diff mbox series

[1/3] math: math: x86 ceill traps when FE_INEXACT is enabled (BZ 31600)

Message ID 20240403193919.1533786-2-adhemerval.zanella@linaro.org
State Accepted
Commit 637bfc392f406c6c77f59494ed34949a6161bceb
Headers show
Series Improve x86 rounding implementation when FE_INEXACT trap is enabled | expand

Commit Message

Adhemerval Zanella April 3, 2024, 7:39 p.m. UTC
The implementations of ceil functions using x87 floating point (i386 and
x86_64 long double only) traps when FE_INEXACT is enabled.  Although
this is a GNU extension outside the scope of the C standard, other
architectures that also support traps do not show this behavior.

The fix moves the implementation to a common one that holds any
exceptions with a 'fnclex' (libc_feholdexcept_setround_387).

Checked on x86_64-linux-gnu and i686-linux-gnu.
---
 math/Makefile                               |  3 +
 math/test-ceil-except-2.c                   | 67 +++++++++++++++++++++
 sysdeps/i386/fpu/s_ceil.S                   | 34 -----------
 sysdeps/i386/fpu/s_ceil.c                   | 25 ++++++++
 sysdeps/i386/fpu/s_ceilf.S                  | 34 -----------
 sysdeps/i386/fpu/s_ceilf.c                  | 25 ++++++++
 sysdeps/i386/fpu/s_ceill.S                  | 39 ------------
 sysdeps/x86/fpu/s_ceill.c                   | 25 ++++++++
 sysdeps/x86/fpu/s_nearestint_387_template.c | 36 +++++++++++
 sysdeps/x86_64/fpu/s_ceill.S                | 34 -----------
 10 files changed, 181 insertions(+), 141 deletions(-)
 create mode 100644 math/test-ceil-except-2.c
 delete mode 100644 sysdeps/i386/fpu/s_ceil.S
 create mode 100644 sysdeps/i386/fpu/s_ceil.c
 delete mode 100644 sysdeps/i386/fpu/s_ceilf.S
 create mode 100644 sysdeps/i386/fpu/s_ceilf.c
 delete mode 100644 sysdeps/i386/fpu/s_ceill.S
 create mode 100644 sysdeps/x86/fpu/s_ceill.c
 create mode 100644 sysdeps/x86/fpu/s_nearestint_387_template.c
 delete mode 100644 sysdeps/x86_64/fpu/s_ceill.S

