mirror of
https://sourceware.org/git/glibc.git
synced 2024-11-22 13:00:06 +00:00
0edbf12301
Current allocate_stack logic for create stacks is to first mmap all
the required memory with the desirable memory and then mprotect the
guard area with PROT_NONE if required. Although it works as expected,
it pessimizes the allocation because it requires the kernel to actually
increase commit charge (it counts against the available physical/swap
memory available for the system).
The only issue is to actually check this change since side-effects are
really Linux specific and to actually account them it would require a
kernel specific tests to parse the system wide information. On the kernel
I checked /proc/self/statm does not show any meaningful difference for
vmm and/or rss before and after thread creation. I could only see
really meaningful information checking on system wide /proc/meminfo
between thread creation: MemFree, MemAvailable, and Committed_AS shows
large difference without the patch. I think trying to use these
kind of information on a testcase is fragile.
The BZ#18988 reports shows that the commit pages are easily seen with
mlockall (MCL_FUTURE) (with lock all pages that become mapped in the
process) however a more straighfoward testcase shows that pthread_create
could be faster using this patch:
--
static const int inner_count = 256;
static const int outer_count = 128;
static
void *thread1(void *arg)
{
return NULL;
}
static
void *sleeper(void *arg)
{
pthread_t ts[inner_count];
for (int i = 0; i < inner_count; i++)
pthread_create (&ts[i], &a, thread1, NULL);
for (int i = 0; i < inner_count; i++)
pthread_join (ts[i], NULL);
return NULL;
}
int main(void)
{
pthread_attr_init(&a);
pthread_attr_setguardsize(&a, 1<<20);
pthread_attr_setstacksize(&a, 1134592);
pthread_t ts[outer_count];
for (int i = 0; i < outer_count; i++)
pthread_create(&ts[i], &a, sleeper, NULL);
for (int i = 0; i < outer_count; i++)
pthread_join(ts[i], NULL);
assert(r == 0);
}
return 0;
}
--
On x86_64 (4.4.0-45-generic, gcc 5.4.0) running the small benchtests
I see:
$ time ./test
real 0m3.647s
user 0m0.080s
sys 0m11.836s
While with the patch I see:
$ time ./test
real 0m0.696s
user 0m0.040s
sys 0m1.152s
So I added a pthread_create benchtest (thread_create) which check
the thread creation latency. As for the simple benchtests, I saw
improvements in thread creation on all architectures I tested the
change.
Checked on x86_64-linux-gnu, i686-linux-gnu, aarch64-linux-gnu,
arm-linux-gnueabihf, powerpc64le-linux-gnu, sparc64-linux-gnu,
and sparcv9-linux-gnu.
[BZ #18988]
* benchtests/thread_create-inputs: New file.
* benchtests/thread_create-source.c: Likewise.
* support/xpthread_attr_setguardsize.c: Likewise.
* support/Makefile (libsupport-routines): Add
xpthread_attr_setguardsize object.
* support/xthread.h: Add xpthread_attr_setguardsize prototype.
* benchtests/Makefile (bench-pthread): Add thread_create.
* nptl/allocatestack.c (allocate_stack): Call mmap with PROT_NONE and
then mprotect the required area.
217 lines
6.9 KiB
Makefile
217 lines
6.9 KiB
Makefile
# Copyright (C) 2013-2017 Free Software Foundation, Inc.
|
||
# This file is part of the GNU C Library.
|
||
|
||
# The GNU C Library is free software; you can redistribute it and/or
|
||
# modify it under the terms of the GNU Lesser General Public
|
||
# License as published by the Free Software Foundation; either
|
||
# version 2.1 of the License, or (at your option) any later version.
|
||
|
||
# The GNU C Library is distributed in the hope that it will be useful,
|
||
# but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
||
# Lesser General Public License for more details.
|
||
|
||
# You should have received a copy of the GNU Lesser General Public
|
||
# License along with the GNU C Library; if not, see
|
||
# <http://www.gnu.org/licenses/>.
|
||
|
||
# Makefile for benchmark tests. The only useful target here is `bench`.
|
||
# Add benchmark functions in alphabetical order.
|
||
|
||
subdir := benchtests
|
||
|
||
include ../Makeconfig
|
||
bench-math := acos acosh asin asinh atan atanh cos cosh exp exp2 log log2 \
|
||
modf pow rint sin sincos sinh sqrt tan tanh fmin fmax fminf \
|
||
fmaxf
|
||
|
||
bench-pthread := pthread_once thread_create
|
||
|
||
bench-string := ffs ffsll
|
||
|
||
bench := $(bench-math) $(bench-pthread) $(bench-string)
