Using 'mffs' instruction to read the Floating Point Status Control Register
(FPSCR) can force a processor flush in some cases, with undesirable
performance impact. If the values of the bits in the FPSCR which force the
flush are not needed, an instruction that is new to POWER9 (ISA version 3.0),
'mffsl' can be used instead.
Cases included: get_rounding_mode, fegetround, fegetmode, fegetexcept.
* sysdeps/powerpc/bits/fenvinline.h (__fegetround): Use
__fegetround_ISA300() or __fegetround_ISA2() as appropriate.
(__fegetround_ISA300) New.
(__fegetround_ISA2) New.
* sysdeps/powerpc/fpu_control.h (IS_ISA300): New.
(_FPU_MFFS): Move implementation...
(_FPU_GETCW): Here.
(_FPU_MFFSL): Move implementation....
(_FPU_GET_RC_ISA300): Here. New.
(_FPU_GET_RC): Use _FPU_GET_RC_ISA300() or _FPU_GETCW() as appropriate.
* sysdeps/powerpc/fpu/fenv_libc.h (fegetenv_status_ISA300): New.
(fegetenv_status): New.
* sysdeps/powerpc/fpu/fegetmode.c (fegetmode): Use fegetenv_status()
instead of fegetenv_register().
* sysdeps/powerpc/fpu/fegetexcept.c (__fegetexcept): Likewise.
Reviewed-by: Tulio Magno Quites Machado Filho <tuliom@linux.ibm.com>
Some benchmarks with a very short runtime show significantly
different results across runs on Aarch64 - up to tens of percents.
Increasing the runtime to 100ms+ makes the deviation under 5%.
Tested on Aarch64 and x86-64.
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
* benchtests/bench-memccpy.c: Replace INNER_LOOP_ITERS
with INNER_LOOP_ITERS_LARGE.
* benchtests/bench-memchr.c: Likewise.
* benchtests/bench-rawmemchr.c: Likewise.
* benchtests/bench-strcat.c: Likewise.
* benchtests/bench-strchr.c: Likewise.
* benchtests/bench-string.h: Likewise.
* benchtests/bench-strlen.c: Likewise.
* benchtests/bench-strncpy.c: Likewise.
* benchtests/bench-strnlen.c: Likewise.
This change should be fully backwards-compatible because the old
code aborted the load if a soname mismatch was encountered
(instead of searching further for a matching symbol). This means
that no different symbols are found.
The soname check was explicitly disabled for the skip_map != NULL
case. However, this only happens with dl(v)sym and RTLD_NEXT,
and those lookups do not come with a verneed entry that could be used
for the check.
The error check was already explicitly disabled for the skip_map !=
NULL case, that is, when dl(v)sym was called with RTLD_NEXT. But
_dl_vsym always sets filename in the struct r_found_version argument
to NULL, so the check was not active anyway. This means that
symbol lookup results for the skip_map != NULL case do not change,
either.
The kernel is evolving this interface (e.g., removal of the
restriction on cross-device copies), and keeping up with that
is difficult. Applications which need the function should
run kernels which support the system call instead of relying on
the imperfect glibc emulation.
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
On powerpc64le, long double can currently take two formats: the same as
double (-mlong-double-64) or IBM Extended Precision (default with
-mlong-double-128 or explicitly with -mabi=ibmlongdouble). The internal
implementation of printf-like functions is aware of these possibilities
and properly parses floating-point values from the variable arguments,
before making calls to __printf_fp and __printf_fphex. These functions
are also aware of the format possibilities and know how to convert both
formats to string.
When library support for TS 18661-3 was added to glibc, __printf_fp and
__printf_fphex were extended with support for an additional type
(__float128/_Float128) with a different format (binary128). Now that
powerpc64le is getting support for its third long double format, and
taking into account that this format is the same as the format of
__float128/_Float128, this patch extends __vfprintf_internal to properly
call __printf_fp and __printf_fphex with this new format.
Tested for powerpc64le (with additional patches to actually enable the
use of these preparations) and for x86_64.
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
The kernel interface uses type unsigned int, but there is an
internal conversion to int, so INT_MAX is the correct limit.
Part of the buffer will always be unused, but this is not a
problem. Such huge buffers do not occur in practice anyway.
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
Since sysdeps/i386/dl-lookupcfg.h and sysdeps/x86_64/dl-lookupcfg.h are
identical, we can replace them with sysdeps/x86/dl-lookupcfg.h.
* sysdeps/i386/dl-lookupcfg.h: Moved to ...
* sysdeps/x86/dl-lookupcfg.h: Here.
* sysdeps/x86_64/dl-lookupcfg.h: Removed.
The nds32 creates two specific syscalls, udftrap and fp_udfiex_crtl, in
kernel v5.0 and v5.2, respectively. Add these two syscalls to
syscall-names.list.
