Thu May 30 11:24:05 1996 Roland McGrath <roland@delasyd.gnu.ai.mit.edu>
* po/header.pot: Replace with exact boilerplate pinard dictates.
* sysdeps/i386/strtok.S (Lillegal_argument): Remove this code to set
errno and the check that jumped to it.
* sysdeps/mach/hurd/Makefile (errnos.d): Use $(sed-remove-objpfx).
Thu May 30 03:21:57 1996 Ulrich Drepper <drepper@cygnus.com>
* FAQ: Document need of gperf program for developers.
* elf/elf.h: Fix typos in comments.
* libio/stdio.h [!__STRICT_ANSI__ || _POSIX_SOURCE]: Add
prototypes for `ctermid' and `cuserid'.
* locale/programs/locale.c: Switch to user selected locale
before printing variables.
* math/Makefile [$(long-double-fcts)==yes]: Define long-m-routines
and long-c-routines. Only if the `long double' data type is
available we need to compile the functions.
(libm-routines): Add $(long-m-routines).
(routines): Remove isinfl, isnanl. Use new file s_isinfl and
s_isnanl instead if `long double' is available.
* math/math.h: Include <mathcalls.h> again to define `long double'
functions.
* math/math_private.h: Define data types, prototypes and access
macros for `long double'.
* stdlib/stdlib.h: Add prototypes for `strtoll' and `strtoull'.
[GCC2 && OPTIMIZE]: Define strto{,u}ll as inline function which
calls __strto{,u}q_internal.
* stdlib/strfmon.c: Replace PTR by `void *'.
* stdlib/strtoq.c: Define strtoll as weak alias.
* stdlib/strtouq.c: Define strtoull as weak alias.
* string/tester.c: Correct `strsep' test.
* sysdeps/generic/strsep.c: Make compatible with BSD version.
Trailing characters of skip set are not skipped. In this case
empty tokens are returned.
* sysdeps/i386/isinfl.c, sysdeps/i386/isnanl.c,
sysdeps/ieee754/isinf.c, sysdeps/ieee754/isinfl.c,
sysdeps/ieee754/isnan.c, sysdeps/ieee754/isnanl.c: Removed. We
now use the versions part of libm.
* sysdeps/i386/strsep.S: Removed. Generic C version is of
similar speed.
* sysdeps/i386/strtok.S: Remove support for `strsep'.
* sysdeps/libm-i387/e_acosl.S, sysdeps/libm-i387/s_ceill.S,
sysdeps/libm-i387/s_copysignl.S, sysdeps/libm-i387/s_finitel.S,
sysdeps/libm-i387/s_floorl.S, sysdeps/libm-i387/s_isinfl.c,
sysdeps/libm-i387/s_isnanl.c, sysdeps/libm-i387/s_nextafterl.c,
sysdeps/libm-i387/s_rintl.S, sysdeps/libm-i387/s_significandl.S:
New i387 specific math functions implementing `long double'
versions.
* sysdeps/libm-ieee754/s_ceill.c,
sysdeps/libm-ieee754/s_copysignl.c,
sysdeps/libm-ieee754/s_fabsl.c, sysdeps/libm-ieee754/s_finitel.c,
sysdeps/libm-ieee754/s_floorl.c, sysdeps/libm-ieee754/s_isinfl.c,
sysdeps/libm-ieee754/s_isnanl.c,
sysdeps/libm-ieee754/s_nextafterl.c,
sysdeps/libm-ieee754/s_rintl.c, sysdeps/libm-ieee754/s_scalbnl.c,
sysdeps/libm-ieee754/s_significandl.c: New generic `long double'
versions of libm functions.
* sysdeps/libm-i387/e_exp.S: Add a few comments to explain the
Intel FPU nonsense.
* sysdeps/libm-i387/s_ceil.S, sysdeps/libm-i387/s_ceilf.S,
sysdeps/libm-i387/s_floor.S, sysdeps/libm-i387/s_floorf.S: Correct
handling of local variables. The old version created a stack
frame but stored the values outside.
* sysdeps/libm-ieee754/s_isinf.c, sysdeps/libm-ieee754/s_isnan.c
[!NO_LONG_DOUBLE]: Define alias with `long double' versions name.
* login/pututline_r.c: Include sys/stat.h. Fix typos.
according to currently used locale for category LC_CTYPE by
inet_nsap_ntoa. Now in <arpa/inet.h>.
_IO_dup2 to contain complete parameter list.
1996-05-30 16:12:42 +00:00
|
|
|
/*
|
|
|
|
|
* Written by J.T. Conklin <jtc@netbsd.org>.
|
|
|
|
|
* Changes for long double by Ulrich Drepper <drepper@cygnus.com>
|
|
|
|
|
* Public domain.
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
#include <machine/asm.h>
|
|
|
|
|
|
|
|
|
|
RCSID("$NetBSD: $")
|
|
|
|
|
|
|
|
|
|
ENTRY(__floorl)
|
1997-01-01 15:28:18 +00:00
|
|
|
fldt 4(%esp)
|
Avoid "inexact" exceptions in i386/x86_64 floor functions (bug 15479).
