benchtests/Makefile

   1 # Copyright (C) 2013-2024 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   fmod \
  52   fmodf \
  53   hypot \
  54   hypotf \
  55   ilogb \
  56   ilogbf \
  57   isfinite \
  58   isinf \
  59   isnan \
  60   j0 \
  61   j1 \
  62   lgamma \
  63   log \
  64   log10 \
  65   log1p \
  66   log2 \
  67   log2f \
  68   logb \
  69   logbf \
  70   logf \
  71   modf \
  72   pow \
  73   powf \
  74   rint \
  75   roundeven \
  76   roundevenf \
  77   sin \
  78   sincos \
  79   sincosf \
  80   sinf \
  81   sinh \
  82   sqrt \
  83   tan \
  84   tanh \
  85   tgamma \
  86   trunc \
  87   truncf \
  88   y0 \
  89   y1 \
  90   # bench-math
  91
  92 ifneq (,$(filter yes,$(float96-fcts)))
  93 bench-math += \
  94   cbrtl \
  95   # bench-math
  96 endif
  97
  98 ifneq (,$(filter yes,$(float128-fcts) $(float128-alias-fcts)))
  99 bench-math += \
 100   expf128 \
 101   ilogbf128 \
 102   powf128 \
 103   sinf128 \
 104   # bench-math
 105 endif
 106
 107 bench-pthread := \
 108   pthread-locks \
 109   pthread-mutex-lock \
 110   pthread-mutex-trylock \
 111   pthread-spin-lock \
 112   pthread-spin-trylock \
 113   pthread_once \
 114   thread_create \
 115   # bench-pthread
 116
 117 LDLIBS-bench-pthread-mutex-lock += -lm
 118 LDLIBS-bench-pthread-mutex-trylock += -lm
 119 LDLIBS-bench-pthread-spin-lock += -lm
 120 LDLIBS-bench-pthread-spin-trylock += -lm
 121
 122 bench-string := \
 123   ffs \
 124   ffsll \
 125   # bench-string
 126
 127 # String function benchmarks.
 128 string-benchset := \
 129   bzero \
 130   bzero-large \
 131   bzero-walk \
 132   memccpy \
 133   memchr \
 134   memcmp \
 135   memcmpeq \
 136   memcpy \
 137   memcpy-large \
 138   memcpy-random \
 139   memcpy-walk \
 140   memmem \
 141   memmove \
 142   memmove-large \
 143   memmove-walk \
 144   mempcpy \
 145   memrchr \
 146   memset \
 147   memset-large \
 148   memset-walk \
 149   memset-zero \
 150   memset-zero-large \
 151   memset-zero-walk \
 152   rawmemchr \
 153   stpcpy \
 154   stpcpy_chk \
 155   stpncpy \
 156   strcasecmp \
 157   strcasestr \
 158   strcat \
 159   strchr \
 160   strchrnul \
 161   strcmp \
 162   strcoll \
 163   strcpy \
 164   strcpy_chk \
 165   strcspn \
 166   strlen \
 167   strncasecmp \
 168   strncat \
 169   strncmp \
 170   strncpy \
 171   strnlen \
 172   strpbrk \
 173   strrchr \
 174   strsep \
 175   strspn \
 176   strstr \
 177   strtok \
 178   # string-benchset
 179
 180 # Build and run locale-dependent benchmarks only if we're building natively.
