This patch significantly improves performance of memmem using a novel
modified Horspool algorithm. Needles up to size 256 use a bad-character
table indexed by hashed pairs of characters to quickly skip past mismatches.
Long needles use a self-adapting filtering step to avoid comparing the whole
needle repeatedly.
By limiting the needle length to 256, the shift table only requires 8 bits
per entry, lowering preprocessing overhead and minimizing cache effects.
This limit also implies worst-case performance is linear.
Small needles up to size 2 use a dedicated linear search. Very long needles
use the Two-Way algorithm (to avoid increasing stack size or slowing down
the common case, inlining is disabled).
The performance gain is 6.6 times on English text on AArch64 using random
needles with average size 8.
Tested against GLIBC testsuite and randomized tests.
Reviewed-by: Szabolcs Nagy <szabolcs.nagy@arm.com>
* string/memmem.c (__memmem): Rewrite to improve performance.
This patch significantly improves performance of strstr using a novel
modified Horspool algorithm. Needles up to size 256 use a bad-character
table indexed by hashed pairs of characters to quickly skip past mismatches.
Long needles use a self-adapting filtering step to avoid comparing the whole
needle repeatedly.
By limiting the needle length to 256, the shift table only requires 8 bits
per entry, lowering preprocessing overhead and minimizing cache effects.
This limit also implies worst-case performance is linear.
Small needles up to size 3 use a dedicated linear search. Very long needles
use the Two-Way algorithm.
The performance gain using the improved bench-strstr on Cortex-A72 is 5.8
times basic_strstr and 3.7 times twoway_strstr.
Tested against GLIBC testsuite, randomized tests and the GNULIB strstr test
(https://git.savannah.gnu.org/cgit/gnulib.git/tree/tests/test-strstr.c).
Reviewed-by: Szabolcs Nagy <szabolcs.nagy@arm.com>
* string/str-two-way.h (two_way_short_needle): Add inline to avoid
warning.
(two_way_long_needle): Block inlining.
* string/strstr.c (strstr2): Add new function.
(strstr3): Likewise.
(STRSTR): Completely rewrite strstr to improve performance.
Benchmark needles which exhibit worst-case performance. This shows that
basic_strstr is quadratic and thus unsuitable for large needles.
On the other hand the Two-way and new strstr implementations are linear with
increasing needle sizes. The slowest cases of the two implementations are
within a factor of 2 on several different microarchitectures. Two-way is
slowest on inputs which cause a branch mispredict on almost every character.
The new strstr is slowest on inputs which almost match and result in many
calls to memcmp. Thanks to Szabolcs for providing various hard needles.
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
* benchtests/bench-strstr.c (test_hard_needle): New function.
GCC mainline has recently added warn_unused_result attributes to some
malloc-like built-in functions, where glibc previously had them in its
headers only for __USE_FORTIFY_LEVEL > 0. This results in those
attributes being newly in effect for building the glibc testsuite, so
resulting in new warnings that break the build where tests
deliberately call such functions and ignore the result. Thus patch
duly adds calls to DIAG_* macros around those calls to disable the
warning.
Tested with build-many-glibcs.py for aarch64-linux-gnu.
* malloc/tst-calloc.c: Include <libc-diag.h>.
(null_test): Ignore -Wunused-result around calls to calloc.
* malloc/tst-mallocfork.c: Include <libc-diag.h>.
(do_test): Ignore -Wunused-result around call to malloc.
No 32-bit system call wrapper is added because the interface
is problematic because it cannot deal with 64-bit inode numbers
and 64-bit directory hashes.
A future commit will deprecate the undocumented getdirentries
and getdirentries64 functions.
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
Add support to use 'mffsl' instruction if compiled for POWER9 (or later).
Also, mask the result to avoid bleeding unrelated bits into the result of
_FPU_GET_RC().
Reviewed-by: Tulio Magno Quites Machado Filho <tuliom@linux.ibm.com>
fegetexcept() included code which exactly duplicates the code in
fenv_reg_to_exceptions(). Replace with a call to that function.
2019-06-05 Paul A. Clarke <pc@us.ibm.com>
* sysdeps/powerpc/fpu/fegetexcept.c (__fegetexcept): Replace code
with call to equivalent function.
