[PATCH 2/8] x86: Do not use __builtin_fpclassify for _Float64x/long double

Adhemerval Zanella Netto adhemerval.zanella@linaro.org
Thu Dec 18 12:39:00 GMT 2025



On 17/12/25 20:54, H.J. Lu wrote:
> On Thu, Dec 18, 2025 at 7:22 AM H.J. Lu <hjl.tools@gmail.com> wrote:
>>
>> On Thu, Dec 18, 2025 at 1:56 AM Adhemerval Zanella
>> <adhemerval.zanella@linaro.org> wrote:
>>>
>>> Neither gcc [1] nor clang [2] handles pseudo-normal numbers correctly
>>> with the __builtin_fpclassify, so disable its usage for _Float64x and
>>> long double types.
>>
>> Please add tests to verify this fix.
>>
>>>
>>> This only affects x86, so add a new header, fp-builtin-denormal.h, that
>>> defines whether the architecture requires disabling the optimization
>>> through a new glibc define (__FP_BUILTIN_FPCLASSIFY_DENORMAL).
>>>
>>> It fixes the regression on test-ldouble-fpclassify and
>>> test-float64x-fpclassify when built with clang:
>>
> 
> Why don't they fail with GCC?

I forgot to add on the commit message, but gcc builds the fpclassify tests
with -fsignaling-nans.  With gcc this defines __SUPPORT_SNAN__, which will
always calls the glibc symbol.  The clang does not support the options,
nor define __SUPPORT_SNAN__; and then will always use the builtin.

> 
>>> Failure: fpclassify (pseudo_zero): Exception "Invalid operation" set
>>> Failure: fpclassify (pseudo_inf): Exception "Invalid operation" set
>>> Failure: fpclassify (pseudo_qnan): Exception "Invalid operation" set
>>> Failure: fpclassify (pseudo_snan): Exception "Invalid operation" set
>>> Failure: fpclassify (pseudo_unnormal): Exception "Invalid operation" set
>>> Failure: fpclassify_downward (pseudo_zero): Exception "Invalid operation" set
>>> Failure: fpclassify_downward (pseudo_inf): Exception "Invalid operation" set
>>> Failure: fpclassify_downward (pseudo_qnan): Exception "Invalid operation" set
>>> Failure: fpclassify_downward (pseudo_snan): Exception "Invalid operation" set
>>> Failure: fpclassify_downward (pseudo_unnormal): Exception "Invalid operation" set
>>> Failure: fpclassify_towardzero (pseudo_zero): Exception "Invalid operation" set
>>> Failure: fpclassify_towardzero (pseudo_inf): Exception "Invalid operation" set
>>> Failure: fpclassify_towardzero (pseudo_qnan): Exception "Invalid operation" set
>>> Failure: fpclassify_towardzero (pseudo_snan): Exception "Invalid operation" set
>>> Failure: fpclassify_towardzero (pseudo_unnormal): Exception "Invalid operation" set
>>> Failure: fpclassify_upward (pseudo_zero): Exception "Invalid operation" set
>>> Failure: fpclassify_upward (pseudo_inf): Exception "Invalid operation" set
>>> Failure: fpclassify_upward (pseudo_qnan): Exception "Invalid operation" set
>>> Failure: fpclassify_upward (pseudo_snan): Exception "Invalid operation" set
>>> Failure: fpclassify_upward (pseudo_unnormal): Exception "Invalid operation" set
>>>
>>> Checked on x86_64-linux-gnu with gcc-15 and clang-18.
>>>
>>> [1] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123161
>>> [2] https://github.com/llvm/llvm-project/issues/172533
>>> ---
>>>  bits/fp-builtin-denormal.h             | 28 ++++++++
>>>  math/Makefile                          |  1 +
>>>  math/math.h                            | 89 +++++++++++++++++++++++++-
>>>  sysdeps/x86/bits/fp-builtin-denormal.h | 25 ++++++++
>>>  4 files changed, 141 insertions(+), 2 deletions(-)
>>>  create mode 100644 bits/fp-builtin-denormal.h
>>>  create mode 100644 sysdeps/x86/bits/fp-builtin-denormal.h
>>>
>>> diff --git a/bits/fp-builtin-denormal.h b/bits/fp-builtin-denormal.h
>>> new file mode 100644
>>> index 0000000000..77b7184975
>>> --- /dev/null
>>> +++ b/bits/fp-builtin-denormal.h
>>> @@ -0,0 +1,28 @@
>>> +/* Denormal number definitions.
