mirror of
https://sourceware.org/git/glibc.git
synced 2024-11-21 20:40:05 +00:00
5c22fd25c1
The CORE-MATH implementation is correctly rounded (for any rounding mode)
and shows better performance compared to the generic exp2m1f.
The code was adapted to glibc style and to use the definition of
math_config.h (to handle errno, overflow, and underflow). The
only change is to handle FLT_MAX_EXP for FE_DOWNWARD or FE_TOWARDZERO.
The benchmark inputs are based on exp2f ones.
Benchtest on x64_64 (Ryzen 9 5900X, gcc 14.2.1), aarch64 (Neoverse-N1,
gcc 13.3.1), and powerpc (POWER10, gcc 13.2.1):
Latency master patched improvement
x86_64 40.6042 48.7104 -19.96%
x86_64v2 40.7506 35.9032 11.90%
x86_64v3 35.2301 31.7956 9.75%
i686 102.094 94.6657 7.28%
aarch64 18.2704 15.1387 17.14%
power10 11.9444 8.2402 31.01%
reciprocal-throughput master patched improvement
x86_64 20.8683 16.1428 22.64%
x86_64v2 19.5076 10.4474 46.44%
x86_64v3 19.2106 10.4014 45.86%
i686 56.4054 59.3004 -5.13%
aarch64 12.0781 7.3953 38.77%
power10 6.5306 5.9388 9.06%
The generic implementation calls __ieee754_exp2f and x86_64 provides
an optimized ifunc version (built with -mfma -mavx2, not correctly
rounded). This explains the performance difference for x86_64.
Same for i686, where the ABI provides an optimized __ieee754_exp2f
version built with '-msse2 -mfpmath=sse'. When built wth same
flags, the new algorithm shows a better performance:
master patched improvement
latency 102.094 91.2823 10.59%
reciprocal-throughput 56.4054 52.7984 6.39%
Signed-off-by: Alexei Sibidanov <sibid@uvic.ca>
Signed-off-by: Paul Zimmermann <Paul.Zimmermann@inria.fr>
Signed-off-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
Reviewed-by: DJ Delorie <dj@redhat.com>
251 lines
7.6 KiB
Plaintext
251 lines
7.6 KiB
Plaintext
# Files shared with other projects. Pass a file path to the
|
|
# get_glibc_shared_code() function in the python library
|
|
# scripts/glibc_shared_code.py to get a dict object with this information. See
|
|
# the library sources for more information.
|
|
|
|
# The headers on most of these files indicate that glibc is the canonical
|
|
# source for these files, although in many cases there seem to be useful
|
|
# changes in the gnulib versions that could be merged back in. Not all gnulib
|
|
# files contain such a header and it is not always consistent in its format, so
|
|
# it would be useful to make sure that all gnulib files that are using glibc as
|
|
# upstream have a greppable header.
|
|
#
|
|
# These files are quite hard to find without a header to grep for and each file
|
|
# has to be compared manually so this list is likely incomplete or may contain
|
|
# errors.
|
|
gnulib:
|
|
argp/argp-ba.c
|
|
argp/argp-ba.c
|
|
argp/argp-eexst.c
|
|
argp/argp-fmtstream.c
|
|
argp/argp-fmtstream.h
|
|
argp/argp-fs-xinl.c
|
|
argp/argp-help.c
|
|
argp/argp-namefrob.h
|
|
argp/argp-parse.c
|
|
argp/argp-pv.c
|
|
argp/argp-pvh.c
|
|
argp/argp-xinl.c
|
|
argp/argp.h
|
|
dirent/alphasort.c
|
|
dirent/scandir.c
|
|
# Merged from gnulib 2021-09-21
|
|
include/intprops.h
|
|
# Merged from gnulib 2021-09-21
|
|
include/regex.h
|
|
locale/programs/3level.h
|
|
# Merged from gnulib 2014-6-23
|
|
malloc/obstack.c
|
|
# Merged from gnulib 2014-6-23
|
|
malloc/obstack.h
|
|
# Merged from gnulib 2014-07-10
|
|
misc/error.c
|
|
misc/error.h
|
|
misc/getpass.c
|
|
misc/mkdtemp.c
|
|
# Merged from gnulib 2021-09-21
|
|
misc/sys/cdefs.h
|
|
posix/fnmatch_loop.c
|
|
# Intended to be the same. Gnulib copy contains glibc changes.
|
|
posix/getopt.c
|
|
# Intended to be the same. Gnulib copy contains glibc changes.
|
|
posix/getopt1.c
|
|
# Intended to be the same. Gnulib copy contains glibc changes.
|
|
posix/getopt_int.h
|
|
posix/glob.c
|
|
# Merged from gnulib 2021-09-21
|
|
posix/regcomp.c
|
|
# Merged from gnulib 2021-09-21
|
|
posix/regex.c
|
|
# Merged from gnulib 2021-09-21
|
|
posix/regex.h
|
|
# Merged from gnulib 2021-09-21
|
|
posix/regex_internal.c
|
|
# Merged from gnulib 2021-09-21
|
|
posix/regex_internal.h
|
|
# Merged from gnulib 2021-09-21
|
|
posix/regexec.c
|
|
posix/spawn.c
|
|
posix/spawn_faction_addclose.c
|
|
posix/spawn_faction_adddup2.c
|
|
posix/spawn_faction_addopen.c
|
|
posix/spawn_faction_destroy.c
|
|
posix/spawn_faction_init.c
|
|
posix/spawn_int.h
|
|
posix/spawnattr_destroy.c
|
|
posix/spawnattr_getdefault.c
|
|
posix/spawnattr_getflags.c
|
|
posix/spawnattr_getpgroup.c
|
|
posix/spawnattr_getschedparam.c
|
|
posix/spawnattr_getschedpolicy.c
|
|
posix/spawnattr_getsigmask.c
|
|
posix/spawnattr_init.c
|
|
posix/spawnattr_setdefault.c
|
|
posix/spawnattr_setflags.c
|
|
posix/spawnattr_setpgroup.c
|
|
posix/spawnattr_setschedparam.c
|
|
posix/spawnattr_setschedpolicy.c
|
|
posix/spawnattr_setsigmask.c
|
|
posix/spawnp.c
|
|
stdlib/atoll.c
|
|
stdlib/getsubopt.c
|
|
stdlib/setenv.c
|
|
stdlib/strtoll.c
|
|
stdlib/strtoul.c
|
|
# Merged from gnulib 2014-6-26, needs merge back
|
|
string/memchr.c
|
|
string/memcmp.c
|
|
string/memmem.c
|
|
string/mempcpy.c
|
|
string/memrchr.c
|
|
string/rawmemchr.c
|
|
string/stpcpy.c
|
|
string/stpncpy.c
|
|
string/str-two-way.h
|
|
string/strcasestr.c
|
|
string/strcspn.c
|
|
string/strdup.c
|
|
string/strndup.c
|
|
string/strpbrk.c
|
|
string/strsignal.c
|
|
string/strstr.c
|
|
string/strtok_r.c
|
|
string/strverscmp.c
|
|
# Merged from gnulib 2024-04-08 (gnulib commit 3238349628)
|
|
stdio-common/tmpdir.c
|
|
stdio-common/tmpdir.h
|
|
sysdeps/generic/pty-private.h
|
|
sysdeps/generic/siglist.h
|
|
sysdeps/posix/euidaccess.c
|
|
sysdeps/posix/gai_strerror.c
|
|
sysdeps/posix/getcwd.c
|
|
sysdeps/posix/pwrite.c
|
|
sysdeps/posix/spawni.c
|
|
# Merged from gnulib 2024-04-08 (gnulib commit 3238349628)
|
|
sysdeps/posix/tempname.c
|
|
# Merged from gnulib 2014-6-27
|
|
time/mktime.c
|
|
time/mktime-internal.h
|
|
time/strptime.c
|
|
time/timegm.c
|
|
|
|
# The last merge was 2014-12-11 and merged gettext 0.19.3 into glibc with a
|
|
# patch submitted to the gettext mailing list for changes that could be merged
|
|
# back.
|
|
#
|
|
# This commit was omitted from the merge as it does not appear to be compatible
|
|
# with how glibc expects things to work:
|
|
#
|
|
# commit 279b57fc367251666f00e8e2b599b83703451afb
|
|
# Author: Bruno Haible <bruno@clisp.org>
|
|
# Date: Fri Jun 14 12:03:49 2002 +0000
|
|
#
|
|
# Make absolute pathnames inside $LANGUAGE work.
|
|
gettext:
|
|
intl/bindtextdom.c
|
|
intl/dcgettext.c
|
|
intl/dcigettext.c
|
|
intl/dcngettext.c
|
|
intl/dgettext.c
|
|
intl/dngettext.c
|
|
intl/explodename.c
|
|
intl/finddomain.c
|
|
intl/gettext.c
|
|
intl/gettextP.h
|
|
intl/gmo.h
|
|
intl/hash-string.c
|
|
intl/hash-string.h
|
|
intl/l10nflist.c
|
|
intl/loadinfo.h
|
|
intl/loadmsgcat.c
|
|
intl/locale.alias
|
|
intl/localealias.c
|
|
intl/ngettext.c
|
|
intl/plural-exp.c
|
|
intl/plural-exp.h
|
|
intl/plural.y
|
|
intl/textdomain.c
|
|
|
|
# The following files are bundled from upstream Linux 6.10 for FUSE testing
|
|
# support.
|
|
linux:
|
|
support/bundled/linux/COPYING
|
|
support/bundled/linux/LICENSES
|
|
support/bundled/linux/LICENSES/exceptions/Linux-syscall-note
|
|
support/bundled/linux/LICENSES/preferred/GPL-2.0
|
|
support/bundled/linux/include/uapi/linux/fuse.h
|
|
|
|
# The following files are shared with the upstream Unicode project and must be
|
|
# updated regularly to stay in sync with the upstream unicode releases.
|
|
#
|
|
# Merged from Unicode 16.0.0 release.
|
|
unicode:
|
|
localedata/unicode-gen/UnicodeData.txt
|
|
localedata/unicode-gen/unicode-license.txt
|
|
localedata/unicode-gen/DerivedCoreProperties.txt
|
|
localedata/unicode-gen/EastAsianWidth.txt
|
|
localedata/unicode-gen/HangulSyllableType.txt
|
|
|
|
# The following files are shared with the upstream tzcode project and must be
|
|
# updated regularly to stay in sync with the upstream releases.
|
|
#
|
|
# Currently synced to TZDB 2024b, announced and distributed here:
|
|
# https://github.com/eggert/tz/releases/tag/2024b
|
|
tzcode:
|
|
timezone/private.h
|
|
timezone/tzfile.h
|
|
timezone/tzselect.ksh
|
|
timezone/version
|
|
timezone/zdump.c
|
|
timezone/zic.c
|
|
|
|
# The following files are shared with the upstream tzdata project but is not
|
|
# synchronized regularly. The data files themselves are used only for testing
|
|
# purposes and their data is never used to generate any output. We synchronize
|
|
# them only to stay on top of newer data that might help with testing.
|
|
#
|
|
# Currently synced to tzcode 2009i, announced and distributed here:
|
|
# https://mm.icann.org/pipermail/tz/2009-June/040697.html
|
|
# https://data.iana.org/time-zones/releases/tzdata2009i.tar.gz
|
|
tzdata:
|
|
timezone/africa
|
|
timezone/antarctica
|
|
timezone/asia
|
|
timezone/australasia
|
|
timezone/europe
|
|
timezone/northamerica
|
|
timezone/southamerica
|
|
timezone/pacificnew
|
|
timezone/etcetera
|
|
timezone/factory
|
|
timezone/backward
|
|
timezone/systemv
|
|
timezone/solar87
|
|
timezone/solar88
|
|
timezone/solar89
|
|
timezone/iso3166.tab
|
|
timezone/zone.tab
|
|
timezone/leapseconds
|
|
# This is yearistype.sh in the parent project
|
|
timezone/yearistype
|
|
|
|
# The following files are shared with the CORE-MATH project
|
|
sysdeps/ieee754/flt-32/e_gammaf_r.c
|
|
(file src/binary32/tgamma/tgammaf.c in CORE-MATH)
|
|
Instructions to merge new versions:
|
|
- change the function name from cr_tgammaf to __ieee754_gammaf_r
|
|
- add "int *signgamp" as 2nd argument and add at the beginning:
|
|
if (signgamp != NULL) *signgamp = 1;
|
|
- remove the errno stuff (this is done by the wrapper)
|
|
- replace 0x1p127f * 0x1p127f by math_narrow_eval (x * 0x1p127f)
|
|
- replace 0x1p-127f * sgn[k&1] by math_narrow_eval (0x1p-127f * sgn[k&1])
|
|
- add libm_alias_finite (__ieee754_gammaf_r, __gammaf_r) at the end
|
|
sysdeps/ieee754/flt-32/s_exp10m1f.c
|
|
(file src/binary32/exp10m1/exp10m1f.c in CORE-MATH)
|
|
- The code was adapted to use glibc code style and internal
|
|
functions to handle errno, overflow, and underflow.
|
|
sysdeps/ieee754/flt-32/s_exp2m1f.c
|
|
(file src/binary32/exp2m1/exp2m1f.c in CORE-MATH)
|
|
- The code was adapted to use glibc code style and internal
|
|
functions to handle errno, overflow, and underflow.
|