benchtests/Makefile

   1 # Copyright (C) 2013-2023 Free Software Foundation, Inc.
   2 # Copyright The GNU Toolchain Authors.
   3 # This file is part of the GNU C Library.
   4
   5 # The GNU C Library is free software; you can redistribute it and/or
   6 # modify it under the terms of the GNU Lesser General Public
   7 # License as published by the Free Software Foundation; either
   8 # version 2.1 of the License, or (at your option) any later version.
   9
  10 # The GNU C Library is distributed in the hope that it will be useful,
  11 # but WITHOUT ANY WARRANTY; without even the implied warranty of
  12 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13 # Lesser General Public License for more details.
  14
  15 # You should have received a copy of the GNU Lesser General Public
  16 # License along with the GNU C Library; if not, see
  17 # <https://www.gnu.org/licenses/>.
  18
  19 # Makefile for benchmark tests.  The only useful target here is `bench`.
  20 # Add benchmark functions in alphabetical order.
  21
  22 subdir := benchtests
  23
  24 include ../Makeconfig
  25
  26 bench-math := \
  27   acos \
  28   acosh \
  29   asin \
  30   asinh \
  31   atan \
  32   atan2 \
  33   atanh \
  34   cbrt \
  35   cos \
  36   cosf \
  37   cosh \
  38   erf \
  39   erfc \
  40   exp \
  41   exp10 \
  42   exp10f \
  43   exp2 \
  44   exp2f \
  45   expf \
  46   expm1 \
  47   fmax \
  48   fmaxf \
  49   fmin \
  50   fminf \
  51   hypot \
  52   hypotf \
  53   ilogb \
  54   ilogbf \
  55   isfinite \
  56   isinf \
  57   isnan \
  58   j0 \
  59   j1 \
  60   lgamma \
  61   log \
  62   log10 \
  63   log1p \
  64   log2 \
  65   log2f \
  66   logb \
  67   logbf \
  68   logf \
  69   modf \
  70   pow \
  71   powf \
  72   rint \
  73   roundeven \
  74   roundevenf \
  75   sin \
  76   sincos \
  77   sincosf \
  78   sinf \
  79   sinh \
  80   sqrt \
  81   tan \
  82   tanh \
  83   tgamma \
  84   trunc \
  85   truncf \
  86   y0 \
  87   y1 \
  88
  89 ifneq (,$(filter yes,$(float96-fcts)))
  90 bench-math += \
  91   cbrtl \
  92 # bench-math
  93 endif
  94
  95 ifneq (,$(filter yes,$(float128-fcts) $(float128-alias-fcts)))
  96 bench-math += \
  97   expf128 \
  98   ilogbf128 \
  99   powf128 \
 100   sinf128 \
 101 # bench-math
 102 endif
 103
 104 bench-pthread := \
 105   pthread-locks \
 106   pthread-mutex-lock \
 107   pthread-mutex-trylock \
 108   pthread-spin-lock \
 109   pthread-spin-trylock \
 110   pthread_once \
 111   thread_create \
 112 # bench-pthread
 113
 114 LDLIBS-bench-pthread-mutex-lock += -lm
 115 LDLIBS-bench-pthread-mutex-trylock += -lm
 116 LDLIBS-bench-pthread-spin-lock += -lm
 117 LDLIBS-bench-pthread-spin-trylock += -lm
 118
 119 bench-string := \
 120   ffs \
 121   ffsll \
 122 # bench-string
 123
 124 # String function benchmarks.
 125 string-benchset := \
 126   bzero \
 127   bzero-large \
 128   bzero-walk \
 129   memccpy \
 130   memchr \
 131   memcmp \
 132   memcmpeq \
 133   memcpy \
 134   memcpy-large \
 135   memcpy-random \
 136   memcpy-walk \
 137   memmem \
 138   memmove \
 139   memmove-large \
 140   memmove-walk \
 141   mempcpy \
 142   memrchr \
 143   memset \
 144   memset-large \
 145   memset-walk \
 146   memset-zero \
 147   memset-zero-large \
 148   memset-zero-walk \
 149   rawmemchr \
 150   stpcpy \
 151   stpcpy_chk \
 152   stpncpy \
 153   strcasecmp \
 154   strcasestr \
 155   strcat \
 156   strchr \
 157   strchrnul \
 158   strcmp \
 159   strcoll \
 160   strcpy \
 161   strcpy_chk \
 162   strcspn \
 163   strlen \
 164   strncasecmp \
 165   strncat \
 166   strncmp \
 167   strncpy \
 168   strnlen \
 169   strpbrk \
 170   strrchr \
 171   strsep \
 172   strspn \
 173   strstr \
 174   strtok \
 175 # string-benchset
 176
 177 # Build and run locale-dependent benchmarks only if we're building natively.
