The CORE-MATH exp2m1f implementation showed slight worse latency
when using x86_64 baseline ABI. This patch adds a ifunc variant
with similar performance for x86_64-v3.
Reviewed-by: Noah Goldstein <goldstein.w.n@gmail.com>
Reviewed-by: DJ Delorie <dj@redhat.com>
The CORE-MATH exp10m1f implementation showed slight worse latency
when using x86_64 baseline ABI. This patch adds a ifunc variant
with similar performance for x86_64-v3.
Reviewed-by: Noah Goldstein <goldstein.w.n@gmail.com>
Reviewed-by: DJ Delorie <dj@redhat.com>
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>
The CORE-MATH implementation is correctly rounded (for any rounding mode)
and shows better performance compared to the generic exp10m1f.
The code was adapted to glibc style and to use the definition of
math_config.h (to handle errno, overflow, and underflow). I mostly
fixed some small issues in corner cases (sNaN handling, -INFINITY,
a specific overflow check).
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 45.4690 49.5845 -9.05%
x86_64v2 46.1604 36.2665 21.43%
x86_64v3 37.8442 31.0359 17.99%
i686 121.367 93.0079 23.37%
aarch64 21.1126 15.0165 28.87%
power10 12.7426 8.4929 33.35%
reciprocal-throughput master patched improvement
x86_64 19.6005 17.4005 11.22%
x86_64v2 19.6008 11.1977 42.87%
x86_64v3 17.5427 10.2898 41.34%
i686 59.4215 60.9675 -2.60%
aarch64 13.9814 7.9173 43.37%
power10 6.7814 6.4258 5.24%
The generic implementation calls __ieee754_exp10f which has an
optimized version, although it is not correctly rounded, which is
the main culprit of the the latency difference for x86_64 and
throughp for i686.
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>
The CORE-MATH implementation is correctly rounded (for any rounding mode).
This can be checked by exhaustive tests in a few minutes since there are
less than 2^32 values to check against for example GNU MPFR.
This patch also adds some bench values for tgammaf.
Tested on x86_64 and x86 (cfarm26).
With the initial GNU libc code it gave on an Intel(R) Core(TM) i7-8700:
"tgammaf": {
"": {
"duration": 3.50188e+09,
"iterations": 2e+07,
"max": 602.891,
"min": 65.1415,
"mean": 175.094
}
}
With the new code:
"tgammaf": {
"": {
"duration": 3.30825e+09,
"iterations": 5e+07,
"max": 211.592,
"min": 32.0325,
"mean": 66.1649
}
}
With the initial GNU libc code it gave on cfarm26 (i686):
"tgammaf": {
"": {
"duration": 3.70505e+09,
"iterations": 6e+06,
"max": 2420.23,
"min": 243.154,
"mean": 617.509
}
}
With the new code:
"tgammaf": {
"": {
"duration": 3.24497e+09,
"iterations": 1.8e+07,
"max": 1238.15,
"min": 101.155,
"mean": 180.276
}
}
Signed-off-by: Alexei Sibidanov <sibid@uvic.ca>
Signed-off-by: Paul Zimmermann <Paul.Zimmermann@inria.fr>
Changes in v2:
- include <math.h> (fix the linknamespace failures)
- restored original benchtests/strcoll-inputs/filelist#en_US.UTF-8 file
- restored original wrapper code (math/w_tgammaf_compat.c),
except for the dealing with the sign
- removed the tgammaf/float entries in all libm-test-ulps files
- address other comments from Joseph Myers
(https://sourceware.org/pipermail/libc-alpha/2024-July/158736.html)
Changes in v3:
- pass NULL argument for signgam from w_tgammaf_compat.c
- use of math_narrow_eval
- added more comments
Changes in v4:
- initialize local_signgam to 0 in math/w_tgamma_template.c
- replace sysdeps/ieee754/dbl-64/gamma_productf.c by dummy file
Changes in v5:
- do not mention local_signgam any more in math/w_tgammaf_compat.c
- initialize local_signgam to 1 instead of 0 in w_tgamma_template.c
and added comment
Changes in v6:
- pass NULL as 2nd argument of __ieee754_gammaf_r in
w_tgammaf_compat.c, and check for NULL in e_gammaf_r.c
Changes in v7:
- added Signed-off-by line for Alexei Sibidanov (author of the code)
Changes in v8:
- added Signed-off-by line for Paul Zimmermann (submitted of the patch)
Changes in v9:
- address comments from review by Adhemerval Zanella
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
Several copies of the licenses in files contained whitespace related
problems. Two cases are addressed here, the first is two spaces
after a period which appears between "PURPOSE." and "See". The other
is a space after the last forward slash in the URL. Both issues are
corrected and the licenses now match the official textual description
of the license (and the other license in the sources).
Since these whitespaces changes do not alter the paragraph structure of
the license, nor create new sentences, they do not change the license.