|
||
|
||
# String function benchmarks.
|
||
string-benchset := bcopy bzero memccpy memchr memcmp memcpy memmem memmove \
|
||
mempcpy memset rawmemchr stpcpy stpncpy strcasecmp strcasestr \
|
||
strcat strchr strchrnul strcmp strcpy strcspn strlen \
|
||
strncasecmp strncat strncmp strncpy strnlen strpbrk strrchr \
|
||
strspn strstr strcpy_chk stpcpy_chk memrchr strsep strtok \
|
||
strcoll memcpy-large memcpy-random memmove-large memset-large
|
||
|
||
# Build and run locale-dependent benchmarks only if we're building natively.
|
||
ifeq (no,$(cross-compiling))
|
||
wcsmbs-benchset := wcslen wcsnlen wcscpy wcpcpy wcsncpy wcpncpy wcscat wcsncat \
|
||
wcscmp wcsncmp wcschr wcschrnul wcsrchr wcsspn wcspbrk wcscspn \
|
||
wmemchr wmemset wmemcmp
|
||
else
|
||
wcsmbs-benchset :=
|
||
endif
|
||
|
||
string-benchset-all := $(string-benchset) ${wcsmbs-benchset}
|
||
|
||
ifeq (no,$(cross-compiling))
|
||
# We have to generate locales
|
||
LOCALES := en_US.UTF-8 tr_TR.UTF-8 cs_CZ.UTF-8 fa_IR.UTF-8 fr_FR.UTF-8 \
|
||
ja_JP.UTF-8 si_LK.UTF-8 en_GB.UTF-8 vi_VN.UTF-8 ar_SA.UTF-8 \
|
||
da_DK.UTF-8 pl_PL.UTF-8 pt_PT.UTF-8 el_GR.UTF-8 ru_RU.UTF-8 \
|
||
he_IL.UTF-8 is_IS.UTF-8 es_ES.UTF-8 hi_IN.UTF-8 sv_SE.UTF-8 \
|
||
hu_HU.UTF-8 it_IT.UTF-8 sr_RS.UTF-8 zh_CN.UTF-8
|
||
include ../gen-locales.mk
|
||
endif
|
||
|
||
stdlib-benchset := strtod
|
||
|
||
stdio-common-benchset := sprintf
|
||
|
||
math-benchset := math-inlines
|
||
|
||
benchset := $(string-benchset-all) $(stdlib-benchset) $(stdio-common-benchset) \
|
||
$(math-benchset)
|
||
|
||
CFLAGS-bench-ffs.c += -fno-builtin
|
||
CFLAGS-bench-ffsll.c += -fno-builtin
|
||
CFLAGS-bench-sqrt.c += -fno-builtin
|
||
CFLAGS-bench-fmin.c += -fno-builtin
|
||
CFLAGS-bench-fminf.c += -fno-builtin
|
||
CFLAGS-bench-fmax.c += -fno-builtin
|
||
CFLAGS-bench-fmaxf.c += -fno-builtin
|
||
|
||
bench-malloc := malloc-thread
|
||
|
||
$(addprefix $(objpfx)bench-,$(bench-math)): $(libm)
|
||
$(addprefix $(objpfx)bench-,$(math-benchset)): $(libm)
|
||
$(addprefix $(objpfx)bench-,$(bench-pthread)): $(shared-thread-library)
|
||
$(objpfx)bench-malloc-thread: $(shared-thread-library)
|
||
|
||
|
||
|
||
# Rules to build and execute the benchmarks. Do not put any benchmark
|
||
# parameters beyond this point.
|
||
|
||
# We don't want the benchmark programs to run in parallel since that could
|
||
# affect their performance.
|
||
.NOTPARALLEL:
|
||
|
||
include ../Rules
|
||
|
||
binaries-bench := $(addprefix $(objpfx)bench-,$(bench))
|
||
binaries-benchset := $(addprefix $(objpfx)bench-,$(benchset))
|
||
binaries-bench-malloc := $(addprefix $(objpfx)bench-,$(bench-malloc))
|
||
|
||
# The default duration: 10 seconds.
|
||
ifndef BENCH_DURATION
|
||
BENCH_DURATION := 10
|
||
endif
|
||
|
||
CPPFLAGS-nonlib += -DDURATION=$(BENCH_DURATION)
