARY=OFF \
         -DLIBCUDACXX_ENABLE_LIBCUDACXX_TESTS=ON \
         -DLIBCUDACXX_ENABLE_LIBCXX_TESTS=ON \
@@ -88,8 +89,8 @@ RUN cmake -S /libcudacxx -B /build \
         -DCMAKE_CUDA_COMPILER=${CUDACXX_PATH} \
         -DCMAKE_CUDA_FLAGS="-allow-unsupported-compiler"
 
-RUN make -j -C /build/libcxx
-RUN make -j -C /build/test/host_only
+RUN ninja -C /build libcudacxx_tu_tests && ninja -C /build clean
+RUN ninja -C /build cxx
 
 ENV LIBCUDACXX_SITE_CONFIG=/build/test/lit.site.cfg
 ENV LIBCXX_SITE_CONFIG=/build/libcxx/test/lit.site.cfg

include/cuda/std/detail/libcxx/include/span

@@ -226,11 +226,11 @@ public:
 #endif
     template <class _OtherElementType, enable_if_t<
                 is_convertible<_OtherElementType(*)[], element_type (*)[]>::value, nullptr_t> = nullptr>
-    _LIBCUDACXX_INLINE_VISIBILITY _LIBCUDACXX_CONSTEXPR_AFTER_CXX14 
+    _LIBCUDACXX_INLINE_VISIBILITY _LIBCUDACXX_CONSTEXPR_AFTER_CXX14
     span(      array<_OtherElementType, _Extent>& __arr) noexcept : __data{__arr.data()} {}
     template <class _OtherElementType, enable_if_t<
                 is_convertible<_OtherElementType(*)[], element_type (*)[]>::value, nullptr_t> = nullptr>
-    _LIBCUDACXX_INLINE_VISIBILITY _LIBCUDACXX_CONSTEXPR_AFTER_CXX14 
+    _LIBCUDACXX_INLINE_VISIBILITY _LIBCUDACXX_CONSTEXPR_AFTER_CXX14
     span(const array<_OtherElementType, _Extent>& __arr) noexcept : __data{__arr.data()} {}
 
     template <class _OtherElementType, enable_if_t<
@@ -501,7 +501,7 @@ public:
     _LIBCUDACXX_INLINE_VISIBILITY constexpr iterator end() const noexcept { return iterator(data() + size()); }
     _LIBCUDACXX_INLINE_VISIBILITY _LIBCUDACXX_CONSTEXPR_AFTER_CXX14 reverse_iterator rbegin() const noexcept { return reverse_iterator(end()); }
     _LIBCUDACXX_INLINE_VISIBILITY _LIBCUDACXX_CONSTEXPR_AFTER_CXX14 reverse_iterator rend() const noexcept { return reverse_iterator(begin()); }
-    
+
     _LIBCUDACXX_INLINE_VISIBILITY span<const byte, dynamic_extent> __as_bytes() const noexcept
     { return {reinterpret_cast<const byte *>(data()), size_bytes()}; }
 
@@ -528,7 +528,7 @@ auto as_writable_bytes(span<_Tp, _Extent> __s) noexcept
 
 #endif // _LIBCUDACXX_STD_VER > 11
 
-#if _LIBCUDACXX_STD_VER > 14
+#if _LIBCUDACXX_STD_VER > 14 && !defined(_LIBCUDACXX_HAS_NO_DEDUCTION_GUIDES)
 
 //  Deduction guides
 template<class _Tp, size_t _Sz>

include/cuda/std/detail/libcxx/include/support/win32/limits_msvc_win32.h

@@ -64,8 +64,14 @@
 #define __LDBL_DENORM_MIN__ 3.64519953188247460253e-4951L
 
 // __builtin replacements/workarounds
+#if _MSC_VER < 1934
 #define __builtin_huge_vall()    _LInf._Long_double
 #define __builtin_nanl(__dummmy) _LNan._Long_double
 #define __builtin_nansl(__dummy) _LSnan._Long_double
+#else
+#define __builtin_huge_vall()    __builtin_huge_val()
+#define __builtin_nanl(__v)      __builtin_nan(__v)
+#define __builtin_nansl(__v)     __builtin_nans(__v)
+#endif
 
 #endif // _LIBCUDACXX_SUPPORT_WIN32_LIMITS_MSVC_WIN32_H