Comments

H.J. Lu April 3, 2024, 8:03 p.m. UTC | #1
On Wed, Apr 3, 2024 at 12:39 PM Adhemerval Zanella
<adhemerval.zanella@linaro.org> wrote:
>
> The implementations of ceil functions using x87 floating point (i386 and
> x86_64 long double only) traps when FE_INEXACT is enabled.  Although
> this is a GNU extension outside the scope of the C standard, other
> architectures that also support traps do not show this behavior.
>
> The fix moves the implementation to a common one that holds any
> exceptions with a 'fnclex' (libc_feholdexcept_setround_387).
>
> Checked on x86_64-linux-gnu and i686-linux-gnu.
> ---
>  math/Makefile                               |  3 +
>  math/test-ceil-except-2.c                   | 67 +++++++++++++++++++++
>  sysdeps/i386/fpu/s_ceil.S                   | 34 -----------
>  sysdeps/i386/fpu/s_ceil.c                   | 25 ++++++++
>  sysdeps/i386/fpu/s_ceilf.S                  | 34 -----------
>  sysdeps/i386/fpu/s_ceilf.c                  | 25 ++++++++
>  sysdeps/i386/fpu/s_ceill.S                  | 39 ------------
>  sysdeps/x86/fpu/s_ceill.c                   | 25 ++++++++
>  sysdeps/x86/fpu/s_nearestint_387_template.c | 36 +++++++++++
>  sysdeps/x86_64/fpu/s_ceill.S                | 34 -----------
>  10 files changed, 181 insertions(+), 141 deletions(-)
>  create mode 100644 math/test-ceil-except-2.c
>  delete mode 100644 sysdeps/i386/fpu/s_ceil.S
>  create mode 100644 sysdeps/i386/fpu/s_ceil.c
>  delete mode 100644 sysdeps/i386/fpu/s_ceilf.S
>  create mode 100644 sysdeps/i386/fpu/s_ceilf.c
>  delete mode 100644 sysdeps/i386/fpu/s_ceill.S
>  create mode 100644 sysdeps/x86/fpu/s_ceill.c
>  create mode 100644 sysdeps/x86/fpu/s_nearestint_387_template.c
>  delete mode 100644 sysdeps/x86_64/fpu/s_ceill.S
>
> diff --git a/math/Makefile b/math/Makefile
> index 121a709121..d2a740eebe 100644
> --- a/math/Makefile
> +++ b/math/Makefile
> @@ -498,6 +498,7 @@ tests = \
>    bug-nextafter \
>    bug-nexttoward \
>    bug-tgmath1 \
> +  test-ceil-except-2 \
>    test-femode \
>    test-femode-traps \
>    test-fenv basic-test \
> @@ -989,6 +990,8 @@ CFLAGS-test-fe-snans-always-signal.c += $(config-cflags-signaling-nans)
>
>  CFLAGS-test-nan-const.c += -fno-builtin
>
> +CFLAGS-test-ceil-except-2.c += -fno-builtin
> +
>  include ../Rules
>
>  gen-all-calls = $(gen-libm-calls) $(gen-calls)
> diff --git a/math/test-ceil-except-2.c b/math/test-ceil-except-2.c
> new file mode 100644
> index 0000000000..394a272d89
> --- /dev/null
> +++ b/math/test-ceil-except-2.c
> @@ -0,0 +1,67 @@
> +/* Test ceil functions do not disable exception traps.
> +   Copyright (C) 2024 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 <fenv.h>
> +#include <math.h>
> +#include <stdio.h>
> +
> +#ifndef FE_INEXACT
> +# define FE_INEXACT 0
> +#endif
> +
> +#define TEST_FUNC(NAME, FLOAT, SUFFIX)                                 \
> +static int                                                             \
> +NAME (void)                                                            \
> +{                                                                      \
> +  int result = 0;                                                      \
> +  volatile FLOAT a, b __attribute__ ((unused));                                \
> +  a = 1.5;                                                             \
> +  /* ceil must work when traps on "inexact" are enabled.  */   \
> +  b = ceil ## SUFFIX (a);                                              \
