benchtests/Makefile

   1 # Copyright (C) 2013-2022 Free Software Foundation, Inc.
   2 # This file is part of the GNU C Library.
   3
   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.
   8
   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.
  13
  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/>.
  17
  18 # Makefile for benchmark tests.  The only useful target here is `bench`.
  19 # Add benchmark functions in alphabetical order.
  20
  21 subdir := benchtests
  22
  23 include ../Makeconfig
  24
  25 bench-math := \
  26   acos \
  27   acosh \
  28   asin \
  29   asinh \
  30   atan \
  31   atan2 \
  32   atanh \
  33   cbrt \
  34   cos \
  35   cosf \
  36   cosh \
  37   erf \
  38   erfc \
  39   exp \
  40   exp10 \
  41   exp10f \
  42   exp2 \
  43   exp2f \
  44   expf \
  45   expm1 \
  46   fmax \
  47   fmaxf \
  48   fmin \
  49   fminf \
  50   hypot \
  51   hypotf \
  52   ilogb \
  53   ilogbf \
  54   isfinite \
  55   isinf \
  56   isnan \
  57   j0 \
  58   j1 \
  59   lgamma \
  60   log \
  61   log10 \
  62   log1p \
  63   log2 \
  64   log2f \
  65   logb \
  66   logbf \
  67   logf \
  68   modf \
  69   pow \
  70   powf \
  71   rint \
  72   roundeven \
  73   roundevenf \
  74   sin \
  75   sincos \
  76   sincosf \
  77   sinf \
  78   sinh \
  79   sqrt \
  80   tan \
  81   tanh \
  82   tgamma \
  83   trunc \
  84   truncf \
  85   y0 \
  86   y1 \
  87
  88 ifneq (,$(filter yes,$(float96-fcts)))
  89 bench-math += \
  90   cbrtl \
  91 # bench-math
  92 endif
  93
  94 ifneq (,$(filter yes,$(float128-fcts) $(float128-alias-fcts)))
  95 bench-math += \
  96   expf128 \
  97   ilogbf128 \
  98   powf128 \
  99   sinf128 \
 100 # bench-math
 101 endif
 102
 103 bench-pthread := \
 104   pthread-locks \
 105   pthread-mutex-locks \
 106   pthread_once \
 107   thread_create \
 108 # bench-pthread
 109
 110 bench-string := \
 111   ffs \
 112   ffsll \
 113 # bench-string
 114
 115 ifeq (${BENCHSET},)
 116 bench := $(bench-math) $(bench-pthread) $(bench-string)
 117 else
 118 bench := $(foreach B,$(filter bench-%,${BENCHSET}), ${${B}})
 119 endif
 120
 121 # String function benchmarks.
 122 string-benchset := \
 123   bzero \
 124   bzero-large \
 125   bzero-walk \
 126   memccpy \
 127   memchr \
 128   memcmp \
 129   memcmpeq \
 130   memcpy \
 131   memcpy-large \
 132   memcpy-random \
 133   memcpy-walk \
 134   memmem \
 135   memmove \
 136   memmove-large \
 137   memmove-walk \
 138   mempcpy \
 139   memrchr \
 140   memset \
 141   memset-large \
 142   memset-walk \
 143   memset-zero \
 144   memset-zero-large \
 145   memset-zero-walk \
 146   rawmemchr \
 147   stpcpy \
 148   stpcpy_chk \
 149   stpncpy \
 150   strcasecmp \
 151   strcasestr \
 152   strcat \
 153   strchr \
 154   strchrnul \
 155   strcmp \
 156   strcoll \
 157   strcpy \
 158   strcpy_chk \
 159   strcspn \
 160   strlen \
 161   strncasecmp \
 162   strncat \
 163   strncmp \
 164   strncpy \
 165   strnlen \
 166   strpbrk \
 167   strrchr \
 168   strsep \
 169   strspn \
 170   strstr \
 171   strtok \
 172 # string-benchset
 173
 174 # Build and run locale-dependent benchmarks only if we're building natively.
