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 (,$(filter yes,$(float96-fcts)))
35 ifneq (,$(filter yes,$(float128-fcts) $(float128-alias-fcts)))
36 bench-math += expf128 powf128 sinf128 ilogbf128
39 bench-pthread := pthread_once thread_create pthread-locks
41 bench-string := ffs ffsll
44 bench := $(bench-math) $(bench-pthread) $(bench-string)
46 bench := $(foreach B,$(filter bench-%,${BENCHSET}), ${${B}})
49 # String function benchmarks.
50 string-benchset := memccpy memchr memcmp memcpy memmem memmove \
51 mempcpy memset rawmemchr stpcpy stpncpy strcasecmp strcasestr \
52 strcat strchr strchrnul strcmp strcpy strcspn strlen \
53 strncasecmp strncat strncmp strncpy strnlen strpbrk strrchr \
54 strspn strstr strcpy_chk stpcpy_chk memrchr strsep strtok \
55 strcoll memcpy-large memcpy-random memmove-large memset-large \
56 memcpy-walk memset-walk memmove-walk
58 # Build and run locale-dependent benchmarks only if we're building natively.
59 ifeq (no,$(cross-compiling))
60 wcsmbs-benchset := wcslen wcsnlen wcscpy wcpcpy wcsncpy wcpncpy wcscat wcsncat \
61 wcscmp wcsncmp wcschr wcschrnul wcsrchr wcsspn wcspbrk wcscspn \
62 wmemchr wmemset wmemcmp
67 string-benchset-all := $(string-benchset) ${wcsmbs-benchset}
69 ifeq (no,$(cross-compiling))
70 # We have to generate locales
71 LOCALES := en_US.UTF-8 tr_TR.UTF-8 cs_CZ.UTF-8 fa_IR.UTF-8 fr_FR.UTF-8 \
72 ja_JP.UTF-8 si_LK.UTF-8 en_GB.UTF-8 vi_VN.UTF-8 ar_SA.UTF-8 \
73 da_DK.UTF-8 pl_PL.UTF-8 pt_PT.UTF-8 el_GR.UTF-8 ru_RU.UTF-8 \
74 he_IL.UTF-8 is_IS.UTF-8 es_ES.UTF-8 hi_IN.UTF-8 sv_SE.UTF-8 \
75 hu_HU.UTF-8 it_IT.UTF-8 sr_RS.UTF-8 zh_CN.UTF-8
76 include ../gen-locales.mk
79 stdlib-benchset := strtod
81 stdio-common-benchset := sprintf
83 math-benchset := math-inlines
86 benchset := $(string-benchset-all) $(stdlib-benchset) $(stdio-common-benchset) \
89 benchset := $(foreach B,$(filter %-benchset,${BENCHSET}), ${${B}})
92 CFLAGS-bench-ffs.c += -fno-builtin
93 CFLAGS-bench-ffsll.c += -fno-builtin
94 CFLAGS-bench-sqrt.c += -fno-builtin
95 CFLAGS-bench-fmin.c += -fno-builtin
96 CFLAGS-bench-fminf.c += -fno-builtin
97 CFLAGS-bench-fmax.c += -fno-builtin
98 CFLAGS-bench-fmaxf.c += -fno-builtin
99 CFLAGS-bench-trunc.c += -fno-builtin
100 CFLAGS-bench-truncf.c += -fno-builtin
101 CFLAGS-bench-roundeven.c += -fno-builtin
102 CFLAGS-bench-roundevenf.c += -fno-builtin
103 CFLAGS-bench-isnan.c += -fsignaling-nans
104 CFLAGS-bench-isinf.c += -fsignaling-nans
105 CFLAGS-bench-isfinite.c += -fsignaling-nans
108 bench-malloc := malloc-thread malloc-simple
110 bench-malloc := $(filter malloc-%,${BENCHSET})
113 ifeq (${STATIC-BENCHTESTS},yes)
114 +link-benchtests = $(+link-static-tests)
115 link-libc-benchtests = $(link-libc-static)
116 libm-benchtests = $(common-objpfx)math/libm.a
117 thread-library-benchtests = $(static-thread-library)
119 link-libc-benchtests = $(link-libc)
120 +link-benchtests = $(+link-tests)
121 thread-library-benchtests = $(shared-thread-library)
122 libm-benchtests = $(libm)
125 $(addprefix $(objpfx)bench-,$(bench-math)): $(libm-benchtests)
126 $(addprefix $(objpfx)bench-,$(math-benchset)): $(libm-benchtests)
127 $(addprefix $(objpfx)bench-,$(bench-pthread)): $(thread-library-benchtests)
128 $(addprefix $(objpfx)bench-,$(bench-malloc)): $(thread-library-benchtests)
129 $(addprefix $(objpfx)bench-,pthread-locks): $(libm-benchtests)
133 # Rules to build and execute the benchmarks. Do not put any benchmark
134 # parameters beyond this point.
136 # We don't want the benchmark programs to run in parallel since that could
137 # affect their performance.
140 bench-extra-objs = json-lib.o
142 extra-objs += $(bench-extra-objs)
143 others-extras = $(bench-extra-objs)
147 binaries-bench := $(addprefix $(objpfx)bench-,$(bench))
148 binaries-benchset := $(addprefix $(objpfx)bench-,$(benchset))
149 binaries-bench-malloc := $(addprefix $(objpfx)bench-,$(bench-malloc))
151 # The default duration: 1 seconds.
