benchtests/Makefile

   1 # Copyright (C) 2013-2021 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 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
  30
  31 ifneq (,$(filter yes,$(float96-fcts)))
  32 bench-math += cbrtl
  33 endif
  34
  35 ifneq (,$(filter yes,$(float128-fcts) $(float128-alias-fcts)))
  36 bench-math += expf128 powf128 sinf128 ilogbf128
  37 endif
  38
  39 bench-pthread := pthread_once thread_create pthread-locks
  40
  41 bench-string := ffs ffsll
  42
  43 ifeq (${BENCHSET},)
  44 bench := $(bench-math) $(bench-pthread) $(bench-string)
  45 else
  46 bench := $(foreach B,$(filter bench-%,${BENCHSET}), ${${B}})
  47 endif
  48
  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
  57
  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
  63 else
  64 wcsmbs-benchset :=
  65 endif
  66
  67 string-benchset-all := $(string-benchset) ${wcsmbs-benchset}
  68
  69 ifeq (no,$(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
  77 endif
  78
  79 stdlib-benchset := strtod
  80
  81 stdio-common-benchset := sprintf
  82
  83 math-benchset := math-inlines
  84
  85 ifeq (${BENCHSET},)
  86 benchset := $(string-benchset-all) $(stdlib-benchset) $(stdio-common-benchset) \
  87             $(math-benchset)
  88 else
  89 benchset := $(foreach B,$(filter %-benchset,${BENCHSET}), ${${B}})
  90 endif
  91
  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
 106
 107 ifeq (${BENCHSET},)
 108 bench-malloc := malloc-thread malloc-simple
 109 else
 110 bench-malloc := $(filter malloc-%,${BENCHSET})
 111 endif
 112
 113 $(addprefix $(objpfx)bench-,$(bench-math)): $(libm)
 114 $(addprefix $(objpfx)bench-,$(math-benchset)): $(libm)
 115 $(addprefix $(objpfx)bench-,$(bench-pthread)): $(shared-thread-library)
 116 $(addprefix $(objpfx)bench-,$(bench-malloc)): $(shared-thread-library)
 117 $(addprefix $(objpfx)bench-,pthread-locks): $(libm)
 118
 119 \f
 120
 121 # Rules to build and execute the benchmarks.  Do not put any benchmark
 122 # parameters beyond this point.
 123
 124 # We don't want the benchmark programs to run in parallel since that could
 125 # affect their performance.
 126 .NOTPARALLEL:
 127
 128 bench-extra-objs = json-lib.o
 129
 130 extra-objs += $(bench-extra-objs)
 131 others-extras = $(bench-extra-objs)
 132
 133 include ../Rules
 134
 135 binaries-bench := $(addprefix $(objpfx)bench-,$(bench))
 136 binaries-benchset := $(addprefix $(objpfx)bench-,$(benchset))
 137 binaries-bench-malloc := $(addprefix $(objpfx)bench-,$(bench-malloc))
 138
 139 # The default duration: 1 seconds.
 140 ifndef BENCH_DURATION
 141 BENCH_DURATION := 1
 142 endif
 143
 144 CPPFLAGS-nonlib += -DDURATION=$(BENCH_DURATION) -D_ISOMAC
 145
 146 # Use clock_gettime to measure performance of functions.  The default is
 147 # to use the architecture-specific high precision timing instructions.
 148 ifdef USE_CLOCK_GETTIME
 149 CPPFLAGS-nonlib += -DUSE_CLOCK_GETTIME
 150 else
 151 # On x86 processors, use RDTSCP, instead of RDTSC, to measure performance
 152 # of functions.  All x86 processors since 2010 support RDTSCP instruction.
 153 ifdef USE_RDTSCP
 154 CPPFLAGS-nonlib += -DUSE_RDTSCP
 155 endif
 156 endif
 157
 158 DETAILED_OPT :=
 159
 160 ifdef DETAILED
 161 DETAILED_OPT := -d
 162 endif
 163
 164 bench-deps := bench-skeleton.c bench-timing.h Makefile
 165
 166 run-bench = $(test-wrapper-env) \
 167             $(run-program-env) \
 168             $($*-ENV) $(test-via-rtld-prefix) $${run}
 169
 170 timing-type := $(objpfx)bench-timing-type
 171
 172 # This makes sure CPPFLAGS-nonlib and CFLAGS-nonlib are passed
 173 # for all these modules.