 175 ifeq (no,$(cross-compiling))
 176 wcsmbs-benchset := \
 177   wcpcpy \
 178   wcpncpy \
 179   wcscat \
 180   wcschr \
 181   wcschrnul \
 182   wcscmp \
 183   wcscpy \
 184   wcscspn \
 185   wcslen \
 186   wcsncat \
 187   wcsncmp \
 188   wcsncpy \
 189   wcsnlen \
 190   wcspbrk \
 191   wcsrchr \
 192   wcsspn \
 193   wmemchr \
 194   wmemcmp \
 195   wmemset \
 196 # wcsmbs-benchset
 197 else
 198 wcsmbs-benchset :=
 199 endif
 200
 201 string-benchset-all := $(string-benchset) ${wcsmbs-benchset}
 202
 203 ifeq (no,$(cross-compiling))
 204 # We have to generate locales
 205 LOCALES := \
 206   ar_SA.UTF-8 \
 207   cs_CZ.UTF-8 \
 208   da_DK.UTF-8 \
 209   el_GR.UTF-8 \
 210   en_GB.UTF-8 \
 211   en_US.UTF-8 \
 212   es_ES.UTF-8 \
 213   fa_IR.UTF-8 \
 214   fr_FR.UTF-8 \
 215   he_IL.UTF-8 \
 216   hi_IN.UTF-8 \
 217   hu_HU.UTF-8 \
 218   is_IS.UTF-8 \
 219   it_IT.UTF-8 \
 220   ja_JP.UTF-8 \
 221   pl_PL.UTF-8 \
 222   pt_PT.UTF-8 \
 223   ru_RU.UTF-8 \
 224   si_LK.UTF-8 \
 225   sr_RS.UTF-8 \
 226   sv_SE.UTF-8 \
 227   tr_TR.UTF-8 \
 228   vi_VN.UTF-8 \
 229   zh_CN.UTF-8 \
 230 # LOCALES
 231 include ../gen-locales.mk
 232 endif
 233
 234 stdlib-benchset := strtod
 235
 236 stdio-common-benchset := sprintf
 237
 238 math-benchset := math-inlines
 239
 240 ifeq (${BENCHSET},)
 241 benchset := $(string-benchset-all) $(stdlib-benchset) $(stdio-common-benchset) \
 242             $(math-benchset)
 243 else
 244 benchset := $(foreach B,$(filter %-benchset,${BENCHSET}), ${${B}})
 245 endif
 246
 247 CFLAGS-bench-ffs.c += -fno-builtin
 248 CFLAGS-bench-ffsll.c += -fno-builtin
 249 CFLAGS-bench-sqrt.c += -fno-builtin
 250 CFLAGS-bench-fmin.c += -fno-builtin
 251 CFLAGS-bench-fminf.c += -fno-builtin
 252 CFLAGS-bench-fmax.c += -fno-builtin
 253 CFLAGS-bench-fmaxf.c += -fno-builtin
 254 CFLAGS-bench-trunc.c += -fno-builtin
 255 CFLAGS-bench-truncf.c += -fno-builtin
 256 CFLAGS-bench-roundeven.c += -fno-builtin
 257 CFLAGS-bench-roundevenf.c += -fno-builtin
 258 CFLAGS-bench-isnan.c += -fsignaling-nans
 259 CFLAGS-bench-isinf.c += -fsignaling-nans
 260 CFLAGS-bench-isfinite.c += -fsignaling-nans
 261
 262 ifeq (${BENCHSET},)
 263 bench-malloc := malloc-thread malloc-simple
 264 else
 265 bench-malloc := $(filter malloc-%,${BENCHSET})
 266 endif
 267
 268 ifeq (${STATIC-BENCHTESTS},yes)
 269 +link-benchtests = $(+link-static-tests)
 270 link-libc-benchtests = $(link-libc-static)
 271 libm-benchtests = $(common-objpfx)math/libm.a
 272 thread-library-benchtests = $(static-thread-library)
 273 else
 274 link-libc-benchtests = $(link-libc)
 275 +link-benchtests = $(+link-tests)
 276 thread-library-benchtests = $(shared-thread-library)
 277 libm-benchtests = $(libm)
 278 endif
 279
 280 $(addprefix $(objpfx)bench-,$(bench-math)): $(libm-benchtests)
 281 $(addprefix $(objpfx)bench-,$(math-benchset)): $(libm-benchtests)
 282 $(addprefix $(objpfx)bench-,$(bench-pthread)): $(thread-library-benchtests)
 283 $(addprefix $(objpfx)bench-,$(bench-malloc)): $(thread-library-benchtests)
 284 $(addprefix $(objpfx)bench-,pthread-locks): $(libm-benchtests)
 285 $(addprefix $(objpfx)bench-,pthread-mutex-locks): $(libm-benchtests)
 286
 287 \f
 288
 289 # Rules to build and execute the benchmarks.  Do not put any benchmark
 290 # parameters beyond this point.