This patch fixes the gcc warnings seen with gcc 9 -march>=z13 on s390x:
programs/ld-ctype.c: In function ‘ctype_read’:
programs/ld-ctype.c:1392:13: error: ‘wch’ may be used uninitialized in this function [-Werror=maybe-uninitialized]
1392 | uint32_t wch;
| ^~~
programs/ld-ctype.c:1401:7: error: ‘seq’ may be used uninitialized in this function [-Werror=maybe-uninitialized]
1401 | if (seq != NULL && seq->nbytes == 1)
| ^
programs/ld-ctype.c:1391:20: note: ‘seq’ was declared here
1391 | struct charseq *seq;
| ^~~
Both seq and wch are uninitialized if get_character fails.
Thus we are now returning with an error.
ChangeLog:
* locale/programs/ld-ctype.c (charclass_symbolic_ellipsis):
Return error if get_character fails.
The Unicode sequences in the format <Uxxxx> should be used instead of
non-ASCII characters.
Reported by Piotr Drąg:
https://sourceware.org/bugzilla/show_bug.cgi?id=24652#c8
[BZ #24652]
* localedata/locales/szl_PL (day): Use the correct Unicode
sequences instead of non-ASCII characters.
This commit also provides the correct month names in both nominative
and genitive case for Silesian language, as required by the fix for
the bug 10871.
[BZ #24652]
* localedata/locales/szl_PL (abday): Spelling corrections.
(day): Likewise.
(abmon): Likewise.
(mon): Rename to...
(alt_mon): This, then apply spelling corrections.
(mon): New entry, month names in the genitive case.
According to CLDR 35.1 and the bug report the thousands grouping
separator should be always "." (a single dot) and digits should be
grouped by 3.
[BZ #23831]
* localedata/locales/nl_AW (mon_thousands_sep): Set to ".".
* localedata/locales/nl_NL (mon_thousands_sep): Likewise.
(thousands_sep): Likewise.
(grouping): Set to 3;3.
Define all currently used Linux versions used for
PREPARE_VERSION{,_KNOWN} in sysdeps/unix/sysv/linux/dl-vdso.h and use
them instead of duplicating the versions and precomputed hashes across
architecture specific files.
* sysdeps/unix/sysv/linux/aarch64/gettimeofday.c (INIT_ARCH): Use
PREPARE_VERSION_KNOWN.
* sysdeps/unix/sysv/linux/aarch64/init-first.c: Likewise.
* sysdeps/unix/sysv/linux/dl-vdso.h (VDSO_NAME_LINUX_2_6_39): New
define.
(VDSO_HASH_LINUX_2_6_39): Likewise.
(VDSO_NAME_LINUX_4_9): Likewise.
(VDSO_HASH_LINUX_4_9): Likewise.
* sysdeps/unix/sysv/linux/powerpc/gettimeofday.c (INIT_ARCH): Likewise.
* sysdeps/unix/sysv/linux/powerpc/init-first.c
(_libc_vdso_platform_setup): Likewise.
* sysdeps/unix/sysv/linux/powerpc/time.c (INIT_ARCH): Likewise.
* sysdeps/unix/sysv/linux/s390/init-first.c (_libc_vdso_platform_setup):
Likewise.
* sysdeps/unix/sysv/linux/x86_64/init-first.c (__vdso_platform_setup):
Likewise.
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
Commit a601b74d31 aka glibc-2.23~693
("In preparation for fixing BZ#16734, fix failure in misc/tst-error1-mem
when _G_HAVE_MMAP is turned off.") introduced a regression:
_IO_unbuffer_all now invokes _IO_wsetb to free wide buffers of all
files, including legacy standard files which are small statically
allocated objects that do not have wide buffers and the _mode member,
causing memory corruption.
Another memory corruption in _IO_unbuffer_all happens when -1
is assigned to the _mode member of legacy standard files that
do not have it.
[BZ #24228]
* libio/genops.c (_IO_unbuffer_all)
[SHLIB_COMPAT (libc, GLIBC_2_0, GLIBC_2_1)]: Do not attempt to free wide
buffers and access _IO_FILE_complete members of legacy libio streams.
* libio/tst-bz24228.c: New file.
* libio/tst-bz24228.map: Likewise.
* libio/Makefile [build-shared] (tests): Add tst-bz24228.
[build-shared] (generated): Add tst-bz24228.mtrace and
tst-bz24228.check.
[run-built-tests && build-shared] (tests-special): Add
$(objpfx)tst-bz24228-mem.out.
(LDFLAGS-tst-bz24228, tst-bz24228-ENV): New variables.
($(objpfx)tst-bz24228-mem.out): New rule.
Add 'volatile' keyword to a few asm statements, to force the compiler
to generate the instructions therein.
Some instances were implicitly volatile, but adding keyword for consistency.
2019-06-19 Paul A. Clarke <pc@us.ibm.com>
* sysdeps/powerpc/fpu/fenv_libc.h (relax_fenv_state): Add 'volatile'.