 178 ifeq (no,$(cross-compiling))
 179 wcsmbs-benchset := \
 180   wcpcpy \
 181   wcpncpy \
 182   wcrtomb \
 183   wcscat \
 184   wcschr \
 185   wcschrnul \
 186   wcscmp \
 187   wcscpy \
 188   wcscspn \
 189   wcslen \
 190   wcsncat \
 191   wcsncmp \
 192   wcsncpy \
 193   wcsnlen \
 194   wcspbrk \
 195   wcsrchr \
 196   wcsspn \
 197   wmemchr \
 198   wmemcmp \
 199   wmemset \
 200 # wcsmbs-benchset
 201 else
 202 wcsmbs-benchset :=
 203 endif
 204
 205 string-benchset-all := $(string-benchset) ${wcsmbs-benchset}
 206
 207 ifeq (no,$(cross-compiling))
 208 # We have to generate locales
 209 LOCALES := \
 210   ar_SA.UTF-8 \
 211   cs_CZ.UTF-8 \
 212   da_DK.UTF-8 \
 213   el_GR.UTF-8 \
 214   en_GB.UTF-8 \
 215   en_US.UTF-8 \
 216   es_ES.UTF-8 \
 217   fa_IR.UTF-8 \
 218   fr_FR.UTF-8 \
 219   he_IL.UTF-8 \
 220   hi_IN.UTF-8 \
 221   hu_HU.UTF-8 \
 222   is_IS.UTF-8 \
 223   it_IT.UTF-8 \
 224   ja_JP.UTF-8 \
 225   pl_PL.UTF-8 \
 226   pt_PT.UTF-8 \
 227   ru_RU.UTF-8 \
 228   si_LK.UTF-8 \
 229   sr_RS.UTF-8 \
 230   sv_SE.UTF-8 \
 231   tr_TR.UTF-8 \
 232   vi_VN.UTF-8 \
 233   zh_CN.UTF-8 \
 234 # LOCALES
 235 include ../gen-locales.mk
 236 endif
 237
 238 hash-benchset := \
 239   dl-elf-hash \
 240   dl-new-hash \
 241   nss-hash \
 242 # hash-benchset
 243
 244 stdlib-benchset := \
 245   arc4random \
 246   strtod \
 247   # stdlib-benchset
 248
 249 stdio-common-benchset := sprintf
 250
 251 math-benchset := math-inlines
 252
 253 ifeq (${BENCHSET},)
 254 benchset := $(string-benchset-all) $(stdlib-benchset) $(stdio-common-benchset) \
 255             $(math-benchset) $(hash-benchset)
 256 else
 257 benchset := $(foreach B,$(filter %-benchset,${BENCHSET}), ${${B}})
 258 endif
 259
 260 CFLAGS-bench-ffs.c += -fno-builtin
 261 CFLAGS-bench-ffsll.c += -fno-builtin
 262 CFLAGS-bench-sqrt.c += -fno-builtin
 263 CFLAGS-bench-fmin.c += -fno-builtin
 264 CFLAGS-bench-fminf.c += -fno-builtin
 265 CFLAGS-bench-fmax.c += -fno-builtin
 266 CFLAGS-bench-fmaxf.c += -fno-builtin
 267 CFLAGS-bench-trunc.c += -fno-builtin
 268 CFLAGS-bench-truncf.c += -fno-builtin
 269 CFLAGS-bench-roundeven.c += -fno-builtin
 270 CFLAGS-bench-roundevenf.c += -fno-builtin
 271 CFLAGS-bench-isnan.c += $(config-cflags-signaling-nans)
 272 CFLAGS-bench-isinf.c += $(config-cflags-signaling-nans)
 273 CFLAGS-bench-isfinite.c += $(config-cflags-signaling-nans)
 274
 275 ifeq (${BENCHSET},)
 276 bench-malloc := malloc-thread malloc-simple
 277 else
 278 bench-malloc := $(filter malloc-%,${BENCHSET})
 279 endif
 280
 281 ifeq (${STATIC-BENCHTESTS},yes)
 282 +link-benchtests = $(+link-static-tests)
 283 link-libc-benchtests = $(link-libc-static)
 284 libm-benchtests = $(common-objpfx)math/libm.a
 285 thread-library-benchtests = $(static-thread-library)
 286 else
 287 link-libc-benchtests = $(link-libc)
 288 +link-benchtests = $(+link-tests)
 289 thread-library-benchtests = $(shared-thread-library)
 290 libm-benchtests = $(libm)
 291 endif
 292
 293 $(addprefix $(objpfx)bench-,$(bench-math)): $(libm-benchtests)
 294 $(addprefix $(objpfx)bench-,$(math-benchset)): $(libm-benchtests)
 295 $(addprefix $(objpfx)bench-,$(bench-pthread)): $(thread-library-benchtests)
 296 $(addprefix $(objpfx)bench-,$(bench-malloc)): $(thread-library-benchtests)
 297 $(addprefix $(objpfx)bench-,pthread-locks): $(libm-benchtests)
 298 $(addprefix $(objpfx)bench-,pthread-mutex-locks): $(libm-benchtests)
 299
 300 \f
 301
 302 # Rules to build and execute the benchmarks.  Do not put any benchmark
 303 # parameters beyond this point.
