This patch provides optimized versions of strlen and wcslen with the z13 vector
instructions.
The helper macro IFUNC_VX_IMPL is introduced and is used to register all
__<func>_c() and __<func>_vx() functions within __libc_ifunc_impl_list()
to the ifunc test framework.
ChangeLog:
* sysdeps/s390/multiarch/Makefile: New File.
* sysdeps/s390/multiarch/strlen-c.c: Likewise.
* sysdeps/s390/multiarch/strlen-vx.S: Likewise.
* sysdeps/s390/multiarch/strlen.c: Likewise.
* sysdeps/s390/multiarch/wcslen-c.c: Likewise.
* sysdeps/s390/multiarch/wcslen-vx.S: Likewise.
* sysdeps/s390/multiarch/wcslen.c: Likewise.
* string/strlen.c (STRLEN): Define and use macro.
* sysdeps/s390/multiarch/ifunc-impl-list.c
(IFUNC_VX_IMPL): New macro function.
(__libc_ifunc_impl_list): Add ifunc test for strlen, wcslen.
* benchtests/Makefile (wcsmbs-bench): New variable.
(string-bench-all): Added wcsmbs-bench.
* benchtests/bench-wcslen.c: New File.
I think the last clause of the conditional,
|| __n <= __bos (__dest)
may be backward. The code should call the runtime-checking function
if __n is not constant, or if __n is known to be LARGER than the size
of the destination.
This patches uses the default strcpy/stpcpy implementation for
POWER7/PPC64. This is faster in mostly inputs for benchtests
and for multiarch the implementation uses the POWER7 strlen and
memcpy.
* string/stpcpy.c (__stpcpy): Use STPCPY to redefine symbol name and
cleanup macro usage.
* string/strcpy.c (strcpt): Use STRCPY to redefine symbol name.
* sysdeps/powerpc/powerpc64/multiarch/stpcpy-power7.S: Remove file.
* sysdeps/powerpc/powerpc64/multiarch/stpcpy-ppc64.S: Likewise.
* sysdeps/powerpc/powerpc64/multiarch/strcpy-power7.S: Likewise.
* sysdeps/powerpc/powerpc64/multiarch/strcpy-ppc64.S: Likewise.
* sysdeps/powerpc/powerpc64/power7/stpcpy.S: Likewise.
* sysdeps/powerpc/powerpc64/power7/strcpy.S: Likewise.
* sysdeps/powerpc/powerpc64/power7/strcpy.c: Likewise.
* sysdeps/powerpc/powerpc64/stpcpy.S: Likewise.
* sysdeps/powerpc/powerpc64/strcpy.S: Likewise.
* sysdeps/powerpc/powerpc64/multiarch/stpcpy.c
[SHARED && IS_IN (libc)]: Include <string/strcpy.c>.
* sysdeps/powerpc/powerpc64/multiarch/stpcpy.c
[SHARED && IS_IN (libc)]: Include <string/stpcpy.c>.
* sysdeps/powerpc/powerpc64/multiarch/stpcpy-power7.c: New file.
* sysdeps/powerpc/powerpc64/multiarch/stpcpy-ppc64.c: Likewise.
* sysdeps/powerpc/powerpc64/multiarch/strcpy-power7.c: Likewise.
* sysdeps/powerpc/powerpc64/multiarch/strcpy-ppc64.c: Likewise.
* sysdeps/powerpc/powerpc64/power7/strcpy.c: Likewise.
These tests were skipped by the use-test-skeleton conversion done in
commit 29955b5d because they did not have an `int main (void)'
declaration. Instead their `main' functions were declared with arguments
(i.e. argc, argv) even though they didn't use them.
Remove these arguments and include the test skeleton in these tests.
on bench-strncpy is 1.9-2.1x faster on average. I tried several variations, and using a tailcall and
calling memset conditionally gave the best overall results.
Various code in glibc uses __strnlen instead of strnlen for namespace
reasons. However, __strnlen does not use libc_hidden_proto /
libc_hidden_def (as is normally done for any function defined and
called within the same library, whether or not exported from the
library and whatever namespace it is in), so the compiler does not
know that those calls are to a function within libc.
This patch uses libc_hidden_proto / libc_hidden_def with __strnlen.