 291
 292 # We don't want the benchmark programs to run in parallel since that could
 293 # affect their performance.
 294 .NOTPARALLEL:
 295
 296 bench-extra-objs = json-lib.o
 297
 298 extra-objs += $(bench-extra-objs)
 299 others-extras = $(bench-extra-objs)
 300
 301 # NB: Use "=" instead of ":=" since sysdeps Makefiles may add more
 302 # benches.
 303 binaries-bench = $(addprefix $(objpfx)bench-,$(bench))
 304 extra-objs += $(addsuffix .o,$(addprefix bench-,$(bench)))
 305 binaries-benchset = $(addprefix $(objpfx)bench-,$(benchset))
 306 extra-objs += $(addsuffix .o,$(addprefix bench-,$(benchset)))
 307 binaries-bench-malloc := $(addprefix $(objpfx)bench-,$(bench-malloc))
 308 extra-objs += $(addsuffix .o,$(addprefix bench-,$(bench-malloc)))
 309
 310 # The default duration: 1 seconds.
 311 ifndef BENCH_DURATION
 312 BENCH_DURATION := 1
 313 endif
 314
 315 CPPFLAGS-nonlib += -DDURATION=$(BENCH_DURATION) -D_ISOMAC
 316
 317 # Use clock_gettime to measure performance of functions.  The default is
 318 # to use the architecture-specific high precision timing instructions.
 319 ifdef USE_CLOCK_GETTIME
 320 CPPFLAGS-nonlib += -DUSE_CLOCK_GETTIME
 321 else
 322 # On x86 processors, use RDTSCP, instead of RDTSC, to measure performance
 323 # of functions.  All x86 processors since 2010 support RDTSCP instruction.
 324 ifdef USE_RDTSCP
 325 CPPFLAGS-nonlib += -DUSE_RDTSCP
 326 endif
 327 endif
 328
 329 DETAILED_OPT :=
 330
 331 ifdef DETAILED
 332 DETAILED_OPT := -d
 333 endif
 334
 335 bench-deps := bench-skeleton.c bench-timing.h Makefile
 336
 337 run-bench = $(test-wrapper-env) \
 338             $(run-program-env) \
 339             $($*-ENV) $(test-via-rtld-prefix) $${run}
 340
 341 timing-type := $(objpfx)bench-timing-type
 342 extra-objs += bench-timing-type.o
 343
 344 include ../Rules
 345
 346 # This makes sure CPPFLAGS-nonlib and CFLAGS-nonlib are passed
 347 # for all these modules.
 348 cpp-srcs-left := $(binaries-benchset:=.c) $(binaries-bench:=.c) \
 349                  $(binaries-bench-malloc:=.c) $(timing-type:=.c)
 350 lib := nonlib
 351 include $(patsubst %,$(..)libof-iterator.mk,$(cpp-srcs-left))
 352
 353 bench-clean:
 354         rm -f $(binaries-bench) $(addsuffix .o,$(binaries-bench))
 355         rm -f $(binaries-benchset) $(addsuffix .o,$(binaries-benchset))
 356         rm -f $(binaries-bench-malloc) $(addsuffix .o,$(binaries-bench-malloc))
 357         rm -f $(timing-type) $(addsuffix .o,$(timing-type))
 358         rm -f $(addprefix $(objpfx),$(bench-extra-objs))
 359
 360 # Validate the passed in BENCHSET
 361 ifneq ($(strip ${BENCHSET}),)
 362 VALIDBENCHSETNAMES := bench-pthread bench-math bench-string string-benchset \
 363    wcsmbs-benchset stdlib-benchset stdio-common-benchset math-benchset \
 364    malloc-thread malloc-simple
 365 INVALIDBENCHSETNAMES := $(filter-out ${VALIDBENCHSETNAMES},${BENCHSET})
 366 ifneq (${INVALIDBENCHSETNAMES},)
 367 $(info The following values in BENCHSET are invalid: ${INVALIDBENCHSETNAMES})
 368 $(info The valid ones are: ${VALIDBENCHSETNAMES})
 369 $(error Invalid BENCHSET value)
 370 endif
 371 endif
 372
 373 bench: bench-build bench-set bench-func bench-malloc
 374
 375 # Target to only build the benchmark without running it.  We generate locales
 376 # only if we're building natively.
