From: Carlos O'Donell <carlos@redhat.com>
To: Adhemerval Zanella <adhemerval.zanella@linaro.org>,
libc-alpha@sourceware.org, Bruno Haible <bruno@clisp.org>
Subject: Re: [PATCH v2 2/7] i686: Do not raise exception traps on fesetexcept (BZ 30989)
Date: Mon, 6 Nov 2023 11:14:13 -0500 [thread overview]
Message-ID: <067cc983-cc14-f9f7-5875-88a7dadb77a6@redhat.com> (raw)
In-Reply-To: <20231106132713.953501-3-adhemerval.zanella@linaro.org>
On 11/6/23 08:27, Adhemerval Zanella wrote:
> According to ISO C23 (7.6.4.4), fesetexcept is supposed to set
> floating-point exception flags without raising a trap (unlike
> feraiseexcept, which is supposed to raise a trap if feenableexcept
> was called with the appropriate argument).
>
> The flags can be set in the 387 unit or in the SSE unit. To set
> a flag, it is sufficient to do it in the SSE unit, because that is
> guaranteed to not trap. However, on i386 CPUs that have only a
> 387 unit, set the flags in the 387, as long as this cannot trap.
>
> Checked on i686-linux-gnu.
LGTM.
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
> ---
> math/test-fesetexcept-traps.c | 28 ++++++++++++---
> sysdeps/i386/fpu/fesetexcept.c | 46 +++++++++++++++++++++---
> sysdeps/i386/fpu/math-tests-trap-force.h | 29 +++++++++++++++
> sysdeps/x86/fpu/test-fenv-sse-2.c | 23 +++---------
> 4 files changed, 100 insertions(+), 26 deletions(-)
> create mode 100644 sysdeps/i386/fpu/math-tests-trap-force.h
>
> diff --git a/math/test-fesetexcept-traps.c b/math/test-fesetexcept-traps.c
> index 96f6c4752f..8a5c0bca80 100644
> --- a/math/test-fesetexcept-traps.c
> +++ b/math/test-fesetexcept-traps.c
> @@ -19,6 +19,7 @@
> #include <fenv.h>
> #include <stdio.h>
> #include <math-tests.h>
> +#include <math-barriers.h>
>
> static int
> do_test (void)
> @@ -41,8 +42,28 @@ do_test (void)
>
> /* Verify fesetexcept does not cause exception traps. For architectures
> where setting the exception might result in traps the function should
> - return a nonzero value. */
> - ret = fesetexcept (FE_ALL_EXCEPT);
> + return a nonzero value.
> + Also check if the function does not alter the exception mask. */
> + {
> + int exc_before = fegetexcept ();
> + ret = fesetexcept (FE_ALL_EXCEPT);
> + int exc_after = fegetexcept ();
> + if (exc_before != exc_after)
> + {
> + puts ("fesetexcept (FE_ALL_EXCEPT) changed the exceptions mask");
> + return 1;
> + }
> + }
> +
> + /* Execute some floating-point operations, since on some CPUs exceptions
> + triggers a trap only at the next floating-point instruction. */
Correct, this is called delayed floating point exception handling and hppa has this too.
