glibc/math/w_powf.c

61 lines
1.7 KiB
C
Raw Normal View History

/* Copyright (C) 2011-2016 Free Software Foundation, Inc.
This file is part of the GNU C Library.
Contributed by Ulrich Drepper <drepper@gmail.com>, 2011.
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
<http://www.gnu.org/licenses/>. */
2005-12-14 15:06:39 +00:00
#include <math.h>
#include <math_private.h>
/* wrapper powf */
float
__powf (float x, float y)
{
float z = __ieee754_powf (x, y);
if (__glibc_unlikely (!isfinite (z)))
{
if (_LIB_VERSION != _IEEE_)
{
Fix pow (qNaN, 0) result with -lieee (bug 20919), remove dead parts of wrappers. The dbl-64 implementation of __ieee754_pow returns a NaN for pow (qNaN, 0) when it should return 1. Normally this is covered up by the wrappers ending up calling __kernel_standard which fixes up the result for this case, but for -lieee the wrappers are bypassed and the bad result gets through as a return value. Now, the wrappers fixing this are dealing with variant error handling that wants a result of NaN for pow (qNaN, 0), and only ever call __kernel_standard for this case if NaN resulted from __ieee754_pow. This leads to a question of whether the dbl-64 code might be deliberately returning NaN in order to use those code paths. However, I can find no sign that this is deliberate. If it were deliberate one would expect other implementations to do the same, and would expect the return of NaN to be very old, but it appears it came in by accident when the present e_pow.c implementation replaced an fdlibm implementation in 2001. So it appears to be unintended that this path through the pow wrapper could be used at all. So this patch fixes the implementation to return 1 in this case as expected. This is consistent with all the other implementations. The relevant path through the wrappers is now unreachable, so is removed (which is the main motivation of this patch: to avoid that path becoming accidentally reachable when implementing TS 18661-1 semantics that pow (sNaN, 0) should return qNaN with "invalid" raised). Another path that would require __ieee754_pow (0, 0) to return 0 is also unreachable (as all implementations return 1, in accordance with C99 semantics), so is removed as well. Note: we don't have anything set up to test -lieee, which in any case is obsolescent (at some point we should remove the ability for new programs to access _LIB_VERSION or define matherr and have it called by glibc). So testing will be implicit through sNaN tests added when making sNaN inputs work correctly for pow functions. Tested for x86_64 and x86. [BZ #20919] * sysdeps/ieee754/dbl-64/e_pow.c (__ieee754_pow): Do not return NaN first argument when raised to power 0. * math/w_pow.c (__pow): Do not check for NaN or zero results from raising to power zero. * math/w_powf.c (__powf): Likewise. * math/w_powl.c (__powl): Likewise. * sysdeps/ieee754/k_standard.c (__kernel_standard): Do not handle pow (0, 0) or pow (NaN, 0).
2016-12-02 22:50:46 +00:00
if (isfinite (x) && isfinite (y))
{
if (isnan (z))
/* pow neg**non-int */
return __kernel_standard_f (x, y, 124);
else if (x == 0.0f && y < 0.0f)
{
if (signbit (x) && signbit (z))
/* pow(-0.0,negative) */
return __kernel_standard_f (x, y, 123);
else
/* pow(+0.0,negative) */
return __kernel_standard_f (x, y, 143);
}