* sysdeps/powerpc/fpu/fpu_control.h (__FPU_MFFS): Likewise.
(__FPU_MFFSL): Likewise.
(_FPU_SETCW): Likewise.
__ppc_get_timebase_freq() always return 0 when using static linked
glibc.
This is a minimal example.c to reproduce:
/******************************/
#include <inttypes.h>
#include <stdint.h>
#include <stdio.h>
#include <sys/platform/ppc.h>
int main() {
uint64_t freq = __ppc_get_timebase_freq();
printf("Time Base frequency = %"PRIu64" Hz\n", freq);
if (freq == 0)
return -1;
return 0;
}
/******************************/
Compile command: gcc -static example.c
This bug has been reproduced, fixed and tested on all powerpc platforms
(ppc32, ppc64 and ppc64le).
The underlying code of __ppc_get_timebase_freq uses __get_timebase_freq
that has a different implementation for shared and static version of
glibc. In the static version, there is an incorrect sense in the if
check for the fd returned when opening /proc/cpuinfo.
This solution is mostly a cherry-pick from:
commit 4791e4f773d060c1a37b27aac5b03cdfa9327afc
Author: Stan Shebs <stanshebs@google.com>
Date: Fri May 17 12:25:19 2019 -0700
Subject: Fix sense of a test in the static-linking version of ppc get_clockfreq
That is in branch glibc/google/grte/v5-2.27/master and was mentioned for
inclusion on master here:
https://www.sourceware.org/ml/libc-alpha/2019-05/msg00409.html
Adapted from original fix for get_clockfreq. That code was moved to
get_timebase_freq.
Also added a static-build testcase for __ppc_get_timebase_freq since the
underlying function has different implementations for shared and static
build.
[BZ #24640]
* sysdeps/unix/sysv/linux/powerpc/get_timebase_freq.c
[!SHARED] (__get_timebase_freq): Fix sense of a test in the
static-linking version.
* sysdeps/unix/sysv/linux/powerpc/Makefile
(tests-static): Add test-gettimebasefreq-static.
(tests): Likewise.
* sysdeps/unix/sysv/linux/powerpc/test-gettimebasefreq-static.c:
New file.
Follow the same changes as made in the commit 02d8b5ab1c because the
respective entries in nl_NL and nl_AW had been the same before the change
so they should be the same after. CLDR does not provide complete data
for nl_AW, it says it is missing and displays a copy of nl_NL.
[BZ #24614]
* localedata/locales/nl_AW (n_sep_by_space): Set to 2 (a space
between the currency symbol and the minus sign).
(n_sign_posn): Set to 4 (the minus sign after the currency symbol).
This patch fixes the following gcc 9 warnings for "make xcheck" / "make bench":
-string/tst-strcasestr.c:
../include/bits/../../misc/bits/error.h:42:5: error: ‘%s’ directive argument is null [-Werror=format-overflow=]
-argp/argp-test.c:
argp-test.c:130:20: error: ‘%d’ directive writing between 1 and 11 bytes into a region of size 10 [-Werror=format-overflow=]
argp-test.c:130:19: note: directive argument in the range [-2147483648, 122]
argp-test.c:130:5: note: ‘sprintf’ output between 2 and 12 bytes into a destination of size 10
-nss/tst-field.c:
tst-field.c:52:7: error: ‘%s’ directive argument is null [-Werror=format-overflow=]
-benchtests/bench-strstr.c:
../include/bits/../../misc/bits/error.h:42:5: error: ‘%s’ directive argument is null [-Werror=format-overflow=]
-benchtests/bench-malloc-simple.c:
bench-malloc-simple.c:93:16: error: iteration 3 invokes undefined behavior [-Werror=aggressive-loop-optimizations]
ChangeLog:
[BZ #24556]
* string/test-strcasestr.c (check_result): Add NULL check.
* nss/tst-field.c (check_rewrite): Likewise.
* benchtests/bench-strstr.c (do_one_test): Likewise.
* string/test-strstr.c (check_result): Likewise.
* argp/argp-test.c (popt): Increase size of buf to 12.
* benchtests/bench-malloc-simple.c (bench):
Do not initialize tests array out of bounds.
The dls_serpath path field, as an array of length 1, introduces
unexpected array subscript checks with some compilers.
GCC versions before 3.0 treat the nested anonymous union as a
declaration of an unnamed type, and not as a member declaration,
so this construct cannot be used for these compilers.
According to CLDR 35.1 and the bug report the correct monetary format
for negative amounts should be "EUR -1 234,56" while previously it was
"EUR 1 234,56-".
This patch does not change the thousands (grouping) separator.
[BZ #24614]
* localedata/Makefile (LOCALES): Add nl_NL.UTF-8.
* localedata/locales/nl_NL (n_sep_by_space): Set to 2 (a space
between the currency symbol and the minus sign).