The loop should be aligned to 32-bytes so that it can ideally run out
the DSB. This is particularly important on Skylake-Server where
deficiencies in it's DSB implementation make it prone to not being
able to run loops out of the DSB.
For example running strcmp-evex on 200Mb string:
32-byte aligned loop:
- 43,399,578,766 idq.dsb_uops
not 32-byte aligned loop:
- 6,060,139,704 idq.dsb_uops
This results in a 25% performance degradation for the non-aligned
version.
The fix is to just ensure the code layout is such that the loop is
aligned. (Which was previously the case but was accidentally dropped
in 84e7c46df).
NB: The fix was actually 64-byte alignment. This is because 64-byte
alignment generally produces more stable performance than 32-byte
aligned code (cache line crosses can affect perf), so if we are going
past 16-byte alignmnent, might as well go to 64. 64-byte alignment
also matches most other functions we over-align, so it creates a
common point of optimization.
Times are reported as ratio of Time_With_Patch /
Time_Without_Patch. Lower is better.
The values being reported is the geometric mean of the ratio across
all tests in bench-strcmp and bench-strncmp.
Note this patch is only attempting to improve the Skylake-Server
strcmp for long strings. The rest of the numbers are only to test for
regressions.
Tigerlake Results Strings <= 512:
strcmp : 1.026
strncmp: 0.949
Tigerlake Results Strings > 512:
strcmp : 0.994
strncmp: 0.998
Skylake-Server Results Strings <= 512:
strcmp : 0.945
strncmp: 0.943
Skylake-Server Results Strings > 512:
strcmp : 0.778
strncmp: 1.000
The 2.6% regression on TGL-strcmp is due to slowdowns caused by
changes in alignment of code handling small sizes (most on the
page-cross logic). These should be safe to ignore because 1) We
previously only 16-byte aligned the function so this behavior is not
new and was essentially up to chance before this patch and 2) this
type of alignment related regression on small sizes really only comes
up in tight micro-benchmark loops and is unlikely to have any affect
on realworld performance.
Reviewed-by: H.J. Lu <hjl.tools@gmail.com>
The current racy approach is to enable asynchronous cancellation
before making the syscall and restore the previous cancellation
type once the syscall returns, and check if cancellation has happen
during the cancellation entrypoint.
As described in BZ#12683, this approach shows 2 problems:
1. Cancellation can act after the syscall has returned from the
kernel, but before userspace saves the return value. It might
result in a resource leak if the syscall allocated a resource or a
side effect (partial read/write), and there is no way to program
handle it with cancellation handlers.
2. If a signal is handled while the thread is blocked at a cancellable
syscall, the entire signal handler runs with asynchronous
cancellation enabled. This can lead to issues if the signal
handler call functions which are async-signal-safe but not
async-cancel-safe.
For the cancellation to work correctly, there are 5 points at which the
cancellation signal could arrive:
[ ... )[ ... )[ syscall ]( ...
1 2 3 4 5
1. Before initial testcancel, e.g. [*... testcancel)
2. Between testcancel and syscall start, e.g. [testcancel...syscall start)
3. While syscall is blocked and no side effects have yet taken
place, e.g. [ syscall ]
4. Same as 3 but with side-effects having occurred (e.g. a partial
read or write).
5. After syscall end e.g. (syscall end...*]
And libc wants to act on cancellation in cases 1, 2, and 3 but not
in cases 4 or 5. For the 4 and 5 cases, the cancellation will eventually
happen in the next cancellable entrypoint without any further external
event.
The proposed solution for each case is:
1. Do a conditional branch based on whether the thread has received
a cancellation request;
2. It can be caught by the signal handler determining that the saved
program counter (from the ucontext_t) is in some address range
beginning just before the "testcancel" and ending with the
syscall instruction.
3. SIGCANCEL can be caught by the signal handler and determine that
the saved program counter (from the ucontext_t) is in the address
range beginning just before "testcancel" and ending with the first
uninterruptable (via a signal) syscall instruction that enters the
kernel.
4. In this case, except for certain syscalls that ALWAYS fail with
EINTR even for non-interrupting signals, the kernel will reset
the program counter to point at the syscall instruction during
signal handling, so that the syscall is restarted when the signal
handler returns. So, from the signal handler's standpoint, this
looks the same as case 2, and thus it's taken care of.
5. For syscalls with side-effects, the kernel cannot restart the
syscall; when it's interrupted by a signal, the kernel must cause
the syscall to return with whatever partial result is obtained
(e.g. partial read or write).
6. The saved program counter points just after the syscall
instruction, so the signal handler won't act on cancellation.
This is similar to 4. since the program counter is past the syscall
instruction.
