Optimize strlen by unrolling the main loop. Large strings are 64% faster on
modern CPUs.
Reviewed-by: Szabolcs Nagy <szabolcs.nagy@arm.com>
(cherry picked from commit 03c8ce5000198947a4dd7b2c14e5131738fda62b)
Unroll the main loop. Large strings are around 20% faster on modern CPUs.
Reviewed-by: Szabolcs Nagy <szabolcs.nagy@arm.com>
(cherry picked from commit 349e48c01e85bd96006860084e76d322e6ca02f1)
Unroll the main loop, which improves performance slightly.
Reviewed-by: Szabolcs Nagy <szabolcs.nagy@arm.com>
(cherry picked from commit 09ebd8549b2ce5a3a6c0c7c5f3e62227faf50a99)
Simplify calculation of the mask using shrn. Unroll the main loop.
Small strings are 20% faster on modern CPUs.
Reviewed-by: Szabolcs Nagy <szabolcs.nagy@arm.com>
(cherry picked from commit 51541a229740801882490177fa178e49264b13fb)
Use shrn for the mask, merge tst+bne into cbnz, and tweak code alignment.
Performance improves slightly as a result.
Reviewed-by: Szabolcs Nagy <szabolcs.nagy@arm.com>
(cherry picked from commit 1bbb1a2022e126f21810d3d0ebe0a975d5243e43)
Optimize the main loop - large strings are 43% faster on modern CPUs.
Reviewed-by: Szabolcs Nagy <szabolcs.nagy@arm.com>
(cherry picked from commit 00776241776e67fc666b896c1e85770f4f3ec1e1)
Optimize the main loop - large strings are 40% faster on modern CPUs.
Reviewed-by: Szabolcs Nagy <szabolcs.nagy@arm.com>
(cherry picked from commit ce758d4f063820c2bc743e12797d7454c66be718)
We found that string functions were using AND+ADDP
to find the nibble/syndrome mask but there is an easier
opportunity through `SHRN dst.8b, src.8h, 4` (shift
right every 2 bytes by 4 and narrow to 1 byte) and has
same latency on all SIMD ARMv8 targets as ADDP. There
are also possible gaps for memcmp but that's for
another patch.
We see 10-20% savings for small-mid size cases (<=128)
which are primary cases for general workloads.
(cherry picked from commit 3c9980698988ef64072f1fac339b180f52792faf)
Rewrite memcmp to improve performance. On small and medium inputs performance
is 10-20% better. Large inputs use a SIMD loop processing 64 bytes per
iteration, which is 30-50% faster depending on the size.
Reviewed-by: Szabolcs Nagy <szabolcs.nagy@arm.com>
(cherry picked from commit b51eb35c572b015641f03e3682c303f7631279b7)
Optimize strnlen by avoiding UMINV which is slow on most cores. On Neoverse N1
large strings are 1.8x faster than the current version, and bench-strnlen is
50% faster overall. This version is MTE compatible.
Reviewed-by: Szabolcs Nagy <szabolcs.nagy@arm.com>
(cherry picked from commit 252cad02d4c63540501b9b8c988cb91248563224)
DELOUSE was added to asm code to make them compatible with non-LP64
ABIs, but it is an unfortunate name and the code was not compatible
with ABIs where pointer and size_t are different. Glibc currently
only supports the LP64 ABI so these macros are not really needed or
tested, but for now the name is changed to be more meaningful instead
of removing them completely.
Some DELOUSE macros were dropped: clone, strlen and strnlen used it
unnecessarily.
The out of tree ILP32 patches are currently not maintained and will
likely need a rework to rebase them on top of the time64 changes.
(cherry picked from commit 45b1e17e9150dbd9ac2d578579063fbfa8e1b327)
Ffsll function randomly regress by ~20%, depending on how code gets
aligned in memory. Ffsll function code size is 17 bytes. Since default
function alignment is 16 bytes, it can load on 16, 32, 48 or 64 bytes
aligned memory. When ffsll function load at 16, 32 or 64 bytes aligned
memory, entire code fits in single 64 bytes cache line. When ffsll
function load at 48 bytes aligned memory, it splits in two cache line,
hence random regression.
Ffsll function size reduction from 17 bytes to 12 bytes ensures that it
will always fit in single 64 bytes cache line.
This patch fixes ffsll function random performance regression.
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
(cherry picked from commit 9d94997b5f9445afd4f2bccc5fa60ff7c4361ec1)
The:
```
if (shared_per_thread > 0 && threads > 0)
shared_per_thread /= threads;
```
Code was accidentally moved to inside the else scope. This doesn't
match how it was previously (before af992e7abd).
This patch fixes that by putting the division after the `else` block.