 377 ifeq (no,$(cross-compiling))
 378 bench-build: $(gen-locales) $(timing-type) $(binaries-bench) \
 379         $(binaries-benchset) $(binaries-bench-malloc)
 380 else
 381 bench-build: $(timing-type) $(binaries-bench) $(binaries-benchset) \
 382         $(binaries-bench-malloc)
 383 endif
 384
 385 bench-set: $(binaries-benchset)
 386         for run in $^; do \
 387           echo "Running $${run}"; \
 388           $(run-bench) > $${run}.out; \
 389         done
 390
 391 bench-malloc: $(binaries-bench-malloc)
 392         for run in $^; do \
 393           echo "$${run}"; \
 394           if [ `basename $${run}` = "bench-malloc-thread" ]; then \
 395                 for thr in 1 8 16 32; do \
 396                         echo "Running $${run} $${thr}"; \
 397                         $(run-bench) $${thr} > $${run}-$${thr}.out; \
 398                 done;\
 399           else \
 400                 for thr in 8 16 32 64 128 256 512 1024 2048 4096; do \
 401                   echo "Running $${run} $${thr}"; \
 402                   $(run-bench) $${thr} > $${run}-$${thr}.out; \
 403                 done;\
 404           fi;\
 405         done
 406
 407 # Build and execute the benchmark functions.  This target generates JSON
 408 # formatted bench.out.  Each of the programs produce independent JSON output,
 409 # so one could even execute them individually and process it using any JSON
 410 # capable language or tool.
 411 bench-func: $(binaries-bench)
 412         if [ -n '$^' ] ; then \
 413         { timing_type=$$($(test-wrapper-env) \
 414                          $(run-program-env) \
 415                          $(test-via-rtld-prefix) \
 416                          $(timing-type)); \
 417           echo "{\"timing_type\": \"$${timing_type}\","; \
 418           echo " \"functions\": {"; \
 419           for run in $^; do \
 420             if ! [ "x$${run}" = "x$<" ]; then \
 421               echo ","; \
 422             fi; \
 423             echo "Running $${run}" >&2; \
 424             $(run-bench) $(DETAILED_OPT); \
 425           done; \
 426           echo; \
 427           echo " }"; \
 428           echo "}"; \
 429           } > $(objpfx)bench.out-tmp; \
 430           if [ -f $(objpfx)bench.out ]; then \
 431             mv -f $(objpfx)bench.out $(objpfx)bench.out.old; \
 432           fi; \
 433           mv -f $(objpfx)bench.out-tmp $(objpfx)bench.out; \
 434           $(PYTHON) scripts/validate_benchout.py $(objpfx)bench.out \
 435           scripts/benchout.schema.json; \
 436         fi
 437
 438 ifeq ($(bind-now),yes)
 439 link-bench-bind-now = -Wl,-z,now
 440 endif
 441
 442 bench-link-targets = $(timing-type) $(binaries-bench) $(binaries-benchset) \
 443         $(binaries-bench-malloc)
 444
 445 $(bench-link-targets): %: %.o $(objpfx)json-lib.o \
 446         $(link-extra-libs-tests) \
 447   $(sort $(filter $(common-objpfx)lib%,$(link-libc-benchtests))) \
 448   $(addprefix $(csu-objpfx),start.o) $(+preinit) $(+postinit)
 449         $(+link-benchtests)
 450
 451 $(bench-link-targets): LDFLAGS += $(link-bench-bind-now)
 452
 453 $(objpfx)bench-%.c: %-inputs $(bench-deps)
 454         { if [ -n "$($*-INCLUDE)" ]; then \
 455           cat $($*-INCLUDE); \
 456         fi; \
 457         $(PYTHON) scripts/bench.py $(patsubst %-inputs,%,$<); } > $@-tmp
 458         mv -f $@-tmp $@