[gcc r13-8841] libstdc++: Fix simd<char> conversion for -fno-signed-char for Clang

[Matthias Kretz]

https://gcc.gnu.org/g:ef2169090d0868e4718c2ebf25365aaa52c22139

commit r13-8841-gef2169090d0868e4718c2ebf25365aaa52c22139
Author: Matthias Kretz <m.kretz@gsi.de>
Date:   Mon Jun 3 12:02:07 2024 +0200

    libstdc++: Fix simd<char> conversion for -fno-signed-char for Clang
    
    The special case for Clang in the trait producing a signed integer type
    lead to the trait returning 'char' where it should have been 'signed
    char'. This workaround was introduced because on Clang the return type
    of vector compares was not convertible to '_SimdWrapper<
    __int_for_sizeof_t<...' unless '__int_for_sizeof_t<char>' was an alias
    for 'char'. In order to not rewrite the complete mask type code (there
    is code scattered around the implementation assuming signed integers),
    this needs to be 'signed char'; so the special case for Clang needs to
    be removed.
    The conversion issue is now solved in _SimdWrapper, which now
    additionally allows conversion from vector types with compatible
    integral type.
    
    Signed-off-by: Matthias Kretz <m.kretz@gsi.de>
    
    libstdc++-v3/ChangeLog:
    
            PR libstdc++/115308
            * include/experimental/bits/simd.h (__int_for_sizeof): Remove
            special cases for __clang__.
            (_SimdWrapper): Change constructor overload set to allow
            conversion from vector types with integral conversions via bit
            reinterpretation.
    
    (cherry picked from commit 8e36cf4c5c9140915d0019999db132a900b48037)

Diff:
---
 libstdc++-v3/include/experimental/bits/simd.h | 45 ++++++++++++++++-----------
 1 file changed, 27 insertions(+), 18 deletions(-)

diff --git a/libstdc++-v3/include/experimental/bits/simd.h b/libstdc++-v3/include/experimental/bits/simd.h
index ce7805e43e5..e6c407d86fa 100644
--- a/libstdc++-v3/include/experimental/bits/simd.h
+++ b/libstdc++-v3/include/experimental/bits/simd.h
@@ -584,19 +584,12 @@ template <size_t _Bytes>
     static_assert(_Bytes > 0);
     if constexpr (_Bytes == sizeof(int))
       return int();
-  #ifdef __clang__
-    else if constexpr (_Bytes == sizeof(char))
-      return char();
-  #else
     else if constexpr (_Bytes == sizeof(_SChar))
       return _SChar();
-  #endif
     else if constexpr (_Bytes == sizeof(short))
       return short();
-  #ifndef __clang__
     else if constexpr (_Bytes == sizeof(long))
       return long();
-  #endif
     else if constexpr (_Bytes == sizeof(_LLong))
       return _LLong();
   #ifdef __SIZEOF_INT128__
@@ -2712,6 +2705,8 @@ template <typename _BuiltinType>
 
 // }}}
 // _SimdWrapper{{{
+struct _DisabledSimdWrapper;
+
 template <typename _Tp, size_t _Width>
   struct _SimdWrapper<
     _Tp, _Width,
@@ -2721,16 +2716,17 @@ template <typename _Tp, size_t _Width>
 			      == sizeof(__vector_type_t<_Tp, _Width>),
 		       __vector_type_t<_Tp, _Width>>
   {
-    using _Base
-      = _SimdWrapperBase<__has_iec559_behavior<__signaling_NaN, _Tp>::value
-			   && sizeof(_Tp) * _Width
-				== sizeof(__vector_type_t<_Tp, _Width>),
-			 __vector_type_t<_Tp, _Width>>;
+    static constexpr bool _S_need_default_init
+      = __has_iec559_behavior<__signaling_NaN, _Tp>::value
+	  and sizeof(_Tp) * _Width == sizeof(__vector_type_t<_Tp, _Width>);
+
+    using _BuiltinType = __vector_type_t<_Tp, _Width>;
+
+    using _Base = _SimdWrapperBase<_S_need_default_init, _BuiltinType>;
 
     static_assert(__is_vectorizable_v<_Tp>);
     static_assert(_Width >= 2); // 1 doesn't make sense, use _Tp directly then
 
-    using _BuiltinType = __vector_type_t<_Tp, _Width>;
     using value_type = _Tp;
 
     static inline constexpr size_t _S_full_size
@@ -2766,13 +2762,26 @@ template <typename _Tp, size_t _Width>
     _GLIBCXX_SIMD_INTRINSIC constexpr _SimdWrapper&
     operator=(_SimdWrapper&&) = default;
 
-    template <typename _V, typename = enable_if_t<disjunction_v<
-			     is_same<_V, __vector_type_t<_Tp, _Width>>,
-			     is_same<_V, __intrinsic_type_t<_Tp, _Width>>>>>
+    // Convert from exactly matching __vector_type_t
+    using _SimdWrapperBase<_S_need_default_init, _BuiltinType>::_SimdWrapperBase;
+
+    // Convert from __intrinsic_type_t if __intrinsic_type_t and __vector_type_t differ, otherwise
+    // this ctor should not exist. Making the argument type unusable is our next best solution.
+    _GLIBCXX_SIMD_INTRINSIC constexpr
+    _SimdWrapper(conditional_t<is_same_v<_BuiltinType, __intrinsic_type_t<_Tp, _Width>>,
+			       _DisabledSimdWrapper, __intrinsic_type_t<_Tp, _Width>> __x)
+    : _Base(__vector_bitcast<_Tp, _Width>(__x)) {}
+
+    // Convert from different __vector_type_t, but only if bit reinterpretation is a correct
+    // conversion of the value_type
+    template <typename _V, typename _TVT = _VectorTraits<_V>,
+	      typename = enable_if_t<sizeof(typename _TVT::value_type) == sizeof(_Tp)
+				       and sizeof(_V) == sizeof(_BuiltinType)
+				       and is_integral_v<_Tp>
+				       and is_integral_v<typename _TVT::value_type>>>
       _GLIBCXX_SIMD_INTRINSIC constexpr
       _SimdWrapper(_V __x)
-      // __vector_bitcast can convert e.g. __m128 to __vector(2) float
-      : _Base(__vector_bitcast<_Tp, _Width>(__x)) {}
+      : _Base(reinterpret_cast<_BuiltinType>(__x)) {}
 
     template <typename... _As,
 	      typename = enable_if_t<((is_same_v<simd_abi::scalar, _As> && ...)