From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-yw1-x112a.google.com (mail-yw1-x112a.google.com [IPv6:2607:f8b0:4864:20::112a]) by sourceware.org (Postfix) with ESMTPS id 038AD3847725 for ; Wed, 3 Apr 2024 20:03:47 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.2 sourceware.org 038AD3847725 Authentication-Results: sourceware.org; dmarc=pass (p=none dis=none) header.from=gmail.com Authentication-Results: sourceware.org; spf=pass smtp.mailfrom=gmail.com ARC-Filter: OpenARC Filter v1.0.0 sourceware.org 038AD3847725 Authentication-Results: server2.sourceware.org; arc=none smtp.remote-ip=2607:f8b0:4864:20::112a ARC-Seal: i=1; a=rsa-sha256; d=sourceware.org; s=key; t=1712174629; cv=none; b=KicqehNHKHFdjwfPo3C4pLvPW5bewQ7tDh4EwCy51EBCbSKD5FjBEanJU55p07Ye5t917B3Tw2QO40lo4zf+w3RME9rrJLq7P7gsNJRmVIYg0Ez7ds4DQwMi55JtmEk74QvQACHUBlgfK2cFG50T9BuJCDhO8brbDmyGjaBYMbs= ARC-Message-Signature: i=1; a=rsa-sha256; d=sourceware.org; s=key; t=1712174629; c=relaxed/simple; bh=odzvMeXRh1AxobTA+f84U/RqZ/Z3VWxg11zzxrM/mQY=; h=DKIM-Signature:MIME-Version:From:Date:Message-ID:Subject:To; b=azdYQ3wLJy4uQ1i+fFjFEqoJESQgrPkybQmOUQxt6iYYjsPfyFyl43d1rgjTTey3IjiDgx+v/y3d3OaZYOT+Eg77pQi0UCFlvGaQl6ktnIREpD2eVgWTW/38+y6M74tDRe1HcQcoS72RmfV/4QIIhuQprkpP8NyROCpH/PZYxlE= ARC-Authentication-Results: i=1; server2.sourceware.org Received: by mail-yw1-x112a.google.com with SMTP id 00721157ae682-6150c1fa3daso2862887b3.1 for ; Wed, 03 Apr 2024 13:03:47 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20230601; t=1712174626; x=1712779426; darn=sourceware.org; h=content-transfer-encoding:cc:to:subject:message-id:date:from :in-reply-to:references:mime-version:from:to:cc:subject:date :message-id:reply-to; bh=L1G2fcQ+GPFYZXlqIaXUXKa+vvcJ7Xcl3kvAwxE86SM=; b=azTB38oe46hBa/SeMNOtpfN/PIEPh8xddAsV4ZN5FLHyxmMJca+sb2tTSP8NfXLC/l +mZabkPjdrPC+gYuMcxNqD/B1CKP/Vwzo/H3LNgNVrdpHBfzKG5crtd6uqG/z73XqnLT r2oVHajMicnj4NNx96GPSnSzpVSlxU6zX2UwP1MgboBBHiAwwbIbvN77ll++yAJuJnRK xqG7f0iVLDZA2dY82Kh3/4An4BmqxlZ23PxhfBME7UOzjjqJIgfMdzEOylw5elDPDjH0 qnFOj2QunyH9xeBeLS2fsgC8cmfIeYR5UkXNB13oDHnOtuLN43qv0I3q0sDFiWWXDzU2 UNww== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20230601; t=1712174626; x=1712779426; h=content-transfer-encoding:cc:to:subject:message-id:date:from :in-reply-to:references:mime-version:x-gm-message-state:from:to:cc :subject:date:message-id:reply-to; bh=L1G2fcQ+GPFYZXlqIaXUXKa+vvcJ7Xcl3kvAwxE86SM=; b=OJ5eKsXYj/d/TKEHqRRLD8mITWkw+7yzXC7cBfN7uSCwqb8V3gnFQ2Ku//qik6uBfR dHtr62bTevho0rARAKGvqBh5bYPXw1lqf5yr4vBiDC/khi7hDeMv+tD5ZBLtfs4OHxxm fm4b46XvpSpZRKno6/XED1ssLZdrNJp/33NREVYg2G9BTFpYmCq1yHWoYOSTB1sH7SD4 lQXElCXdiv31nxlxmRv8cMC0yltoM2pduN8ZWo2pC5XDcXu+ytTAetKaZ7CLaLmhTJAS xlAwhwQjcC3whqu03FB4vuDjamvzVDoABZCuA9B9vvb4r0Gc1lhEMmxYEqUZODo7lQwU y9Yg== X-Gm-Message-State: AOJu0YzsY7NJ+NOp6xJS0cuUzE+fLh10l5s9OFSmm2QGtp9MHhl6e9A6 4OSbeDMBRAcWmlp4KU+nZu9Ss8gFopuzLiNfZ5Buc8Q1SJlzfqdSzOz2zwWg10bC/TcL6YRhvMQ lxSxEROKXCQ/JlqAFRUiZlo1Lq6s= X-Google-Smtp-Source: AGHT+IFph/L7ASXs7qRbxhe4wd2FvQDrKZwQHQeAcMuX9KbYLbztjF9S5qdIYvZ8DDjESL4LDOQ1dgUBID1jCLAmuMo= X-Received: by 2002:a5b:288:0:b0:dc7:46e7:7aea with SMTP id