> + volatile double a = 1.0;
> + volatile double b = a + a;
> + math_force_eval (b);
> + volatile long double al = 1.0L;
> + volatile long double bl = al + al;
> + math_force_eval (bl);
> +
> if (ret == 0)
> puts ("fesetexcept (FE_ALL_EXCEPT) succeeded");
> else if (!EXCEPTION_SET_FORCES_TRAP)
> @@ -61,5 +82,4 @@ do_test (void)
> return result;
> }
>
> -#define TEST_FUNCTION do_test ()
> -#include "../test-skeleton.c"
> +#include <support/test-driver.c>
> diff --git a/sysdeps/i386/fpu/fesetexcept.c b/sysdeps/i386/fpu/fesetexcept.c
> index 18949e982a..58f577d93d 100644
> --- a/sysdeps/i386/fpu/fesetexcept.c
> +++ b/sysdeps/i386/fpu/fesetexcept.c
> @@ -17,15 +17,53 @@
> <https://www.gnu.org/licenses/>. */
>
> #include <fenv.h>
> +#include <ldsodefs.h>
>
> int
> fesetexcept (int excepts)
> {
> - fenv_t temp;
> + /* The flags can be set in the 387 unit or in the SSE unit. To set a flag,
> + it is sufficient to do it in the SSE unit, because that is guaranteed to
> + not trap. However, on i386 CPUs that have only a 387 unit, set the flags
> + in the 387, as long as this cannot trap. */
>
> - __asm__ ("fnstenv %0" : "=m" (*&temp));
> - temp.__status_word |= excepts & FE_ALL_EXCEPT;
> - __asm__ ("fldenv %0" : : "m" (*&temp));
> + excepts &= FE_ALL_EXCEPT;
> +
> + if (CPU_FEATURE_USABLE (SSE))
> + {
> + /* Get the control word of the SSE unit. */
> + unsigned int mxcsr;
> + __asm__ ("stmxcsr %0" : "=m" (*&mxcsr));
> +
> + /* Set relevant flags. */
> + mxcsr |= excepts;
> +
> + /* Put the new data in effect. */
> + __asm__ ("ldmxcsr %0" : : "m" (*&mxcsr));
> + }
> + else
> + {
> + fenv_t temp;
> +
> + /* Note: fnstenv masks all floating-point exceptions until the fldenv
> + or fldcw below. */
> + __asm__ ("fnstenv %0" : "=m" (*&temp));
> +
> + /* Set relevant flags. */
> + temp.__status_word |= excepts;
> +
> + if ((~temp.__control_word) & excepts)
> + {
> + /* Setting the exception flags may trigger a trap (at the next
> + floating-point instruction, but that does not matter).
> + ISO C23 (7.6.4.4) does not allow it. */
> + __asm__ volatile ("fldcw %0" : : "m" (*&temp.__control_word));
> + return -1;
> + }
> +
> + /* Store the new status word (along with the rest of the environment). */
> + __asm__ ("fldenv %0" : : "m" (*&temp));
> + }
>
> return 0;
> }
> diff --git a/sysdeps/i386/fpu/math-tests-trap-force.h b/sysdeps/i386/fpu/math-tests-trap-force.h
> new file mode 100644
> index 0000000000..f41e1ffc2d
> --- /dev/null
> +++ b/sysdeps/i386/fpu/math-tests-trap-force.h
> @@ -0,0 +1,29 @@
> +/* Configuration for math tests: support for setting exception flags
> + without causing enabled traps. i686 version.
> + Copyright (C) 2023 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 I386_FPU_MATH_TESTS_TRAP_FORCE_H
> +#define I386_FPU_MATH_TESTS_TRAP_FORCE_H 1
> +
> +#include <cpu-features.h>
> +
> +/* Setting exception flags in FPU Status Register results in enabled traps for
> + those exceptions being taken. */
> +#define EXCEPTION_SET_FORCES_TRAP !CPU_FEATURE_USABLE (SSE)
> +
> +#endif /* math-tests-trap-force.h. */
> diff --git a/sysdeps/x86/fpu/test-fenv-sse-2.c b/sysdeps/x86/fpu/test-fenv-sse-2.c
> index f3e820b6ed..7a0503790f 100644
> --- a/sysdeps/x86/fpu/test-fenv-sse-2.c
> +++ b/sysdeps/x86/fpu/test-fenv-sse-2.c
> @@ -22,17 +22,8 @@
> #include <stdbool.h>
> #include <stdint.h>
> #include <stdio.h>
> -
> -static bool
> -have_sse2 (void)
> -{
> - unsigned int eax, ebx, ecx, edx;
> -
> - if (!__get_cpuid (1, &eax, &ebx, &ecx, &edx))
> - return false;
> -
> - return (edx & bit_SSE2) != 0;
> -}
> +#include <cpu-features.h>
> +#include <support/check.h>
>
> static uint32_t
> get_sse_mxcsr (void)
> @@ -164,13 +155,9 @@ sse_tests (void)
> static int
> do_test (void)
> {
> - if (!have_sse2 ())
> - {
> - puts ("CPU does not support SSE2, cannot test");
> - return 0;
> - }
> + if (!CPU_FEATURE_USABLE (SSE2))
> + FAIL_UNSUPPORTED ("CPU does not support SSE2");
> return sse_tests ();
> }
>
> -#define TEST_FUNCTION do_test ()
> -#include <test-skeleton.c>
> +#include <support/test-driver.c>
--
Cheers,
Carlos.
