[PATCH 2/8] x86: Do not use __builtin_fpclassify for _Float64x/long double
H.J. Lu
hjl.tools@gmail.com
Thu Dec 18 23:36:56 GMT 2025
On Thu, Dec 18, 2025 at 8:39 PM Adhemerval Zanella Netto
<adhemerval.zanella@linaro.org> wrote:
>
>
>
> 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.
We should add some tests without -fsignaling-nans.
> >
> >>> 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.
> >
> >
> >
>
--
H.J.
More information about the Libc-alpha
mailing list