1 # Copyright (C) 2013-2021 Free Software Foundation, Inc.
2 # This file is part of the GNU C Library.
4 # The GNU C Library is free software; you can redistribute it and/or
5 # modify it under the terms of the GNU Lesser General Public
6 # License as published by the Free Software Foundation; either
7 # version 2.1 of the License, or (at your option) any later version.
9 # The GNU C Library is distributed in the hope that it will be useful,
10 # but WITHOUT ANY WARRANTY; without even the implied warranty of
11 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12 # Lesser General Public License for more details.
14 # You should have received a copy of the GNU Lesser General Public
15 # License along with the GNU C Library; if not, see
16 # <https://www.gnu.org/licenses/>.
18 # Makefile for benchmark tests. The only useful target here is `bench`.
19 # Add benchmark functions in alphabetical order.
24 bench-math
:= acos acosh asin asinh atan atanh cos cosh exp exp2 log log2 \
25 modf pow rint sin sincos sinh sqrt tan tanh fmin fmax fminf \
26 fmaxf powf trunc truncf roundeven roundevenf expf exp2f logf \
27 log2f sincosf sinf cosf isnan isinf isfinite hypot logb logbf \
28 exp10f ilogb ilogbf cbrt erf erfc exp10 expm1 j0 j1 lgamma \
29 log10 log1p tgamma y0 y1 atan2
31 ifneq (,$(filter yes
,$(float128-fcts
) $(float128-alias-fcts
)))
32 bench-math
+= expf128 powf128 sinf128 ilogbf128
35 bench-pthread
:= pthread_once thread_create pthread-locks
37 bench-string
:= ffs ffsll
40 bench
:= $(bench-math
) $(bench-pthread
) $(bench-string
)
42 bench
:= $(foreach B
,$(filter bench-
%,${BENCHSET}), ${${B}})
45 # String function benchmarks.
46 string-benchset
:= memccpy memchr memcmp memcpy memmem memmove \
47 mempcpy memset rawmemchr stpcpy stpncpy strcasecmp strcasestr \
48 strcat strchr strchrnul strcmp strcpy strcspn strlen \
49 strncasecmp strncat strncmp strncpy strnlen strpbrk strrchr \
50 strspn strstr strcpy_chk stpcpy_chk memrchr strsep strtok \
51 strcoll memcpy-large memcpy-random memmove-large memset-large \
52 memcpy-walk memset-walk memmove-walk
54 # Build and run locale-dependent benchmarks only if we're building natively.
55 ifeq (no
,$(cross-compiling
))
56 wcsmbs-benchset
:= wcslen wcsnlen wcscpy wcpcpy wcsncpy wcpncpy wcscat wcsncat \
57 wcscmp wcsncmp wcschr wcschrnul wcsrchr wcsspn wcspbrk wcscspn \
58 wmemchr wmemset wmemcmp
63 string-benchset-all
:= $(string-benchset
) ${wcsmbs-benchset
}
65 ifeq (no
,$(cross-compiling
))
66 # We have to generate locales
67 LOCALES
:= en_US.UTF-8 tr_TR.UTF-8 cs_CZ.UTF-8 fa_IR.UTF-8 fr_FR.UTF-8 \
68 ja_JP.UTF-8 si_LK.UTF-8 en_GB.UTF-8 vi_VN.UTF-8 ar_SA.UTF-8 \
69 da_DK.UTF-8 pl_PL.UTF-8 pt_PT.UTF-8 el_GR.UTF-8 ru_RU.UTF-8 \
70 he_IL.UTF-8 is_IS.UTF-8 es_ES.UTF-8 hi_IN.UTF-8 sv_SE.UTF-8 \
71 hu_HU.UTF-8 it_IT.UTF-8 sr_RS.UTF-8 zh_CN.UTF-8
72 include ..