As discussed in
<https://sourceware.org/ml/libc-alpha/2016-05/msg00577.html>, TS
18661-1 disallows ceil, floor, round and trunc functions from raising
the "inexact" exception, in accordance with general IEEE 754 semantics
for when that exception is raised. Fixing this for x87 floating point
is more complicated than for the other versions of these functions,
because they use the frndint instruction that raises "inexact" and
this can only be avoided by saving and restoring the whole
floating-point environment.
As I noted in
<https://sourceware.org/ml/libc-alpha/2016-06/msg00128.html>, I have
now implemented a GCC option -fno-fp-int-builtin-inexact for GCC 7,
such that GCC will inline these functions on x86, without caring about
"inexact", when the default -ffp-int-builtin-inexact is in effect.
This allows users to get optimized code depending on the options they
pass to the compiler, while making the out-of-line functions follow TS
18661-1 semantics and avoid "inexact".
This patch duly fixes the out-of-line floor function implementations
to avoid "inexact", in the same way as the nearbyint implementations.
I do not know how the performance of implementations such as these
based on saving the environment and changing the rounding mode
temporarily compares to that of the C versions or SSE 4.1 versions (of
course, for 32-bit x86 SSE implementations still need to get the
return value in an x87 register); it's entirely possible other
implementations could be faster in some cases.
Tested for x86_64 and x86.
[BZ #15479]
* sysdeps/i386/fpu/s_floor.S (__floor): Save and restore
floating-point environment rather than just control word.
* sysdeps/i386/fpu/s_floorf.S (__floorf): Likewise.
* sysdeps/i386/fpu/s_floorl.S (__floorl): Save and restore
floating-point environment, with "invalid" exceptions merged in,
rather than just control word.
* sysdeps/x86_64/fpu/s_floorl.S (__floorl): Likewise.
* math/libm-test.inc (floor_test_data): Do not allow spurious
"inexact" exceptions.
2016-06-27 17:25:47 +00:00
|
|
|
subl $32,%esp
|
|
|
|
|
cfi_adjust_cfa_offset (32)
|
Thu May 30 11:24:05 1996 Roland McGrath <roland@delasyd.gnu.ai.mit.edu>
* po/header.pot: Replace with exact boilerplate pinard dictates.
* sysdeps/i386/strtok.S (Lillegal_argument): Remove this code to set
errno and the check that jumped to it.
* sysdeps/mach/hurd/Makefile (errnos.d): Use $(sed-remove-objpfx).
Thu May 30 03:21:57 1996 Ulrich Drepper <drepper@cygnus.com>
* FAQ: Document need of gperf program for developers.
* elf/elf.h: Fix typos in comments.
* libio/stdio.h [!__STRICT_ANSI__ || _POSIX_SOURCE]: Add
prototypes for `ctermid' and `cuserid'.
* locale/programs/locale.c: Switch to user selected locale
before printing variables.
* math/Makefile [$(long-double-fcts)==yes]: Define long-m-routines
and long-c-routines. Only if the `long double' data type is
available we need to compile the functions.
(libm-routines): Add $(long-m-routines).
(routines): Remove isinfl, isnanl. Use new file s_isinfl and
s_isnanl instead if `long double' is available.
* math/math.h: Include <mathcalls.h> again to define `long double'
functions.
* math/math_private.h: Define data types, prototypes and access
macros for `long double'.
* stdlib/stdlib.h: Add prototypes for `strtoll' and `strtoull'.
[GCC2 && OPTIMIZE]: Define strto{,u}ll as inline function which
calls __strto{,u}q_internal.
* stdlib/strfmon.c: Replace PTR by `void *'.
* stdlib/strtoq.c: Define strtoll as weak alias.
* stdlib/strtouq.c: Define strtoull as weak alias.
* string/tester.c: Correct `strsep' test.
* sysdeps/generic/strsep.c: Make compatible with BSD version.
Trailing characters of skip set are not skipped. In this case
empty tokens are returned.
* sysdeps/i386/isinfl.c, sysdeps/i386/isnanl.c,
sysdeps/ieee754/isinf.c, sysdeps/ieee754/isinfl.c,
sysdeps/ieee754/isnan.c, sysdeps/ieee754/isnanl.c: Removed. We
now use the versions part of libm.
* sysdeps/i386/strsep.S: Removed. Generic C version is of
similar speed.
* sysdeps/i386/strtok.S: Remove support for `strsep'.
* sysdeps/libm-i387/e_acosl.S, sysdeps/libm-i387/s_ceill.S,
sysdeps/libm-i387/s_copysignl.S, sysdeps/libm-i387/s_finitel.S,
sysdeps/libm-i387/s_floorl.S, sysdeps/libm-i387/s_isinfl.c,
sysdeps/libm-i387/s_isnanl.c, sysdeps/libm-i387/s_nextafterl.c,
sysdeps/libm-i387/s_rintl.S, sysdeps/libm-i387/s_significandl.S:
New i387 specific math functions implementing `long double'
versions.