> +  /* And it must have left those traps enabled.  */                    \
> +  if (fegetexcept () == FE_INEXACT)                                    \
> +    puts ("PASS: " #FLOAT);                                            \
> +  else                                                                 \
> +    {                                                                  \
> +      puts ("FAIL: " #FLOAT);                                          \
> +      result = 1;                                                      \
> +    }                                                                  \
> +  return result;                                                       \
> +}
> +
> +TEST_FUNC (float_test, float, f)
> +TEST_FUNC (double_test, double, )
> +TEST_FUNC (ldouble_test, long double, l)
> +
> +static int
> +do_test (void)
> +{
> +  if (feenableexcept (FE_INEXACT) == -1)
> +    {
> +      puts ("enabling FE_INEXACT traps failed, cannot test");
> +      return 77;
> +    }
> +  int result = float_test ();
> +  feenableexcept (FE_INEXACT);
> +  result |= double_test ();
> +  feenableexcept (FE_INEXACT);
> +  result |= ldouble_test ();
> +  return result;
> +}
> +
> +#include <support/test-driver.c>
> diff --git a/sysdeps/i386/fpu/s_ceil.S b/sysdeps/i386/fpu/s_ceil.S
> deleted file mode 100644
> index 99984f9b8d..0000000000
> --- a/sysdeps/i386/fpu/s_ceil.S
> +++ /dev/null
> @@ -1,34 +0,0 @@
> -/*
> - * Public domain.
> - */
> -
> -#include <machine/asm.h>
> -#include <libm-alias-double.h>
> -
> -RCSID("$NetBSD: s_ceil.S,v 1.4 1995/05/08 23:52:13 jtc Exp $")
> -
> -ENTRY(__ceil)
> -       fldl    4(%esp)
> -       subl    $32,%esp
> -       cfi_adjust_cfa_offset (32)
> -
> -       fnstenv 4(%esp)                 /* store fpu environment */
> -
> -       /* We use here %edx although only the low 1 bits are defined.
> -          But none of the operations should care and they are faster
> -          than the 16 bit operations.  */
> -       movl    $0x0800,%edx            /* round towards +oo */
> -       orl     4(%esp),%edx
> -       andl    $0xfbff,%edx
> -       movl    %edx,(%esp)
> -       fldcw   (%esp)                  /* load modified control word */
> -
> -       frndint                         /* round */
> -
> -       fldenv  4(%esp)                 /* restore original environment */
> -
> -       addl    $32,%esp
> -       cfi_adjust_cfa_offset (-32)
> -       ret
> -END (__ceil)
> -libm_alias_double (__ceil, ceil)
> diff --git a/sysdeps/i386/fpu/s_ceil.c b/sysdeps/i386/fpu/s_ceil.c
> new file mode 100644
> index 0000000000..349135c5d3
> --- /dev/null
> +++ b/sysdeps/i386/fpu/s_ceil.c
> @@ -0,0 +1,25 @@
> +/* Return smallest integral value not less than argument.  i386 version.
> +   Copyright (C) 2024 Free Software Foundation, Inc.
> +   This file is part of the GNU C Library.
> +
> +   The GNU C Library is free software; you can redistribute it and/or
> +   modify it under the terms of the GNU Lesser General Public
> +   License as published by the Free Software Foundation; either
> +   version 2.1 of the License, or (at your option) any later version.
> +
> +   The GNU C Library is distributed in the hope that it will be useful,
> +   but WITHOUT ANY WARRANTY; without even the implied warranty of
> +   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
> +   Lesser General Public License for more details.
> +