(n_sign_posn): Set to 4 (the minus sign after the currency symbol).
* localedata/tst-strfmon1.c (tests): Add test data for nl_NL.UTF-8.
Although defined in initial TLS/NPTL ABI for m68k and ColdFire [1], kernel
support was never pushed upstream. This patch removes the unused m68k
vDSO support.
Checked with a build against m68k and m68k-coldfire and some basic
tests on ARAnyM.
* sysdeps/unix/sysv/linux/m68k/Makefile (sysdep_routines,
sysdep-rtld-routines): Remove rules.
* sysdeps/unix/sysv/linux/m68k/Versions (libc) [GLIBC_PRIVATE]:
Remove __vdso_atomic_cmpxchg_32 and __vdso_atomic_barrier.
(ld) [GLIBC_PRIVATE]: __rtld___vdso_read_tp,
__rtld___vdso_atomic_cmpxchg_32, and __rtld___vdso_atomic_barrier.
* sysdeps/unix/sysv/linux/m68k/coldfire/atomic-machine.h
(atomic_compare_and_exchange_val_acq, atomic_full_barrier): Remove
vDSO path for SHARED.
* sysdeps/unix/sysv/linux/m68k/init-first.c: Remove file.
* sysdeps/unix/sysv/linux/m68k/libc-m68k-vdso.c: Likewise.
* sysdeps/unix/sysv/linux/m68k/m68k-helpers.S: Likewise.
* sysdeps/unix/sysv/linux/m68k/m68k-vdso.c: Likewise.
* sysdeps/unix/sysv/linux/m68k/m68k-vdso.h: Likewise.
* sysdeps/unix/sysv/linux/m68k/m68k-helpers.c: New file.
[1] https://lists.debian.org/debian-68k/2007/11/msg00071.html
This patches consolidates all the powerpc llrint{f} implementations on
the generic sysdeps/powerpc/fpu/s_llrint{f}.
The IFUNC support is also moved only to powerpc64 only, since for
powerpc64le generic implementation resulting in optimized code.
Checked on powerpc-linux-gnu (built without --with-cpu, with
--with-cpu=power4 and with --with-cpu=power5+ and --disable-multi-arch),
powerpc64-linux-gnu (built without --with-cp and with --with-cpu=power5+
and --disable-multi-arch).
* sysdeps/powerpc/powerpc64/be/fpu/multiarch/Makefile
(libm-sysdep_routines): Add s_llrint-power8, s_llrint-power6x, and
s_llrint-ppc64.
(CFLAGS-s_llrint-power8.c, CFLAGS-s_llrint-power6x.c): New rule.
* sysdeps/powerpc/powerpc64/be/fpu/multiarch/s_llrint-power6x.c: New
file.
* sysdeps/powerpc/powerpc64/be/fpu/multiarch/s_llrint-power8.c:
Likewise.
* sysdeps/powerpc/powerpc64/be/fpu/multiarch/s_llrint-ppc64.c:
Likewise.
* sysdeps/powerpc/powerpc64/fpu/multiarch/s_lrint.c: Move to ...
* sysdeps/powerpc/powerpc64/be/fpu/multiarch/s_lrint.c: ... here.
* sysdeps/powerpc/powerpc64/fpu/multiarch/s_llrint.c: Move to ...
* sysdeps/powerpc/powerpc64/be/fpu/multiarch/s_llrint.c: ... here.
* sysdeps/powerpc/powerpc64/fpu/multiarch/s_llrintf.c: Move to ...
* sysdeps/powerpc/powerpc64/be/fpu/multiarch/s_llrintf.c: ... here.
* sysdeps/powerpc/powerpc64/be/fpu/multiarch/s_lrint.c: New file.
* sysdeps/powerpc/powerpc64/fpu/Makefile: Likewise.
* sysdeps/powerpc/powerpc64/fpu/multiarch/Makefile
(libm-sysdep_routines): Remove s_llrint-* objects.
* sysdeps/powerpc/powerpc64/fpu/multiarch/s_llrint-power6x.S: Remove
file.
* sysdeps/powerpc/powerpc64/fpu/multiarch/s_llrint-power8.S:
Likewise.
* sysdeps/powerpc/powerpc64/fpu/multiarch/s_llrint-ppc64.S: Likewise.
* sysdeps/powerpc/powerpc64/fpu/s_llrint.c: New file.
* sysdeps/powerpc/powerpc64/fpu/s_llrintf.c: Likewise.
* sysdeps/powerpc/powerpc64/fpu/s_lrint.c: Likewise.
* sysdeps/powerpc/powerpc64/fpu/s_lrintf.c: Likewise.
* sysdeps/powerpc/powerpc64/fpu/s_llrint.S: Remove file.
* sysdeps/powerpc/powerpc64/fpu/s_llrintf.S: Likewise.
* sysdeps/powerpc/powerpc64/fpu/s_lrint.S: Likewise.