Linux only supports the required ISA sysctls on StrongARM devices,
which are armv4 and no longer tested during glibc development
and probably bit-rotted by this point. (No reported test results,
and the last discussion of armv4 support was in the glibc 2.19
release notes.)
<asm/unistd.h> on arm defines the following macros:
#define __ARM_NR_breakpoint (__ARM_NR_BASE+1)
#define __ARM_NR_cacheflush (__ARM_NR_BASE+2)
#define __ARM_NR_usr26 (__ARM_NR_BASE+3)
#define __ARM_NR_usr32 (__ARM_NR_BASE+4)
#define __ARM_NR_set_tls (__ARM_NR_BASE+5)
#define __ARM_NR_get_tls (__ARM_NR_BASE+6)
These do not follow the regular __NR_* naming convention and
have so far been ignored by the syscall-names.list consistency
checks. This commit adds these names to the file, preparing
for the availability of these names in the regular __NR_*
namespace.
Since GCC commit 271500 (svn), also known as the following commit on the
git mirror:
commit 61edec870f9fdfb5df3fa4e40f28cbaede28a5b1
Author: amodra <amodra@138bc75d-0d04-0410-961f-82ee72b054a4>
Date: Wed May 22 04:34:26 2019 +0000
[RS6000] Don't pass -many to the assembler
glibc builds are failing when an assembly implementation does not
declare the correct '.machine' directive, or when no such directive is
declared at all. For example, when a POWER6 instruction is used, but
'.machine power6' is not declared, the assembler will fail with an error
similar to the following:
../sysdeps/powerpc/powerpc64/power8/strcmp.S: Assembler messages:
24 ../sysdeps/powerpc/powerpc64/power8/strcmp.S:55: Error: unrecognized opcode: `cmpb'
This patch adds '.machine powerN' directives where none existed, as well
as it updates '.machine power7' directives on POWER8 files, because the
minimum binutils version required to build glibc (binutils 2.25) now
provides this machine version. It also adds '-many' to the assembler
command used to build tst-set_ppr.c.
Tested for powerpc, powerpc64, and powerpc64le, as well as with
build-many-glibcs.py for powerpc targets.
Reviewed-by: Tulio Magno Quites Machado Filho <tuliom@linux.ibm.com>
The patch 6e8ba7fd57 meant to remove the all get_clockfreq.c. This
patch removes the missing files for sparcv9 and x86_64.
Checked against a build to x86_64-linux-gnu and sparcv9-linux-gnu.
* sysdeps/unix/sysv/linux/sparc/sparc32/sparcv9/get_clockfreq.c:
Remove file.
* sysdeps/unix/sysv/linux/x86_64/get_clockfreq.c: Likewise.
This commit fixes some errors and converts all month names to lowercase.
The content is synchronized with CLDR-35.1 now but trailing dots are
removed from abmon values in order to maintain consistency with the
previous values and with many other locales which do the same.
[BZ #24369]
* localedata/locales/tt_RU (mon): Update from CLDR-35.1, fix errors.
(abmon): Likewise, but remove the trailing dots.
This patch adds the new F_SEAL_FUTURE_WRITE constant from Linux 5.1 to
bits/fcntl-linux.h.
Tested for x86_64.
* sysdeps/unix/sysv/linux/bits/fcntl-linux.h [__USE_GNU]
(F_SEAL_FUTURE_WRITE): New macro.
This test corrupts /var/cache/ldconfig/aux-cache and executes ldconfig
to check it will not segfault using the corrupted aux_cache. The test
uses the test-in-container framework. Verified no regressions on
x86_64.
GCC 9 dropped support for the SPE extensions to PowerPC, which means
powerpc*-*-*gnuspe* configurations are no longer buildable with that
compiler. This ISA extension was peculiar to the “e500” line of
embedded PowerPC chips, which, as far as I can tell, are no longer
being manufactured, so I think we should follow suit.
This patch was developed by grepping for “e500”, “__SPE__”, and
“__NO_FPRS__”, and may not eliminate every vestige of SPE support.
Most uses of __NO_FPRS__ are left alone, as they are relevant to
normal embedded PowerPC with soft-float.
* sysdeps/powerpc/preconfigure: Error out on powerpc-*-*gnuspe*
host type.