 304
 305 # We don't want the benchmark programs to run in parallel since that could
 306 # affect their performance.
 307 .NOTPARALLEL:
 308
 309 bench-extra-objs = json-lib.o
 310
 311 extra-objs += $(bench-extra-objs)
 312 others-extras = $(bench-extra-objs)
 313
 314 # The default duration: 1 seconds.
 315 ifndef BENCH_DURATION
 316 BENCH_DURATION := 1
 317 endif
 318
 319 CPPFLAGS-nonlib += -DDURATION=$(BENCH_DURATION) -D_ISOMAC
 320
 321 # Use clock_gettime to measure performance of functions.  The default is
 322 # to use the architecture-specific high precision timing instructions.
 323 ifdef USE_CLOCK_GETTIME
 324 CPPFLAGS-nonlib += -DUSE_CLOCK_GETTIME
 325 else
 326 # On x86 processors, use RDTSCP, instead of RDTSC, to measure performance
 327 # of functions.  All x86 processors since 2010 support RDTSCP instruction.
 328 ifdef USE_RDTSCP
 329 CPPFLAGS-nonlib += -DUSE_RDTSCP
 330 endif
 331 endif
 332
 333 DETAILED_OPT :=
 334
 335 ifdef DETAILED
 336 DETAILED_OPT := -d
 337 endif
 338
 339 bench-deps := bench-skeleton.c bench-timing.h Makefile
 340
 341 run-bench = $(test-wrapper-env) \
 342             $(run-program-env) \
 343             $($*-ENV) $(test-via-rtld-prefix) $${run}
 344
 345 timing-type := $(objpfx)bench-timing-type
 346 extra-objs += bench-timing-type.o
 347
 348 include ../Rules
 349
 350 bench-math += $(bench-libmvec)
 351
 352 ifeq (${BENCHSET},)
 353 bench := $(bench-math) $(bench-pthread) $(bench-string)
 354 else
 355 bench := $(foreach B,$(filter bench-%,${BENCHSET}), ${${B}})
 356 endif
 357
 358 # NB: Use "=" instead of ":=" since sysdeps Makefiles may add more
 359 # benches.
 360 binaries-bench = $(addprefix $(objpfx)bench-,$(bench))
 361 extra-objs += $(addsuffix .o,$(addprefix bench-,$(bench)))
 362 binaries-benchset = $(addprefix $(objpfx)bench-,$(benchset))
 363 extra-objs += $(addsuffix .o,$(addprefix bench-,$(benchset)))
 364 binaries-bench-malloc := $(addprefix $(objpfx)bench-,$(bench-malloc))
 365 extra-objs += $(addsuffix .o,$(addprefix bench-,$(bench-malloc)))
 366
 367 # This makes sure CPPFLAGS-nonlib and CFLAGS-nonlib are passed
 368 # for all these modules.
 369 cpp-srcs-left := $(binaries-benchset:=.c) $(binaries-bench:=.c) \
 370                  $(binaries-bench-malloc:=.c) $(timing-type:=.c)
 371 lib := nonlib
 372 include $(patsubst %,$(..)libof-iterator.mk,$(cpp-srcs-left))
 373
 374 bench-clean:
 375         rm -f $(binaries-bench) $(addsuffix .o,$(binaries-bench))
 376         rm -f $(binaries-benchset) $(addsuffix .o,$(binaries-benchset))
 377         rm -f $(binaries-bench-malloc) $(addsuffix .o,$(binaries-bench-malloc))
 378         rm -f $(timing-type) $(addsuffix .o,$(timing-type))
 379         rm -f $(addprefix $(objpfx),$(bench-extra-objs))
 380
 381 # Validate the passed in BENCHSET
 382 ifneq ($(strip ${BENCHSET}),)
 383 VALIDBENCHSETNAMES := \
 384   bench-math \
 385   bench-pthread \
 386   bench-string \
 387   hash-benchset \
 388   malloc-simple \
 389   malloc-thread \
 390   math-benchset \
 391   stdio-common-benchset \
 392   stdlib-benchset \
 393   string-benchset \
 394   wcsmbs-benchset \
 395 # VALIDBENCHSETNAMES
 396
 397 INVALIDBENCHSETNAMES := $(filter-out ${VALIDBENCHSETNAMES},${BENCHSET})
 398 ifneq (${INVALIDBENCHSETNAMES},)
 399 $(info The following values in BENCHSET are invalid: ${INVALIDBENCHSETNAMES})
 400 $(info The valid ones are: ${VALIDBENCHSETNAMES})
 401 $(error Invalid BENCHSET value)
 402 endif
 403 endif
 404
 405 bench: bench-build bench-set bench-func bench-malloc
 406
 407 # Target to only build the benchmark without running it.  We generate locales
 408 # only if we're building natively.