* sysdeps/powerpc/powerpc64/power6x/fpu/s_llrint.S: Likewise.
* sysdeps/powerpc/powerpc64/power8/fpu/s_llrint.S: Likewise.
Reviewed-by: Gabriel F. T. Gomes <gabrielftg@linux.ibm.com>
The identifier linux is used as a predefined macro, so the actually
used path is 1/stat.h or 1/stat64.h. Using the quote-based version
triggers a file lookup for /usr/include/bits/linux/stat.h (or whatever
directory is used to store bits/statx.h), but since bits/ is pretty
much reserved by glibc, this appears to be acceptable.
This is related to GCC PR 80005: incorrect macro expansion of the
argument of __has_include.
Suggested by Zack Weinberg.
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
This is currently ineffective with GCC because of GCC PR 80005, but
it makes sense to anticipate a fix for this defect.
Suggested by Zack Weinberg.
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
This patch adds the new constant IPV6_ROUTER_ALERT_ISOLATE from Linux
5.1 to sysdeps/unix/sysv/linux/bits/in.h.
Tested for x86_64.
* sysdeps/unix/sysv/linux/bits/in.h (IPV6_ROUTER_ALERT_ISOLATE):
New macro.
Some recent change on GCC mainline resulted in the localplt test
failing for powerpc soft-float (not sure exactly when, as the failure
appeared when there were other build test failures as well;
<https://sourceware.org/ml/libc-testresults/2019-q2/msg00261.html>
shows it remaining when other failures went away). The problem is a
call to memset that GCC now generates in the libgcc long double code.
Since memset is documented as a function GCC may always implicitly
generate calls to, it seems reasonable to allow that local PLT
reference (just like those for libgcc functions that GCC implicitly
generates calls to and that are also exported from libc.so), which
this patch does.
Tested for powerpc soft-float with build-many-glibcs.py.
* sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/localplt.data:
Allow memset in libc.so.
Avoid lazy binding of symbols that may follow a variant PCS with different
register usage convention from the base PCS.
Currently the lazy binding entry code does not preserve all the registers
required for AdvSIMD and SVE vector calls. Saving and restoring all
registers unconditionally may break existing binaries, even if they never
use vector calls, because of the larger stack requirement for lazy
resolution, which can be significant on an SVE system.
The solution is to mark all symbols in the symbol table that may follow
a variant PCS so the dynamic linker can handle them specially. In this
patch such symbols are always resolved at load time, not lazily.
So currently LD_AUDIT for variant PCS symbols are not supported, for that
the _dl_runtime_profile entry needs to be changed e.g. to unconditionally
save/restore all registers (but pass down arg and retval registers to
pltentry/exit callbacks according to the base PCS).
This patch also removes a __builtin_expect from the modified code because
the branch prediction hint did not seem useful.
* sysdeps/aarch64/dl-dtprocnum.h: New file.
* sysdeps/aarch64/dl-machine.h (DT_AARCH64): Define.
(elf_machine_runtime_setup): Handle DT_AARCH64_VARIANT_PCS.
(elf_machine_lazy_rel): Check STO_AARCH64_VARIANT_PCS and bind such
symbols at load time.
* sysdeps/aarch64/linkmap.h (struct link_map_machine): Add variant_pcs.
STO_AARCH64_VARIANT_PCS is a non-visibility st_other flag for marking
symbols that reference functions that may follow a variant PCS with
different register usage convention from the base PCS.
DT_AARCH64_VARIANT_PCS is a dynamic tag that marks ELF modules that
have R_*_JUMP_SLOT relocations for symbols marked with
STO_AARCH64_VARIANT_PCS (i.e. have variant PCS calls via a PLT).
* elf/elf.h (STO_AARCH64_VARIANT_PCS): Define.
(DT_AARCH64_VARIANT_PCS): Define.
The powerpc finite optimization do not show much gain:
- GCC will call libm iff -fsignaling-nans is used. This usage pattern
is usually not performance oriented and for such calls PLT overhead
should dominate execution time.
- The power7 uses ftdiv to optimize for some input patterns, but at
cost of others. Comparing against generic C implementation built
for powerpc64-linux-gnu-power7 (--with-cpu=power7):
- Generic sysdeps/ieee754 implementation:
"isfinite": {
"": {
"duration": 5.0082e+09,
"iterations": 2.45299e+09,
"max": 43.824,
"min": 2.008,
"mean": 2.04167
},
"INF": {
"duration": 4.66554e+09,
"iterations": 2.28288e+09,
"max": 35.73,
"min": 2.008,
"mean": 2.04371
},
"NAN": {
"duration": 4.66274e+09,
"iterations": 2.28716e+09,
"max": 34.161,
"min": 2.009,
"mean": 2.03866
}
}
- power7 optimized one:
"isfinite": {
"": {
"duration": 4.99111e+09,
"iterations": 2.65566e+09,
"max": 25.015,
"min": 1.716,
"mean": 1.87942
},
"INF": {
"duration": 4.6783e+09,
"iterations": 2.0999e+09,
"max": 35.264,
"min": 1.868,
"mean": 2.22787
},
"NAN": {
"duration": 4.67915e+09,
"iterations": 2.08678e+09,
"max": 38.099,
"min": 1.869,
"mean": 2.24228
}
}
So it basically optimizes marginally for normal numbers while
increasing the latency for other kind of FP.
