From 932544efa469568525455bcb53d4eabeba0eb295 Mon Sep 17 00:00:00 2001 From: Adhemerval Zanella Date: Wed, 3 Apr 2024 16:39:18 -0300 Subject: math: x86 floor traps when FE_INEXACT is enabled (BZ 31601) The implementations of floor functions using x87 floating point (i386 and 86_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. Reviewed-by: H.J. Lu --- sysdeps/i386/fpu/s_floor.S | 34 ---------------------------------- sysdeps/i386/fpu/s_floor.c | 25 +++++++++++++++++++++++++ sysdeps/i386/fpu/s_floorf.S | 34 ---------------------------------- sysdeps/i386/fpu/s_floorf.c | 25 +++++++++++++++++++++++++ sysdeps/i386/fpu/s_floorl.S | 39 --------------------------------------- sysdeps/x86/fpu/s_floorl.c | 25 +++++++++++++++++++++++++ sysdeps/x86_64/fpu/s_floorl.S | 33 --------------------------------- 7 files changed, 75 insertions(+), 140 deletions(-) delete mode 100644 sysdeps/i386/fpu/s_floor.S create mode 100644 sysdeps/i386/fpu/s_floor.c delete mode 100644 sysdeps/i386/fpu/s_floorf.S create mode 100644 sysdeps/i386/fpu/s_floorf.c delete mode 100644 sysdeps/i386/fpu/s_floorl.S create mode 100644 sysdeps/x86/fpu/s_floorl.c delete mode 100644 sysdeps/x86_64/fpu/s_floorl.S (limited to 'sysdeps') diff --git a/sysdeps/i386/fpu/s_floor.S b/sysdeps/i386/fpu/s_floor.S deleted file mode 100644 index 7143fdcc9a..0000000000 --- a/sysdeps/i386/fpu/s_floor.S +++ /dev/null @@ -1,34 +0,0 @@ -/* - * Public domain. - */ - -#include -#include - -RCSID("$NetBSD: s_floor.S,v 1.4 1995/05/09 00:01:59 jtc Exp $") - -ENTRY(__floor) - 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 $0x400,%edx /* round towards -oo */ - orl 4(%esp),%edx - andl $0xf7ff,%edx - movl %edx,(%esp) - fldcw (%esp) /* load modified control word */ - - frndint /* round */ - - fldenv 4(%esp) /* restore original environment */ - - addl $32,%esp - cfi_adjust_cfa_offset (-32) - ret -END (__floor) -libm_alias_double (__floor, floor) diff --git a/sysdeps/i386/fpu/s_floor.c b/sysdeps/i386/fpu/s_floor.c new file mode 100644 index 0000000000..cc50e33b59 --- /dev/null +++ b/sysdeps/i386/fpu/s_floor.c @@ -0,0 +1,25 @@ +/* Return smallest integral value not less than argument. i386 version. + Copyright (C) 2024 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + . */ + +#include + +#define FUNC __floor +#define TYPE double +#define FE_OPTION FE_DOWNWARD +#include "s_nearestint_387_template.c" +libm_alias_double (__floor, floor) diff --git a/sysdeps/i386/fpu/s_floorf.S b/sysdeps/i386/fpu/s_floorf.S deleted file mode 100644 index 8fad9c0698..0000000000 --- a/sysdeps/i386/fpu/s_floorf.S +++ /dev/null @@ -1,34 +0,0 @@ -/* - * Public domain. - */ - -#include -#include - -RCSID("$NetBSD: s_floorf.S,v 1.3 1995/05/09 00:04:32 jtc Exp $") - -ENTRY(__floorf) - 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 $0x400,%edx /* round towards -oo */ - orl 4(%esp),%edx - andl $0xf7ff,%edx - movl %edx,(%esp) - fldcw (%esp) /* load modified control word */ - - frndint /* round */ - - fldenv 4(%esp) /* restore original environment */ - - addl $32,%esp - cfi_adjust_cfa_offset (-32) - ret -END (__floorf) -libm_alias_float (__floor, floor) diff --git a/sysdeps/i386/fpu/s_floorf.c b/sysdeps/i386/fpu/s_floorf.c new file mode 100644 index 0000000000..fa9454e56b --- /dev/null +++ b/sysdeps/i386/fpu/s_floorf.c @@ -0,0 +1,25 @@ +/* Largest integral value not greater than argument i386 version. + Copyright (C) 2024 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + . */ + +#include + +#define FUNC __floorf +#define TYPE float +#define FE_OPTION FE_DOWNWARD +#include "s_nearestint_387_template.c" +libm_alias_float (__floor, floor) diff --git a/sysdeps/i386/fpu/s_floorl.S b/sysdeps/i386/fpu/s_floorl.S deleted file mode 100644 index 3ec28b477b..0000000000 --- a/sysdeps/i386/fpu/s_floorl.S +++ /dev/null @@ -1,39 +0,0 @@ -/* - * Public domain. - */ - -#include -#include - -RCSID("$NetBSD: $") - -ENTRY(__floorl) - 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 $0x400,%edx /* round towards -oo */ - orl 4(%esp),%edx - andl $0xf7ff,%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 (__floorl) -libm_alias_ldouble (__floor, floor) diff --git a/sysdeps/x86/fpu/s_floorl.c b/sysdeps/x86/fpu/s_floorl.c new file mode 100644 index 0000000000..9c92d33fbe --- /dev/null +++ b/sysdeps/x86/fpu/s_floorl.c @@ -0,0 +1,25 @@ +/* Return largest 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 + . */ + +#include + +#define FUNC __floorl +#define TYPE long double +#define FE_OPTION FE_DOWNWARD +#include "s_nearestint_387_template.c" +libm_alias_ldouble (__floor, floor) diff --git a/sysdeps/x86_64/fpu/s_floorl.S b/sysdeps/x86_64/fpu/s_floorl.S deleted file mode 100644 index b74d1a4d6b..0000000000 --- a/sysdeps/x86_64/fpu/s_floorl.S +++ /dev/null @@ -1,33 +0,0 @@ -/* - * Public domain. - */ - -#include -#include - -ENTRY(__floorl) - 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 $0x400,%edx /* round towards -oo */ - orl -28(%rsp),%edx - andl $0xf7ff,%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 (__floorl) -libm_alias_ldouble (__floor, floor) -- cgit 1.4.1