 181 ifeq (no,$(cross-compiling))
 182 wcsmbs-benchset := \
 183   wcpcpy \
 184   wcpncpy \
 185   wcrtomb \
 186   wcscat \
 187   wcschr \
 188   wcschrnul \
 189   wcscmp \
 190   wcscpy \
 191   wcscspn \
 192   wcslen \
 193   wcsncat \
 194   wcsncmp \
 195   wcsncpy \
 196   wcsnlen \
 197   wcspbrk \
 198   wcsrchr \
 199   wcsspn \
 200   wmemchr \
 201   wmemcmp \
 202   wmemset \
 203   # wcsmbs-benchset
 204 else
 205 wcsmbs-benchset :=
 206 endif
 207
 208 string-benchset-all := $(string-benchset) ${wcsmbs-benchset}
 209
 210 ifeq (no,$(cross-compiling))
 211 # We have to generate locales
 212 LOCALES := \
 213   ar_SA.UTF-8 \
 214   cs_CZ.UTF-8 \
 215   da_DK.UTF-8 \
 216   el_GR.UTF-8 \
 217   en_GB.UTF-8 \
 218   en_US.UTF-8 \
 219   es_ES.UTF-8 \
 220   fa_IR.UTF-8 \
 221   fr_FR.UTF-8 \
 222   he_IL.UTF-8 \
 223   hi_IN.UTF-8 \
 224   hu_HU.UTF-8 \
 225   is_IS.UTF-8 \
 226   it_IT.UTF-8 \
 227   ja_JP.UTF-8 \
 228   pl_PL.UTF-8 \
 229   pt_PT.UTF-8 \
 230   ru_RU.UTF-8 \
 231   si_LK.UTF-8 \
 232   sr_RS.UTF-8 \
 233   sv_SE.UTF-8 \
 234   tr_TR.UTF-8 \
 235   vi_VN.UTF-8 \
 236   zh_CN.UTF-8 \
 237   # LOCALES
 238 include ../gen-locales.mk
 239 endif
 240
 241 hash-benchset := \
 242   dl-elf-hash \
 243   dl-new-hash \
 244   nss-hash \
 245   # hash-benchset
 246
 247 stdlib-benchset := \
 248   arc4random \
 249   strtod \
 250   # stdlib-benchset
 251
 252 stdio-common-benchset := sprintf
 253
 254 math-benchset := math-inlines
 255
 256 ifeq (${BENCHSET},)
 257 benchset := \
 258   $(hash-benchset) \
 259   $(math-benchset) \
 260   $(stdio-common-benchset) \
 261   $(stdlib-benchset) \
 262   $(string-benchset-all) \
 263   # benchset
 264 else
 265 benchset := $(foreach B,$(filter %-benchset,${BENCHSET}), ${${B}})
 266 endif
 267
 268 CFLAGS-bench-ffs.c += -fno-builtin
 269 CFLAGS-bench-ffsll.c += -fno-builtin
 270 CFLAGS-bench-sqrt.c += -fno-builtin
 271 CFLAGS-bench-fmin.c += -fno-builtin
 272 CFLAGS-bench-fminf.c += -fno-builtin
 273 CFLAGS-bench-fmax.c += -fno-builtin
 274 CFLAGS-bench-fmaxf.c += -fno-builtin
 275 CFLAGS-bench-trunc.c += -fno-builtin
 276 CFLAGS-bench-truncf.c += -fno-builtin
 277 CFLAGS-bench-roundeven.c += -fno-builtin
 278 CFLAGS-bench-roundevenf.c += -fno-builtin
 279 CFLAGS-bench-isnan.c += $(config-cflags-signaling-nans)
 280 CFLAGS-bench-isinf.c += $(config-cflags-signaling-nans)
 281 CFLAGS-bench-isfinite.c += $(config-cflags-signaling-nans)
 282
 283 ifeq (${BENCHSET},)
 284 bench-malloc := \
 285   malloc-simple \
 286   malloc-thread \
 287   # bench-malloc
 288 else
 289 bench-malloc := $(filter malloc-%,${BENCHSET})
 290 endif
 291
 292 ifeq (${STATIC-BENCHTESTS},yes)
 293 +link-benchtests = $(+link-static-tests)
 294 link-libc-benchtests = $(link-libc-static)
 295 libm-benchtests = $(common-objpfx)math/libm.a
 296 thread-library-benchtests = $(static-thread-library)
 297 else
 298 link-libc-benchtests = $(link-libc)
 299 +link-benchtests = $(+link-tests)
 300 thread-library-benchtests = $(shared-thread-library)
 301 libm-benchtests = $(libm)
 302 endif
 303
 304 $(addprefix $(objpfx)bench-,$(bench-math)): $(libm-benchtests)
 305 $(addprefix $(objpfx)bench-,$(math-benchset)): $(libm-benchtests)
 306 $(addprefix $(objpfx)bench-,$(bench-pthread)): $(thread-library-benchtests)
 307 $(addprefix $(objpfx)bench-,$(bench-malloc)): $(thread-library-benchtests)
 308 $(addprefix $(objpfx)bench-,pthread-locks): $(libm-benchtests)
 309 $(addprefix $(objpfx)bench-,pthread-mutex-locks): $(libm-benchtests)
 310
 311 \f
 312
 313 # Rules to build and execute the benchmarks.  Do not put any benchmark
 314 # parameters beyond this point.
 315
 316 # We don't want the benchmark programs to run in parallel since that could
 317 # affect their performance.