* sysdeps/libm-ieee754/s_ceill.c,
sysdeps/libm-ieee754/s_copysignl.c,
sysdeps/libm-ieee754/s_fabsl.c, sysdeps/libm-ieee754/s_finitel.c,
sysdeps/libm-ieee754/s_floorl.c, sysdeps/libm-ieee754/s_isinfl.c,
sysdeps/libm-ieee754/s_isnanl.c,
sysdeps/libm-ieee754/s_nextafterl.c,
sysdeps/libm-ieee754/s_rintl.c, sysdeps/libm-ieee754/s_scalbnl.c,
sysdeps/libm-ieee754/s_significandl.c: New generic `long double'
versions of libm functions.
* sysdeps/libm-i387/e_exp.S: Add a few comments to explain the
Intel FPU nonsense.
* sysdeps/libm-i387/s_ceil.S, sysdeps/libm-i387/s_ceilf.S,
sysdeps/libm-i387/s_floor.S, sysdeps/libm-i387/s_floorf.S: Correct
handling of local variables. The old version created a stack
frame but stored the values outside.
* sysdeps/libm-ieee754/s_isinf.c, sysdeps/libm-ieee754/s_isnan.c
[!NO_LONG_DOUBLE]: Define alias with `long double' versions name.
* login/pututline_r.c: Include sys/stat.h. Fix typos.
according to currently used locale for category LC_CTYPE by
inet_nsap_ntoa. Now in <arpa/inet.h>.
_IO_dup2 to contain complete parameter list.
1996-05-30 16:12:42 +00:00
|
|
|
|
Avoid "inexact" exceptions in i386/x86_64 floor functions (bug 15479).
As discussed in
<https://sourceware.org/ml/libc-alpha/2016-05/msg00577.html>, TS
18661-1 disallows ceil, floor, round and trunc functions from raising
the "inexact" exception, in accordance with general IEEE 754 semantics
for when that exception is raised. Fixing this for x87 floating point
is more complicated than for the other versions of these functions,
because they use the frndint instruction that raises "inexact" and
this can only be avoided by saving and restoring the whole
floating-point environment.
As I noted in
<https://sourceware.org/ml/libc-alpha/2016-06/msg00128.html>, I have
now implemented a GCC option -fno-fp-int-builtin-inexact for GCC 7,
such that GCC will inline these functions on x86, without caring about
"inexact", when the default -ffp-int-builtin-inexact is in effect.
This allows users to get optimized code depending on the options they
pass to the compiler, while making the out-of-line functions follow TS
18661-1 semantics and avoid "inexact".
This patch duly fixes the out-of-line floor function implementations
to avoid "inexact", in the same way as the nearbyint implementations.
I do not know how the performance of implementations such as these
based on saving the environment and changing the rounding mode
temporarily compares to that of the C versions or SSE 4.1 versions (of
course, for 32-bit x86 SSE implementations still need to get the
return value in an x87 register); it's entirely possible other
implementations could be faster in some cases.
Tested for x86_64 and x86.
[BZ #15479]
* sysdeps/i386/fpu/s_floor.S (__floor): Save and restore
floating-point environment rather than just control word.
* sysdeps/i386/fpu/s_floorf.S (__floorf): Likewise.
* sysdeps/i386/fpu/s_floorl.S (__floorl): Save and restore
floating-point environment, with "invalid" exceptions merged in,
rather than just control word.
* sysdeps/x86_64/fpu/s_floorl.S (__floorl): Likewise.
* math/libm-test.inc (floor_test_data): Do not allow spurious
"inexact" exceptions.
2016-06-27 17:25:47 +00:00
|
|
|
fnstenv 4(%esp) /* store fpu environment */
|
Thu May 30 11:24:05 1996 Roland McGrath <roland@delasyd.gnu.ai.mit.edu>
* po/header.pot: Replace with exact boilerplate pinard dictates.
* sysdeps/i386/strtok.S (Lillegal_argument): Remove this code to set
errno and the check that jumped to it.
* sysdeps/mach/hurd/Makefile (errnos.d): Use $(sed-remove-objpfx).
Thu May 30 03:21:57 1996 Ulrich Drepper <drepper@cygnus.com>
* FAQ: Document need of gperf program for developers.
* elf/elf.h: Fix typos in comments.
* libio/stdio.h [!__STRICT_ANSI__ || _POSIX_SOURCE]: Add
prototypes for `ctermid' and `cuserid'.
* locale/programs/locale.c: Switch to user selected locale
before printing variables.
* math/Makefile [$(long-double-fcts)==yes]: Define long-m-routines
and long-c-routines. Only if the `long double' data type is
available we need to compile the functions.
(libm-routines): Add $(long-m-routines).
(routines): Remove isinfl, isnanl. Use new file s_isinfl and
s_isnanl instead if `long double' is available.
* math/math.h: Include <mathcalls.h> again to define `long double'
functions.
* math/math_private.h: Define data types, prototypes and access
macros for `long double'.
* stdlib/stdlib.h: Add prototypes for `strtoll' and `strtoull'.
[GCC2 && OPTIMIZE]: Define strto{,u}ll as inline function which
calls __strto{,u}q_internal.
* stdlib/strfmon.c: Replace PTR by `void *'.
* stdlib/strtoq.c: Define strtoll as weak alias.