> +   You should have received a copy of the GNU Lesser General Public
> +   License along with the GNU C Library; if not, see
> +   <https://www.gnu.org/licenses/>.  */
> +
> +#include <libm-alias-double.h>
> +
> +#define FUNC       __ceil
> +#define TYPE       double
> +#define FE_OPTION  FE_UPWARD
> +#include "s_nearestint_387_template.c"
> +libm_alias_double (__ceil, ceil)
> diff --git a/sysdeps/i386/fpu/s_ceilf.S b/sysdeps/i386/fpu/s_ceilf.S
> deleted file mode 100644
> index 03e8e22609..0000000000
> --- a/sysdeps/i386/fpu/s_ceilf.S
> +++ /dev/null
> @@ -1,34 +0,0 @@
> -/*
> - * Public domain.
> - */
> -
> -#include <machine/asm.h>
> -#include <libm-alias-float.h>
> -
> -RCSID("$NetBSD: s_ceilf.S,v 1.3 1995/05/08 23:52:44 jtc Exp $")
> -
> -ENTRY(__ceilf)
> -       flds    4(%esp)
> -       subl    $32,%esp
> -       cfi_adjust_cfa_offset (32)
> -
> -       fnstenv 4(%esp)                 /* store fpu environment */
> -
> -       /* We use here %edx although only the low 1 bits are defined.
> -          But none of the operations should care and they are faster
> -          than the 16 bit operations.  */
> -       movl    $0x0800,%edx            /* round towards +oo */
> -       orl     4(%esp),%edx
> -       andl    $0xfbff,%edx
> -       movl    %edx,(%esp)
> -       fldcw   (%esp)                  /* load modified control word */
> -
> -       frndint                         /* round */
> -
> -       fldenv  4(%esp)                 /* restore original environment */
> -
> -       addl    $32,%esp
> -       cfi_adjust_cfa_offset (-32)
> -       ret
> -END (__ceilf)
> -libm_alias_float (__ceil, ceil)
> diff --git a/sysdeps/i386/fpu/s_ceilf.c b/sysdeps/i386/fpu/s_ceilf.c
> new file mode 100644
> index 0000000000..e73a20fd71
> --- /dev/null
> +++ b/sysdeps/i386/fpu/s_ceilf.c
> @@ -0,0 +1,25 @@
> +/* Return largest integral value not less than argument.  i386 version.
> +   Copyright (C) 2024 Free Software Foundation, Inc.
> +   This file is part of the GNU C Library.
> +
> +   The GNU C Library is free software; you can redistribute it and/or
> +   modify it under the terms of the GNU Lesser General Public
> +   License as published by the Free Software Foundation; either
> +   version 2.1 of the License, or (at your option) any later version.
> +
> +   The GNU C Library is distributed in the hope that it will be useful,
> +   but WITHOUT ANY WARRANTY; without even the implied warranty of
> +   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
> +   Lesser General Public License for more details.
> +
> +   You should have received a copy of the GNU Lesser General Public
> +   License along with the GNU C Library; if not, see
> +   <https://www.gnu.org/licenses/>.  */
> +
> +#include <libm-alias-float.h>
> +
> +#define FUNC       __ceilf
> +#define TYPE       float
> +#define FE_OPTION  FE_UPWARD
> +#include "s_nearestint_387_template.c"
> +libm_alias_float (__ceil, ceil)
> diff --git a/sysdeps/i386/fpu/s_ceill.S b/sysdeps/i386/fpu/s_ceill.S
> deleted file mode 100644
> index a551fce7f9..0000000000
> --- a/sysdeps/i386/fpu/s_ceill.S
> +++ /dev/null
> @@ -1,39 +0,0 @@
> -/*
> - * Public domain.
> - */
> -
> -#include <libm-alias-ldouble.h>
> -#include <machine/asm.h>
> -
> -RCSID("$NetBSD: $")
> -
> -ENTRY(__ceill)
> -       fldt    4(%esp)
> -       subl    $32,%esp
> -       cfi_adjust_cfa_offset (32)
> -
> -       fnstenv 4(%esp)                 /* store fpu environment */
> -