/gen-locales.mk
75 stdlib-benchset
:= strtod
77 stdio-common-benchset
:= sprintf
79 math-benchset
:= math-inlines
82 benchset
:= $(string-benchset-all
) $(stdlib-benchset
) $(stdio-common-benchset
) \
85 benchset
:= $(foreach B
,$(filter %-benchset
,${BENCHSET}), ${${B}})
88 CFLAGS-bench-ffs.c
+= -fno-builtin
89 CFLAGS-bench-ffsll.c
+= -fno-builtin
90 CFLAGS-bench-sqrt.c
+= -fno-builtin
91 CFLAGS-bench-fmin.c
+= -fno-builtin
92 CFLAGS-bench-fminf.c
+= -fno-builtin
93 CFLAGS-bench-fmax.c
+= -fno-builtin
94 CFLAGS-bench-fmaxf.c
+= -fno-builtin
95 CFLAGS-bench-trunc.c
+= -fno-builtin
96 CFLAGS-bench-truncf.c
+= -fno-builtin
97 CFLAGS-bench-roundeven.c
+= -fno-builtin
98 CFLAGS-bench-roundevenf.c
+= -fno-builtin
99 CFLAGS-bench-isnan.c
+= -fsignaling-nans
100 CFLAGS-bench-isinf.c
+= -fsignaling-nans
101 CFLAGS-bench-isfinite.c
+= -fsignaling-nans
104 bench-malloc
:= malloc-thread malloc-simple
106 bench-malloc
:= $(filter malloc-
%,${BENCHSET})
109 $(addprefix $(objpfx
)bench-
,$(bench-math
)): $(libm
)
110 $(addprefix $(objpfx
)bench-
,$(math-benchset
)): $(libm
)
111 $(addprefix $(objpfx
)bench-
,$(bench-pthread
)): $(shared-thread-library
)
112 $(addprefix $(objpfx
)bench-
,$(bench-malloc
)): $(shared-thread-library
)
113 $(addprefix $(objpfx
)bench-
,pthread-locks
): $(libm
)
117 # Rules to build and execute the benchmarks. Do not put any benchmark
118 # parameters beyond this point.
120 # We don't want the benchmark programs to run in parallel since that could
121 # affect their performance.
124 bench-extra-objs
= json-lib.o
126 extra-objs
+= $(bench-extra-objs
)
127 others-extras
= $(bench-extra-objs
)
131 binaries-bench
:= $(addprefix $(objpfx
)bench-
,$(bench
))
132 binaries-benchset
:= $(addprefix $(objpfx
)bench-
,$(benchset
))
133 binaries-bench-malloc
:= $(addprefix $(objpfx
)bench-
,$(bench-malloc
))
135 # The default duration: 1 seconds.
136 ifndef BENCH_DURATION
140 CPPFLAGS-nonlib
+= -DDURATION
=$(BENCH_DURATION
) -D_ISOMAC
142 # Use clock_gettime to measure performance of functions. The default is
143 # to use the architecture-specific high precision timing instructions.
144 ifdef USE_CLOCK_GETTIME
145 CPPFLAGS-nonlib
+= -DUSE_CLOCK_GETTIME
147 # On x86 processors, use RDTSCP, instead of RDTSC, to measure performance
148 # of functions. All x86 processors since 2010 support RDTSCP instruction.
150 CPPFLAGS-nonlib
+= -DUSE_RDTSCP
160 bench-deps
:= bench-skeleton.c bench-timing.h Makefile
162 run-bench
= $(test-wrapper-env
) \
164 $($*-ENV
) $(test-via-rtld-prefix
) $${run}
166 timing-type
:= $(objpfx
)bench-timing-type
168 # This makes sure CPPFLAGS-nonlib and CFLAGS-nonlib are passed
169 # for all these modules.
170 cpp-srcs-left
:= $(binaries-benchset
:=.c
) $(binaries-bench
:=.c
) \
171 $(binaries-bench-malloc
:=.c
) $(timing-type
:=.c
)
173 include $(patsubst %,$(..