On x86_64, it makes no difference to the installed stripped shared
libraries. On 32-bit x86, it causes __strnlen calls to go to the same
place as strnlen calls (the fallback strnlen implementation), rather
than through a PLT entry for the strnlen IFUNC; I'm not sure of the
logic behind when calls from within libc should use IFUNCs versus when
they should go direct to a particular function implementation, but
clearly it doesn't make sense for strnlen and __strnlen to be handled
differently in this regard.
Tested for x86_64 and x86 (testsuite, and comparison of installed
shared libraries as described above).
* string/strnlen.c [!STRNLEN] (__strnlen): Use libc_hidden_def.
* include/string.h (__strnlen): Use libc_hidden_proto.
* sysdeps/aarch64/strnlen.S (__strnlen): Use libc_hidden_def.
* sysdeps/i386/i686/multiarch/strnlen-c.c [SHARED]
(libc_hidden_def): Define __GI___strnlen as well as __GI_strnlen.
* sysdeps/powerpc/powerpc32/power4/multiarch/strnlen-power7.S
(libc_hidden_def): Undefine and redefine.
* sysdeps/powerpc/powerpc32/power4/multiarch/strnlen-ppc32.c
[SHARED] (libc_hidden_def): Define __GI___strnlen as well as
__GI_strnlen.
* sysdeps/powerpc/powerpc32/power7/strnlen.S (__strnlen): Use
libc_hidden_def.
* sysdeps/tile/tilegx/strnlen.c (__strnlen): Likewise.
pathconf (sysdeps/unix/sysv/linux/pathconf.c) uses basename. But
pathconf is in POSIX back to 1990 while basename is only reserved with
external linkage in those standards including XPG functions. This
patch fixes this namespace issue in the usual way, renaming basename
to __basename and making it into a weak alias.
Tested for x86_64 and x86 (testsuite, and that disassembly of
installed shared libraries is unchanged by the patch).
[BZ #18444]
* string/basename.c (basename): Rename to __basename and define as
weak alias of __basename. Use libc_hidden_weak.
* include/string.h (__basename): Declare. Use libc_hidden_proto.
* sysdeps/unix/sysv/linux/pathconf.c (distinguish_extX): Call
__basename instead of basename.
* conform/Makefile (test-xfail-POSIX2008/unistd.h/linknamespace):
Remove variable.
(test-xfail-XOPEN2K8/unistd.h/linknamespace): Likewise.
[BZ #18206]
* wcsmbs/wcsncmp.c (wcsncmp): Compare as wchar_t, not wint_t.
Use signed comparision instead of substraction to avoid
overflow bug.
* localedata/tests-mbwc/tst_wcsncmp.c (tst_wcsncmp):
Take the sign of ret.
* localedata/tests-mbwc/dat_wcsncmp.c (tst_wcsncmp_loc):
Do not expect precise return values. Only the sign matters.
* wcsmbs/Makefile (strop-tests): Add wcsncmp.
* wcsmbs/test-wcsncmp.c: New File.
* string/test-strncmp.c: Add wcsncmp support.
[Modified from the original email by Siddhesh Poyarekar]
This patch solves bug #16009 by implementing an additional path in
strxfrm that does not depend on caching the weight and rule indices.
In detail the following changed:
* The old main loop was factored out of strxfrm_l into the function
do_xfrm_cached to be able to alternativly use the non-caching version
do_xfrm.
* strxfrm_l allocates a a fixed size array on the stack. If this is not
sufficiant to store the weight and rule indices, the non-caching path is
taken. As the cache size is not dependent on the input there can be no
problems with integer overflows or stack allocations greater than
__MAX_ALLOCA_CUTOFF. Note that malloc-ing is not possible because the
definition of strxfrm does not allow an oom errorhandling.
* The uncached path determines the weight and rule index for every char
and for every pass again.
* Passing all the locale data array by array resulted in very long
parameter lists, so I introduced a structure that holds them.
* Checking for zero src string has been moved a bit upwards, it is
before the locale data initialization now.
* To verify that the non-caching path works correct I added a test run
to localedata/sort-test.sh & localedata/xfrm-test.c where all strings
are patched up with spaces so that they are too large for the caching path.
gcc now warns when the arguments to memset may have been accidentally
transposed (i.e. length set to zero instead of the byte), so we don't
need that bit of the code in glibc headers anymore.
Tested on x86_64. Coe generated by gcc 4.8 is identical with or
without the patch. I also tested gcc master, which does not result in
any new failures. It does fail quite a few FORTIFY_SOURCE tests, but
those failures are not due to this patch.