* stdlib/strtouq.c: Define strtoull as weak alias.
* string/tester.c: Correct `strsep' test.
* sysdeps/generic/strsep.c: Make compatible with BSD version.
Trailing characters of skip set are not skipped. In this case
empty tokens are returned.
* sysdeps/i386/isinfl.c, sysdeps/i386/isnanl.c,
sysdeps/ieee754/isinf.c, sysdeps/ieee754/isinfl.c,
sysdeps/ieee754/isnan.c, sysdeps/ieee754/isnanl.c: Removed. We
now use the versions part of libm.
* sysdeps/i386/strsep.S: Removed. Generic C version is of
similar speed.
* sysdeps/i386/strtok.S: Remove support for `strsep'.
* sysdeps/libm-i387/e_acosl.S, sysdeps/libm-i387/s_ceill.S,
sysdeps/libm-i387/s_copysignl.S, sysdeps/libm-i387/s_finitel.S,
sysdeps/libm-i387/s_floorl.S, sysdeps/libm-i387/s_isinfl.c,
sysdeps/libm-i387/s_isnanl.c, sysdeps/libm-i387/s_nextafterl.c,
sysdeps/libm-i387/s_rintl.S, sysdeps/libm-i387/s_significandl.S:
New i387 specific math functions implementing `long double'
versions.
* sysdeps/libm-ieee754/s_ceill.c,
sysdeps/libm-ieee754/s_copysignl.c,
sysdeps/libm-ieee754/s_fabsl.c, sysdeps/libm-ieee754/s_finitel.c,
sysdeps/libm-ieee754/s_floorl.c, sysdeps/libm-ieee754/s_isinfl.c,
sysdeps/libm-ieee754/s_isnanl.c,
sysdeps/libm-ieee754/s_nextafterl.c,
sysdeps/libm-ieee754/s_rintl.c, sysdeps/libm-ieee754/s_scalbnl.c,
sysdeps/libm-ieee754/s_significandl.c: New generic `long double'
versions of libm functions.
* sysdeps/libm-i387/e_exp.S: Add a few comments to explain the
Intel FPU nonsense.
* sysdeps/libm-i387/s_ceil.S, sysdeps/libm-i387/s_ceilf.S,
sysdeps/libm-i387/s_floor.S, sysdeps/libm-i387/s_floorf.S: Correct
handling of local variables. The old version created a stack
frame but stored the values outside.
* sysdeps/libm-ieee754/s_isinf.c, sysdeps/libm-ieee754/s_isnan.c
[!NO_LONG_DOUBLE]: Define alias with `long double' versions name.
* login/pututline_r.c: Include sys/stat.h. Fix typos.
according to currently used locale for category LC_CTYPE by
inet_nsap_ntoa. Now in <arpa/inet.h>.
_IO_dup2 to contain complete parameter list.
1996-05-30 16:12:42 +00:00
|
|
|
|
1997-01-01 15:28:18 +00:00
|
|
|
/* 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 */
|
Thu May 30 11:24:05 1996 Roland McGrath <roland@delasyd.gnu.ai.mit.edu>
* po/header.pot: Replace with exact boilerplate pinard dictates.
* sysdeps/i386/strtok.S (Lillegal_argument): Remove this code to set
errno and the check that jumped to it.
* sysdeps/mach/hurd/Makefile (errnos.d): Use $(sed-remove-objpfx).
Thu May 30 03:21:57 1996 Ulrich Drepper <drepper@cygnus.com>
* FAQ: Document need of gperf program for developers.
* elf/elf.h: Fix typos in comments.
* libio/stdio.h [!__STRICT_ANSI__ || _POSIX_SOURCE]: Add
prototypes for `ctermid' and `cuserid'.
* locale/programs/locale.c: Switch to user selected locale
before printing variables.
* math/Makefile [$(long-double-fcts)==yes]: Define long-m-routines
and long-c-routines. Only if the `long double' data type is
available we need to compile the functions.
(libm-routines): Add $(long-m-routines).
(routines): Remove isinfl, isnanl. Use new file s_isinfl and
s_isnanl instead if `long double' is available.
* math/math.h: Include <mathcalls.h> again to define `long double'
functions.
* math/math_private.h: Define data types, prototypes and access
macros for `long double'.
* stdlib/stdlib.h: Add prototypes for `strtoll' and `strtoull'.
[GCC2 && OPTIMIZE]: Define strto{,u}ll as inline function which
calls __strto{,u}q_internal.
* stdlib/strfmon.c: Replace PTR by `void *'.
* stdlib/strtoq.c: Define strtoll as weak alias.
* stdlib/strtouq.c: Define strtoull as weak alias.
* string/tester.c: Correct `strsep' test.
* sysdeps/generic/strsep.c: Make compatible with BSD version.
Trailing characters of skip set are not skipped. In this case
empty tokens are returned.
* sysdeps/i386/isinfl.c, sysdeps/i386/isnanl.c,
sysdeps/ieee754/isinf.c, sysdeps/ieee754/isinfl.c,
sysdeps/ieee754/isnan.c, sysdeps/ieee754/isnanl.c: Removed. We
now use the versions part of libm.