x8-20020a5b0288000000b00dc746e77aeamr526021ybl.47.1712174626145; Wed, 03 Apr 2024 13:03:46 -0700 (PDT) MIME-Version: 1.0 References: <20240403193919.1533786-1-adhemerval.zanella@linaro.org> <20240403193919.1533786-2-adhemerval.zanella@linaro.org> In-Reply-To: <20240403193919.1533786-2-adhemerval.zanella@linaro.org> From: "H.J. Lu" Date: Wed, 3 Apr 2024 13:03:09 -0700 Message-ID: Subject: Re: [PATCH 1/3] math: math: x86 ceill traps when FE_INEXACT is enabled (BZ 31600) To: Adhemerval Zanella Cc: libc-alpha@sourceware.org, Joseph Myers Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: quoted-printable X-Spam-Status: No, score=-3019.7 required=5.0 tests=BAYES_00,DKIM_SIGNED,DKIM_VALID,DKIM_VALID_AU,DKIM_VALID_EF,FREEMAIL_FROM,GIT_PATCH_0,KAM_SHORT,RCVD_IN_DNSWL_NONE,SPF_HELO_NONE,SPF_PASS,TXREP autolearn=ham autolearn_force=no version=3.4.6 X-Spam-Checker-Version: SpamAssassin 3.4.6 (2021-04-09) on server2.sourceware.org List-Id: On Wed, Apr 3, 2024 at 12:39=E2=80=AFPM Adhemerval Zanella 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 =3D \ > 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 +=3D $(config-cf= lags-signaling-nans) > > CFLAGS-test-nan-const.c +=3D -fno-builtin > > +CFLAGS-test-ceil-except-2.c +=3D -fno-builtin > + > include ../Rules > > gen-all-calls =3D $(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 > + . */ > + > +#include > +#include > +#include > + > +#ifndef FE_INEXACT > +# define FE_INEXACT 0 > +#endif > + > +#define TEST_FUNC(NAME, FLOAT, SUFFIX) \ > +static int \ > +NAME (void) \ > +{ \ > + int result =3D 0; = \ > + volatile FLOAT a, b __attribute__ ((unused)); = \ > + a =3D 1.5; = \ > + /* ceil must work when traps on "inexact" are enabled. */ \ > + b =3D ceil ## SUFFIX (a); = \ > + /* And it must have left those traps enabled. */ \ > + if (fegetexcept () =3D=3D FE_INEXACT) = \ > + puts ("PASS: " #FLOAT); \ > + else \ > + { \ > + puts ("FAIL: " #FLOAT); \ > + result =3D 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) =3D=3D -1) > + { > + puts ("enabling FE_INEXACT traps failed, cannot test"); > + return 77; > + } > + int result =3D float_test (); > + feenableexcept (FE_INEXACT); > + result |=3D double_test (); > + feenableexcept (FE_INEXACT); > + result |=3D ldouble_test (); > + return result; > +} > + > +#include > 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 > -#include > - > -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 $32,%esp > - cfi_adjust_cfa_offset (-32) > - ret > -END (__ceil) > -libm_alias_double (__ceil, ceil) > diff --git a/sysdeps/i386/fpu/s_ceil.c b/sysdeps/i386/fpu/s_ceil.c > new file mode 100644 > index 0000000000..349135c5d3 > --- /dev/null > +++ b/sysdeps/i386/fpu/s_ceil.c > @@ -0,0 +1,25 @@ > +/* Return smallest integral value not less than argument. i386 version. > + Copyright (C) 2024 Free Software Foundation, Inc. > + This file is part of the GNU C Library. > + > + The GNU C Library is free software; you can redistribute it and/or > + modify it under the terms of the GNU Lesser General Public > + License as published by the Free Software Foundation; either > + version 2.1 of the License, or (at your option) any later