- The power8 implementation is just the generic implementation using
ISA 2.07 mfvsrd instruction (which GCC uses for generic implementation).
So generic implementation is the best option for powerpc64le.
Checked on powerpc-linux-gnu (built without --with-cpu, with
--with-cpu=power4 and with --with-cpu=power5+ and --disable-multi-arch),
powerpc64-linux-gnu (built without --with-cp and with --with-cpu=power5+
and --disable-multi-arch).
* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/Makefile
(sysdeps_routines, libm-sysdep_routines): Remove s_finite*
objects.
* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_finite-power7.S:
Remove file.
* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_finite-ppc32.c:
Likewise.
* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_finite.c: Likewise.
* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_finitef-ppc32.c:
Likewise.
* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_finitef.c: Likewise.
* sysdeps/powerpc/powerpc32/power7/fpu/s_finite.S: Likewise.
* sysdeps/powerpc/powerpc32/power7/fpu/s_finitef.S: Likewise.
* sysdeps/powerpc/powerpc64/fpu/multiarch/Makefile (sysdep_call):
Remove s_finite* objects.
* sysdeps/powerpc/powerpc64/fpu/multiarch/s_finite-power7.S: Remove file.
* sysdeps/powerpc/powerpc64/fpu/multiarch/s_finite-power8.S: Likewise.
* sysdeps/powerpc/powerpc64/fpu/multiarch/s_finite-ppc64.c: Likewise.
* sysdeps/powerpc/powerpc64/fpu/multiarch/s_finite.c: Likewise.
* sysdeps/powerpc/powerpc64/fpu/multiarch/s_finitef-ppc64.c: Likewise.
* sysdeps/powerpc/powerpc64/fpu/multiarch/s_finitef.c: Likewise.
* sysdeps/powerpc/powerpc64/power7/fpu/s_finite.S: Likewise.
* sysdeps/powerpc/powerpc64/power7/fpu/s_finitef.S: Likewise.
* sysdeps/powerpc/powerpc64/power8/fpu/s_finite.S: Likewise.
* sysdeps/powerpc/powerpc64/power8/fpu/s_finitef.S: Likewise.
Reviewed-by: Gabriel F. T. Gomes <gabrielftg@linux.ibm.com>
- math.h will use compiler builtin for gcc 4.4 when built without
-fsignaling-nans and the builtin is expanded inline for all
support architectures. As an example, there is no intra finite
call on libm for the architecture I checked, x86, arm, aarch64,
and powerpc.
- The resulting binary difference on 32 bits architecture is minimum
for the non hotspot symbol.
- It helps wordsize-64 architectures that use ldbl-opt.
- It add some code simplification with reduction of duplicated
implementations.
Checked on powerpc-linux-gnu (built without --with-cpu, with
--with-cpu=power4 and with --with-cpu=power5+ and --disable-multi-arch),
powerpc64-linux-gnu (built without --with-cp and with --with-cpu=power5+
and --disable-multi-arch).
* sysdeps/ieee754/dbl-64/wordsize-64/s_finite.c: Move to ...
* sysdeps/ieee754/dbl-64/s_finite.c: ... here and format code.
Reviewed-by: Gabriel F. T. Gomes <gabrielftg@linux.ibm.com>
The powerpc isinf optimizations onyl adds complexity:
- GCC will call libm iff -fsignaling-nans is used. This usage pattern
is usually not performance oriented and for such calls PLT overhead
should dominate execution time.
- The power7 uses ftdiv to optimize for some input pattern and branch
implementation for INF and denormal that does:
return (ix & UINT64_C (0x7fffffffffffffff)) == UINT64_C (0x7ff0000000000000)
Although it does show slight better latency than generic algorithm
(as below), it is only for power7 and requires it to override it
for power8.
- The power8 implementation is just the generic implementation using
ISA 2.07 mfvsrd instruction (which GCC uses for generic implementation).
So generic implementation is the best option for powerpc64le.
Checked on powerpc-linux-gnu (built without --with-cpu, with
--with-cpu=power4 and with --with-cpu=power5+ and --disable-multi-arch),
powerpc64-linux-gnu (built without --with-cp and with --with-cpu=power5+
and --disable-multi-arch).
* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/Makefile
(sysdeps_routines, libm-sysdep_routines): Remove s_isinf* and s_isinf*
objects.
* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_isinf-power7.S:
Remove file.
* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_isinf-ppc32.c:
Likewise.
* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_isinf.c: Likewise.
* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_isinff-ppc32.c:
Likewise.
* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_isinff.c: Likewise.
* sysdeps/powerpc/powerpc32/power7/fpu/s_isinf.S: Likewise.
* sysdeps/powerpc/powerpc32/power7/fpu/s_isinff.S: Likewise.
* sysdeps/powerpc/powerpc64/fpu/multiarch/Makefile (sysdep_call):
Remove s_isinf* and s_isinf* objects.
* sysdeps/powerpc/powerpc64/fpu/multiarch/s_isinf-power7.S: Likewise.
* sysdeps/powerpc/powerpc64/fpu/multiarch/s_isinf-power8.S: Likewise.
* sysdeps/powerpc/powerpc64/fpu/multiarch/s_isinf-ppc64.c: Likewise.
* sysdeps/powerpc/powerpc64/fpu/multiarch/s_isinf.c: Likewise.
* sysdeps/powerpc/powerpc64/fpu/multiarch/s_isinff-ppc64.c: Likewise.
* sysdeps/powerpc/powerpc64/fpu/multiarch/s_isinff.c: Likewise.
* sysdeps/powerpc/powerpc64/power7/fpu/s_isinf.S: Likewise.
* sysdeps/powerpc/powerpc64/power7/fpu/s_isinff.S: Likewise.
* sysdeps/powerpc/powerpc64/power8/fpu/s_isinf.S: Likewise.
* sysdeps/powerpc/powerpc64/power8/fpu/s_isinff.S: Likewise.
Reviewed-by: Gabriel F. T. Gomes <gabrielftg@linux.ibm.com>
- math.h will use compiler builtin for gcc 4.4 when built without
-fsignaling-nans and the builtin is expanded inline for all
support architectures. As an example, there is no intra isinf
call on libm for the architecture I checked, x86, arm, aarch64,
and powerpc.
- The resulting binary difference on 32 bits architecture is minimum
for the non hotspot symbol.
- It helps wordsize-64 architectures that use ldbl-opt.
- It add some code simplification with reduction of duplicated
implementations.
Checked on powerpc-linux-gnu (built without --with-cpu, with
--with-cpu=power4 and with --with-cpu=power5+ and --disable-multi-arch),
powerpc64-linux-gnu (built without --with-cp and with --with-cpu=power5+
and --disable-multi-arch).
* sysdeps/ieee754/dbl-64/wordsize-64/s_isinf.c: Move to ...
* sysdeps/ieee754/dbl-64/s_isinf.c: ... here and format code.
Reviewed-by: Gabriel F. T. Gomes <gabrielftg@linux.ibm.com>
The powerpc isnan optimizations are not really a gain:
- GCC will call libm iff -fsignaling-nans is used. This usage pattern
is usually not performance oriented and for such calls PLT overhead
should dominate execution time.
- The power5, power6, and power6x are just micro-optimization to
improve the Load-Hit-Store hazards from floating-point to general
register transfer, and current GCC already has support to minimize
it by inserting either extra nops or group dispatch instructions.
- The power7 uses ftdiv to optimize for some input patterns, but at
cost of others. Comparing against generic C implementation built
for powerpc-linux-gnu-power4 (which uses the hp-timing support on
benchtests):
- Generic sysdeps/ieee754 implementation:
"isnan": {
"": {
"duration": 4.98415e+09,
"iterations": 2.34516e+09,
"max": 45.925,
"min": 2.052,
"mean": 2.12529
},
"INF": {
"duration": 4.74057e+09,
"iterations": 1.69761e+09,
"max": 91.01,
"min": 2.052,
"mean": 2.79249
},
"NAN": {
"duration": 4.74071e+09,
"iterations": 1.68768e+09,
"max": 282.343,
"min": 2.052,
"mean": 2.809
}
}
- power7 optimized one:
$ ./testrun.sh benchtests/bench-isnan
"isnan": {
"": {
"duration": 4.96842e+09,
"iterations": 2.56297e+09,
"max": 50.048,
"min": 1.872,
"mean": 1.93854
},
"INF": {
"duration": 4.76648e+09,
"iterations": 1.54213e+09,
"max": 373.408,
"min": 2.661,
"mean": 3.09084
},
"NAN": {
"duration": 4.76845e+09,
"iterations": 1.54515e+09,
"max": 51.016,
"min": 2.736,
"mean": 3.08607
}
}
So it basically optimizes marginally for normal numbers while
increasing the latency for other kind of FP.
- The generic implementation requires getting the floating point
status, disable the invalid operation bit, and restore the
floating-point status. Each operation is costly and requires
flushing the FP pipeline.