So The proposed fixes are:
1. Remove the enable_asynccancel/disable_asynccancel function usage in
cancellable syscall definition and instead make them call a common
symbol that will check if cancellation is enabled (__syscall_cancel
at nptl/cancellation.c), call the arch-specific cancellable
entry-point (__syscall_cancel_arch), and cancel the thread when
required.
2. Provide an arch-specific generic system call wrapper function
that contains global markers. These markers will be used in
SIGCANCEL signal handler to check if the interruption has been
called in a valid syscall and if the syscalls has side-effects.
A reference implementation sysdeps/unix/sysv/linux/syscall_cancel.c
is provided. However, the markers may not be set on correct
expected places depending on how INTERNAL_SYSCALL_NCS is
implemented by the architecture. It is expected that all
architectures add an arch-specific implementation.
3. Rewrite SIGCANCEL asynchronous handler to check for both canceling
type and if current IP from signal handler falls between the global
markers and act accordingly.
4. Adjust libc code to replace LIBC_CANCEL_ASYNC/LIBC_CANCEL_RESET to
use the appropriate cancelable syscalls.
5. Adjust 'lowlevellock-futex.h' arch-specific implementations to
provide cancelable futex calls.
Some architectures require specific support on syscall handling:
* On i386 the syscall cancel bridge needs to use the old int80
instruction because the optimized vDSO symbol the resulting PC value
for an interrupted syscall points to an address outside the expected
markers in __syscall_cancel_arch. It has been discussed in LKML [1]
on how kernel could help userland to accomplish it, but afaik
discussion has stalled.
Also, sysenter should not be used directly by libc since its calling
convention is set by the kernel depending of the underlying x86 chip
(check kernel commit 30bfa7b3488bfb1bb75c9f50a5fcac1832970c60).
* mips o32 is the only kABI that requires 7 argument syscall, and to
avoid add a requirement on all architectures to support it, mips
support is added with extra internal defines.
Checked on aarch64-linux-gnu, arm-linux-gnueabihf, powerpc-linux-gnu,
powerpc64-linux-gnu, powerpc64le-linux-gnu, i686-linux-gnu, and
x86_64-linux-gnu.
[1] https://lkml.org/lkml/2016/3/8/1105
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
This commit uses a common implementation 'strnlen-evex-base.S' for both
'strnlen-evex' and 'strnlen-evex512'
This patch serves both to reduce the number of implementations, and it also does some small optimizations that benefit strnlen-evex and strnlen-evex512.
All tests pass on x86.
Benchmarks were taken on SKX.
https://www.intel.com/content/www/us/en/products/sku/123613/intel-core-i97900x-xseries-processor-13-75m-cache-up-to-4-30-ghz/specifications.html
Geometric mean for strnlen-evex over all benchmarks (N=10) was (new/old) 0.881
Geometric mean for strnlen-evex512 over all benchmarks (N=10) was (new/old) 0.953
Code Size Changes:
strnlen-evex : +31 bytes
strnlen-evex512 : +156 bytes
Reviewed-by: Noah Goldstein <goldstein.w.n@gmail.com>
The goal of this flag is to allow targets which don't prefer/have ERMS
to still access the non-temporal memset implementation.
There are 4 cases for tuning memset:
1) `Avoid_STOSB && Avoid_Non_Temporal_Memset`
- Memset with temporal stores
2) `Avoid_STOSB && !Avoid_Non_Temporal_Memset`
- Memset with temporal/non-temporal stores. Non-temporal path
goes through `rep stosb` path. We accomplish this by setting
`x86_rep_stosb_threshold` to
`x86_memset_non_temporal_threshold`.
3) `!Avoid_STOSB && Avoid_Non_Temporal_Memset`
- Memset with temporal stores/`rep stosb`
3) `!Avoid_STOSB && !Avoid_Non_Temporal_Memset`
- Memset with temporal stores/`rep stosb`/non-temporal stores.
Reviewed-by: H.J. Lu <hjl.tools@gmail.com>
Issue was we were expecting not matches with CHAR before the start of
the string in the page cross case.
The check code in the page cross case:
```
and $0xffffffffffffffc0,%rax
vmovdqa64 (%rax),%zmm17
vpcmpneqb %zmm17,%zmm16,%k1
vptestmb %zmm17,%zmm17,%k0{%k1}
kmovq %k0,%rax
inc %rax
shr %cl,%rax
je L(continue)
```
expects that all characters that neither match null nor CHAR will be
1s in `rax` prior to the `inc`. Then the `inc` will overflow all of
the 1s where no relevant match was found.
This is incorrect in the page-cross case, as the
`vmovdqa64 (%rax),%zmm17` loads from before the start of the input
string.
If there are matches with CHAR before the start of the string, `rax`
won't properly overflow.
The fix is quite simple. Just replace:
```
inc %rax
shr %cl,%rax
```
With:
```
sar %cl,%rax
inc %rax
```
The arithmetic shift will clear any matches prior to the start of the
string while maintaining the signbit so the 1s can properly overflow
to zero in the case of no matches.