(cherry picked from commit 084fb31bc2c5f95ae0b9e6df4d3cf0ff43471ede)
On some machines we end up with incomplete cache information. This can
make the new calculation of `sizeof(total-L3)/custom-divisor` end up
lower than intended (and lower than the prior value). So reintroduce
the old bound as a lower bound to avoid potentially regressing code
where we don't have complete information to make the decision.
Reviewed-by: DJ Delorie <dj@redhat.com>
(cherry picked from commit 8b9a0af8ca012217bf90d1dc0694f85b49ae09da)
After:
```
commit af992e7abdc9049714da76cae1e5e18bc4838fb8
Author: Noah Goldstein <goldstein.w.n@gmail.com>
Date: Wed Jun 7 13:18:01 2023 -0500
x86: Increase `non_temporal_threshold` to roughly `sizeof_L3 / 4`
```
Split `shared` (cumulative cache size) from `shared_per_thread` (cache
size per socket), the `shared_per_thread` *can* be slightly off from
the previous calculation.
Previously we added `core` even if `threads_l2` was invalid, and only
used `threads_l2` to divide `core` if it was present. The changed
version only included `core` if `threads_l2` was valid.
This change restores the old behavior if `threads_l2` is invalid by
adding the entire value of `core`.
Reviewed-by: DJ Delorie <dj@redhat.com>
(cherry picked from commit 47f747217811db35854ea06741be3685e8bbd44d)
Current `non_temporal_threshold` set to roughly '3/4 * sizeof_L3 /
ncores_per_socket'. This patch updates that value to roughly
'sizeof_L3 / 4`
The original value (specifically dividing the `ncores_per_socket`) was
done to limit the amount of other threads' data a `memcpy`/`memset`
could evict.
Dividing by 'ncores_per_socket', however leads to exceedingly low
non-temporal thresholds and leads to using non-temporal stores in
cases where REP MOVSB is multiple times faster.
Furthermore, non-temporal stores are written directly to main memory
so using it at a size much smaller than L3 can place soon to be
accessed data much further away than it otherwise could be. As well,
modern machines are able to detect streaming patterns (especially if
REP MOVSB is used) and provide LRU hints to the memory subsystem. This
in affect caps the total amount of eviction at 1/cache_associativity,
far below meaningfully thrashing the entire cache.
As best I can tell, the benchmarks that lead this small threshold
where done comparing non-temporal stores versus standard cacheable
stores. A better comparison (linked below) is to be REP MOVSB which,
on the measure systems, is nearly 2x faster than non-temporal stores
at the low-end of the previous threshold, and within 10% for over
100MB copies (well past even the current threshold). In cases with a
low number of threads competing for bandwidth, REP MOVSB is ~2x faster
up to `sizeof_L3`.
The divisor of `4` is a somewhat arbitrary value. From benchmarks it
seems Skylake and Icelake both prefer a divisor of `2`, but older CPUs
such as Broadwell prefer something closer to `8`. This patch is meant
to be followed up by another one to make the divisor cpu-specific, but
in the meantime (and for easier backporting), this patch settles on
`4` as a middle-ground.
Benchmarks comparing non-temporal stores, REP MOVSB, and cacheable
stores where done using:
https://github.com/goldsteinn/memcpy-nt-benchmarks
Sheets results (also available in pdf on the github):
https://docs.google.com/spreadsheets/d/e/2PACX-1vS183r0rW_jRX6tG_E90m9qVuFiMbRIJvi5VAE8yYOvEOIEEc3aSNuEsrFbuXw5c3nGboxMmrupZD7K/pubhtml
Reviewed-by: DJ Delorie <dj@redhat.com>
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
(cherry picked from commit af992e7abdc9049714da76cae1e5e18bc4838fb8)
The signal handler installed in the ELF constructor cannot easily
be removed again (because the program may have changed handlers
in the meantime). Mark the object as NODELETE so that the registered
handler function is never unloaded.
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
(cherry picked from commit 23ee92deea4c99d0e6a5f48fa7b942909b123ec5)
Previous implementation was adjusting length (rsi) to match
bytes (eax), but since there is no bound to length this can cause
overflow.
Fix is to just convert the byte-count (eax) to length by dividing by
sizeof (wchar_t) before the comparison.
Full check passes on x86-64 and build succeeds w/ and w/o multiarch.
(cherry picked from commit b0969fa53a28b4ab2159806bf6c99a98999502ee)
The sunrpc function svcunix_create suffers from a stack-based buffer
overflow with overlong pathname arguments.
Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
(cherry picked from commit f545ad4928fa1f27a3075265182b38a4f939a5f7)
This is helpful for testing compat symbols in cases where _ISOMAC
is activated implicitly due to -DMODULE_NAME=testsuite and cannot
be disabled easily.
