From: Matthias Kretz Date: Wed, 27 Mar 2024 07:49:43 +0000 (+0100) Subject: libstdc++: Fix call signature of builtins from masked ++/-- X-Git-Tag: releases/gcc-11.5.0~155 X-Git-Url: http://git.ipfire.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=f0b7583dab56c78824344e00eb6230eaaeb09a44;p=thirdparty%2Fgcc.git libstdc++: Fix call signature of builtins from masked ++/-- 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. (cherry picked from commit 0ac2c0f0687b321ab54de271d788b4e0a287b4e2) --- diff --git a/libstdc++-v3/include/experimental/bits/simd_x86.h b/libstdc++-v3/include/experimental/bits/simd_x86.h index 2e7ea238811c..fb0d6977832a 100644 --- a/libstdc++-v3/include/experimental/bits/simd_x86.h +++ b/libstdc++-v3/include/experimental/bits/simd_x86.h @@ -3537,17 +3537,23 @@ template } 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__