Reviewed-by: H.J. Lu <hjl.tools@gmail.com>
These functions are exp10m1, exp2m1, log10p1, log2p1.
Also regenerated ulps on x86_64.
For each format, there are 4 values, one for each rounding mode.
(For the intel96 format, there are 8 values, 4 for Intel hardware,
and 4 for AMD hardware. However, regen-ulps was only run on Intel.
It should be run in a separate patch on a AMD x86_64.)
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
Remove sysdeps/x86_64/x32/dl-machine.h by folding x32 ARCH_LA_PLTENTER,
ARCH_LA_PLTEXIT and RTLD_START into sysdeps/x86_64/dl-machine.h. There
are no regressions on x86-64 nor x32. There are no changes in x86-64
_dl_start_user. On x32, _dl_start_user changes are
<_dl_start_user>:
mov %eax,%r12d
+ mov %esp,%r13d
mov (%rsp),%edx
mov %edx,%esi
- mov %esp,%r13d
and $0xfffffff0,%esp
mov 0x0(%rip),%edi # <_dl_start_user+0x14>
lea 0x8(%r13,%rdx,4),%ecx
Signed-off-by: H.J. Lu <hjl.tools@gmail.com>
Reviewed-by: Noah Goldstein <goldstein.w.n@gmail.com>
Xfail elf/tst-platform-1 on x32 since kernel passes i686 in AT_PLATFORM.
See https://sourceware.org/bugzilla/show_bug.cgi?id=22363
Signed-off-by: H.J. Lu <hjl.tools@gmail.com>
Reviewed-by: Sam James <sam@gentoo.org>
It turns out that quite a few applications use bundled mallocs that
have been built to use global-dynamic TLS (instead of the recommended
initial-exec TLS). The previous workaround from
commit afe42e935b ("elf: Avoid some
free (NULL) calls in _dl_update_slotinfo") does not fix all
encountered cases unfortunatelly.
This change avoids the TLS generation update for recursive use
of TLS from a malloc that was called during a TLS update. This
is possible because an interposed malloc has a fixed module ID and
TLS slot. (It cannot be unloaded.) If an initially-loaded module ID
is encountered in __tls_get_addr and the dynamic linker is already
in the middle of a TLS update, use the outdated DTV, thus avoiding
another call into malloc. It's still necessary to update the
DTV to the most recent generation, to get out of the slow path,
which is why the check for recursion is needed.
The bookkeeping is done using a global counter instead of per-thread
flag because TLS access in the dynamic linker is tricky.
All this will go away once the dynamic linker stops using malloc
for TLS, likely as part of a change that pre-allocates all TLS
during pthread_create/dlopen.
Fixes commit d2123d6827 ("elf: Fix slow
tls access after dlopen [BZ #19924]").
Reviewed-by: Szabolcs Nagy <szabolcs.nagy@arm.com>
This patch optimizes large size copy using normal store when src > dst
and overlap. Make it the same as the logic in memmove-vec-unaligned-erms.S.
Current memmove-ssse3 use '__x86_shared_cache_size_half' as the non-
temporal threshold, this patch updates that value to
'__x86_shared_non_temporal_threshold'. Currently, the
__x86_shared_non_temporal_threshold is cpu-specific, and different CPUs
will have different values based on the related nt-benchmark results.
However, in memmove-ssse3, the nontemporal threshold uses
'__x86_shared_cache_size_half', which sounds unreasonable.
The performance is not changed drastically although shows overall
improvements without any major regressions or gains.
Results on Zhaoxin KX-7000:
bench-memcpy geometric_mean(N=20) New / Original: 0.999
bench-memcpy-random geometric_mean(N=20) New / Original: 0.999
bench-memcpy-large geometric_mean(N=20) New / Original: 0.978
bench-memmove geometric_mean(N=20) New / Original: 1.000
bench-memmmove-large geometric_mean(N=20) New / Original: 0.962
Results on Intel Core i5-6600K:
bench-memcpy geometric_mean(N=20) New / Original: 1.001
bench-memcpy-random geometric_mean(N=20) New / Original: 0.999
bench-memcpy-large geometric_mean(N=20) New / Original: 1.001
bench-memmove geometric_mean(N=20) New / Original: 0.995
bench-memmmove-large geometric_mean(N=20) New / Original: 0.936
Reviewed-by: Noah Goldstein <goldstein.w.n@gmail.com>
Remove the definitions of _DL_PLATFORMS_COUNT as those are not used
anymore after removal in elf/dl-cache.c:search_cache().
Note: On x86, we can also get rid of the definitions
HWCAP_PLATFORMS_START and HWCAP_PLATFORMS_COUNT.