 318 .NOTPARALLEL:
 319
 320 bench-extra-objs = json-lib.o
 321
 322 extra-objs += $(bench-extra-objs)
 323 others-extras = $(bench-extra-objs)
 324
 325 # The default duration: 1 seconds.
 326 ifndef BENCH_DURATION
 327 BENCH_DURATION := 1
 328 endif
 329
 330 CPPFLAGS-nonlib += -DDURATION=$(BENCH_DURATION) -D_ISOMAC
 331
 332 # Use clock_gettime to measure performance of functions.  The default is
 333 # to use the architecture-specific high precision timing instructions.
 334 ifdef USE_CLOCK_GETTIME
 335 CPPFLAGS-nonlib += -DUSE_CLOCK_GETTIME
 336 else
 337 # On x86 processors, use RDTSCP, instead of RDTSC, to measure performance
 338 # of functions.  All x86 processors since 2010 support RDTSCP instruction.
 339 ifdef USE_RDTSCP
 340 CPPFLAGS-nonlib += -DUSE_RDTSCP
 341 endif
 342 endif
 343
 344 DETAILED_OPT :=
 345
 346 ifdef DETAILED
 347 DETAILED_OPT := -d
 348 endif
 349
 350 bench-deps := bench-skeleton.c bench-timing.h Makefile
 351
 352 run-bench = $(test-wrapper-env) \
 353             $(run-program-env) \
 354             $($*-ENV) $(test-via-rtld-prefix) $${run}
 355
 356 timing-type := $(objpfx)bench-timing-type
 357 extra-objs += bench-timing-type.o
 358
 359 include ../Rules
 360
 361 bench-math += $(bench-libmvec)
 362
 363 ifeq (${BENCHSET},)
 364 bench := \
 365   $(bench-math) \
 366   $(bench-pthread) \
 367   $(bench-string) \
 368   # bench
 369 else
 370 bench := $(foreach B,$(filter bench-%,${BENCHSET}), ${${B}})
 371 endif
 372
 373 # NB: Use "=" instead of ":=" since sysdeps Makefiles may add more
 374 # benches.
 375 binaries-bench = $(addprefix $(objpfx)bench-,$(bench))
 376 extra-objs += $(addsuffix .o,$(addprefix bench-,$(bench)))
 377 binaries-benchset = $(addprefix $(objpfx)bench-,$(benchset))
 378 extra-objs += $(addsuffix .o,$(addprefix bench-,$(benchset)))
 379 binaries-bench-malloc := $(addprefix $(objpfx)bench-,$(bench-malloc))
 380 extra-objs += $(addsuffix .o,$(addprefix bench-,$(bench-malloc)))
 381
 382 # This makes sure CPPFLAGS-nonlib and CFLAGS-nonlib are passed
 383 # for all these modules.
 384 cpp-srcs-left := \
 385   $(binaries-bench-malloc:=.c) \
 386   $(binaries-bench:=.c) \
 387   $(binaries-benchset:=.c) \
 388   $(timing-type:=.c) \
 389   # cpp-srcs-left
 390 lib := nonlib
 391 include $(patsubst %,$(..)libof-iterator.mk,$(cpp-srcs-left))
 392
 393 bench-clean:
 394         rm -f $(binaries-bench) $(addsuffix .o,$(binaries-bench))
 395         rm -f $(binaries-benchset) $(addsuffix .o,$(binaries-benchset))
 396         rm -f $(binaries-bench-malloc) $(addsuffix .o,$(binaries-bench-malloc))
 397         rm -f $(timing-type) $(addsuffix .o,$(timing-type))
 398         rm -f $(addprefix $(objpfx),$(bench-extra-objs))
 399
 400 # Validate the passed in BENCHSET
 401 ifneq ($(strip ${BENCHSET}),)
 402 VALIDBENCHSETNAMES := \
 403   bench-math \
 404   bench-pthread \
 405   bench-string \
 406   hash-benchset \
 407   malloc-simple \
 408   malloc-thread \
 409   math-benchset \
 410   stdio-common-benchset \
 411   stdlib-benchset \
 412   string-benchset \
 413   wcsmbs-benchset \
 414   # VALIDBENCHSETNAMES
 415
 416 INVALIDBENCHSETNAMES := $(filter-out ${VALIDBENCHSETNAMES},${BENCHSET})
 417 ifneq (${INVALIDBENCHSETNAMES},)
 418 $(info The following values in BENCHSET are invalid: ${INVALIDBENCHSETNAMES})
 419 $(info The valid ones are: ${VALIDBENCHSETNAMES})
 420 $(error Invalid BENCHSET value)
 421 endif
 422 endif
 423
 424 bench: bench-build bench-set bench-func bench-malloc
 425
 426 # Target to only build the benchmark without running it.  We generate locales
 427 # only if we're building natively.
