[PATCH v2 1/5] x86: Do not use __builtin_fpclassify for _Float64x/long double

H.J. Lu hjl.tools@gmail.com
Fri Dec 19 23:58:37 GMT 2025


On Sat, Dec 20, 2025 at 2:54 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.
>
> 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:
>
> 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 +++++++
>  sysdeps/x86/fpu/Makefile                |  1 +
>  sysdeps/x86/fpu/test-builtin-denormal.c | 53 +++++++++++++++
>  6 files changed, 195 insertions(+), 2 deletions(-)
>  create mode 100644 bits/fp-builtin-denormal.h
>  create mode 100644 sysdeps/x86/bits/fp-builtin-denormal.h
>  create mode 100644 sysdeps/x86/fpu/test-builtin-denormal.c
>
> 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
> diff --git a/sysdeps/x86/fpu/Makefile b/sysdeps/x86/fpu/Makefile
> index 91b2444f3a..2914be887e 100644
> --- a/sysdeps/x86/fpu/Makefile
> +++ b/sysdeps/x86/fpu/Makefile
> @@ -9,6 +9,7 @@ tests += \
>    test-fenv-sse-2 \
>    test-fenv-x87 \
>    test-flt-eval-method-sse \
> +  test-builtin-denormal \
>  # tests
>  ifeq ($(have-test-cc-cflags-mfpmath-387),yes)
>  tests += test-flt-eval-method-387
> diff --git a/sysdeps/x86/fpu/test-builtin-denormal.c b/sysdeps/x86/fpu/test-builtin-denormal.c
> new file mode 100644
> index 0000000000..1fb8aeedf6
> --- /dev/null
> +++ b/sysdeps/x86/fpu/test-builtin-denormal.c
> @@ -0,0 +1,53 @@
> +/* Ccheck if math.h optimizations to call compiler builtin
> +   does not trigger FE_INVALID on x86 denormal numbers.
> +   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/>.  */
> +
> +#include <array_length.h>
> +#include <fenv.h>
> +#include <math.h>
> +#include <math_ldbl.h>
> +#include <support/check.h>
> +
> +#define pseudo_inf { .parts = { 0x00000000, 0x00000000, 0x7fff }}
> +#define pseudo_zero { .parts = { 0x00000000, 0x00000000, 0x0100 }}
> +#define pseudo_qnan { .parts = { 0x00000001, 0x00000000, 0x7fff }}
> +#define pseudo_snan { .parts = { 0x00000001, 0x40000000, 0x7fff }}
> +#define pseudo_unnormal { .parts = { 0x00000001, 0x40000000, 0x0100 }}
> +
> +static const ieee_long_double_shape_type inputs[] = {
> +  pseudo_inf,
> +  pseudo_zero,
> +  pseudo_qnan,
> +  pseudo_snan,
> +  pseudo_unnormal
> +};
> +
> +static int
> +do_test (void)
> +{
> +  for (int i = 0; i < array_length (inputs); i++)
> +    {
> +      TEST_COMPARE (feclearexcept (FE_INVALID), 0);
> +      TEST_COMPARE (fpclassify (inputs[i].value), FP_NAN);
> +      TEST_COMPARE (fetestexcept (FE_INVALID), 0);
> +    }
> +
> +  return 0;
> +}
> +
> +#include <support/test-driver.c>
> --
> 2.43.0
>

LGTM.

Reviewed-by: H.J. Lu <hjl.tools@gmail.com>

Thanks.

-- 
H.J.


More information about the Libc-alpha mailing list