>>> +   Copyright (C) 2025 Free Software Foundation, Inc.
>>> +   This file is part of the GNU C Library.
>>> +
>>> +   The GNU C Library is free software; you can redistribute it and/or
>>> +   modify it under the terms of the GNU Lesser General Public
>>> +   License as published by the Free Software Foundation; either
>>> +   version 2.1 of the License, or (at your option) any later version.
>>> +
>>> +   The GNU C Library is distributed in the hope that it will be useful,
>>> +   but WITHOUT ANY WARRANTY; without even the implied warranty of
>>> +   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
>>> +   Lesser General Public License for more details.
>>> +
>>> +   You should have received a copy of the GNU Lesser General Public
>>> +   License along with the GNU C Library; if not, see
>>> +   <https://www.gnu.org/licenses/>.  */
>>> +
>>> +#ifndef _MATH_H
>>> +# error "Never use <bits/fp-builtin-denormal.h> directly; include <math.h> instead."
>>> +#endif
>>> +
>>> +/*  __FP_BUILTIN_FPCLASSIFY_DENORMAL is defined to 1 if compiler supports
>>> +    handling pseudo-denormal numbers with fpclassify builtin.  Pseudo-denormal
>>> +    is a non-standard denormalized floating-point number only supported by
>>> +    Intel double extended-precision (long double).  By default assume 1 to
>>> +    enable the usage of compiler builtin on math.h.  */
>>> +#define __FP_BUILTIN_FPCLASSIFY_DENORMAL 1
>>> diff --git a/math/Makefile b/math/Makefile
>>> index f7a1d9016c..8fe2540ce0 100644
>>> --- a/math/Makefile
>>> +++ b/math/Makefile
>>> @@ -28,6 +28,7 @@ headers := \
>>>    bits/floatn-common.h \
>>>    bits/floatn.h \
>>>    bits/flt-eval-method.h \
>>> +  bits/fp-builtin-denormal.h \
>>>    bits/fp-fast.h \
>>>    bits/fp-logb.h \
>>>    bits/iscanonical.h \
>>> diff --git a/math/math.h b/math/math.h
>>> index 74b064d96e..cbec2f406a 100644
>>> --- a/math/math.h
>>> +++ b/math/math.h
>>> @@ -1064,6 +1064,86 @@ extern int signgam;
>>>     : FUNC ## l ARGS)
>>>  #endif
>>>
>>> +
>>> +/* Depending on the type of TG_ARG and extra DEFINE to check, either call the
>>> +   BUILTIN with ARGS_B or an appropriately suffixed version of FUNC with
>>> +   arguments (including parentheses) ARGS_B.  The function call is used for
>>> +   long double and/or  _Float64x is the builtin can not be safely used on all
>>> +   arguments (defined by DEFINE).  */
>>> +
>>> +#include <bits/fp-builtin-denormal.h>
>>> +
>>> +#ifdef __NO_LONG_DOUBLE_MATH
>>> +# define __MATH_TG_BUILTIN_CLASSIFY(TG_ARG, BUILTIN, ARGS_B, FUNC, ARGS_F,    \
>>> +                                   DEFINE)    \
>>> +  BUILTIN ARGS_B
>>> +#elif __HAVE_DISTINCT_FLOAT128
>>> +# if __HAVE_GENERIC_SELECTION
>>> +#  if __HAVE_FLOATN_NOT_TYPEDEF && __HAVE_FLOAT32
>>> +#   define __MATH_TG_BUILTIN_CLASSIFY_F32(BUILTIN, ARGS_B, FUNC, ARGS_F,      \
>>> +                                         DEFINE)                             \
>>> +  _Float32: BUILTIN ARGS_B,
>>> +#  else
>>> +#   define __MATH_TG_BUILTIN_CLASSIFY_F32(BUILTIN, ARGS_B, FUNC, ARGS_F,      \
>>> +                                         DEFINE)
>>> +#  endif
>>> +#  define __MATH_TG_BUILTIN_CLASSIFY_LDOUBLE(BUILTIN, ARGS_B, FUNC, ARGS_F,   \