* sysdeps/i386/strsep.S: Removed. Generic C version is of
similar speed.
* sysdeps/i386/strtok.S: Remove support for `strsep'.
* sysdeps/libm-i387/e_acosl.S, sysdeps/libm-i387/s_ceill.S,
sysdeps/libm-i387/s_copysignl.S, sysdeps/libm-i387/s_finitel.S,
sysdeps/libm-i387/s_floorl.S, sysdeps/libm-i387/s_isinfl.c,
sysdeps/libm-i387/s_isnanl.c, sysdeps/libm-i387/s_nextafterl.c,
sysdeps/libm-i387/s_rintl.S, sysdeps/libm-i387/s_significandl.S:
New i387 specific math functions implementing `long double'
versions.
* sysdeps/libm-ieee754/s_ceill.c,
sysdeps/libm-ieee754/s_copysignl.c,
sysdeps/libm-ieee754/s_fabsl.c, sysdeps/libm-ieee754/s_finitel.c,
sysdeps/libm-ieee754/s_floorl.c, sysdeps/libm-ieee754/s_isinfl.c,
sysdeps/libm-ieee754/s_isnanl.c,
sysdeps/libm-ieee754/s_nextafterl.c,
sysdeps/libm-ieee754/s_rintl.c, sysdeps/libm-ieee754/s_scalbnl.c,
sysdeps/libm-ieee754/s_significandl.c: New generic `long double'
versions of libm functions.
* sysdeps/libm-i387/e_exp.S: Add a few comments to explain the
Intel FPU nonsense.
* sysdeps/libm-i387/s_ceil.S, sysdeps/libm-i387/s_ceilf.S,
sysdeps/libm-i387/s_floor.S, sysdeps/libm-i387/s_floorf.S: Correct
handling of local variables. The old version created a stack
frame but stored the values outside.
* sysdeps/libm-ieee754/s_isinf.c, sysdeps/libm-ieee754/s_isnan.c
[!NO_LONG_DOUBLE]: Define alias with `long double' versions name.
* login/pututline_r.c: Include sys/stat.h. Fix typos.
according to currently used locale for category LC_CTYPE by
inet_nsap_ntoa. Now in <arpa/inet.h>.
_IO_dup2 to contain complete parameter list.
1996-05-30 16:12:42 +00:00
|
|
|
|
1997-01-01 15:28:18 +00:00
|
|
|
frndint /* round */
|
Thu May 30 11:24:05 1996 Roland McGrath <roland@delasyd.gnu.ai.mit.edu>
* po/header.pot: Replace with exact boilerplate pinard dictates.
* sysdeps/i386/strtok.S (Lillegal_argument): Remove this code to set
errno and the check that jumped to it.
* sysdeps/mach/hurd/Makefile (errnos.d): Use $(sed-remove-objpfx).
Thu May 30 03:21:57 1996 Ulrich Drepper <drepper@cygnus.com>
* FAQ: Document need of gperf program for developers.
* elf/elf.h: Fix typos in comments.
* libio/stdio.h [!__STRICT_ANSI__ || _POSIX_SOURCE]: Add
prototypes for `ctermid' and `cuserid'.
* locale/programs/locale.c: Switch to user selected locale
before printing variables.
* math/Makefile [$(long-double-fcts)==yes]: Define long-m-routines
and long-c-routines. Only if the `long double' data type is
available we need to compile the functions.
(libm-routines): Add $(long-m-routines).
(routines): Remove isinfl, isnanl. Use new file s_isinfl and
s_isnanl instead if `long double' is available.
* math/math.h: Include <mathcalls.h> again to define `long double'
functions.
* math/math_private.h: Define data types, prototypes and access
macros for `long double'.
* stdlib/stdlib.h: Add prototypes for `strtoll' and `strtoull'.
[GCC2 && OPTIMIZE]: Define strto{,u}ll as inline function which
calls __strto{,u}q_internal.
* stdlib/strfmon.c: Replace PTR by `void *'.
* stdlib/strtoq.c: Define strtoll as weak alias.
* stdlib/strtouq.c: Define strtoull as weak alias.
* string/tester.c: Correct `strsep' test.
* sysdeps/generic/strsep.c: Make compatible with BSD version.
Trailing characters of skip set are not skipped. In this case
empty tokens are returned.
* sysdeps/i386/isinfl.c, sysdeps/i386/isnanl.c,
sysdeps/ieee754/isinf.c, sysdeps/ieee754/isinfl.c,
sysdeps/ieee754/isnan.c, sysdeps/ieee754/isnanl.c: Removed. We
now use the versions part of libm.
* sysdeps/i386/strsep.S: Removed. Generic C version is of
similar speed.
* sysdeps/i386/strtok.S: Remove support for `strsep'.
* sysdeps/libm-i387/e_acosl.S, sysdeps/libm-i387/s_ceill.S,
sysdeps/libm-i387/s_copysignl.S, sysdeps/libm-i387/s_finitel.S,
sysdeps/libm-i387/s_floorl.S, sysdeps/libm-i387/s_isinfl.c,
sysdeps/libm-i387/s_isnanl.c, sysdeps/libm-i387/s_nextafterl.c,
sysdeps/libm-i387/s_rintl.S, sysdeps/libm-i387/s_significandl.S:
New i387 specific math functions implementing `long double'
versions.