Update. 1997-03-16 18:43 Ulrich Drepper <drepper@cygnus.com> * manual/filesys.texi: Add documentation for scandir and alphasort. * math/math.c (fpclassify): Correct stupid typos. * math/libm-test.c: New file. libm test suite by Andreas Jaeger. * nss/nss_files/files-hosts.c: Add gethostbyname2 imlementation. * posix/Makefile (routines): Add bsd-getpgrp. * posix/bsd-getpgrp.c: New file. * posix/unistd.h [__FAVOR_BSD]: Define macro getpgrp which maps calls to __bsd_getpgrp. * sysdeps/generic/getpgrp.c: De-ANSI-declfy. * sysdeps/i386/huge_val.h: New file. ix87 specific infinity values. * sysdeps/m68k/huge_val.h: New file. m68k specific infinity values. * sysdeps/generic/huge_val.h: Remove definition of long double definition. Make it the same as the double definition. * sysdeps/libm-i387/e_acos.S: Fix bug in FPU stack handling. * sysdeps/libm-i387/e_acosf.S: Likewise. * sysdeps/libm-i387/e_acosl.S: Likewise. * sysdeps/libm-i387/e_asin.S: Likewise. * sysdeps/libm-i387/e_asinf.S: Likewise. * sysdeps/libm-i387/e_asinl.S: Likewise. * sysdeps/libm-i387/e_exp.S: Likewise. * sysdeps/libm-i387/e_expf.S: Likewise. * sysdeps/libm-i387/e_expl.S: Likewise. * sysdeps/libm-i387/e_scalbn.S: Likewise. * sysdeps/libm-i387/e_scalbnf.S: Likewise. * sysdeps/libm-i387/e_scalbnl.S: Likewise. * sysdeps/libm-i387/e_log.S: Optimize branch code. * sysdeps/libm-i387/e_logf.S: Likewise. * sysdeps/libm-i387/e_logl.S: Likewise. * sysdeps/libm-i387/e_log10.S: Likewise. * sysdeps/libm-i387/e_log10f.S: Likewise. * sysdeps/libm-i387/e_log10l.S: Likewise. * sysdeps/libm-i387/e_pow.S: Major rewrite to handle special cases. * sysdeps/libm-i387/e_powf.S: Likewise. * sysdeps/libm-i387/e_powl.S: Likewise. * sysdeps/libm-i387/e_expm1.S: Change return value for -inf argument to -1.0. * sysdeps/libm-i387/e_expm1f.S: Likewise. * sysdeps/libm-i387/e_expm1l.S: Likewise. * sysdeps/libm-i387/e_isinfl.c: Return -1 for -inf. * sysdeps/libm-i387/e_logbl.S: Correct return value. Discard first stack element after fxtract. * sysdeps/libm-ieee754/e_atan2l.c: New file. `long double' implementation for atan2 function. * sysdeps/libm-ieee754/k_standard.c: Return NAN for libm not in _SVID_ mode when acos, asin, atan2, log, log10 is called with argument out of range. Add new error case for pow(+0,neg). * sysdeps/libm-ieee754/s_fpclassifyf.c: Correct recognition of NaN and +-inf. * sysdeps/libm-ieee754/s_fpclassifyl.c: Mask out explicit leading digit in stupid 80 bit formats. * sysdeps/libm-ieee754/s_isinf.c: Rewrite to return -1 for -inf. * sysdeps/libm-ieee754/s_isinff.c: Likewise. * sysdeps/libm-ieee754/s_isinfl.c: Likewise. * sysdeps/libm-ieee754/s_scalbnl.c (huge, tiny): Adapt values for long double type. * sysdeps/libm-ieee754/w_atan2.c: Do not raise exception expect when in SVID mode. * sysdeps/libm-ieee754/w_atan2f.c: Likewise. * sysdeps/libm-ieee754/w_atan2l.c: Likewise. * sysdeps/libm-ieee754/w_pow.c: Distinguish error cases for x is +0 or -0. * sysdeps/posix/isfdtype.c: Add cast to prevent warning. * sysdeps/stub/fcntlbits.h: Update copyright. * sysdeps/unix/bsd/fcntlbits.h: Likewise. * sysdeps/unix/bsd/bsd4.4/fcntlbits.h: Likewise. * sysdeps/unix/bsd/sun/sunos4/fcntlbits.h: Likewise. * sysdeps/unix/bsd/ultrix4/fcntlbits.h: Likewise. * sysdeps/unix/common/fcntlbits.h: Likewise. * sysdeps/unix/sysv/fcntlbits.h: Likewise. Define O_FSYNC as alias of O_SYNC. Add BSD compatibility macros FAPPEND, FFSYNC, FNONBLOCK, and FNDELAY. * sysdeps/unix/sysv/irix4/fcntlbits.h: Likewise. * sysdeps/unix/readdir_r.c: Don't copy whole `struct dirent' record, only reclen bytes. * sysdeps/unix/sysv/linux/fcntlbits.h [__USE_GNU]: Add O_READ, O_WRITE and O_NORW. * sysdeps/unix/sysv/linux/alpha/fcntlbits.h: Likewise. * sysdeps/unix/sysv/linux/init-first.h: Add copyright. * sysdeps/unix/sysv/linux/fxstat.c: New file. Rewrite kernel-level struct stat to user-level form. * sysdeps/unix/sysv/linux/lxstat: New file. * sysdeps/unix/sysv/linux/xstat: New file. * sysdeps/unix/sysv/linux/kernel_stat.h: Define struct stat used in kernel. * sysdeps/unix/sysv/linux/statbuf.h (struct stat): Change definition to use prescribed types for elements. (_STAT_VER): Change to value 3. * sysdeps/unix/sysv/linux/alph/statbuf.h: Likewise. * sysdeps/unix/sysv/linux/Dist: Add kernel_stat.h. * sysdeps/unix/sysv/linux/alpha/Dist: Likewise. * time/Makefile: Correct dependencies for test-tz. 1997-03-16 14:59 Philip Blundell <phil@london.uk.eu.org> * resolv/netdb.h: Add prototypes for gai_strerror and getnameinfo (needed for IPv6 basic sockets API). 1997-03-16 15:02 a sun <asun@zoology.washington.edu> * sysdeps/unix/sysv/linux/net/if_ppp.h: Don't use incompatible kernel header. * sysdeps/unix/sysv/linux/net/ppp_defs.h: Likewise. 1997-03-14 17:15 Ulrich Drepper <drepper@cygnus.com> * db/hash/hash_bigkey.c (__big_delete): Don't call __free_ovflpage without testing for last_bfp to be NULL. Reported by fabsoft@fabserver1.zarm.uni-bremen.de. 1997-03-13 11:42 Jim Meyering <meyering@asic.sc.ti.com> * time/mktime.c (TIME_T_MIN): Work around a bug in Cray C 5.0.3.0. 1997-03-14 04:00 Kurt Garloff <garloff@kg1.ping.de> * sysdeps/unix/sysv/linux/fcntlbits.h (O_FSYNC): Make alias for O_SYNC. (FASYNC): Move to __USE_BSD section. Create new macro O_ASYNC. 1997-03-14 02:50 Ulrich Drepper <drepper@cygnus.com> * nis/nss_nis/nis-hosts.c (_nss_nis_gethostbyname2_r): New functions. Compare result for correct address type. (_nss_nis_gethostbyname_r): Use _nss_nis_gethostbyname2_r. Reported by Mirko Streckenbach <mirko@marian.hil.de>. 1997-02-17 01:40 Zlatko Calusic <zcalusic@srce.hr> * time/strptime.c (recursive): Return rp to caller. (strptime_internal): First check for long names, then abbreviated (month & weekday). 1997-03-10 19:44 Andreas Schwab <schwab@issan.informatik.uni-dortmund.de> * Makeconfig: Remove useless definitions of ASFLAGS-%. * config.make.in (ASFLAGS-.so): Remove. * configure.in: Don't substitute ASFLAGS_SO. * sysdeps/sparc/configure.in: Remove file. * sysdeps/sparc/Makefile (ASFLAGS-.so): Define. 1997-03-11 17:00 Andreas Schwab <schwab@issan.informatik.uni-dortmund.de> * time/strptime.c (strptime_internal) [case 'Y']: Always subtract 1900 from year, regardless of century. 1997-03-12 05:43 Ulrich Drepper <drepper@cygnus.com> * stdlib/strtod.c (_tens_in_limb) [BITS_PER_MP_LIMB > 32]: Make all numbers unsigned to make buggy gccs happy. Patch by Bryan W. Headley <bheadley@interaccess.com>. * sysdeps/unix/sysv/linux/netinet/ip.h: Add backward-compatibility definitions. Patch by a sun <asun@zoology.washington.edu>. Pretty print header. * Makerules (build-shlib): Also create symlink if library is versioned. based on a patch by H.J. Lu <hjl@gnu.ai.mit.edu>. Remove special rule to libc.so symlink. 1997-03-11 20:16 Andreas Jaeger <aj@arthur.pfalz.de> * manual/math.texi (Domain and Range Errors): Change descriptions according to recent changes for ISO C 9X. 1997-03-11 22:39 Ulrich Drepper <drepper@cygnus.com> * sysdeps/libm-ieee754/k_standard.c (__kernel_standard): Correct return values for acos, asin, and atan2. Reported by Andreas Jaeger <aj@arthur.pfalz.de>. 1997-03-10 18:16 Thorsten Kukuk <kukuk@vt.uni-paderborn.de> * ypclnt.c (__yp_bind): Fix possible buffer overflow. 