> -       /* We use here %edx although only the low 1 bits are defined.
> -          But none of the operations should care and they are faster
> -          than the 16 bit operations.  */
> -       movl    $0x0800,%edx            /* round towards +oo */
> -       orl     4(%esp),%edx
> -       andl    $0xfbff,%edx
> -       movl    %edx,(%esp)
> -       fldcw   (%esp)                  /* load modified control word */
> -
> -       frndint                         /* round */
> -
> -       /* Preserve "invalid" exceptions from sNaN input.  */
> -       fnstsw
> -       andl    $0x1, %eax
> -       orl     %eax, 8(%esp)
> -
> -       fldenv  4(%esp)                 /* restore original environment */
> -
> -       addl    $32,%esp
> -       cfi_adjust_cfa_offset (-32)
> -       ret
> -END (__ceill)
> -libm_alias_ldouble (__ceil, ceil)
> diff --git a/sysdeps/x86/fpu/s_ceill.c b/sysdeps/x86/fpu/s_ceill.c
> new file mode 100644
> index 0000000000..860dd2c960
> --- /dev/null
> +++ b/sysdeps/x86/fpu/s_ceill.c
> @@ -0,0 +1,25 @@
> +/* Return smallest integral value not less than argument.  x86 version.
> +   Copyright (C) 2024 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 <libm-alias-ldouble.h>
> +
> +#define FUNC       __ceill
> +#define TYPE       long double
> +#define FE_OPTION  FE_UPWARD
> +#include "s_nearestint_387_template.c"
> +libm_alias_ldouble (__ceil, ceil)
> diff --git a/sysdeps/x86/fpu/s_nearestint_387_template.c b/sysdeps/x86/fpu/s_nearestint_387_template.c
> new file mode 100644
> index 0000000000..95fca93f87
> --- /dev/null
> +++ b/sysdeps/x86/fpu/s_nearestint_387_template.c
> @@ -0,0 +1,36 @@
> +/* Nearest integet template for x86.
> +   Copyright (C) 2024 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/>.  */
> +
> +#define NO_MATH_REDIRECT
> +#include <math.h>
> +#include <fenv_private.h>
> +
> +TYPE
> +FUNC (TYPE x)
> +{
> +  fenv_t fenv;
> +  TYPE r;
> +
> +  libc_feholdexcept_setround_387 (&fenv, FE_OPTION);
> +  asm volatile ("frndint" : "=t" (r) : "0" (x));
> +  /* Preserve "invalid" exceptions from sNaN input.  */
> +  fenv.__status_word |= libc_fetestexcept_387 (FE_INVALID);
> +  libc_fesetenv_387 (&fenv);
> +
> +  return r;
> +}
> diff --git a/sysdeps/x86_64/fpu/s_ceill.S b/sysdeps/x86_64/fpu/s_ceill.S
> deleted file mode 100644
> index 16dbecd56d..0000000000
> --- a/sysdeps/x86_64/fpu/s_ceill.S
> +++ /dev/null
> @@ -1,34 +0,0 @@
> -/*
> - * Public domain.
> - */
> -
> -#include <libm-alias-ldouble.h>
> -#include <machine/asm.h>
> -
> -
> -ENTRY(__ceill)
> -       fldt    8(%rsp)
> -
> -       fnstenv -28(%rsp)               /* store fpu environment */
> -
> -       /* We use here %edx although only the low 1 bits are defined.
> -          But none of the operations should care and they are faster
> -          than the 16 bit operations.  */
> -       movl    $0x0800,%edx            /* round towards +oo */
> -       orl     -28(%rsp),%edx
> -       andl    $0xfbff,%edx
> -       movl    %edx,-32(%rsp)
> -       fldcw   -32(%rsp)               /* load modified control word */
> -
> -       frndint                         /* round */
> -
> -       /* Preserve "invalid" exceptions from sNaN input.  */
> -       fnstsw
> -       andl    $0x1, %eax
> -       orl     %eax, -24(%rsp)
> -
> -       fldenv  -28(%rsp)               /* restore original environment */
> -
> -       ret
> -END (__ceill)
> -libm_alias_ldouble (__ceil, ceil)
> --
> 2.34.1
>

