Re: [PATCH] libstdc++: Fix call signature of builtins from masked ++/--

[Jonathan Wakely]

On Wed, 27 Mar 2024 at 08:51, Matthias Kretz  wrote:
>
> This is broken on GCC 13 and 14 (https://compiler-explorer.com/z/GPKGPGs6T)
>
> Tested on x86_64-linux-gnu.
>
> OK for trunk and 13?

OK for both, thanks.


>
> --- 8< -
>
> This resolves failures in the "expensive" where-* test of check-simd
> when targeting AVX-512.
>
> Signed-off-by: Matthias Kretz 
>
> libstdc++-v3/ChangeLog:
>
> * include/experimental/bits/simd_x86.h (_S_masked_unary): Call
> the 4- and 8-byte variants of __builtin_ia32_subp[ds] without
> rounding direction argument.
> ---
>  libstdc++-v3/include/experimental/bits/simd_x86.h | 12 +---
>  1 file changed, 9 insertions(+), 3 deletions(-)
>
>
> --
> ──
>  Dr. Matthias Kretz   https://mattkretz.github.io
>  GSI Helmholtz Centre for Heavy Ion Research   https://gsi.de
>  stdₓ::simd
> ──

[PATCH] libstdc++: Fix call signature of builtins from masked ++/--

[Matthias Kretz]

This is broken on GCC 13 and 14 (https://compiler-explorer.com/z/GPKGPGs6T)

Tested on x86_64-linux-gnu.

OK for trunk and 13?

--- 8< -

This resolves failures in the "expensive" where-* test of check-simd
when targeting AVX-512.

Signed-off-by: Matthias Kretz 

libstdc++-v3/ChangeLog:

* include/experimental/bits/simd_x86.h (_S_masked_unary): Call
the 4- and 8-byte variants of __builtin_ia32_subp[ds] without
rounding direction argument.
---
 libstdc++-v3/include/experimental/bits/simd_x86.h | 12 +---
 1 file changed, 9 insertions(+), 3 deletions(-)


--
──
 Dr. Matthias Kretz   https://mattkretz.github.io
 GSI Helmholtz Centre for Heavy Ion Research   https://gsi.de
 stdₓ::simd
──diff --git a/libstdc++-v3/include/experimental/bits/simd_x86.h b/libstdc++-v3/include/experimental/bits/simd_x86.h
index 16b207be2a3..6b414486fee 100644
--- a/libstdc++-v3/include/experimental/bits/simd_x86.h
+++ b/libstdc++-v3/include/experimental/bits/simd_x86.h
@@ -3538,17 +3538,23 @@ _S_masked_unary(const _SimdWrapper<_K, _Np> __k, const _SimdWrapper<_Tp, _Np> __
 	  }
 	else
 	  {
-#define _GLIBCXX_SIMD_MASK_SUB(_Sizeof, _Width, _Instr)\
+#define _GLIBCXX_SIMD_MASK_SUB_512(_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, _MM_FROUND_CUR_DIRECTION)
-		_GLIBCXX_SIMD_MASK_SUB(4, 64, subps512);
+#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)
+		_GLIBCXX_SIMD_MASK_SUB_512(4, 64, subps512);
 		_GLIBCXX_SIMD_MASK_SUB(4, 32, subps256);
 		_GLIBCXX_SIMD_MASK_SUB(4, 16, subps128);
-		_GLIBCXX_SIMD_MASK_SUB(8, 64, subpd512);
+		_GLIBCXX_SIMD_MASK_SUB_512(8, 64, subpd512);
 		_GLIBCXX_SIMD_MASK_SUB(8, 32, subpd256);
 		_GLIBCXX_SIMD_MASK_SUB(8, 16, subpd128);
+#undef _GLIBCXX_SIMD_MASK_SUB_512
 #undef _GLIBCXX_SIMD_MASK_SUB
 	  }
 #endif // __clang__