)libof-iterator.mk
,$(cpp-srcs-left
))
176 rm -f
$(binaries-bench
) $(addsuffix .o
,$(binaries-bench
))
177 rm -f
$(binaries-benchset
) $(addsuffix .o
,$(binaries-benchset
))
178 rm -f
$(binaries-bench-malloc
) $(addsuffix .o
,$(binaries-bench-malloc
))
179 rm -f
$(timing-type
) $(addsuffix .o
,$(timing-type
))
180 rm -f
$(addprefix $(objpfx
),$(bench-extra-objs
))
182 # Validate the passed in BENCHSET
183 ifneq ($(strip ${BENCHSET}),)
184 VALIDBENCHSETNAMES
:= bench-pthread bench-math bench-string string-benchset \
185 wcsmbs-benchset stdlib-benchset stdio-common-benchset math-benchset \
186 malloc-thread malloc-simple
187 INVALIDBENCHSETNAMES
:= $(filter-out ${VALIDBENCHSETNAMES},${BENCHSET})
188 ifneq (${INVALIDBENCHSETNAMES},)
189 $(info The following values in BENCHSET are invalid
: ${INVALIDBENCHSETNAMES})
190 $(info The valid ones are
: ${VALIDBENCHSETNAMES})
191 $(error Invalid BENCHSET value
)
195 bench
: bench-build bench-set bench-func bench-malloc
197 # Target to only build the benchmark without running it. We generate locales
198 # only if we're building natively.
199 ifeq (no
,$(cross-compiling
))
200 bench-build
: $(gen-locales
) $(timing-type
) $(binaries-bench
) \
201 $(binaries-benchset
) $(binaries-bench-malloc
)
203 bench-build
: $(timing-type
) $(binaries-bench
) $(binaries-benchset
) \
204 $(binaries-bench-malloc
)
207 bench-set
: $(binaries-benchset
)
209 echo
"Running $${run}"; \
210 $(run-bench
) > $${run}.out
; \
213 bench-malloc
: $(binaries-bench-malloc
)
216 if
[ `basename $${run}` = "bench-malloc-thread" ]; then \
217 for thr in
1 8 16 32; do \
218 echo
"Running $${run} $${thr}"; \
219 $(run-bench
) $${thr} > $${run}-$${thr}.out
; \
222 for thr in
8 16 32 64 128 256 512 1024 2048 4096; do \
223 echo
"Running $${run} $${thr}"; \
224 $(run-bench
) $${thr} > $${run}-$${thr}.out
; \
229 # Build and execute the benchmark functions. This target generates JSON
230 # formatted bench.out. Each of the programs produce independent JSON output,
231 # so one could even execute them individually and process it using any JSON
232 # capable language or tool.
233 bench-func
: $(binaries-bench
)
234 if
[ -n
'$^' ] ; then \
235 { timing_type
=$$($(test-wrapper-env
) \
237 $(test-via-rtld-prefix
) \
239 echo
"{\"timing_type\": \"$${timing_type}\","; \
240 echo
" \"functions\": {"; \
242 if
! [ "x$${run}" = "x$<" ]; then \
245 echo
"Running $${run}" >&2; \
246 $(run-bench
) $(DETAILED_OPT
); \
251 } > $(objpfx
)bench.out-tmp
; \
252 if
[ -f
$(objpfx
)bench.out
]; then \
253 mv
-f
$(objpfx
)bench.out
$(objpfx
)bench.out.old
; \
255 mv
-f
$(objpfx
)bench.out-tmp
$(objpfx
)bench.out
; \
256 $(PYTHON
) scripts
/validate_benchout.py
$(objpfx
)bench.out \
257 scripts
/benchout.schema.json
; \
260 ifeq ($(bind-now
),yes
)
261 link-bench-bind-now
= -Wl
,-z
,now
264 bench-link-targets
= $(timing-type
) $(binaries-bench
) $(binaries-benchset
) \
265 $(binaries-bench-malloc
)
267 $(bench-link-targets
): %: %.o
$(objpfx
)json-lib.o \
268 $(link-extra-libs-tests
) \
269 $(sort $(filter $(common-objpfx
)lib
%,$(link-libc
))) \
270 $(addprefix $(csu-objpfx
),start.o
) $(+preinit
) $(+postinit
)
273 $(bench-link-targets
): LDFLAGS
+= $(link-bench-bind-now
)
275 $(objpfx
)bench-
%.c
: %-inputs
$(bench-deps
)
276 { if
[ -n
"$($*-INCLUDE)" ]; then \
279 $(PYTHON
) scripts
/bench.py
$(patsubst %-inputs
,%,$<); } > $@
-tmp