 174 cpp-srcs-left := $(binaries-benchset:=.c) $(binaries-bench:=.c) \
 175                  $(binaries-bench-malloc:=.c) $(timing-type:=.c)
 176 lib := nonlib
 177 include $(patsubst %,$(..)libof-iterator.mk,$(cpp-srcs-left))
 178
 179 bench-clean:
 180         rm -f $(binaries-bench) $(addsuffix .o,$(binaries-bench))
 181         rm -f $(binaries-benchset) $(addsuffix .o,$(binaries-benchset))
 182         rm -f $(binaries-bench-malloc) $(addsuffix .o,$(binaries-bench-malloc))
 183         rm -f $(timing-type) $(addsuffix .o,$(timing-type))
 184         rm -f $(addprefix $(objpfx),$(bench-extra-objs))
 185
 186 # Validate the passed in BENCHSET
 187 ifneq ($(strip ${BENCHSET}),)
 188 VALIDBENCHSETNAMES := bench-pthread bench-math bench-string string-benchset \
 189    wcsmbs-benchset stdlib-benchset stdio-common-benchset math-benchset \
 190    malloc-thread malloc-simple
 191 INVALIDBENCHSETNAMES := $(filter-out ${VALIDBENCHSETNAMES},${BENCHSET})
 192 ifneq (${INVALIDBENCHSETNAMES},)
 193 $(info The following values in BENCHSET are invalid: ${INVALIDBENCHSETNAMES})
 194 $(info The valid ones are: ${VALIDBENCHSETNAMES})
 195 $(error Invalid BENCHSET value)
 196 endif
 197 endif
 198
 199 bench: bench-build bench-set bench-func bench-malloc
 200
 201 # Target to only build the benchmark without running it.  We generate locales
 202 # only if we're building natively.
 203 ifeq (no,$(cross-compiling))
 204 bench-build: $(gen-locales) $(timing-type) $(binaries-bench) \
 205         $(binaries-benchset) $(binaries-bench-malloc)
 206 else
 207 bench-build: $(timing-type) $(binaries-bench) $(binaries-benchset) \
 208         $(binaries-bench-malloc)
 209 endif
 210
 211 bench-set: $(binaries-benchset)
 212         for run in $^; do \
 213           echo "Running $${run}"; \
 214           $(run-bench) > $${run}.out; \
 215         done
 216
 217 bench-malloc: $(binaries-bench-malloc)
 218         for run in $^; do \
 219           echo "$${run}"; \
 220           if [ `basename $${run}` = "bench-malloc-thread" ]; then \
 221                 for thr in 1 8 16 32; do \
 222                         echo "Running $${run} $${thr}"; \
 223                         $(run-bench) $${thr} > $${run}-$${thr}.out; \
 224                 done;\
 225           else \
 226                 for thr in 8 16 32 64 128 256 512 1024 2048 4096; do \
 227                   echo "Running $${run} $${thr}"; \
 228                   $(run-bench) $${thr} > $${run}-$${thr}.out; \
 229                 done;\
 230           fi;\
 231         done
 232
 233 # Build and execute the benchmark functions.  This target generates JSON
 234 # formatted bench.out.  Each of the programs produce independent JSON output,
 235 # so one could even execute them individually and process it using any JSON
 236 # capable language or tool.
 237 bench-func: $(binaries-bench)
 238         if [ -n '$^' ] ; then \
 239         { timing_type=$$($(test-wrapper-env) \
 240                          $(run-program-env) \
 241                          $(test-via-rtld-prefix) \
 242                          $(timing-type)); \
 243           echo "{\"timing_type\": \"$${timing_type}\","; \
 244           echo " \"functions\": {"; \
 245           for run in $^; do \
 246             if ! [ "x$${run}" = "x$<" ]; then \
 247               echo ","; \
 248             fi; \
 249             echo "Running $${run}" >&2; \
 250             $(run-bench) $(DETAILED_OPT); \
 251           done; \
 252           echo; \
 253           echo " }"; \
 254           echo "}"; \
 255           } > $(objpfx)bench.out-tmp; \
 256           if [ -f $(objpfx)bench.out ]; then \
 257             mv -f $(objpfx)bench.out $(objpfx)bench.out.old; \
 258           fi; \
 259           mv -f $(objpfx)bench.out-tmp $(objpfx)bench.out; \
 260           $(PYTHON) scripts/validate_benchout.py $(objpfx)bench.out \
 261           scripts/benchout.schema.json; \
 262         fi
 263
 264 ifeq ($(bind-now),yes)
 265 link-bench-bind-now = -Wl,-z,now
 266 endif
 267
 268 bench-link-targets = $(timing-type) $(binaries-bench) $(binaries-benchset) \
 269         $(binaries-bench-malloc)
 270
 271 $(bench-link-targets): %: %.o $(objpfx)json-lib.o \
 272         $(link-extra-libs-tests) \
 273   $(sort $(filter $(common-objpfx)lib%,$(link-libc))) \
 274   $(addprefix $(csu-objpfx),start.o) $(+preinit) $(+postinit)
 275         $(+link-tests)
 276
 277 $(bench-link-targets): LDFLAGS += $(link-bench-bind-now)
 278
 279 $(objpfx)bench-%.c: %-inputs $(bench-deps)
 280         { if [ -n "$($*-INCLUDE)" ]; then \
 281           cat $($*-INCLUDE); \
 282         fi; \
 283         $(PYTHON) scripts/bench.py $(patsubst %-inputs,%,$<); } > $@-tmp
 284         mv -f $@-tmp $@