* sysdeps/libm-ieee754/s_ceill.c,
sysdeps/libm-ieee754/s_copysignl.c,
sysdeps/libm-ieee754/s_fabsl.c, sysdeps/libm-ieee754/s_finitel.c,
sysdeps/libm-ieee754/s_floorl.c, sysdeps/libm-ieee754/s_isinfl.c,
sysdeps/libm-ieee754/s_isnanl.c,
sysdeps/libm-ieee754/s_nextafterl.c,
sysdeps/libm-ieee754/s_rintl.c, sysdeps/libm-ieee754/s_scalbnl.c,
sysdeps/libm-ieee754/s_significandl.c: New generic `long double'
versions of libm functions.
* sysdeps/libm-i387/e_exp.S: Add a few comments to explain the
Intel FPU nonsense.
* sysdeps/libm-i387/s_ceil.S, sysdeps/libm-i387/s_ceilf.S,
sysdeps/libm-i387/s_floor.S, sysdeps/libm-i387/s_floorf.S: Correct
handling of local variables. The old version created a stack
frame but stored the values outside.
* sysdeps/libm-ieee754/s_isinf.c, sysdeps/libm-ieee754/s_isnan.c
[!NO_LONG_DOUBLE]: Define alias with `long double' versions name.
* login/pututline_r.c: Include sys/stat.h. Fix typos.
according to currently used locale for category LC_CTYPE by
inet_nsap_ntoa. Now in <arpa/inet.h>.
_IO_dup2 to contain complete parameter list.
1996-05-30 16:12:42 +00:00
|
|
|
|
Avoid "inexact" exceptions in i386/x86_64 floor functions (bug 15479).
As discussed in
<https://sourceware.org/ml/libc-alpha/2016-05/msg00577.html>, TS
18661-1 disallows ceil, floor, round and trunc functions from raising
the "inexact" exception, in accordance with general IEEE 754 semantics
for when that exception is raised. Fixing this for x87 floating point
is more complicated than for the other versions of these functions,
because they use the frndint instruction that raises "inexact" and
this can only be avoided by saving and restoring the whole
floating-point environment.
As I noted in
<https://sourceware.org/ml/libc-alpha/2016-06/msg00128.html>, I have
now implemented a GCC option -fno-fp-int-builtin-inexact for GCC 7,
such that GCC will inline these functions on x86, without caring about
"inexact", when the default -ffp-int-builtin-inexact is in effect.
This allows users to get optimized code depending on the options they
pass to the compiler, while making the out-of-line functions follow TS
18661-1 semantics and avoid "inexact".
This patch duly fixes the out-of-line floor function implementations
to avoid "inexact", in the same way as the nearbyint implementations.
I do not know how the performance of implementations such as these
based on saving the environment and changing the rounding mode
temporarily compares to that of the C versions or SSE 4.1 versions (of
course, for 32-bit x86 SSE implementations still need to get the
return value in an x87 register); it's entirely possible other
implementations could be faster in some cases.
Tested for x86_64 and x86.
[BZ #15479]
* sysdeps/i386/fpu/s_floor.S (__floor): Save and restore
floating-point environment rather than just control word.
* sysdeps/i386/fpu/s_floorf.S (__floorf): Likewise.
* sysdeps/i386/fpu/s_floorl.S (__floorl): Save and restore
floating-point environment, with "invalid" exceptions merged in,
rather than just control word.
* sysdeps/x86_64/fpu/s_floorl.S (__floorl): Likewise.
* math/libm-test.inc (floor_test_data): Do not allow spurious
"inexact" exceptions.
2016-06-27 17:25:47 +00:00
|
|
|
/* Preserve "invalid" exceptions from sNaN input. */
|
|
|
|
|
fnstsw
|
|
|
|
|
andl $0x1, %eax
|
|
|
|
|
orl %eax, 8(%esp)
|
1997-01-01 15:28:18 +00:00
|
|
|
|
Avoid "inexact" exceptions in i386/x86_64 floor functions (bug 15479).
As discussed in
<https://sourceware.org/ml/libc-alpha/2016-05/msg00577.html>, TS
18661-1 disallows ceil, floor, round and trunc functions from raising
the "inexact" exception, in accordance with general IEEE 754 semantics
for when that exception is raised. Fixing this for x87 floating point
is more complicated than for the other versions of these functions,
because they use the frndint instruction that raises "inexact" and
this can only be avoided by saving and restoring the whole
floating-point environment.
As I noted in
<https://sourceware.org/ml/libc-alpha/2016-06/msg00128.html>, I have
now implemented a GCC option -fno-fp-int-builtin-inexact for GCC 7,
such that GCC will inline these functions on x86, without caring about
"inexact", when the default -ffp-int-builtin-inexact is in effect.
This allows users to get optimized code depending on the options they
pass to the compiler, while making the out-of-line functions follow TS
18661-1 semantics and avoid "inexact".
This patch duly fixes the out-of-line floor function implementations
to avoid "inexact", in the same way as the nearbyint implementations.