>>> +                                            DEFINE)                          \
>>> +  long double: DEFINE ? BUILTIN ARGS_B : __ ## FUNC ## l ARGS_F,
>>> +#  if __HAVE_FLOATN_NOT_TYPEDEF && __HAVE_FLOAT64X
>>> +#   if __HAVE_FLOAT64X_LONG_DOUBLE
>>> +#    define __MATH_TG_BUILTIN_CLASSIFY_F64X(BUILTIN, ARGS_B, FUNC, ARGS_F,    \
>>> +                                           DEFINE)                           \
>>> +  _Float64x: DEFINE ? BUILTIN ARGS_B : __ ## FUNC ## l ARGS_F,
>>> +#   else
>>> +#    define __MATH_TG_BUILTIN_CLASSIFY_F64X(BUILTIN, ARGS_B, FUNC, ARGS_F,    \
>>> +                                           DEFINE)                           \
>>> +  _Float64x: DEFINE ? BUILTIN ARGS_B : __ ## FUNC ## f128 ARGS_F,
>>> +#   endif
>>> +#  else
>>> +#   define __MATH_TG_BUILTIN_CLASSIFY_F64X(BUILTIN, ARGS_B, FUNC, ARGS_F,     \
>>> +                                          DEFINE)
>>> +#  endif
>>> +#  define __MATH_TG_BUILTIN_CLASSIFY_F128(BUILTIN, ARGS_B, FUNC, ARGS_F,      \
>>> +                                         DEFINE)                             \
>>> +  _Float128: BUILTIN ARGS_B
>>> +#  define __MATH_TG_BUILTIN_CLASSIFY(TG_ARG, BUILTIN, ARGS_B, FUNC, ARGS_F,   \
>>> +                                    DEFINE)                                  \
>>> +     _Generic ((TG_ARG),                                                     \
>>> +              float: BUILTIN ARGS_B,                                         \
>>> +              __MATH_TG_BUILTIN_CLASSIFY_F32 (BUILTIN, ARGS_B, FUNC, ARGS_F, \
>>> +                                              DEFINE)                        \
>>> +              default: BUILTIN ARGS_B,                                       \
>>> +              __MATH_TG_BUILTIN_CLASSIFY_LDOUBLE (BUILTIN, ARGS_B, FUNC,     \
>>> +                                                  ARGS_F, DEFINE)            \
>>> +              __MATH_TG_BUILTIN_CLASSIFY_F64X (BUILTIN, ARGS_B, FUNC, ARGS_F,\
>>> +                                               DEFINE)                       \
>>> +              __MATH_TG_BUILTIN_CLASSIFY_F128 (BUILTIN, ARGS_B, FUNC, ARGS_F,\
>>> +                                               DEFINE))
>>> +# else
>>> +#  define __MATH_TG_BUILTIN_CLASSIFY(TG_ARG, BUILTIN, ARGS_B, FUNC, ARGS_F,   \
>>> +                                    DEFINE)                                  \
>>> +     __builtin_choose_expr                                                   \
>>> +     (__builtin_types_compatible_p (__typeof (TG_ARG), float),               \
>>> +      __builtin ## BUILTIN ARGS_B,                                           \
>>> +      __builtin_choose_expr                                                  \
>>> +      (__builtin_types_compatible_p (__typeof (TG_ARG), double),             \
>>> +       __builtin ## BUILTIN ARGS_B,                                          \
>>> +       __builtin_choose_expr                                                 \
>>> +       (__builtin_types_compatible_p (__typeof (TG_ARG), long double),       \
>>> +       DEFINE ? BUILTIN ARGS_B : __ ## FUNC ## l ARGS_F,                     \
>>> +       BUILTIN ARGS_B)))
>>> +# endif
>>> +#else
>>> +# define __MATH_TG_BUILTIN_CLASSIFY(TG_ARG, BUILTIN, ARGS_B, FUNC, ARGS_F,    \
