[PATCH] libstdc++: Add masked ++/-- implementation for sizeof < 16

[Matthias Kretz <m.kretz@gsi.de>]

[-- Attachment #1: Type: text/plain, Size: 1264 bytes --]

And one more to fix follow-up / remaining failures. Tested on x86_64-linux-
gnu.

OK for trunk and 13?

------------ 8< --------------

This resolves further failures (-Wreturn-type warnings) and test
failures for where-* tests targeting AVX-512.

Signed-off-by: Matthias Kretz <m.kretz@gsi.de>

libstdc++-v3/ChangeLog:

	* include/experimental/bits/simd_x86.h (_S_masked_unary):
	Cast inputs < 16 bytes to 16 byte vectors before calling the
	right subtraction builtin. Before returning, truncate to the
	return vector type.
---
 .../include/experimental/bits/simd_x86.h      | 24 +++++++++++--------
 1 file changed, 14 insertions(+), 10 deletions(-)


--
──────────────────────────────────────────────────────────────────────────
 Dr. Matthias Kretz                           https://mattkretz.github.io
 GSI Helmholtz Centre for Heavy Ion Research               https://gsi.de
 stdₓ::simd
──────────────────────────────────────────────────────────────────────────

[-- Attachment #2: 0001-libstdc-Add-masked-implementation-for-sizeof-16.patch --]
[-- Type: text/x-patch, Size: 3328 bytes --]

diff --git a/libstdc++-v3/include/experimental/bits/simd_x86.h b/libstdc++-v3/include/experimental/bits/simd_x86.h
index 6b414486fee..517c4b4a5be 100644
--- a/libstdc++-v3/include/experimental/bits/simd_x86.h
+++ b/libstdc++-v3/include/experimental/bits/simd_x86.h
@@ -3508,6 +3508,9 @@ _S_masked_unary(const _SimdWrapper<_K, _Np> __k, const _SimdWrapper<_Tp, _Np> __
 #ifdef __clang__
 	    return __movm<_Np, _Tp>(__k._M_data) ? __v._M_data - __pm_one : __v._M_data;
 #else // __clang__
+	    using _TV = __vector_type_t<_Tp, _Np>;
+	    constexpr size_t __bytes = sizeof(__v) < 16 ? 16 : sizeof(__v);
+	    constexpr size_t __width = __bytes / sizeof(_Tp);
 	    if constexpr (is_integral_v<_Tp>)
 	      {
 		constexpr bool __lp64 = sizeof(long) == sizeof(long long);
@@ -3517,11 +3520,11 @@ _S_masked_unary(const _SimdWrapper<_K, _Np> __k, const _SimdWrapper<_Tp, _Np> __
 			      std::conditional_t<__lp64, long long, int>,
 			      std::conditional_t<
 				std::is_same_v<_Ip, signed char>, char, _Ip>>;
-		const auto __value = __vector_bitcast<_Up>(__v._M_data);
+		const auto __value = __intrin_bitcast<__vector_type_t<_Up, __width>>(__v._M_data);
 #define _GLIBCXX_SIMD_MASK_SUB(_Sizeof, _Width, _Instr)                        \
-  if constexpr (sizeof(_Tp) == _Sizeof && sizeof(__v) == _Width)               \
-    return __vector_bitcast<_Tp>(__builtin_ia32_##_Instr##_mask(__value,       \
-	     __vector_broadcast<_Np>(_Up(__pm_one)), __value, __k._M_data))
+  if constexpr (sizeof(_Tp) == _Sizeof && sizeof(__value) == _Width)           \
+    return __intrin_bitcast<_TV>(__builtin_ia32_##_Instr##_mask(__value,       \
+	     __vector_broadcast<__width>(_Up(__pm_one)), __value, __k._M_data))
 		_GLIBCXX_SIMD_MASK_SUB(1, 64, psubb512);
 		_GLIBCXX_SIMD_MASK_SUB(1, 32, psubb256);
 		_GLIBCXX_SIMD_MASK_SUB(1, 16, psubb128);
@@ -3538,16 +3541,17 @@ _S_masked_unary(const _SimdWrapper<_K, _Np> __k, const _SimdWrapper<_Tp, _Np> __
 	      }
 	    else
 	      {
+		const auto __value = __intrin_bitcast<__vector_type_t<_Tp, __width>>(__v._M_data);
 #define _GLIBCXX_SIMD_MASK_SUB_512(_Sizeof, _Width, _Instr)                    \
-  if constexpr (sizeof(_Tp) == _Sizeof && sizeof(__v) == _Width)               \
+  if constexpr (sizeof(_Tp) == _Sizeof && sizeof(__value) == _Width)           \
     return __builtin_ia32_##_Instr##_mask(                                     \
-	     __v._M_data, __vector_broadcast<_Np>(_Tp(__pm_one)), __v._M_data, \
+	     __value, __vector_broadcast<__width>(_Tp(__pm_one)), __value,     \
 	     __k._M_data, _MM_FROUND_CUR_DIRECTION)
 #define _GLIBCXX_SIMD_MASK_SUB(_Sizeof, _Width, _Instr)                        \
-  if constexpr (sizeof(_Tp) == _Sizeof && sizeof(__v) == _Width)               \
-    return __builtin_ia32_##_Instr##_mask(                                     \
-	     __v._M_data, __vector_broadcast<_Np>(_Tp(__pm_one)), __v._M_data, \
-	     __k._M_data)
+  if constexpr (sizeof(_Tp) == _Sizeof && sizeof(__value) == _Width)           \
+    return __intrin_bitcast<_TV>(__builtin_ia32_##_Instr##_mask(               \
+	     __value, __vector_broadcast<__width>(_Tp(__pm_one)), __value,     \
+	     __k._M_data))
 		_GLIBCXX_SIMD_MASK_SUB_512(4, 64, subps512);
 		_GLIBCXX_SIMD_MASK_SUB(4, 32, subps256);
 		_GLIBCXX_SIMD_MASK_SUB(4, 16, subps128);