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
As discussed at the patch review meeting
Signed-off-by: Andreas K. Hüttel <dilfridge@gentoo.org>
Reviewed-by: Simon Chopin <simon.chopin@canonical.com>
C23 adds various <math.h> function families originally defined in TS
18661-4. Add the exp2m1 and exp10m1 functions (exp2(x)-1 and
exp10(x)-1, like expm1).
As with other such functions, these use type-generic templates that
could be replaced with faster and more accurate type-specific
implementations in future. Test inputs are copied from those for
expm1, plus some additions close to the overflow threshold (copied
from exp2 and exp10) and also some near the underflow threshold.
exp2m1 has the unusual property of having an input (M_MAX_EXP) where
whether the function overflows (under IEEE semantics) depends on the
rounding mode. Although these could reasonably be XFAILed in the
testsuite (as we do in some cases for arguments very close to a
function's overflow threshold when an error of a few ulps in the
implementation can result in the implementation not agreeing with an
ideal one on whether overflow takes place - the testsuite isn't smart
enough to handle this automatically), since these functions aren't
required to be correctly rounding, I made the implementation check for
and handle this case specially.
The Makefile ordering expected by lint-makefiles for the new functions
is a bit peculiar, but I implemented it in this patch so that the test
passes; I don't know why log2 also needed moving in one Makefile
variable setting when it didn't in my previous patches, but the
failure showed a different place was expected for that function as
well.
The powerpc64le IFUNC setup seems not to be as self-contained as one
might hope; it shouldn't be necessary to add IFUNCs for new functions
such as these simply to get them building, but without setting up
IFUNCs for the new functions, there were undefined references to
__GI___expm1f128 (that IFUNC machinery results in no such function
being defined, but doesn't stop include/math.h from doing the
redirection resulting in the exp2m1f128 and exp10m1f128
implementations expecting to call it).
Tested for x86_64 and x86, and with build-many-glibcs.py.
C23 adds various <math.h> function families originally defined in TS
18661-4. Add the log10p1 functions (log10(1+x): like log1p, but for
base-10 logarithms).
This is directly analogous to the log2p1 implementation (except that
whereas log2p1 has a smaller underflow range than log1p, log10p1 has a
larger underflow range). The test inputs are copied from those for
log1p and log2p1, plus a few more inputs in that wider underflow
range.
Tested for x86_64 and x86, and with build-many-glibcs.py.
C23 adds various <math.h> function families originally defined in TS
18661-4. Add the logp1 functions (aliases for log1p functions - the
name is intended to be more consistent with the new log2p1 and
log10p1, where clearly it would have been very confusing to name those
functions log21p and log101p). As aliases rather than new functions,
the content of this patch is somewhat different from those actually
adding new functions.
Tests are shared with log1p, so this patch *does* mechanically update
all affected libm-test-ulps files to expect the same errors for both
functions.
The vector versions of log1p on aarch64 and x86_64 are *not* updated
to have logp1 aliases (and thus there are no corresponding header,
tests, abilist or ulps changes for vector functions either). It would
be reasonable for such vector aliases and corresponding changes to
other files to be made separately. For now, the log1p tests instead
avoid testing logp1 in the vector case (a Makefile change is needed to
avoid problems with grep, used in generating the .c files for vector
function tests, matching more than one ALL_RM_TEST line in a file
testing multiple functions with the same inputs, when it assumes that
the .inc file only has a single such line).
Tested for x86_64 and x86, and with build-many-glibcs.py.
The tuning for non-temporal stores for memset vs memcpy is not always
the same. This includes both the exact value and whether non-temporal
stores are profitable at all for a given arch.
This patch add `x86_memset_non_temporal_threshold`. Currently we
disable non-temporal stores for non Intel vendors as the only
benchmarks showing its benefit have been on Intel hardware.
Reviewed-by: H.J. Lu <hjl.tools@gmail.com>
Previously we use `rep stosb` for all medium/large memsets. This is
notably worse than non-temporal stores for large (above a
few MBs) memsets.
See:
https://docs.google.com/spreadsheets/d/1opzukzvum4n6-RUVHTGddV6RjAEil4P2uMjjQGLbLcU/edit?usp=sharing
For data using different stategies for large memset on ICX and SKX.
Using non-temporal stores can be up to 3x faster on ICX and 2x faster
on SKX. Historically, these numbers would not have been so good
because of the zero-over-zero writeback optimization that `rep stosb`
is able to do. But, the zero-over-zero writeback optimization has been
removed as a potential side-channel attack, so there is no longer any
good reason to only rely on `rep stosb` for large memsets. On the flip
size, non-temporal writes can avoid data in their RFO requests saving
memory bandwidth.
All of the other changes to the file are to re-organize the
code-blocks to maintain "good" alignment given the new code added in
the `L(stosb_local)` case.
