[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