>>> +                                   DEFINE)                                   \
>>> +  (sizeof (TG_ARG) == sizeof (float)                                         \
>>> +   ? BUILTIN ARGS_B                                                          \
>>> +   : sizeof (TG_ARG) == sizeof (double)                                              \
>>> +   ? BUILTIN ARGS_B                                                          \
>>> +   : DEFINE ? BUILTIN ARGS_B : __ ## FUNC ## l ARGS_F)
>>> +#endif
>>> +
>>>  /* ISO C99 defines some generic macros which work on any data type.  */
>>>  #ifdef __USE_ISOC99
>>>
>>> @@ -1101,8 +1181,13 @@ enum
>>>         with -Os.  No further use of this definition of fpclassify is
>>>         expected in C++ mode, since libstdc++ provides its own version
>>>         of fpclassify in cmath (which undefines fpclassify).  */
>>> -#  define fpclassify(x) __builtin_fpclassify (FP_NAN, FP_INFINITE,           \
>>> -     FP_NORMAL, FP_SUBNORMAL, FP_ZERO, x)
>>> +#  define fpclassify(x)                                                              \
>>> +  __MATH_TG_BUILTIN_CLASSIFY ((x),                                           \
>>> +                             __builtin_fpclassify, (FP_NAN, FP_INFINITE,     \
>>> +                                                    FP_NORMAL, FP_SUBNORMAL, \
>>> +                                                    FP_ZERO, x),             \
>>> +                             fpclassify, (x),                                \
>>> +                             __FP_BUILTIN_FPCLASSIFY_DENORMAL)
>>>  # else
>>>  #  define fpclassify(x) __MATH_TG ((x), __fpclassify, (x))
>>>  # endif
>>> diff --git a/sysdeps/x86/bits/fp-builtin-denormal.h b/sysdeps/x86/bits/fp-builtin-denormal.h
>>> new file mode 100644
>>> index 0000000000..bfb8f54237
>>> --- /dev/null
>>> +++ b/sysdeps/x86/bits/fp-builtin-denormal.h
>>> @@ -0,0 +1,25 @@
>>> +/* Define __FP_BUILTIN_DENORMAL.
>>> +   Copyright (C) 2025 Free Software Foundation, Inc.
>>> +   This file is part of the GNU C Library.
>>> +
>>> +   The GNU C Library is free software; you can redistribute it and/or
>>> +   modify it under the terms of the GNU Lesser General Public
>>> +   License as published by the Free Software Foundation; either
>>> +   version 2.1 of the License, or (at your option) any later version.
>>> +
>>> +   The GNU C Library is distributed in the hope that it will be useful,
>>> +   but WITHOUT ANY WARRANTY; without even the implied warranty of
>>> +   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
>>> +   Lesser General Public License for more details.
>>> +
>>> +   You should have received a copy of the GNU Lesser General Public
>>> +   License along with the GNU C Library; if not, see
>>> +   <https://www.gnu.org/licenses/>.  */
>>> +
>>> +#ifndef _MATH_H
>>> +# error "Never use <bits/fp-builtin-denormal.h> directly; include <math.h> instead."
>>> +#endif
>>> +
>>> +/* Neither GCC (bug 123161) nor clang (issue 172533) handles pseudo-normal
>>> +   numbers correctly with fpclassify builtin.  */
>>> +#define __FP_BUILTIN_FPCLASSIFY_DENORMAL 0
>>> --
>>> 2.43.0
>>>
>>
>>
>> --
>> H.J.
> 
> 
> 



More information about the Libc-alpha mailing list