1997-03-10 18:06 Bernd Schmidt <crux@Pool.Informatik.RWTH-Aachen.DE> * dirent/alphasort.c (alphasort): Interpret arguments as pointers to pointers to directory entries so that alphasort really can be used as argument for scandir. 1997-03-09 23:33 Andreas Jaeger <aj@arthur.pfalz.de> * string/strdup.c: Declare memcpy if !(_LIBC || STDC_HEADERS) instead of strcpy. 1997-03-10 03:34 Ulrich Drepper <drepper@cygnus.com> * catgets/catgets.c (catopen): Always add NLSPATH to search path for catalogs, not only if the envvar NLSPATH is not available. Reported by Andries.Brouwer@cwi.nl. 1997-03-10 02:46 Ulrich Drepper <drepper@cygnus.com> * Makeconfig (localtime-file): Don't define using installation directory. (inst_localtime-file): New variable. * time/Makefile (installed-localtime-file): Use inst_localtime-file. Reported by Edward Seidl <seidl@janed.com>. 1997-03-10 02:31 H.J. Lu <hjl@gnu.ai.mit.edu> * time/Makefile: Add source files to dependencies for test data. 1997-03-09 22:53 Thorsten Kukuk <kukuk@weber.uni-paderborn.de> * nis/nss_nis/nis-ethers.c: Don't ignore return value of yp_all. * nis/nss_nis/nis-proto.c: Likewise. * nis/nss_nis/nis-rpc.c: Likewise. * nis/nss_nis/nis-service.c: Likewise. 1997-03-08 14:37 Miguel de Icaza <miguel@nuclecu.unam.mx> * sysdeps/sparc/dl-machine.h (elf_machine_rela): Upgrade to versioning; Added missing R_SPARC_WDISP30 handling. (RTLD_START): Implement it. * sysdeps/unix/sysv/linux/sparc/brk.c: Fix. * sysdeps/unix/sysv/linux/sparc/start.c: Startup code for Linux/SPARC. 1997-03-02 18:06 Miguel de Icaza <miguel@nuclecu.unam.mx> * sysdeps/sparc/dl-machine.h (RTLD_START): Make arg as expected by the dynamic linker instead of having a new conditional define. Thanks to Richard Henderson for pointing this out. * elf/rtld.c: Remove usage of ELF_ADJUST_ARG. 1997-03-20 20:44 Thomas Bushnell, n/BSG <thomas@gnu.ai.mit.edu> * sysdeps/mach/hurd/euidaccess.c: Define as __euidaccess and make euidaccess weak alias. 1997-03-07 10:30 Thomas Bushnell, n/BSG <thomas@gnu.ai.mit.edu> * stdio-common/printf_fphex.c (MIN): New macro. * sysdeps/generic/netinet/in.h: Include <sys/types.h>. * sysdeps/generic/sys/mman.h (msync): Mention third arg. * sysdeps/generic/netinet/in.h: Add definitions for IPv6 basic API. (See change by Philip Blundell on Feb 16, 1997.) 1997-03-05 10:40 Thomas Bushnell, n/BSG <thomas@gnu.ai.mit.edu> * hurd/hurd.h (vpprintf): Include <stdarg.h>. New declaration. * hurd/set-host.c (_hurd_set_host_config): Cast second arg to __file_name_split. * mach/mach_error.c (mach_error_string_int): Give full prototype. * mach/errstring.c (mach_error_string_int): Likewise. * mach/error_compat.c (__mach_error_map_compat): Likewise. * hurd/vpprintf.c (pwrite, vpprintf): Likewise. * stdio/vasprintf.c (vasprintf): Likewise. * mach/mach/mach_traps.h: Include <mach/kern_return.h>. * mach/spin-solid.c: Include <mach/mach_traps.h>. * mach/spin-solid.c (__spin_lock_solid): Provide arg to swtch_pri. * mach/mach_init.c: Include <mach/mig_support.h>. * mach/mach_error.h (mach_error_string, mach_error, mach_error_type): Always provide prototypes. * mach/mach/error.h (mach_error_fn_t): Comment out declaration; it appears to be entirely unused dead code. * stdio/stdio.h (freopen): Fix spelling error. 1997-03-02 13:38 Miles Bader <miles@gnu.ai.mit.edu> * string/argz.h (__need_error_t): New macro, before including <errno.h> [!__const] (__const): New macro. [!__error_t_defined] (error_t): New typedef. * sysdeps/generic/socketbits.h: Add PF_FILE as synonym for PF_LOCAL * sysdeps/unix/sysv/linux/socketbits.h: Likewise.