I do not know how the performance of implementations such as these
based on saving the environment and changing the rounding mode
temporarily compares to that of the C versions or SSE 4.1 versions (of
course, for 32-bit x86 SSE implementations still need to get the
return value in an x87 register); it's entirely possible other
implementations could be faster in some cases.
Tested for x86_64 and x86.
[BZ #15479]
* sysdeps/i386/fpu/s_floor.S (__floor): Save and restore
floating-point environment rather than just control word.
* sysdeps/i386/fpu/s_floorf.S (__floorf): Likewise.
* sysdeps/i386/fpu/s_floorl.S (__floorl): Save and restore
floating-point environment, with "invalid" exceptions merged in,
rather than just control word.
* sysdeps/x86_64/fpu/s_floorl.S (__floorl): Likewise.
* math/libm-test.inc (floor_test_data): Do not allow spurious
"inexact" exceptions.
2016-06-27 17:25:47 +00:00
|
|
|
fldenv 4(%esp) /* restore original environment */
|
|
|
|
|
|
|
|
|
|
addl $32,%esp
|
|
|
|
|
cfi_adjust_cfa_offset (-32)
|
Thu May 30 11:24:05 1996 Roland McGrath <roland@delasyd.gnu.ai.mit.edu>
* po/header.pot: Replace with exact boilerplate pinard dictates.
* sysdeps/i386/strtok.S (Lillegal_argument): Remove this code to set
errno and the check that jumped to it.
* sysdeps/mach/hurd/Makefile (errnos.d): Use $(sed-remove-objpfx).
Thu May 30 03:21:57 1996 Ulrich Drepper <drepper@cygnus.com>
* FAQ: Document need of gperf program for developers.
* elf/elf.h: Fix typos in comments.
* libio/stdio.h [!__STRICT_ANSI__ || _POSIX_SOURCE]: Add
prototypes for `ctermid' and `cuserid'.
* locale/programs/locale.c: Switch to user selected locale
before printing variables.
* math/Makefile [$(long-double-fcts)==yes]: Define long-m-routines
and long-c-routines. Only if the `long double' data type is
available we need to compile the functions.
(libm-routines): Add $(long-m-routines).
(routines): Remove isinfl, isnanl. Use new file s_isinfl and
s_isnanl instead if `long double' is available.
* math/math.h: Include <mathcalls.h> again to define `long double'
functions.
* math/math_private.h: Define data types, prototypes and access
macros for `long double'.
* stdlib/stdlib.h: Add prototypes for `strtoll' and `strtoull'.
[GCC2 && OPTIMIZE]: Define strto{,u}ll as inline function which
calls __strto{,u}q_internal.
* stdlib/strfmon.c: Replace PTR by `void *'.
* stdlib/strtoq.c: Define strtoll as weak alias.
* stdlib/strtouq.c: Define strtoull as weak alias.
* string/tester.c: Correct `strsep' test.
* sysdeps/generic/strsep.c: Make compatible with BSD version.
Trailing characters of skip set are not skipped. In this case
empty tokens are returned.
* sysdeps/i386/isinfl.c, sysdeps/i386/isnanl.c,
sysdeps/ieee754/isinf.c, sysdeps/ieee754/isinfl.c,
sysdeps/ieee754/isnan.c, sysdeps/ieee754/isnanl.c: Removed. We
now use the versions part of libm.
* sysdeps/i386/strsep.S: Removed. Generic C version is of
similar speed.
* sysdeps/i386/strtok.S: Remove support for `strsep'.
* sysdeps/libm-i387/e_acosl.S, sysdeps/libm-i387/s_ceill.S,
sysdeps/libm-i387/s_copysignl.S, sysdeps/libm-i387/s_finitel.S,
sysdeps/libm-i387/s_floorl.S, sysdeps/libm-i387/s_isinfl.c,
sysdeps/libm-i387/s_isnanl.c, sysdeps/libm-i387/s_nextafterl.c,
sysdeps/libm-i387/s_rintl.S, sysdeps/libm-i387/s_significandl.S:
New i387 specific math functions implementing `long double'
versions.
* sysdeps/libm-ieee754/s_ceill.c,
sysdeps/libm-ieee754/s_copysignl.c,
sysdeps/libm-ieee754/s_fabsl.c, sysdeps/libm-ieee754/s_finitel.c,
sysdeps/libm-ieee754/s_floorl.c, sysdeps/libm-ieee754/s_isinfl.c,
sysdeps/libm-ieee754/s_isnanl.c,
sysdeps/libm-ieee754/s_nextafterl.c,
sysdeps/libm-ieee754/s_rintl.c, sysdeps/libm-ieee754/s_scalbnl.c,
sysdeps/libm-ieee754/s_significandl.c: New generic `long double'
versions of libm functions.
* sysdeps/libm-i387/e_exp.S: Add a few comments to explain the
Intel FPU nonsense.
* sysdeps/libm-i387/s_ceil.S, sysdeps/libm-i387/s_ceilf.S,
sysdeps/libm-i387/s_floor.S, sysdeps/libm-i387/s_floorf.S: Correct
handling of local variables. The old version created a stack
frame but stored the values outside.