* scripts/build-many-glibcs.py: Remove powerpc-*-linux-gnuspe
and powerpc-*-linux-gnuspe-e500v1 from list of build configurations.
* sysdeps/powerpc/powerpc32/e500: Recursively delete.
* sysdeps/unix/sysv/linux/powerpc/powerpc32/e500: Recursively delete.
* sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/context-e500.h:
Delete.
* sysdeps/powerpc/fpu_control.h: Remove SPE variant.
Issue an #error if used with a compiler in SPE-float mode.
* sysdeps/powerpc/powerpc32/__longjmp_common.S
* sysdeps/powerpc/powerpc32/setjmp_common.S
* sysdeps/unix/sysv/linux/powerpc/powerpc32/getcontext-common.S
* sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/getcontext.S
* sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/setcontext.S
* sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/swapcontext.S
* sysdeps/unix/sysv/linux/powerpc/powerpc32/setcontext-common.S
* sysdeps/unix/sysv/linux/powerpc/powerpc32/swapcontext-common.S:
Remove code to preserve SPE register state.
* sysdeps/unix/sysv/linux/powerpc/elision-lock.c
* sysdeps/unix/sysv/linux/powerpc/elision-trylock.c
* sysdeps/unix/sysv/linux/powerpc/elision-unlock.c
Remove __SPE__ ifndefs.
Improve string benchtest timing. Many tests run for 0.01s which is way too
short to give accurate results. Other tests take over 40 seconds which is
way too long. Significantly increase the iterations of the short running
tests. Reduce number of alignment variations in the long running memcpy walk
tests so they take less than 5 seconds.
As a result most tests take at least 0.1s and all finish within 5 seconds.
* benchtests/bench-memcpy-random.c (do_one_test): Use medium iterations.
* benchtests/bench-memcpy-walk.c (test_main): Reduce alignment tests.
* benchtests/bench-memmem.c (do_one_test): Use small iterations.
* benchtests/bench-memmove-walk.c (test_main): Reduce alignment tests.
* benchtests/bench-memset-walk.c (test_main): Reduce alignment tests.
* benchtests/bench-strcasestr.c (do_one_test): Use small iterations.
* benchtests/bench-string.h (INNER_LOOP_ITERS): Increase iterations.
(INNER_LOOP_ITERS_MEDIUM): New define.
(INNER_LOOP_ITERS_SMALL): New define.
* benchtests/bench-strpbrk.c (do_one_test): Use medium iterations.
* benchtests/bench-strsep.c (do_one_test): Use small iterations.
* benchtests/bench-strspn.c (do_one_test): Use medium iterations.
* benchtests/bench-strstr.c (do_one_test): Use small iterations.
* benchtests/bench-strtok.c (do_one_test): Use small iterations.
This patch add the missing SEMTIMEDOP_IPC_ARGS definions on powerpc
and sparc ipc_priv.h.
Checked on powerpc64le-linux-gnu and with a build for sparc64-linux-gnu.
* sysdeps/unix/sysv/linux/powerpc/ipc_priv.h (SEMTIMEDOP_IPC_ARGS):
New define.
* sysdeps/unix/sysv/linux/sparc/sparc64/ipc_priv.h
(SEMTIMEDOP_IPC_ARGS): Likewise.
struct gconv_fcts for the C locale is statically allocated,
and __gconv_close_transform deallocates the steps object.
Therefore this commit introduces __wcsmbs_close_conv to avoid
freeing the statically allocated steps objects.
The codecvt vtable is not a real vtable because it also contains the
conversion state data. Furthermore, wide stream support was added to
GCC 3.0, after a C++ ABI bump, so there is no compatibility
requirement with libstdc++.
This change removes several unmangled function pointers which could
be used with a corrupted FILE object to redirect execution. (libio
vtable verification did not cover the codecvt vtable.)
Reviewed-by: Yann Droneaud <ydroneaud@opteya.com>
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
The original implementations of test_timespec_before_impl and
test_timespec_equal_or_after in 5198399651
were missing the backslash required for a newline.
Checked on x86_64-linux-gnu.
* support/timespec.c: Add backslash to correct newline in failure
message.