1997-03-16 20:28:07 +00:00
else
/* pow overflow */
return __kernel_standard_f (x, y, 121);
Update. 1998-09-14 11:26 Ulrich Drepper <drepper@cygnus.com> * wcsmbs/wcsmbs-tst1.c: Include stdlib.h. 1998-09-14 Thorsten Kukuk <kukuk@vt.uni-paderborn.de> * libc-work/nis/nss_nisplus/nisplus-service.c (_nss_nisplus_parse_servent): Convert port in network byte order. 1998-09-14 07:53 -0400 Zack Weinberg <zack@rabi.phys.columbia.edu> * stdlib/stdlib.h: Add a switch, __need_malloc_and_calloc, to provide only malloc and calloc. * include/stdlib.h: Support the above. * string/bits/string2.h: Use __need_malloc_and_calloc when including stdlib.h. (__string2_1bptr_p): Avoid -Wbad-function-cast warnings. * iconvdata/iso-2022-jp.c: Include <stdlib.h>. * iconvdata/iso646.c: Include <stdlib.h>. 1998-09-14 07:51 -0400 Zack Weinberg <zack@rabi.phys.columbia.edu> * sunrpc/rpc_cout.c: Add braces around ambiguous else. * sysdeps/libm-ieee754/w_pow.c: Likewise. * sysdeps/libm-ieee754/w_powf.c: Likewise. * sysdeps/libm-ieee754/w_powl.c: Likewise. 1998-09-14 07:57 -0400 Zack Weinberg <zack@rabi.phys.columbia.edu> * stdio-common/stdio_lim.h.in: New file. All parameters are adjustable at build time. * Rules: Add a rule to build bits/stdio_lim.h from stdio-common/stdio_lim.h.in. (It has to be in Rules so that all subdirectories know how to make the file. It can't be in Makerules because then it gets built at top level and the dependencies are wrong.) * stdio-common/Makefile (distribute): Add stdio_lim.h.in. * sysdeps/unix/sysv/linux/Makefile: Delete rules to make stdio_lim.h. * sysdeps/unix/sysv/linux/stdio_lim.h.in: Removed. * sysdeps/unix/sysv/linux/Dist: Take out stdio_lim.h.in. * sysdeps/posix/Makefile: Removed. * sysdeps/posix/mk-stdiolim.c: Removed. * sysdeps/generic/bits/stdio_lim.h: Removed. 1998-09-12 Andreas Schwab <schwab@issan.informatik.uni-dortmund.de> * argp/argp-help.c: Fixup indentation. * nss/nss_files/files-alias.c: Quiet -Wparentheses warning. * resolv/nss_dns/dns-network.c: Likewise. * resolv/res_send.c: Likewise. * rt/aio_cancel.c: Likewise. * rt/aio_misc.c: Likewise. 1998-09-12 Andreas Schwab <schwab@issan.informatik.uni-dortmund.de> * Makefile (install-symbolic-link): Remove the link list file at last. * Makerules (install-clean-symbolic-link-list): Removed. 1998-09-12 Andreas Schwab <schwab@issan.informatik.uni-dortmund.de> * timezone/tst-timezone.c: Print time in UTC to get consistent output. 1998-09-12 Andreas Schwab <schwab@issan.informatik.uni-dortmund.de> * elf/rtld.c (_dl_rpath): Removed. * elf/dl-support.c (_dl_path): Removed. * elf/Makefile ($(objpfx)ld.so): Don't pass -rpath. (CFLAGS-dl-support.c): Removed. * elf/dl-load.c (env_path_list): Renamed from fake_path_list. All uses changed. (_dl_init_paths): Always set env_path_list from LD_LIBRARY_PATH, instead of appending it to the main map's rpath info. (_dl_map_object): Consistently use LD_LIBRARY_PATH after all DT_RPATHs. This makes it effective again. (decompose_rpath): Remove second parameter, callers changed. (fillin_rpath): Allocate enough space in curwd. (expand_dynamic_string_token): Cope with get_origin returning -1. 1998-09-14 Thorsten Kukuk <kukuk@vt.uni-paderborn.de> * libc-work/nis/nss_compat/compat-pwd.c: Add support for passwd.adjunct. * nis/nss_nis/nis-pwd.c (_nss_nis_getpwent_r): Correct test for invalid password. 1998-09-13 18:06 Ulrich Drepper <drepper@cygnus.com> * locale/programs/locale.c: Update dates.