LGTM.

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

Thanks.
diff mbox series

Patch

diff --git a/math/Makefile b/math/Makefile
index 121a709121..d2a740eebe 100644
--- a/math/Makefile
+++ b/math/Makefile
@@ -498,6 +498,7 @@  tests = \
   bug-nextafter \
   bug-nexttoward \
   bug-tgmath1 \
+  test-ceil-except-2 \
   test-femode \
   test-femode-traps \
   test-fenv basic-test \
@@ -989,6 +990,8 @@  CFLAGS-test-fe-snans-always-signal.c += $(config-cflags-signaling-nans)
 
 CFLAGS-test-nan-const.c += -fno-builtin
 
+CFLAGS-test-ceil-except-2.c += -fno-builtin
+
 include ../Rules
 
 gen-all-calls = $(gen-libm-calls) $(gen-calls)
diff --git a/math/test-ceil-except-2.c b/math/test-ceil-except-2.c
new file mode 100644
index 0000000000..394a272d89
--- /dev/null
+++ b/math/test-ceil-except-2.c
@@ -0,0 +1,67 @@ 
+/* Test ceil functions do not disable exception traps.
+   Copyright (C) 2024 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 <fenv.h>
+#include <math.h>
+#include <stdio.h>
+
+#ifndef FE_INEXACT
+# define FE_INEXACT 0
+#endif
+
+#define TEST_FUNC(NAME, FLOAT, SUFFIX)					\
+static int								\
+NAME (void)								\
+{									\
+  int result = 0;							\
+  volatile FLOAT a, b __attribute__ ((unused));				\
+  a = 1.5;								\
+  /* ceil must work when traps on "inexact" are enabled.  */	\
+  b = ceil ## SUFFIX (a);						\
+  /* And it must have left those traps enabled.  */			\
+  if (fegetexcept () == FE_INEXACT)					\
+    puts ("PASS: " #FLOAT);						\
+  else									\
+    {									\
+      puts ("FAIL: " #FLOAT);						\
+      result = 1;							\
+    }									\
+  return result;							\
+}
+
+TEST_FUNC (float_test, float, f)
+TEST_FUNC (double_test, double, )
+TEST_FUNC (ldouble_test, long double, l)
+
+static int
+do_test (void)
+{
+  if (feenableexcept (FE_INEXACT) == -1)
+    {
+      puts ("enabling FE_INEXACT traps failed, cannot test");
+      return 77;
+    }
+  int result = float_test ();
+  feenableexcept (FE_INEXACT);
+  result |= double_test ();
+  feenableexcept (FE_INEXACT);
+  result |= ldouble_test ();
+  return result;
+}
+
+#include <support/test-driver.c>
diff --git a/sysdeps/i386/fpu/s_ceil.S b/sysdeps/i386/fpu/s_ceil.S
deleted file mode 100644
index 99984f9b8d..0000000000
--- a/sysdeps/i386/fpu/s_ceil.S
+++ /dev/null
@@ -1,34 +0,0 @@ 
-/*
- * Public domain.
- */
-
-#include <machine/asm.h>
-#include <libm-alias-double.h>
-
-RCSID("$NetBSD: s_ceil.S,v 1.4 1995/05/08 23:52:13 jtc Exp $")
-
-ENTRY(__ceil)
-	fldl	4(%esp)
-	subl	$32,%esp
-	cfi_adjust_cfa_offset (32)
-
-	fnstenv	4(%esp)			/* store fpu environment */
-
-	/* We use here %edx although only the low 1 bits are defined.
-	   But none of the operations should care and they are faster
-	   than the 16 bit operations.  */
-	movl	$0x0800,%edx		/* round towards +oo */
-	orl	4(%esp),%edx
-	andl	$0xfbff,%edx
-	movl	%edx,(%esp)
-	fldcw	(%esp)			/* load modified control word */
-
-	frndint				/* round */
-
-	fldenv	4(%esp)			/* restore original environment */
-
-	addl	$32,%esp
-	cfi_adjust_cfa_offset (-32)
-	ret
-END (__ceil)
-libm_alias_double (__ceil, ceil)
diff --git a/sysdeps/i386/fpu/s_ceil.c b/sysdeps/i386/fpu/s_ceil.c
new file mode 100644
index 0000000000..349135c5d3
--- /dev/null
+++ b/sysdeps/i386/fpu/s_ceil.c
@@ -0,0 +1,25 @@ 
+/* Return smallest integral value not less than argument.  i386 version.
+   Copyright (C) 2024 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <https://www.gnu.org/licenses/>.  */
+
+#include <libm-alias-double.h>
+
+#define FUNC       __ceil
+#define TYPE       double
+#define FE_OPTION  FE_UPWARD
+#include "s_nearestint_387_template.c"
+libm_alias_double (__ceil, ceil)
diff --git a/sysdeps/i386/fpu/s_ceilf.S b/sysdeps/i386/fpu/s_ceilf.S
deleted file mode 100644
index 03e8e22609..0000000000
--- a/sysdeps/i386/fpu/s_ceilf.S
+++ /dev/null
@@ -1,34 +0,0 @@ 
-/*
- * Public domain.
- */
-
-#include <machine/asm.h>
-#include <libm-alias-float.h>
-
-RCSID("$NetBSD: s_ceilf.S,v 1.3 1995/05/08 23:52:44 jtc Exp $")
-
-ENTRY(__ceilf)
-	flds	4(%esp)
-	subl	$32,%esp
-	cfi_adjust_cfa_offset (32)
-
-	fnstenv	4(%esp)			/* store fpu environment */
-
-	/* We use here %edx although only the low 1 bits are defined.