(cherry picked from commit 36f6e408845c8c539128f3fb9cb132bf1845a2c8)
Processing an overlong pathname in the sunrpc clnt_create function
results in a stack-based buffer overflow.
Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
(cherry picked from commit 226b46770c82899b555986583294b049c6ec9b40)
Previously, in UCS4 conversion routines we limit the number of
characters we examine to the minimum of the number of characters in the
input and the number of characters in the output. This is not the
correct behavior when __GCONV_IGNORE_ERRORS is set, as we do not consume
an output character when we skip a code unit. Instead, track the input
and output pointers and terminate the loop when either reaches its
limit.
This resolves assertion failures when resetting the input buffer in a step of
iconv, which assumes that the input will be fully consumed given sufficient
output space.
(cherry picked from commit 228edd356f03bf62dcf2b1335f25d43c602ee68d)
The fix was backported by commit ff75390ef59823193351ae77584c397c503b7b58
("Use __pthread_attr_copy in mq_notify (bug 27896)")
after glibc 2.32 release.
The fix was backported by commit 050022910be1d1f5c11cd5168f1685ad4f9580d2
("iconv: Accept redundant shift sequences in IBM1364 [BZ #26224]")
after glibc 2.32 release.
Since strchr-avx2.S updated by
commit 1f745ecc2109890886b161d4791e1406fdfc29b8
Author: noah <goldstein.w.n@gmail.com>
Date: Wed Feb 3 00:38:59 2021 -0500
x86-64: Refactor and improve performance of strchr-avx2.S
uses sarx:
c4 e2 72 f7 c0 sarx %ecx,%eax,%eax
for strchr-avx2 family functions, require BMI2 in ifunc-impl-list.c and
ifunc-avx2.h.
This fixes BZ #29611.
(cherry picked from commit 83c5b368226c34a2f0a5287df40fc290b2b34359)
libc_map is never reset to NULL, neither during dlclose nor on a
dlopen call which reuses the namespace structure. As a result, if a
namespace is reused, its libc is not initialized properly. The most
visible result is a crash in the <ctype.h> functions.
To prevent similar bugs on namespace reuse from surfacing,
unconditionally initialize the chosen namespace to zero using memset.
(cherry picked from commit d0e357ff45a75553dee3b17ed7d303bfa544f6fe)
On success, mq_receive() and mq_timedreceive() return the number of
bytes in the received message, so it requires to check if the value
is larger than 0.
Checked on i686-linux-gnu.
(cherry picked from commit 71d87d85bf54f6522813aec97c19bdd24997341e)
Previously TEST_NAME was passing a function pointer. This didn't fail
because of the -Wno-error flag (to allow for overflow sizes passed
to strncmp/wcsncmp)
Reviewed-by: H.J. Lu <hjl.tools@gmail.com>
(cherry picked from commit b98d0bbf747f39770e0caba7e984ce9f8f900330)
In the overflow fallback strncmp-avx2-rtm and wcsncmp-avx2-rtm would
call strcmp-avx2 and wcscmp-avx2 respectively. This would have
not checks around vzeroupper and would trigger spurious
aborts. This commit fixes that.
test-strcmp, test-strncmp, test-wcscmp, and test-wcsncmp all pass on
AVX2 machines with and without RTM.
Reviewed-by: H.J. Lu <hjl.tools@gmail.com>
(cherry picked from commit 7835d611af0854e69a0c71e3806f8fe379282d6f)
In the overflow fallback strncmp-avx2-rtm and wcsncmp-avx2-rtm would
call strcmp-avx2 and wcscmp-avx2 respectively. This would have
not checks around vzeroupper and would trigger spurious
aborts. This commit fixes that.
test-strcmp, test-strncmp, test-wcscmp, and test-wcsncmp all pass on
AVX2 machines with and without RTM.
Co-authored-by: H.J. Lu <hjl.tools@gmail.com>
(cherry picked from commit c6272098323153db373f2986c67786ea8c85f1cf)
Verify that wcsncmp (L("abc"), L("abd"), SIZE_MAX) == 0. The new test
fails without
commit ddf0992cf57a93200e0c782e2a94d0733a5a0b87
Author: Noah Goldstein <goldstein.w.n@gmail.com>
Date: Sun Jan 9 16:02:21 2022 -0600
x86: Fix __wcsncmp_avx2 in strcmp-avx2.S [BZ# 28755]
and
commit 7e08db3359c86c94918feb33a1182cd0ff3bb10b
Author: Noah Goldstein <goldstein.w.n@gmail.com>
Date: Sun Jan 9 16:02:28 2022 -0600
x86: Fix __wcsncmp_evex in strcmp-evex.S [BZ# 28755]
This is for BZ #28755.