* sysdeps/libm-ieee754/s_isinf.c, sysdeps/libm-ieee754/s_isnan.c
[!NO_LONG_DOUBLE]: Define alias with `long double' versions name.
* login/pututline_r.c: Include sys/stat.h. Fix typos.
according to currently used locale for category LC_CTYPE by
inet_nsap_ntoa. Now in <arpa/inet.h>.
_IO_dup2 to contain complete parameter list.
1996-05-30 16:12:42 +00:00
|
|
|
ret
|
1996-12-30 01:47:46 +00:00
|
|
|
END (__floorl)
|
Thu May 30 11:24:05 1996 Roland McGrath <roland@delasyd.gnu.ai.mit.edu>
* po/header.pot: Replace with exact boilerplate pinard dictates.
* sysdeps/i386/strtok.S (Lillegal_argument): Remove this code to set
errno and the check that jumped to it.
* sysdeps/mach/hurd/Makefile (errnos.d): Use $(sed-remove-objpfx).
Thu May 30 03:21:57 1996 Ulrich Drepper <drepper@cygnus.com>
* FAQ: Document need of gperf program for developers.
* elf/elf.h: Fix typos in comments.
* libio/stdio.h [!__STRICT_ANSI__ || _POSIX_SOURCE]: Add
prototypes for `ctermid' and `cuserid'.
* locale/programs/locale.c: Switch to user selected locale
before printing variables.
* math/Makefile [$(long-double-fcts)==yes]: Define long-m-routines
and long-c-routines. Only if the `long double' data type is
available we need to compile the functions.
(libm-routines): Add $(long-m-routines).
(routines): Remove isinfl, isnanl. Use new file s_isinfl and
s_isnanl instead if `long double' is available.
* math/math.h: Include <mathcalls.h> again to define `long double'
functions.
* math/math_private.h: Define data types, prototypes and access
macros for `long double'.
* stdlib/stdlib.h: Add prototypes for `strtoll' and `strtoull'.
[GCC2 && OPTIMIZE]: Define strto{,u}ll as inline function which
calls __strto{,u}q_internal.
* stdlib/strfmon.c: Replace PTR by `void *'.
* stdlib/strtoq.c: Define strtoll as weak alias.
* stdlib/strtouq.c: Define strtoull as weak alias.
* string/tester.c: Correct `strsep' test.
* sysdeps/generic/strsep.c: Make compatible with BSD version.
Trailing characters of skip set are not skipped. In this case
empty tokens are returned.
* sysdeps/i386/isinfl.c, sysdeps/i386/isnanl.c,
sysdeps/ieee754/isinf.c, sysdeps/ieee754/isinfl.c,
sysdeps/ieee754/isnan.c, sysdeps/ieee754/isnanl.c: Removed. We
now use the versions part of libm.
* sysdeps/i386/strsep.S: Removed. Generic C version is of
similar speed.
* sysdeps/i386/strtok.S: Remove support for `strsep'.
* sysdeps/libm-i387/e_acosl.S, sysdeps/libm-i387/s_ceill.S,
sysdeps/libm-i387/s_copysignl.S, sysdeps/libm-i387/s_finitel.S,
sysdeps/libm-i387/s_floorl.S, sysdeps/libm-i387/s_isinfl.c,
sysdeps/libm-i387/s_isnanl.c, sysdeps/libm-i387/s_nextafterl.c,
sysdeps/libm-i387/s_rintl.S, sysdeps/libm-i387/s_significandl.S:
New i387 specific math functions implementing `long double'
versions.
* sysdeps/libm-ieee754/s_ceill.c,
sysdeps/libm-ieee754/s_copysignl.c,
sysdeps/libm-ieee754/s_fabsl.c, sysdeps/libm-ieee754/s_finitel.c,
sysdeps/libm-ieee754/s_floorl.c, sysdeps/libm-ieee754/s_isinfl.c,
sysdeps/libm-ieee754/s_isnanl.c,
sysdeps/libm-ieee754/s_nextafterl.c,
sysdeps/libm-ieee754/s_rintl.c, sysdeps/libm-ieee754/s_scalbnl.c,
sysdeps/libm-ieee754/s_significandl.c: New generic `long double'
versions of libm functions.
* sysdeps/libm-i387/e_exp.S: Add a few comments to explain the
Intel FPU nonsense.
* sysdeps/libm-i387/s_ceil.S, sysdeps/libm-i387/s_ceilf.S,
sysdeps/libm-i387/s_floor.S, sysdeps/libm-i387/s_floorf.S: Correct
handling of local variables. The old version created a stack
frame but stored the values outside.
* sysdeps/libm-ieee754/s_isinf.c, sysdeps/libm-ieee754/s_isnan.c
[!NO_LONG_DOUBLE]: Define alias with `long double' versions name.
* login/pututline_r.c: Include sys/stat.h. Fix typos.
according to currently used locale for category LC_CTYPE by
inet_nsap_ntoa. Now in <arpa/inet.h>.
_IO_dup2 to contain complete parameter list.
1996-05-30 16:12:42 +00:00
|
|
|
weak_alias (__floorl, floorl)
|