TEST_IFUNC is only tested in two headers, bench-string.h and
test-string.h, after it gets defined by those headers, and it never
gets undefined.
Thus no defines of TEST_IFUNC are needed, and the *-ifunc.c tests that
just define TEST_IFUNC and include other tests are also redundant, as
is the code to remove $(tests-ifunc) and $(xtests-ifunc) conditionally
from tests and xtests. This patch removes the useless defines and
tests of TEST_IFUNC and the associated useless tests and makefile
code. It thereby fixes a series of warnings
"../string/test-string.h:21:0: warning: "TEST_IFUNC" redefined" where
test-string.h defines TEST_IFUNC to empty, other files define it to 1
and this produces warnings.
Tested for x86_64.
* debug/test-stpcpy_chk-ifunc.c: Remove file.
* debug/test-strcpy_chk-ifunc.c: Likewise.
* wcsmbs/test-wcschr-ifunc.c: Likewise.
* wcsmbs/test-wcscmp-ifunc.c: Likewise.
* wcsmbs/test-wcscpy-ifunc.c: Likewise.
* wcsmbs/test-wcslen-ifunc.c: Likewise.
* wcsmbs/test-wcsrchr-ifunc.c: Likewise.
* wcsmbs/test-wmemcmp-ifunc.c: Likewise.
* Rules [$(multi-arch) = no] (tests): Do not filter out
$(tests-ifunc).
[$(multi-arch) = no] (xtests): Do not filter out $(xtests-ifunc).
* debug/Makefile (tests-ifunc): Remove variable.
(tests): Do not add $(tests-ifunc).
* wcsmbs/Makefile (tests-ifunc): Remove variable.
(tests): Do not add $(tests-ifunc).
* benchtests/bench-string.h (TEST_IFUNC): Remove macro.
[TEST_IFUNC]: Remove conditionals.
* string/test-string.h (TEST_IFUNC): Remove macro.
[TEST_IFUNC]: Remove conditionals.
string/test-strchr.c is used for both wide and narrow string testing,
but produces a series of warnings for wide string testing because of
hardcoded use of narrow characters in the function check1. This patch
fixes that function to use macros abstracting away the wide / narrow
string choice, adding a new such macro to handle the string and
character constants.
Tested for x86_64.
* string/test-strchr.c [!WIDE] (L): New macro.
[WIDE] (L): Likewise.
(check1): Use CHAR instead of char. Use L on string and character
constants.
This patch fixes the build of C mempcpy and stpcpy by disabling the
redirection to __mempcpy and __stpcpy asm names if
NO_MEMPCPY_STPCPY_REDIRECT is defined, and defining that macro in the
relevant source files.
Tested for powerpc32 that the build is fixed.
* include/string.h [NO_MEMPCPY_STPCPY_REDIRECT] (mempcpy): Do not
redeclare with asm name.
[NO_MEMPCPY_STPCPY_REDIRECT] (stpcpy): Likewise.
* string/mempcpy.c (NO_MEMPCPY_STPCPY_REDIRECT): Define before
including <string.h>.
* string/stpcpy.c (NO_MEMPCPY_STPCPY_REDIRECT): Likewise.
* sysdeps/powerpc/powerpc32/power4/multiarch/mempcpy.c
[!NOT_IN_libc] (NO_MEMPCPY_STPCPY_REDIRECT): Likewise.
* sysdeps/powerpc/powerpc64/multiarch/mempcpy.c
[!NOT_IN_libc] (NO_MEMPCPY_STPCPY_REDIRECT): Likewise.
* sysdeps/powerpc/powerpc64/multiarch/stpcpy.c
[SHARED && !NOT_IN_libc] (NO_MEMPCPY_STPCPY_REDIRECT): Likewise.
Locale code, brought in by ISO C functions, calls memmem, which is not
an ISO C function. This isn't an ISO C conformance bug, because all
mem* names are reserved, but glibc practice is not to rely on that
reservation (thus, memmem is only declared in string.h if __USE_GNU
even though ISO C would allow it to be declared unconditionally, for
example). This patch changes that code to use __memmem.
Note: there are uses of memmem elsewhere in glibc that I didn't
change, although it may turn out some of those also need to use
__memmem.
Tested for x86_64 (testsuite, and that disassembly of installed shared
libraries is unchanged by this patch).
[BZ #17585]
* string/memmem.c [!_LIBC] (__memmem): Define to memmem.