version. > + > + The GNU C Library is distributed in the hope that it will be useful, > + but WITHOUT ANY WARRANTY; without even the implied warranty of > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU > + Lesser General Public License for more details. > + > + You should have received a copy of the GNU Lesser General Public > + License along with the GNU C Library; if not, see > + . */ > + > +#include > + > +#define FUNC __ceil > +#define TYPE double > +#define FE_OPTION FE_UPWARD > +#include "s_nearestint_387_template.c" > +libm_alias_double (__ceil, ceil) > diff --git a/sysdeps/i386/fpu/s_ceilf.S b/sysdeps/i386/fpu/s_ceilf.S > deleted file mode 100644 > index 03e8e22609..0000000000 > --- a/sysdeps/i386/fpu/s_ceilf.S > +++ /dev/null > @@ -1,34 +0,0 @@ > -/* > - * Public domain. > - */ > - > -#include > -#include > - > -RCSID("$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 $32,%esp > - cfi_adjust_cfa_offset (-32) > - ret > -END (__ceilf) > -libm_alias_float (__ceil, ceil) > diff --git a/sysdeps/i386/fpu/s_ceilf.c b/sysdeps/i386/fpu/s_ceilf.c > new file mode 100644 > index 0000000000..e73a20fd71 > --- /dev/null > +++ b/sysdeps/i386/fpu/s_ceilf.c > @@ -0,0 +1,25 @@ > +/* Return largest integral value not less than argument. i386 version. > + Copyright (C) 2024 Free Software Foundation, Inc. > + This file is part of the GNU C Library. > + > + The GNU C Library is free software; you can redistribute it and/or > + modify it under the terms of the GNU Lesser General Public > + License as published by the Free Software Foundation; either > + version 2.1 of the License, or (at your option) any later version. > + > + The GNU C Library is distributed in the hope that it will be useful, > + but WITHOUT ANY WARRANTY; without even the implied warranty of > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU > + Lesser General Public License for more details. > + > + You should have received a copy of the GNU Lesser General Public > + License along with the GNU C Library; if not, see > + . */ > + > +#include > + > +#define FUNC __ceilf > +#define TYPE float > +#define FE_OPTION FE_UPWARD > +#include "s_nearestint_387_template.c" > +libm_alias_float (__ceil, ceil) > diff --git a/sysdeps/i386/fpu/s_ceill.S b/sysdeps/i386/fpu/s_ceill.S > deleted file mode 100644 > index a551fce7f9..0000000000 > --- a/sysdeps/i386/fpu/s_ceill.S > +++ /dev/null > @@ -1,39 +0,0 @@ > -/* > - * Public domain. > - */ > - > -#include > -#include > - > -RCSID("$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 > + . */ > + > +#include > + > +#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/fp= u/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 > + . */ > + > +#define NO_MATH_REDIRECT > +#include > +#include > + > +TYPE > +FUNC (TYPE x) > +{ > + fenv_t fenv; > + TYPE r; > + > + libc_feholdexcept_setround_387 (&fenv, FE_OPTION); > + asm volatile ("frndint" : "=3Dt" (r) : "0" (x)); > + /* Preserve "invalid" exceptions from sNaN input. */ > + fenv.__status_word |=3D 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 > -#include > - > - > -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 Thanks. --=20 H.J.