152 ifndef BENCH_DURATION
156 CPPFLAGS-nonlib += -DDURATION=$(BENCH_DURATION) -D_ISOMAC
158 # Use clock_gettime to measure performance of functions. The default is
159 # to use the architecture-specific high precision timing instructions.
160 ifdef USE_CLOCK_GETTIME
161 CPPFLAGS-nonlib += -DUSE_CLOCK_GETTIME
163 # On x86 processors, use RDTSCP, instead of RDTSC, to measure performance
164 # of functions. All x86 processors since 2010 support RDTSCP instruction.
166 CPPFLAGS-nonlib += -DUSE_RDTSCP
176 bench-deps := bench-skeleton.c bench-timing.h Makefile
178 run-bench = $(test-wrapper-env) \
180 $($*-ENV) $(test-via-rtld-prefix) $${run}
182 timing-type := $(objpfx)bench-timing-type
184 # This makes sure CPPFLAGS-nonlib and CFLAGS-nonlib are passed
185 # for all these modules.
186 cpp-srcs-left := $(binaries-benchset:=.c) $(binaries-bench:=.c) \
187 $(binaries-bench-malloc:=.c) $(timing-type:=.c)
189 include $(patsubst %,$(..)libof-iterator.mk,$(cpp-srcs-left))
192 rm -f $(binaries-bench) $(addsuffix .o,$(binaries-bench))
193 rm -f $(binaries-benchset) $(addsuffix .o,$(binaries-benchset))
194 rm -f $(binaries-bench-malloc) $(addsuffix .o,$(binaries-bench-malloc))
195 rm -f $(timing-type) $(addsuffix .o,$(timing-type))
196 rm -f $(addprefix $(objpfx),$(bench-extra-objs))
198 # Validate the passed in BENCHSET
199 ifneq ($(strip ${BENCHSET}),)
200 VALIDBENCHSETNAMES := bench-pthread bench-math bench-string string-benchset \
201 wcsmbs-benchset stdlib-benchset stdio-common-benchset math-benchset \
202 malloc-thread malloc-simple
203 INVALIDBENCHSETNAMES := $(filter-out ${VALIDBENCHSETNAMES},${BENCHSET})
204 ifneq (${INVALIDBENCHSETNAMES},)
205 $(info The following values in BENCHSET are invalid: ${INVALIDBENCHSETNAMES})
206 $(info The valid ones are: ${VALIDBENCHSETNAMES})
207 $(error Invalid BENCHSET value)
211 bench: bench-build bench-set bench-func bench-malloc
213 # Target to only build the benchmark without running it. We generate locales
214 # only if we're building natively.
215 ifeq (no,$(cross-compiling))
216 bench-build: $(gen-locales) $(timing-type) $(binaries-bench) \
217 $(binaries-benchset) $(binaries-bench-malloc)
219 bench-build: $(timing-type) $(binaries-bench) $(binaries-benchset) \
220 $(binaries-bench-malloc)
223 bench-set: $(binaries-benchset)
225 echo "Running $${run}"; \
226 $(run-bench) > $${run}.out; \
229 bench-malloc: $(binaries-bench-malloc)
232 if [ `basename $${run}` = "bench-malloc-thread" ]; then \
233 for thr in 1 8 16 32; do \
234 echo "Running $${run} $${thr}"; \
235 $(run-bench) $${thr} > $${run}-$${thr}.out; \
238 for thr in 8 16 32 64 128 256 512 1024 2048 4096; do \
239 echo "Running $${run} $${thr}"; \
240 $(run-bench) $${thr} > $${run}-$${thr}.out; \
245 # Build and execute the benchmark functions. This target generates JSON
246 # formatted bench.out. Each of the programs produce independent JSON output,
247 # so one could even execute them individually and process it using any JSON
248 # capable language or tool.
249 bench-func: $(binaries-bench)
250 if [ -n '$^' ] ; then \
251 { timing_type=$$($(test-wrapper-env) \
253 $(test-via-rtld-prefix) \
255 echo "{\"timing_type\": \"$${timing_type}\","; \
256 echo " \"functions\": {"; \
258 if ! [ "x$${run}" = "x$<" ]; then \
261 echo "Running $${run}" >&2; \
262 $(run-bench) $(DETAILED_OPT); \
267 } > $(objpfx)bench.out-tmp; \
268 if [ -f $(objpfx)bench.out ]; then \
269 mv -f $(objpfx)bench.out $(objpfx)bench.out.old; \
271 mv -f $(objpfx)bench.out-tmp $(objpfx)bench.out; \
272 $(PYTHON) scripts/validate_benchout.py $(objpfx)bench.out \
273 scripts/benchout.schema.json; \
276 ifeq ($(bind-now),yes)
277 link-bench-bind-now = -Wl,-z,now
280 bench-link-targets = $(timing-type) $(binaries-bench) $(binaries-benchset) \
281 $(binaries-bench-malloc)
283 $(bench-link-targets): %: %.o $(objpfx)json-lib.o \
284 $(link-extra-libs-tests) \
285 $(sort $(filter $(common-objpfx)lib%,$(link-libc-benchtests))) \
286 $(addprefix $(csu-objpfx),start.o) $(+preinit) $(+postinit)
289 $(bench-link-targets): LDFLAGS += $(link-bench-bind-now)
291 $(objpfx)bench-%.c: %-inputs $(bench-deps)
292 { if [ -n "$($*-INCLUDE)" ]; then \
295 $(PYTHON) scripts/bench.py $(patsubst %-inputs,%,$<); } > $@-tmp