Reviewed-by: Sunil K Pandey <skpgkp2@gmail.com>
(cherry picked from commit aa5a720056d37cf24924c138a3dbe6dace98e97c)
commit 6f573a27b6c8b4236445810a44660612323f5a73
Author: Noah Goldstein <goldstein.w.n@gmail.com>
Date: Wed Jun 23 01:19:34 2021 -0400
x86-64: Add wcslen optimize for sse4.1
added wcsnlen-sse4.1 to the wcslen ifunc implementation list. Since the
random value in the the RSI register is larger than the wide-character
string length in the existing wcslen test, it didn't trigger the wcslen
test failure. Add a test to force 0 into the RSI register before calling
wcslen.
(cherry picked from commit a6e7c3745d73ff876b4ba6991fb00768a938aef5)
The following commit
commit 6f573a27b6c8b4236445810a44660612323f5a73
Author: Noah Goldstein <goldstein.w.n@gmail.com>
Date: Wed Jun 23 01:19:34 2021 -0400
x86-64: Add wcslen optimize for sse4.1
Added wcsnlen-sse4.1 to the wcslen ifunc implementation list and did
not add wcslen-sse4.1 to wcslen ifunc implementation list. This commit
fixes that by removing wcsnlen-sse4.1 from the wcslen ifunc
implementation list and adding wcslen-sse4.1 to the ifunc
implementation list.
Testing:
test-wcslen.c, test-rsi-wcslen.c, and test-rsi-strlen.c are passing as
well as all other tests in wcsmbs and string.
Signed-off-by: Noah Goldstein <goldstein.w.n@gmail.com>
Reviewed-by: H.J. Lu <hjl.tools@gmail.com>
Reviewed-by: H.J. Lu <hjl.tools@gmail.com>
(cherry picked from commit 0679442defedf7e52a94264975880ab8674736b2)
From
https://www.intel.com/content/www/us/en/support/articles/000059422/processors.html
* Intel TSX will be disabled by default.
* The processor will force abort all Restricted Transactional Memory (RTM)
transactions by default.
* A new CPUID bit CPUID.07H.0H.EDX[11](RTM_ALWAYS_ABORT) will be enumerated,
which is set to indicate to updated software that the loaded microcode is
forcing RTM abort.
* On processors that enumerate support for RTM, the CPUID enumeration bits
for Intel TSX (CPUID.07H.0H.EBX[11] and CPUID.07H.0H.EBX[4]) continue to
be set by default after microcode update.
* Workloads that were benefited from Intel TSX might experience a change
in performance.
* System software may use a new bit in Model-Specific Register (MSR) 0x10F
TSX_FORCE_ABORT[TSX_CPUID_CLEAR] functionality to clear the Hardware Lock
Elision (HLE) and RTM bits to indicate to software that Intel TSX is
disabled.
1. Add RTM_ALWAYS_ABORT to CPUID features.
2. Set RTM usable only if RTM_ALWAYS_ABORT isn't set. This skips the
string/tst-memchr-rtm etc. testcases on the affected processors, which
always fail after a microcde update.
3. Check RTM feature, instead of usability, against /proc/cpuinfo.
This fixes BZ #28033.
(cherry picked from commit ea8e465a6b8d0f26c72bcbe453a854de3abf68ec)
Since __strlen_evex and __strnlen_evex added by
commit 1fd8c163a83d96ace1ff78fa6bac7aee084f6f77
Author: H.J. Lu <hjl.tools@gmail.com>
Date: Fri Mar 5 06:24:52 2021 -0800
x86-64: Add ifunc-avx2.h functions with 256-bit EVEX
use sarx:
c4 e2 6a f7 c0 sarx %edx,%eax,%eax
require BMI2 for __strlen_evex and __strnlen_evex in ifunc-impl-list.c.
ifunc-avx2.h already requires BMI2 for EVEX implementation.
(cherry picked from commit 55bf411b451c13f0fb7ff3d3bf9a820020b45df1)
This commit adds tests for a bug in the wide char variant of the
functions where the implementation may assume that maxlen for wcsnlen
or n for wmemchr/strncat will not overflow when multiplied by
sizeof(wchar_t).
These tests show the following implementations failing on x86_64:
wcsnlen-sse4_1
wcsnlen-avx2
wmemchr-sse2
wmemchr-avx2
strncat would fail as well if it where on a system that prefered
either of the wcsnlen implementations that failed as it relies on
wcsnlen.
Signed-off-by: Noah Goldstein <goldstein.w.n@gmail.com>
Reviewed-by: H.J. Lu <hjl.tools@gmail.com>
(cherry picked from commit da5a6fba0febbfc90896ce1b2eb75c6d8a88a72d)