Using the same scenarion for the previous analysis:
"isnan": {
"": {
"duration": 5.08284e+09,
"iterations": 6.2898e+08,
"max": 41.844,
"min": 8.057,
"mean": 8.08108
},
"INF": {
"duration": 4.97904e+09,
"iterations": 6.16176e+08,
"max": 39.661,
"min": 8.057,
"mean": 8.08055
},
"NAN": {
"duration": 4.98695e+09,
"iterations": 5.95866e+08,
"max": 29.728,
"min": 8.345,
"mean": 8.36925
}
}
- The power8 implementation is just the generic implementation using
ISA 2.07 mfvsrd instruction (which GCC uses for generic implementation).
So generic implementation is the best option for powerpc64le.
Checked on powerpc-linux-gnu (built without --with-cpu, with
--with-cpu=power4 and with --with-cpu=power5+ and --disable-multi-arch),
powerpc64-linux-gnu (built without --with-cp and with --with-cpu=power5+
and --disable-multi-arch).
* sysdeps/powerpc/fpu/s_isnan.c: Remove file.
* sysdeps/powerpc/fpu/s_isnanf.S: Likewise.
* sysdeps/powerpc/powerpc32/fpu/s_isnan.S: Likewise.
* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/Makefile
(sysdeps_routines, libm-sysdep_routines): Remove s_isnan-* and
s_isnanf-* objects.
* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_isnan-power5.S:
Remove file
* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_isnan-power6.S:
Likewise.
* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_isnan-power7.S:
Likewise.
* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_isnan-ppc32.S:
Likewise.
* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_isnan.c: Likewise.
* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_isnanf-power5.S:
Likewise.
* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_isnanf-power6.S:
Likewise.
* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_isnanf.c: Likewise.
* sysdeps/powerpc/powerpc32/power5/fpu/s_isnan.S: Likewise.
* sysdeps/powerpc/powerpc32/power5/fpu/s_isnanf.S: Likewise.
* sysdeps/powerpc/powerpc32/power6/fpu/s_isnan.S: Likewise.
* sysdeps/powerpc/powerpc32/power6/fpu/s_isnanf.S: Likewise.
* sysdeps/powerpc/powerpc32/power7/fpu/s_isnan.S: Likewise.
* sysdeps/powerpc/powerpc32/power7/fpu/s_isnanf.S: Likewise.
* sysdeps/powerpc/powerpc64/fpu/multiarch/Makefile (sysdep_calls):
Remove s_isnan-* and s_isnanf-* objects.
* sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan-power5.S: Likewise.
* sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan-power6.S: Likewise.
* sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan-power6x.S:
Likewise.
* sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan-power7.S: Likewise.
* sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan-power8.S: Likewise.
* sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan-ppc64.S: Likewise.
* sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan.c: Likewise.
* sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnanf.c: Likewise.
* sysdeps/powerpc/powerpc64/fpu/s_isnan.S: Likewise.
* sysdeps/powerpc/powerpc64/power5/fpu/s_isnan.S: Likewise.
* sysdeps/powerpc/powerpc64/power6/fpu/s_isnan.S: Likewise.
* sysdeps/powerpc/powerpc64/power6x/fpu/s_isnan.S: Likewise.
* sysdeps/powerpc/powerpc64/power7/fpu/s_isnan.S: Likewise.
* sysdeps/powerpc/powerpc64/power7/fpu/s_isnanf.S: Likewise.
* sysdeps/powerpc/powerpc64/power8/fpu/s_isnan.S: Likewise.
* sysdeps/powerpc/powerpc64/power8/fpu/s_isnanf.S: Likewise.
Reviewed-by: Gabriel F. T. Gomes <gabrielftg@linux.ibm.com>
- math.h will use compiler builtin for gcc 4.4 when built without
-fsignaling-nans and the builtin is expanded inline for all
support architectures. As an example, there is no intra isnan
call on libm for the architecture I checked, x86, arm, aarch64,
and powerpc.
- The resulting binary difference on 32 bits architecture is minimum
for the non hotspot symbol.
- It helps wordsize-64 architectures that use ldbl-opt.
- It add some code simplification with reduction of duplicated
implementations.
Checked on powerpc-linux-gnu (built without --with-cpu, with
--with-cpu=power4 and with --with-cpu=power5+ and --disable-multi-arch),
powerpc64-linux-gnu (built without --with-cp and with --with-cpu=power5+
and --disable-multi-arch).
* sysdeps/ieee754/dbl-64/wordsize-64/s_isnan.c: Move to ...
* sysdeps/ieee754/dbl-64/s_isnan.c: ... here and format code.
Reviewed-by: Gabriel F. T. Gomes <gabrielftg@linux.ibm.com>
* benchtests/Makefile (bench-math): Add isnan, isinf, and isfinite.
(CFLAGS-bench-isnan.c, CFLAGS-bench-isinf.c,
CFLAGS-bench-isfinite.c): New rule.
* benchtests/isnan-input: New file.
* benchtests/isinf-input: New file.
* benchtests/isfinite-input: New file.
Reviewed-by: Gabriel F. T. Gomes <gabrielftg@linux.ibm.com>