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
This patch consolidates the s390-32 semtimedop implementation by defining
a arch-specific SEMTIMEDOP_IPC_ARGS to rearrange the arguments expected
by s390 Linux kABI. The idea is to avoid have multiples semtimedop
implementation changes for Linux v5.1 change to enable wire-up sysvipc
support.
Checked with a s390-linux-gnu and s390x-linux-gnu and checking that
resulting semtimedop objects did not change.
* sysdeps/unix/sysv/linux/ipc_priv.h (SEMTIMEDOP_IPC_ARGS): New
define.
* sysdpes/unix/sysv/linux/s390/ipc_priv.h: New file.
* sysdeps/unix/sysv/linux/s390/semtimedop.c: Remove file.
* sysdeps/unix/sysv/linux/semtimedop.c (semtimedop): Use
SEMTIMEDOP_IPC_ARGS for calls with __NR_ipc.
The __IPC64 flags is meant to be used to enable the new sysv struct
format when the architectures supports it (ARCH_WANT_IPC_PARSE_VERSION
config flag on Linux kernel).
This currently issue only affects alpha.
[BZ #24570]
* sysdeps/unix/sysv/linux/msgctl.c (__old_msgctl): Remove __IPC_64
usage.
This patch adds the new NT_ARM_PACA_KEYS and NT_ARM_PACG_KEYS from
Linux 5.1 to glibc's elf.h.
Tested for x86_64.
* elf/elf.h (NT_ARM_PACA_KEYS): New macro.
(NT_ARM_PACG_KEYS): Likewise.
Change the tcache->counts[] entries to uint16_t - this removes
the limit set by char and allows a larger tcache. Remove a few
redundant asserts.
bench-malloc-thread with 4 threads is ~15% faster on Cortex-A72.
Reviewed-by: DJ Delorie <dj@redhat.com>
* malloc/malloc.c (MAX_TCACHE_COUNT): Increase to UINT16_MAX.
(tcache_put): Remove redundant assert.
(tcache_get): Remove redundant asserts.
(__libc_malloc): Check tcache count is not zero.
* manual/tunables.texi (glibc.malloc.tcache_count): Update maximum.
Linux 5.1 adds missing syscalls to the syscall table for many Linux
kernel architectures. This patch updates the kernel-features.h
headers accordingly. __ASSUME_DIRECT_SYSVIPC_SYSCALLS is not updated
because of the differences between new and old syscalls described in
<https://sourceware.org/ml/libc-alpha/2019-05/msg00235.html>. The
statfs64 structure used by alpha matches what the new kernel syscalls
use.
Tested with build-many-glibcs.py.
* sysdeps/unix/sysv/linux/alpha/kernel-features.h
(__ASSUME_STATFS64): Only undefine if [__LINUX_KERNEL_VERSION <
0x050100].
* sysdeps/unix/sysv/linux/ia64/kernel-features.h (__ASSUME_STATX):
Likewise.
* sysdeps/unix/sysv/linux/sh/kernel-features.h
(__ASSUME_STATX): Likewise.
Provide an explicit diagnostic if the length is positive, and
do not just crash with a null pointer dereference. Null pointers
are only valid if the length is zero, so this can only happen with
a faulty test.
dlerror.c (__dlerror_main_freeres) will try to free resources which only
have been initialized when init () has been called. That function is
called when resources are needed using __libc_once (once, init) where
once is a __libc_once_define (static, once) in the dlerror.c file.
Trying to free those resources if init () hasn't been called will
produce errors under valgrind memcheck. So guard the freeing of those
resources using __libc_once_get (once) and make sure we have a valid
key. Also add a similar guard to __dlerror ().
* dlfcn/dlerror.c (__dlerror_main_freeres): Guard using
__libc_once_get (once) and static_bug == NULL.
(__dlerror): Check we have a valid key, set result to static_buf
otherwise.
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
When computing the length of the converted part of the stdio buffer, use
the number of consumed wide characters, not the (negative) distance to the
end of the wide buffer.
The function uses the internal service_user type, so it is not
really usable from the outside of glibc. Rename the function
to __nss_database_lookup2 for internal use, and change
__nss_database_lookup to always indicate failure to the caller.
__nss_next already was a compatibility symbol. The new
implementation always fails and no longer calls __nss_next2.
unscd, the alternative nscd implementation, does not use
__nss_database_lookup, so it is not affected by this change.