|
||
|
||
# Use clock_gettime to measure performance of functions. The default is to use
|
||
# HP_TIMING if it is available.
|
||
ifdef USE_CLOCK_GETTIME
|
||
CPPFLAGS-nonlib += -DUSE_CLOCK_GETTIME
|
||
endif
|
||
|
||
DETAILED_OPT :=
|
||
|
||
ifdef DETAILED
|
||
DETAILED_OPT := -d
|
||
endif
|
||
|
||
# This makes sure CPPFLAGS-nonlib and CFLAGS-nonlib are passed
|
||
# for all these modules.
|
||
cpp-srcs-left := $(binaries-benchset:=.c) $(binaries-bench:=.c) \
|
||
$(binaries-bench-malloc:=.c)
|
||
lib := nonlib
|
||
include $(patsubst %,$(..)libof-iterator.mk,$(cpp-srcs-left))
|
||
|
||
bench-extra-objs = json-lib.o
|
||
|
||
extra-objs += $(bench-extra-objs)
|
||
|
||
bench-deps := bench-skeleton.c bench-timing.h Makefile
|
||
|
||
run-bench = $(test-wrapper-env) \
|
||
$(run-program-env) \
|
||
$($*-ENV) $(test-via-rtld-prefix) $${run}
|
||
|
||
timing-type := $(objpfx)bench-timing-type
|
||
|
||
bench-clean:
|
||
rm -f $(binaries-bench) $(addsuffix .o,$(binaries-bench))
|
||
rm -f $(binaries-benchset) $(addsuffix .o,$(binaries-benchset))
|
||
rm -f $(binaries-bench-malloc) $(addsuffix .o,$(binaries-bench-malloc))
|
||
rm -f $(timing-type) $(addsuffix .o,$(timing-type))
|
||
rm -f $(addprefix $(objpfx),$(bench-extra-objs))
|
||
|
||
# Define the bench target only if the target has a usable python installation.
|
||
ifdef PYTHON
|
||
bench: bench-build bench-set bench-func bench-malloc
|
||
else
|
||
bench:
|
||
@echo "The bench target needs python to run."
|
||
@exit 1
|
||
endif
|
||
|
||
# Target to only build the benchmark without running it. We generate locales
|
||
# only if we're building natively.
|
||
ifeq (no,$(cross-compiling))
|
||
bench-build: $(gen-locales) $(timing-type) $(binaries-bench) \
|
||
$(binaries-benchset) $(binaries-bench-malloc)
|
||
else
|
||
bench-build: $(timing-type) $(binaries-bench) $(binaries-benchset) \
|
||
$(binaries-bench-malloc)
|
||
endif
|
||
|
||
bench-set: $(binaries-benchset)
|
||
for run in $^; do \
|
||
echo "Running $${run}"; \
|
||
$(run-bench) > $${run}.out; \
|
||
done
|
||
|
||
bench-malloc: $(binaries-bench-malloc)
|
||
run=$(objpfx)bench-malloc-thread; \
|
||
for thr in 1 8 16 32; do \
|
||
echo "Running $${run} $${thr}"; \
|
||
$(run-bench) $${thr} > $${run}-$${thr}.out; \
|
||
done
|
||
|
||
# Build and execute the benchmark functions. This target generates JSON
|
||
# formatted bench.out. Each of the programs produce independent JSON output,
|
||
# so one could even execute them individually and process it using any JSON
|
||
# capable language or tool.
|
||
bench-func: $(binaries-bench)
|
||
{ timing_type=$$($(timing-type)); \
|
||
echo "{\"timing_type\": \"$${timing_type}\","; \
|
||
echo " \"functions\": {"; \
|
||
for run in $^; do \
|
||
if ! [ "x$${run}" = "x$<" ]; then \
|
||
echo ","; \
|
||
fi; \
|
||
echo "Running $${run}" >&2; \
|
||
$(run-bench) $(DETAILED_OPT); \
|
||
done; \
|
||
echo; \
|
||
echo " }"; \
|
||
echo "}"; } > $(objpfx)bench.out-tmp; \
|
||
if [ -f $(objpfx)bench.out ]; then \
|
||
mv -f $(objpfx)bench.out $(objpfx)bench.out.old; \
|
||
fi; \
|
||
mv -f $(objpfx)bench.out-tmp $(objpfx)bench.out
|
||
$(PYTHON) scripts/validate_benchout.py $(objpfx)bench.out \
|
||
scripts/benchout.schema.json
|
||
|
||
$(timing-type) $(binaries-bench) $(binaries-benchset) \
|
||
$(binaries-bench-malloc): %: %.o $(objpfx)json-lib.o \
|
||
$(link-extra-libs-tests) \
|
||
$(sort $(filter $(common-objpfx)lib%,$(link-libc))) \
|
||
$(addprefix $(csu-objpfx),start.o) $(+preinit) $(+postinit)
|
||
$(+link-tests)
|
||
|
||
$(objpfx)bench-%.c: %-inputs $(bench-deps)
|
||
{ if [ -n "$($*-INCLUDE)" ]; then \
|
||
cat $($*-INCLUDE); \
|
||
fi; \
|
||
$(PYTHON) scripts/bench.py $(patsubst %-inputs,%,$<); } > $@-tmp
|
||
mv -f $@-tmp $@
|