-	   But none of the operations should care and they are faster
-	   than the 16 bit operations.  */
-	movl	$0x0800,%edx		/* round towards +oo */
-	orl	4(%esp),%edx
-	andl	$0xfbff,%edx
-	movl	%edx,(%esp)
-	fldcw	(%esp)			/* load modified control word */
-
-	frndint				/* round */
-
-	fldenv	4(%esp)			/* restore original environment */
-
-	addl	$32,%esp
-	cfi_adjust_cfa_offset (-32)
-	ret
-END (__ceilf)
-libm_alias_float (__ceil, ceil)
diff --git a/sysdeps/i386/fpu/s_ceilf.c b/sysdeps/i386/fpu/s_ceilf.c
new file mode 100644
index 0000000000..e73a20fd71
--- /dev/null
+++ b/sysdeps/i386/fpu/s_ceilf.c
@@ -0,0 +1,25 @@ 
+/* Return largest integral value not less than argument.  i386 version.
+   Copyright (C) 2024 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <https://www.gnu.org/licenses/>.  */
+
+#include <libm-alias-float.h>
+
+#define FUNC       __ceilf
+#define TYPE       float
+#define FE_OPTION  FE_UPWARD
+#include "s_nearestint_387_template.c"
+libm_alias_float (__ceil, ceil)
diff --git a/sysdeps/i386/fpu/s_ceill.S b/sysdeps/i386/fpu/s_ceill.S
deleted file mode 100644
index a551fce7f9..0000000000
--- a/sysdeps/i386/fpu/s_ceill.S
+++ /dev/null
@@ -1,39 +0,0 @@ 
-/*
- * Public domain.
- */
-
-#include <libm-alias-ldouble.h>
-#include <machine/asm.h>
-
-RCSID("$NetBSD: $")
-
-ENTRY(__ceill)
-	fldt	4(%esp)
-	subl	$32,%esp
-	cfi_adjust_cfa_offset (32)
-
-	fnstenv	4(%esp)			/* store fpu environment */
-
-	/* We use here %edx although only the low 1 bits are defined.
-	   But none of the operations should care and they are faster
-	   than the 16 bit operations.  */
-	movl	$0x0800,%edx		/* round towards +oo */
-	orl	4(%esp),%edx
-	andl	$0xfbff,%edx
-	movl	%edx,(%esp)
-	fldcw	(%esp)			/* load modified control word */
-
-	frndint				/* round */
-
-	/* Preserve "invalid" exceptions from sNaN input.  */
-	fnstsw
-	andl	$0x1, %eax
-	orl	%eax, 8(%esp)
-
-	fldenv	4(%esp)			/* restore original environment */
-
-	addl	$32,%esp
-	cfi_adjust_cfa_offset (-32)
-	ret
-END (__ceill)
-libm_alias_ldouble (__ceil, ceil)
diff --git a/sysdeps/x86/fpu/s_ceill.c b/sysdeps/x86/fpu/s_ceill.c
new file mode 100644
index 0000000000..860dd2c960
--- /dev/null
+++ b/sysdeps/x86/fpu/s_ceill.c
@@ -0,0 +1,25 @@ 
+/* Return smallest integral value not less than argument.  x86 version.
+   Copyright (C) 2024 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 <libm-alias-ldouble.h>
+
+#define FUNC       __ceill
+#define TYPE       long double
+#define FE_OPTION  FE_UPWARD
+#include "s_nearestint_387_template.c"
+libm_alias_ldouble (__ceil, ceil)
diff --git a/sysdeps/x86/fpu/s_nearestint_387_template.c b/sysdeps/x86/fpu/s_nearestint_387_template.c
new file mode 100644
index 0000000000..95fca93f87
--- /dev/null
+++ b/sysdeps/x86/fpu/s_nearestint_387_template.c
@@ -0,0 +1,36 @@ 
+/* Nearest integet template for x86.
+   Copyright (C) 2024 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/>.  */
+
+#define NO_MATH_REDIRECT
+#include <math.h>
+#include <fenv_private.h>
+
+TYPE
+FUNC (TYPE x)
+{
+  fenv_t fenv;
+  TYPE r;
+
+  libc_feholdexcept_setround_387 (&fenv, FE_OPTION);
+  asm volatile ("frndint" : "=t" (r) : "0" (x));
+  /* Preserve "invalid" exceptions from sNaN input.  */
+  fenv.__status_word |= libc_fetestexcept_387 (FE_INVALID);
+  libc_fesetenv_387 (&fenv);
+
+  return r;
+}
diff --git a/sysdeps/x86_64/fpu/s_ceill.S b/sysdeps/x86_64/fpu/s_ceill.S
deleted file mode 100644
index 16dbecd56d..0000000000
--- a/sysdeps/x86_64/fpu/s_ceill.S
+++ /dev/null
@@ -1,34 +0,0 @@ 
-/*
- * Public domain.
- */
-
-#include <libm-alias-ldouble.h>
-#include <machine/asm.h>
-
-
-ENTRY(__ceill)
-	fldt	8(%rsp)
-
-	fnstenv	-28(%rsp)		/* store fpu environment */
-
-	/* We use here %edx although only the low 1 bits are defined.
-	   But none of the operations should care and they are faster
-	   than the 16 bit operations.  */
-	movl	$0x0800,%edx		/* round towards +oo */
-	orl	-28(%rsp),%edx
-	andl	$0xfbff,%edx
-	movl	%edx,-32(%rsp)
-	fldcw	-32(%rsp)		/* load modified control word */
-
-	frndint				/* round */
-
-	/* Preserve "invalid" exceptions from sNaN input.  */
-	fnstsw
-	andl	$0x1, %eax
-	orl	%eax, -24(%rsp)
-
-	fldenv	-28(%rsp)		/* restore original environment */
-
-	ret
-END (__ceill)
-libm_alias_ldouble (__ceil, ceil)