The results from running the GLIBC memset benchmarks on TGL-client for
N=20 runs:
Geometric Mean across the suite New / Old EXEX256: 0.979
Geometric Mean across the suite New / Old EXEX512: 0.979
Geometric Mean across the suite New / Old AVX2 : 0.986
Geometric Mean across the suite New / Old SSE2 : 0.979
Most of the cases are essentially unchanged, this is mostly to show
that adding the non-temporal case didn't add any regressions to the
other cases.
The results on the memset-large benchmark suite on TGL-client for N=20
runs:
Geometric Mean across the suite New / Old EXEX256: 0.926
Geometric Mean across the suite New / Old EXEX512: 0.925
Geometric Mean across the suite New / Old AVX2 : 0.928
Geometric Mean across the suite New / Old SSE2 : 0.924
So roughly a 7.5% speedup. This is lower than what we see on servers
(likely because clients typically have faster single-core bandwidth so
saving bandwidth on RFOs is less impactful), but still advantageous.
Full test-suite passes on x86_64 w/ and w/o multiarch.
Reviewed-by: H.J. Lu <hjl.tools@gmail.com>
A space is added before the left bracket of the x86_64 elf_machine_rela
function, in order to harmonize with the rest of the implementation of
the function and to make it easier to retrieve the function. The lines
where the function definition is located has been re-indented, as well
as its left curly bracket placed in the correct position.
Signed-off-by: Xin Wang <yw987194828@gmail.com>
Reviewed-by: H.J. Lu <hjl.tools@gmail.com>
C23 adds various <math.h> function families originally defined in TS
18661-4. Add the log2p1 functions (log2(1+x): like log1p, but for
base-2 logarithms).
This illustrates the intended structure of implementations of all
these function families: define them initially with a type-generic
template implementation. If someone wishes to add type-specific
implementations, it is likely such implementations can be both faster
and more accurate than the type-generic one and can then override it
for types for which they are implemented (adding benchmarks would be
desirable in such cases to demonstrate that a new implementation is
indeed faster).
The test inputs are copied from those for log1p. Note that these
changes make gen-auto-libm-tests depend on MPFR 4.2 (or later).
The bulk of the changes are fairly generic for any such new function.
(sysdeps/powerpc/nofpu/Makefile only needs changing for those
type-generic templates that use fabs.)
Tested for x86_64 and x86, and with build-many-glibcs.py.
This code expects the WCSCAT preprocessor macro to be predefined in case
the evex implementation of the function should be defined with a name
different from __wcsncat_evex. However, when glibc is built for
x86-64-v4 without multiarch support, sysdeps/x86_64/wcsncat.S defines
WCSNCAT variable instead of WCSCAT to build it as wcsncat. Rename the
variable to WCSNCAT, as it is actually a better naming choice for the
variable in this case.
Reported-by: Kenton Groombridge
Link: https://bugs.gentoo.org/921945
Fixes: 64b8b6516b ("x86: Add evex optimized functions for the wchar_t strcpy family")
Signed-off-by: Gabi Falk <gabifalk@gmx.com>
Reviewed-by: Sunil K Pandey <skpgkp2@gmail.com>
This reverts commit a1735e0aa8.
The test failure is a real valgrind bug that needs to be fixed before
valgrind is usable with a glibc that has been built with
CC="gcc -march=x86-64-v3". The proposed valgrind patch teaches
valgrind to replace ld.so strcmp with an unoptimized scalar
implementation, thus avoiding any AVX2-related problems.
Valgrind bug: <https://bugs.kde.org/show_bug.cgi?id=485487>
Reviewed-by: H.J. Lu <hjl.tools@gmail.com>
When -mapxf is used to build glibc, the resulting glibc will never run
on FMA4 machines. Exclude FMA4 IFUNC functions when -mapxf is used.
This requires GCC which defines __APX_F__ for -mapxf with commit:
1df56719bd8 x86: Define __APX_F__ for -mapxf
Reviewed-by: Sunil K Pandey <skpgkp2@gmail.com>
The implementations of trunc functions using x87 floating point (i386 and
x86_64 long double only) traps when FE_INEXACT is enabled. Although
this is a GNU extension outside the scope of the C standard, other
architectures that also support traps do not show this behavior.
The fix moves the implementation to a common one that holds any
exceptions with a 'fnclex' (libc_feholdexcept_setround_387).
Checked on x86_64-linux-gnu and i686-linux-gnu.
Reviewed-by: H.J. Lu <hjl.tools@gmail.com>
The implementations of floor functions using x87 floating point (i386 and
86_64 long double only) traps when FE_INEXACT is enabled. Although
this is a GNU extension outside the scope of the C standard, other
architectures that also support traps do not show this behavior.
The fix moves the implementation to a common one that holds any
exceptions with a 'fnclex' (libc_feholdexcept_setround_387).
Checked on x86_64-linux-gnu and i686-linux-gnu.