next prev parent reply other threads:[~2023-11-06 16:14 UTC|newest]
Thread overview: 33+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-11-06 13:27 [PATCH v2 0/7] Multiple floating-point environment fixes Adhemerval Zanella
2023-11-06 13:27 ` [PATCH v2 1/7] powerpc: Do not raise exception traps for fesetexcept/fesetexceptflag (BZ 30988) Adhemerval Zanella
2023-11-06 16:08 ` Carlos O'Donell
2023-11-06 16:50 ` Adhemerval Zanella Netto
2023-11-06 17:02 ` Carlos O'Donell
2023-11-06 17:11 ` Adhemerval Zanella Netto
2023-11-06 17:37 ` Adhemerval Zanella Netto
2023-11-06 17:38 ` Carlos O'Donell
2023-11-06 17:56 ` Adhemerval Zanella Netto
2023-11-06 20:46 ` Adhemerval Zanella Netto
2023-11-23 21:47 ` Carlos O'Donell
2023-11-24 12:28 ` Adhemerval Zanella Netto
2023-11-24 12:37 ` Adhemerval Zanella Netto
2023-11-24 16:22 ` Carlos O'Donell
2023-11-24 17:53 ` Adhemerval Zanella Netto
2023-11-24 18:15 ` Carlos O'Donell
2023-11-24 18:46 ` Adhemerval Zanella Netto
2023-11-27 13:46 ` Adhemerval Zanella Netto
2023-12-19 14:57 ` Carlos O'Donell
2023-11-06 13:27 ` [PATCH v2 2/7] i686: Do not raise exception traps on fesetexcept (BZ 30989) Adhemerval Zanella
2023-11-06 16:14 ` Carlos O'Donell [this message]
2023-11-06 13:27 ` [PATCH v2 3/7] x86: Do not raises floating-point exception traps on fesetexceptflag (BZ 30990) Adhemerval Zanella
2023-11-06 16:16 ` Carlos O'Donell
2023-11-06 13:27 ` [PATCH v2 4/7] manual: Clarify undefined behavior of feenableexcept (BZ 31019) Adhemerval Zanella
2023-11-06 16:17 ` Carlos O'Donell
2023-11-06 13:27 ` [PATCH v2 5/7] riscv: Fix feenvupdate with FE_DFL_ENV (BZ 31022) Adhemerval Zanella
2023-11-06 16:19 ` Carlos O'Donell
2023-11-06 13:27 ` [PATCH v2 6/7] alpha: Fix fesetexceptflag (BZ 30998) Adhemerval Zanella
2023-11-06 16:54 ` Carlos O'Donell
2023-11-06 17:36 ` Bruno Haible
2023-11-06 18:15 ` Carlos O'Donell
2023-11-06 13:27 ` [PATCH v2 7/7] hppa: Fix undefined behaviour in feclearexcept (BZ 30983) Adhemerval Zanella
2023-11-06 16:57 ` Carlos O'Donell
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=067cc983-cc14-f9f7-5875-88a7dadb77a6@redhat.com \
--to=carlos@redhat.com \
--cc=adhemerval.zanella@linaro.org \
--cc=bruno@clisp.org \
--cc=libc-alpha@sourceware.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).