 409 ifeq (no,$(cross-compiling))
 410 bench-build: $(gen-locales) $(timing-type) $(binaries-bench) \
 411         $(binaries-benchset) $(binaries-bench-malloc)
 412 else
 413 bench-build: $(timing-type) $(binaries-bench) $(binaries-benchset) \
 414         $(binaries-bench-malloc)
 415 endif
 416
 417 bench-set: $(binaries-benchset)
 418         for run in $^; do \
 419           echo "Running $${run}"; \
 420           $(run-bench) > $${run}.out; \
 421         done
 422
 423 bench-malloc: $(binaries-bench-malloc)
 424         for run in $^; do \
 425           echo "$${run}"; \
 426           if [ `basename $${run}` = "bench-malloc-thread" ]; then \
 427                 for thr in 1 8 16 32; do \
 428                         echo "Running $${run} $${thr}"; \
 429                         $(run-bench) $${thr} > $${run}-$${thr}.out; \
 430                 done;\
 431           else \
 432                 for thr in 8 16 32 64 128 256 512 1024 2048 4096; do \
 433                   echo "Running $${run} $${thr}"; \
 434                   $(run-bench) $${thr} > $${run}-$${thr}.out; \
 435                 done;\
 436           fi;\
 437         done
 438
 439 # Build and execute the benchmark functions.  This target generates JSON
 440 # formatted bench.out.  Each of the programs produce independent JSON output,
 441 # so one could even execute them individually and process it using any JSON
 442 # capable language or tool.
 443 bench-func: $(binaries-bench)
 444         if [ -n '$^' ] ; then \
 445         { timing_type=$$($(test-wrapper-env) \
 446                          $(run-program-env) \
 447                          $(test-via-rtld-prefix) \
 448                          $(timing-type)); \
 449           echo "{\"timing_type\": \"$${timing_type}\","; \
 450           echo " \"functions\": {"; \
 451           for run in $^; do \
 452             op=$$($(run-bench) $(DETAILED_OPT)); \
 453             ret=$$?; \
 454             case "$${ret}" in \
 455               77) \
 456               echo "UNSUPPORTED $${run}: $${op}" >&2; \
 457                 ;; \
 458               0) \
 459                 echo "Running $${run}" >&2; \
 460                 if [ "$${run}" != "$<" ]; then \
 461                   echo ","; \
 462                 fi; \
 463                 echo "$${op}"; \
 464                 ;; \
 465               *) \
 466                 echo "FAILED $${run}" >&2; \
 467                 ;; \
 468             esac; \
 469           done; \
 470           echo; \
 471           echo " }"; \
 472           echo "}"; \
 473           } > $(objpfx)bench.out-tmp; \
 474           if [ -f $(objpfx)bench.out ]; then \
 475             mv -f $(objpfx)bench.out $(objpfx)bench.out.old; \
 476           fi; \
 477           mv -f $(objpfx)bench.out-tmp $(objpfx)bench.out; \
 478           $(PYTHON) scripts/validate_benchout.py $(objpfx)bench.out \
 479           scripts/benchout.schema.json; \
 480         fi
 481
 482 ifeq ($(bind-now),yes)
 483 link-bench-bind-now = -Wl,-z,now
 484 endif
 485
 486 bench-link-targets = $(timing-type) $(binaries-bench) $(binaries-benchset) \
 487         $(binaries-bench-malloc)
 488
 489 $(bench-link-targets): %: %.o $(objpfx)json-lib.o \
 490         $(link-extra-libs-tests) \
 491   $(sort $(filter $(common-objpfx)lib%,$(link-libc-benchtests))) \
 492   $(addprefix $(csu-objpfx),start.o) $(+preinit) $(+postinit)
 493         $(+link-benchtests)
 494
 495 $(bench-link-targets): LDFLAGS += $(link-bench-bind-now)
 496
 497 $(objpfx)bench-%.c: %-inputs $(bench-deps)
 498         { if [ -n "$($*-INCLUDE)" ]; then \
 499           cat $($*-INCLUDE); \
 500         fi; \
 501         $(PYTHON) scripts/bench.py $(patsubst %-inputs,%,$<); } > $@-tmp
 502         mv -f $@-tmp $@