Reviewed-by: H.J. Lu <hjl.tools@gmail.com>
The implementations of ceil functions using x87 floating point (i386 and
x86_64 long double only) traps when FE_INEXACT is enabled. Although
this is a GNU extension outside the scope of the C standard, other
architectures that also support traps do not show this behavior.
The fix moves the implementation to a common one that holds any
exceptions with a 'fnclex' (libc_feholdexcept_setround_387).
Checked on x86_64-linux-gnu and i686-linux-gnu.
Reviewed-by: H.J. Lu <hjl.tools@gmail.com>
As indicated in a recent thread, this it is a simple brute-force
algorithm that checks the whole needle at a matching character pair
(and does so 1 byte at a time after the first 64 bytes of a needle).
Also it never skips ahead and thus can match at every haystack
position after trying to match all of the needle, which generic
implementation avoids.
As indicated by Wilco, a 4x larger needle and 16x larger haystack gives
a clear 65x slowdown both basic_strstr and __strstr_avx512:
"ifuncs": ["basic_strstr", "twoway_strstr", "__strstr_avx512",
"__strstr_sse2_unaligned", "__strstr_generic"],
{
"len_haystack": 65536,
"len_needle": 1024,
"align_haystack": 0,
"align_needle": 0,
"fail": 1,
"desc": "Difficult bruteforce needle",
"timings": [4.0948e+07, 15094.5, 3.20818e+07, 108558, 10839.2]
},
{
"len_haystack": 1048576,
"len_needle": 4096,
"align_haystack": 0,
"align_needle": 0,
"fail": 1,
"desc": "Difficult bruteforce needle",
"timings": [2.69767e+09, 100797, 2.08535e+09, 495706, 82666.9]
}
PS: I don't have an AVX512 capable machine to verify this issues, but
skimming through the code it does seems to follow what Wilco has
described.
Reviewed-by: Noah Goldstein <goldstein.w.n@gmail.com>
_dl_tlsdesc_dynamic preserves RDI, RSI and RBX before realigning stack.
After realigning stack, it saves RCX, RDX, R8, R9, R10 and R11. Define
TLSDESC_CALL_REGISTER_SAVE_AREA to allocate space for RDI, RSI and RBX
to avoid clobbering saved RDI, RSI and RBX values on stack by xsave to
STATE_SAVE_OFFSET(%rsp).
+==================+<- stack frame start aligned at 8 or 16 bytes
| |<- RDI saved in the red zone
| |<- RSI saved in the red zone
| |<- RBX saved in the red zone
| |<- paddings for stack realignment of 64 bytes
|------------------|<- xsave buffer end aligned at 64 bytes
| |<-
| |<-
| |<-
|------------------|<- xsave buffer start at STATE_SAVE_OFFSET(%rsp)
| |<- 8-byte padding for 64-byte alignment
| |<- 8-byte padding for 64-byte alignment
| |<- R11
| |<- R10
| |<- R9
| |<- R8
| |<- RDX
| |<- RCX
+==================+<- RSP aligned at 64 bytes
Define TLSDESC_CALL_REGISTER_SAVE_AREA, the total register save area size
for all integer registers by adding 24 to STATE_SAVE_OFFSET since RDI, RSI
and RBX are saved onto stack without adjusting stack pointer first, using
the red-zone. This fixes BZ #31501.
Reviewed-by: Sunil K Pandey <skpgkp2@gmail.com>
_dl_tlsdesc_dynamic should also preserve AMX registers which are
caller-saved. Add X86_XSTATE_TILECFG_ID and X86_XSTATE_TILEDATA_ID
to x86-64 TLSDESC_CALL_STATE_SAVE_MASK. Compute the AMX state size
and save it in xsave_state_full_size which is only used by
_dl_tlsdesc_dynamic_xsave and _dl_tlsdesc_dynamic_xsavec. This fixes
the AMX part of BZ #31372. Tested on AMX processor.
AMX test is enabled only for compilers with the fix for
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114098
GCC 14 and GCC 11/12/13 branches have the bug fix.