(memmem): Rename to __memmem and define as weak alias of
__memmem. Use libc_hidden_weak.
(__memmem): Use libc_hidden_def.
* include/string.h (__memmem): Declare. Use libc_hidden_proto.
* locale/findlocale.c (valid_locale_name): Use __memmem instead of
memmem.
Modifies the test examination in test-skeleton.c so that a test can be
successful if it is interrupted or it returns uninterrupted with the
expected status. For this both EXPECTED_SIGNAL and EXPECTED_STATUS
have to be set, as is done in tst-strcoll-overflow.c.
this
will improve performance even on targets which don't have an optimized strlen. It is about twice
as
fast as the original strncat in bench-strncat.
implementation, so this improves performance even on targets which don't have an optimized
strlen and strcpy - it is 25% faster in bench-strcat. On targets which don't provide an
optimized strcat but which do have an optimized strlen and strcpy, performance gain is > 2x.
this is a path that should solve bug 15884. It complains about the performance
of strcoll(). It was found out that the runtime of strcoll() is actually bound
to strlen which is needed for calculating the size of a cache that was
installed to improve the comparison performance.
The idea for this patch was that the cache is only useful in rare cases
(strings of same length and same first-level-chars) and that it would be
better to avoid memory allocation at all. To prove this I wrote a performance
test bench-strcoll.c with test data in benchtests-strcoll.tar.gz. Also
modifications in benchtests/Makefile and localedata/Makefile are necessary to
make it work.
After removing the cache the strcoll method showed the predicted behavior
(getting slightly faster) in all but the test case for hindi word sorting.
This was due the hindi text having much more equal words than the other ones.
For equal strings the performance was worse since all comparison levels were
run through and from the second level on the cache improved the comparison
performance of the original version.
Therefore I added a bytewise test via strcmp iff the first level comparison
found that both strings did match because in this case it is very likely that
equal strings are compared. This solved the problem with the hindi test case
and improved the performance of the others.
Performance comparison:
glibc files -33.77%
vi_VN.UTF-8 -34.12%
en_US.UTF-8 -42.42%
ar_SA.UTF-8 -27.49%
zh_CN.UTF-8 +07.90%
cs_CZ.UTF-8 -29.67%
en_GB.UTF-8 -28.50%
da_DK.UTF-8 -36.57%
pl_PL.UTF-8 -39.31%
fr_FR.UTF-8 -28.57%
pt_PT.UTF-8 -22.82%
el_GR.UTF-8 -26.77%
ru_RU.UTF-8 -35.81%
iw_IL.UTF-8 -35.34%
es_ES.UTF-8 -34.46%
hi_IN.UTF-8 -00.38%
sv_SE.UTF-8 -36.99%
hu_HU.UTF-8 -16.35%
tr_TR.UTF-8 -27.80%
is_IS.UTF-8 -33.24%
it_IT.UTF-8 -24.39%
sr_RS.UTF-8 -37.55%
ja_JP.UTF-8 +02.84%
This patch adds an optimized memmove optimization for POWER7/powerpc64.
Basically the idea is to use the memcpy for POWER7 on non-overlapped
memory regions and a optimized backward memcpy for memory regions
that overlap (similar to the idea of string/memmove.c).
The backward memcpy algorithm used is similar the one use for memcpy for
POWER7, with adjustments done for alignment. The difference is memory
is always aligned to 16 bytes before using VSX/altivec instructions.
Merge most of the gnulib implementation of memchr. The changes that
remain are:
- copyright header
- bp-sym.h removed
- reg_char removed
- allow MEMCHR to be redefined
- non-conforming whitespace changes
The merged code fixes a number of -Wundef warnings and also introduces
an optimized algorithm. I haven't detected any performance difference
in the new code which I believe is down to the quite specific
circumstances required to hit it. However the new code is approximately
half the size of the old code on AArch64 (which uses generic memchr).
ChangeLog:
2014-07-04 Will Newton <will.newton@linaro.org>
* string/memchr.c: Merge from gnulib.
[_LIBC]: Remove conditionals.
(__ptr_t): Remove define.
(LONG_MAX_32_BITS): Likewise.
(LONG_MAX): Likewise.
(MEMCHR): Use ANSI prototype and optimize algorithm.
The PAGE_COPY_THRESHOLD macro is meant to be overridden by
architecture-specific pagecopy.h, but it is currently done only by
mach; all other architectures use the default. Check to see if the
macro is defined in addition to whether it is set to a non-zero value.