 428 ifeq (no,$(cross-compiling))
 429 bench-build: $(gen-locales) $(timing-type) $(binaries-bench) \
 430         $(binaries-benchset) $(binaries-bench-malloc)
 431 else
 432 bench-build: $(timing-type) $(binaries-bench) $(binaries-benchset) \
 433         $(binaries-bench-malloc)
 434 endif
 435
 436 bench-set: $(binaries-benchset)
 437         for run in $^; do \
 438           echo "Running $${run}"; \
 439           $(run-bench) > $${run}.out; \
 440         done
 441
 442 bench-malloc: $(binaries-bench-malloc)
 443         for run in $^; do \
 444           echo "$${run}"; \
 445           if [ `basename $${run}` = "bench-malloc-thread" ]; then \
 446                 for thr in 1 8 16 32; do \
 447                         echo "Running $${run} $${thr}"; \
 448                         $(run-bench) $${thr} > $${run}-$${thr}.out; \
 449                 done;\
 450           else \
 451                 for thr in 8 16 32 64 128 256 512 1024 2048 4096; do \
 452                   echo "Running $${run} $${thr}"; \
 453                   $(run-bench) $${thr} > $${run}-$${thr}.out; \
 454                 done;\
 455           fi;\
 456         done
 457
 458 # Build and execute the benchmark functions.  This target generates JSON
 459 # formatted bench.out.  Each of the programs produce independent JSON output,
 460 # so one could even execute them individually and process it using any JSON
 461 # capable language or tool.
 462 bench-func: $(binaries-bench)
 463         if [ -n '$^' ] ; then \
 464         { timing_type=$$($(test-wrapper-env) \
 465                          $(run-program-env) \
 466                          $(test-via-rtld-prefix) \
 467                          $(timing-type)); \
 468           echo "{\"timing_type\": \"$${timing_type}\","; \
 469           echo " \"functions\": {"; \
 470           for run in $^; do \
 471             op=$$($(run-bench) $(DETAILED_OPT)); \
 472             ret=$$?; \
 473             case "$${ret}" in \
 474               77) \
 475               echo "UNSUPPORTED $${run}: $${op}" >&2; \
 476                 ;; \
 477               0) \
 478                 echo "Running $${run}" >&2; \
 479                 if [ "$${run}" != "$<" ]; then \
 480                   echo ","; \
 481                 fi; \
 482                 echo "$${op}"; \
 483                 ;; \
 484               *) \
 485                 echo "FAILED $${run}" >&2; \
 486                 ;; \
 487             esac; \
 488           done; \
 489           echo; \
 490           echo " }"; \
 491           echo "}"; \
 492           } > $(objpfx)bench.out-tmp; \
 493           if [ -f $(objpfx)bench.out ]; then \
 494             mv -f $(objpfx)bench.out $(objpfx)bench.out.old; \
 495           fi; \
 496           mv -f $(objpfx)bench.out-tmp $(objpfx)bench.out; \
 497           $(PYTHON) scripts/validate_benchout.py $(objpfx)bench.out \
 498           scripts/benchout.schema.json; \
 499         fi
 500
 501 ifeq ($(bind-now),yes)
 502 link-bench-bind-now = -Wl,-z,now
 503 endif
 504
 505 bench-link-targets = $(timing-type) $(binaries-bench) $(binaries-benchset) \
 506         $(binaries-bench-malloc)
 507
 508 $(bench-link-targets): %: %.o $(objpfx)json-lib.o \
 509         $(link-extra-libs-tests) \
 510   $(sort $(filter $(common-objpfx)lib%,$(link-libc-benchtests))) \
 511   $(addprefix $(csu-objpfx),start.o) $(+preinit) $(+postinit)
 512         $(+link-benchtests)
 513
 514 $(bench-link-targets): LDFLAGS += $(link-bench-bind-now)
 515
 516 $(objpfx)bench-%.c: %-inputs $(bench-deps)
 517         { if [ -n "$($*-INCLUDE)" ]; then \
 518           cat $($*-INCLUDE); \
 519         fi; \
 520         $(PYTHON) scripts/bench.py $(patsubst %-inputs,%,$<); } > $@-tmp
 521         mv -f $@-tmp $@