Reviewed-by: Sunil K Pandey <skpgkp2@gmail.com>
When strcmp-avx2.S is used as the default, elf/tst-valgrind-smoke fails
with
==1272761== Conditional jump or move depends on uninitialised value(s)
==1272761== at 0x4022C98: strcmp (strcmp-avx2.S:462)
==1272761== by 0x400B05B: _dl_name_match_p (dl-misc.c:75)
==1272761== by 0x40085F3: _dl_map_object (dl-load.c:1966)
==1272761== by 0x401AEA4: map_doit (rtld.c:644)
==1272761== by 0x4001488: _dl_catch_exception (dl-catch.c:237)
==1272761== by 0x40015AE: _dl_catch_error (dl-catch.c:256)
==1272761== by 0x401B38F: do_preload (rtld.c:816)
==1272761== by 0x401C116: handle_preload_list (rtld.c:892)
==1272761== by 0x401EDF5: dl_main (rtld.c:1842)
==1272761== by 0x401A79E: _dl_sysdep_start (dl-sysdep.c:140)
==1272761== by 0x401BEEE: _dl_start_final (rtld.c:494)
==1272761== by 0x401BEEE: _dl_start (rtld.c:581)
==1272761== by 0x401AD87: ??? (in */elf/ld.so)
The assembly codes are:
0x0000000004022c80 <+144>: vmovdqu 0x20(%rdi),%ymm0
0x0000000004022c85 <+149>: vpcmpeqb 0x20(%rsi),%ymm0,%ymm1
0x0000000004022c8a <+154>: vpcmpeqb %ymm0,%ymm15,%ymm2
0x0000000004022c8e <+158>: vpandn %ymm1,%ymm2,%ymm1
0x0000000004022c92 <+162>: vpmovmskb %ymm1,%ecx
0x0000000004022c96 <+166>: inc %ecx
=> 0x0000000004022c98 <+168>: jne 0x4022c32 <strcmp+66>
strcmp-avx2.S has 32-byte vector loads of strings which are shorter than
32 bytes:
(gdb) p (char *) ($rdi + 0x20)
$6 = 0x1ffeffea20 "memcheck-amd64-linux.so"
(gdb) p (char *) ($rsi + 0x20)
$7 = 0x4832640 "core-amd64-linux.so"
(gdb) call (int) strlen ((char *) ($rsi + 0x20))
$8 = 19
(gdb) call (int) strlen ((char *) ($rdi + 0x20))
$9 = 23
(gdb)
It triggers the valgrind error. The above code is safe since the loads
don't cross the page boundary. Update tst-valgrind-smoke.sh to accept
an optional suppression file and pass a suppression file to valgrind when
strcmp-avx2.S is the default implementation of strcmp.
Reviewed-by: Sunil K Pandey <skpgkp2@gmail.com>
When glibc is built with ISA level 3 or above enabled, SSE resolvers
aren't available and glibc fails to build:
ld: .../elf/librtld.os: in function `init_cpu_features':
.../elf/../sysdeps/x86/cpu-features.c:1200:(.text+0x1445f): undefined reference to `_dl_runtime_resolve_fxsave'
ld: .../elf/librtld.os: relocation R_X86_64_PC32 against undefined hidden symbol `_dl_runtime_resolve_fxsave' can not be used when making a shared object
/usr/local/bin/ld: final link failed: bad value
For ISA level 3 or above, don't use _dl_runtime_resolve_fxsave nor
_dl_tlsdesc_dynamic_fxsave.
This fixes BZ #31429.
Reviewed-by: Noah Goldstein <goldstein.w.n@gmail.com>
Compiler generates the following instruction sequence for GNU2 dynamic
TLS access:
leaq tls_var@TLSDESC(%rip), %rax
call *tls_var@TLSCALL(%rax)
or
leal tls_var@TLSDESC(%ebx), %eax
call *tls_var@TLSCALL(%eax)
CALL instruction is transparent to compiler which assumes all registers,
except for EFLAGS and RAX/EAX, are unchanged after CALL. When
_dl_tlsdesc_dynamic is called, it calls __tls_get_addr on the slow
path. __tls_get_addr is a normal function which doesn't preserve any
caller-saved registers. _dl_tlsdesc_dynamic saved and restored integer
caller-saved registers, but didn't preserve any other caller-saved
registers. Add _dl_tlsdesc_dynamic IFUNC functions for FNSAVE, FXSAVE,
XSAVE and XSAVEC to save and restore all caller-saved registers. This
fixes BZ #31372.
Add GLRO(dl_x86_64_runtime_resolve) with GLRO(dl_x86_tlsdesc_dynamic)
to optimize elf_machine_runtime_setup.
Reviewed-by: Noah Goldstein <goldstein.w.n@gmail.com>
When glibc is built with ISA level 3 or higher by default, the resulting
glibc binaries won't run on SSE or FMA4 processors. Exclude SSE, AVX and
FMA4 variants in libm multiarch when ISA level 3 or higher is enabled by
default.
When glibc is built with ISA level 2 enabled by default, only keep SSE4.1
variant.
Fixes BZ 31335.
NB: elf/tst-valgrind-smoke test fails with ISA level 4, because valgrind
doesn't support AVX512 instructions:
https://bugs.kde.org/show_bug.cgi?id=383010
Reviewed-by: H.J. Lu <hjl.tools@gmail.com>
Some shadow stack test scripts use the '==' operator with the 'test'
command to validate exit codes resulting in the following error:
sysdeps/x86_64/tst-shstk-legacy-1e.sh: 31: test: 139: unexpected operator
The '==' operator is invalid for the 'test' command, use '-eq' like the
previous call to 'test'.
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Reviewed-by: H.J. Lu <hjl.tools@gmail.com>