1998-09-14 11:43:26 +00:00
}
}
}
Fix pow (qNaN, 0) result with -lieee (bug 20919), remove dead parts of wrappers. The dbl-64 implementation of __ieee754_pow returns a NaN for pow (qNaN, 0) when it should return 1. Normally this is covered up by the wrappers ending up calling __kernel_standard which fixes up the result for this case, but for -lieee the wrappers are bypassed and the bad result gets through as a return value. Now, the wrappers fixing this are dealing with variant error handling that wants a result of NaN for pow (qNaN, 0), and only ever call __kernel_standard for this case if NaN resulted from __ieee754_pow. This leads to a question of whether the dbl-64 code might be deliberately returning NaN in order to use those code paths. However, I can find no sign that this is deliberate. If it were deliberate one would expect other implementations to do the same, and would expect the return of NaN to be very old, but it appears it came in by accident when the present e_pow.c implementation replaced an fdlibm implementation in 2001. So it appears to be unintended that this path through the pow wrapper could be used at all. So this patch fixes the implementation to return 1 in this case as expected. This is consistent with all the other implementations. The relevant path through the wrappers is now unreachable, so is removed (which is the main motivation of this patch: to avoid that path becoming accidentally reachable when implementing TS 18661-1 semantics that pow (sNaN, 0) should return qNaN with "invalid" raised). Another path that would require __ieee754_pow (0, 0) to return 0 is also unreachable (as all implementations return 1, in accordance with C99 semantics), so is removed as well. Note: we don't have anything set up to test -lieee, which in any case is obsolescent (at some point we should remove the ability for new programs to access _LIB_VERSION or define matherr and have it called by glibc). So testing will be implicit through sNaN tests added when making sNaN inputs work correctly for pow functions. Tested for x86_64 and x86. [BZ #20919] * sysdeps/ieee754/dbl-64/e_pow.c (__ieee754_pow): Do not return NaN first argument when raised to power 0. * math/w_pow.c (__pow): Do not check for NaN or zero results from raising to power zero. * math/w_powf.c (__powf): Likewise. * math/w_powl.c (__powl): Likewise. * sysdeps/ieee754/k_standard.c (__kernel_standard): Do not handle pow (0, 0) or pow (NaN, 0).
2016-12-02 22:50:46 +00:00
else if (__builtin_expect (z == 0.0f, 0)
&& isfinite (x) && x != 0 && isfinite (y)
&& _LIB_VERSION != _IEEE_)
Fix pow (qNaN, 0) result with -lieee (bug 20919), remove dead parts of wrappers. The dbl-64 implementation of __ieee754_pow returns a NaN for pow (qNaN, 0) when it should return 1. Normally this is covered up by the wrappers ending up calling __kernel_standard which fixes up the result for this case, but for -lieee the wrappers are bypassed and the bad result gets through as a return value. Now, the wrappers fixing this are dealing with variant error handling that wants a result of NaN for pow (qNaN, 0), and only ever call __kernel_standard for this case if NaN resulted from __ieee754_pow. This leads to a question of whether the dbl-64 code might be deliberately returning NaN in order to use those code paths. However, I can find no sign that this is deliberate. If it were deliberate one would expect other implementations to do the same, and would expect the return of NaN to be very old, but it appears it came in by accident when the present e_pow.c implementation replaced an fdlibm implementation in 2001. So it appears to be unintended that this path through the pow wrapper could be used at all. So this patch fixes the implementation to return 1 in this case as expected. This is consistent with all the other implementations. The relevant path through the wrappers is now unreachable, so is removed (which is the main motivation of this patch: to avoid that path becoming accidentally reachable when implementing TS 18661-1 semantics that pow (sNaN, 0) should return qNaN with "invalid" raised). Another path that would require __ieee754_pow (0, 0) to return 0 is also unreachable (as all implementations return 1, in accordance with C99 semantics), so is removed as well. Note: we don't have anything set up to test -lieee, which in any case is obsolescent (at some point we should remove the ability for new programs to access _LIB_VERSION or define matherr and have it called by glibc). So testing will be implicit through sNaN tests added when making sNaN inputs work correctly for pow functions. Tested for x86_64 and x86. [BZ #20919] * sysdeps/ieee754/dbl-64/e_pow.c (__ieee754_pow): Do not return NaN first argument when raised to power 0. * math/w_pow.c (__pow): Do not check for NaN or zero results from raising to power zero. * math/w_powf.c (__powf): Likewise. * math/w_powl.c (__powl): Likewise. * sysdeps/ieee754/k_standard.c (__kernel_standard): Do not handle pow (0, 0) or pow (NaN, 0).
2016-12-02 22:50:46 +00:00
/* pow underflow */
return __kernel_standard_f (x, y, 122);
return z;
}
weak_alias (__powf, powf)