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
  29
  30 ifneq (,$(filter yes,$(float128-fcts) $(float128-alias-fcts)))
  31 bench-math += expf128 powf128 sinf128
  32 endif
  33
  34 bench-pthread := pthread_once thread_create pthread-locks
  35
  36 bench-string := ffs ffsll
  37
  38 ifeq (${BENCHSET},)
  39 bench := $(bench-math) $(bench-pthread) $(bench-string)
  40 else
  41 bench := $(foreach B,$(filter bench-%,${BENCHSET}), ${${B}})
  42 endif
  43
  44 # String function benchmarks.
  45 string-benchset := memccpy memchr memcmp memcpy memmem memmove \
  46                    mempcpy memset rawmemchr stpcpy stpncpy strcasecmp strcasestr \
  47                    strcat strchr strchrnul strcmp strcpy strcspn strlen \
  48                    strncasecmp strncat strncmp strncpy strnlen strpbrk strrchr \
  49                    strspn strstr strcpy_chk stpcpy_chk memrchr strsep strtok \
  50                    strcoll memcpy-large memcpy-random memmove-large memset-large \
  51                    memcpy-walk memset-walk memmove-walk
  52
  53 # Build and run locale-dependent benchmarks only if we're building natively.
  54 ifeq (no,$(cross-compiling))
  55 wcsmbs-benchset := wcslen wcsnlen wcscpy wcpcpy wcsncpy wcpncpy wcscat wcsncat \
  56                    wcscmp wcsncmp wcschr wcschrnul wcsrchr wcsspn wcspbrk wcscspn \
  57                    wmemchr wmemset wmemcmp
  58 else
  59 wcsmbs-benchset :=
  60 endif
  61
  62 string-benchset-all := $(string-benchset) ${wcsmbs-benchset}
  63
  64 ifeq (no,$(cross-compiling))
  65 # We have to generate locales
  66 LOCALES := en_US.UTF-8 tr_TR.UTF-8 cs_CZ.UTF-8 fa_IR.UTF-8 fr_FR.UTF-8 \
  67            ja_JP.UTF-8 si_LK.UTF-8 en_GB.UTF-8 vi_VN.UTF-8 ar_SA.UTF-8 \
  68            da_DK.UTF-8 pl_PL.UTF-8 pt_PT.UTF-8 el_GR.UTF-8 ru_RU.UTF-8 \
  69            he_IL.UTF-8 is_IS.UTF-8 es_ES.UTF-8 hi_IN.UTF-8 sv_SE.UTF-8 \
  70            hu_HU.UTF-8 it_IT.UTF-8 sr_RS.UTF-8 zh_CN.UTF-8
  71 include ../gen-locales.mk
  72 endif
  73
  74 stdlib-benchset := strtod
  75
  76 stdio-common-benchset := sprintf
  77
  78 math-benchset := math-inlines
  79
  80 ifeq (${BENCHSET},)
  81 benchset := $(string-benchset-all) $(stdlib-benchset) $(stdio-common-benchset) \
  82             $(math-benchset)
  83 else
  84 benchset := $(foreach B,$(filter %-benchset,${BENCHSET}), ${${B}})
  85 endif
  86
  87 CFLAGS-bench-ffs.c += -fno-builtin
  88 CFLAGS-bench-ffsll.c += -fno-builtin
  89 CFLAGS-bench-sqrt.c += -fno-builtin
  90 CFLAGS-bench-fmin.c += -fno-builtin
  91 CFLAGS-bench-fminf.c += -fno-builtin
  92 CFLAGS-bench-fmax.c += -fno-builtin
  93 CFLAGS-bench-fmaxf.c += -fno-builtin
  94 CFLAGS-bench-trunc.c += -fno-builtin
  95 CFLAGS-bench-truncf.c += -fno-builtin
  96 CFLAGS-bench-roundeven.c += -fno-builtin
  97 CFLAGS-bench-roundevenf.c += -fno-builtin
  98 CFLAGS-bench-isnan.c += -fsignaling-nans
  99 CFLAGS-bench-isinf.c += -fsignaling-nans
 100 CFLAGS-bench-isfinite.c += -fsignaling-nans
 101
 102 ifeq (${BENCHSET},)
 103 bench-malloc := malloc-thread malloc-simple
 104 else
 105 bench-malloc := $(filter malloc-%,${BENCHSET})
 106 endif
 107
 108 $(addprefix $(objpfx)bench-,$(bench-math)): $(libm)
 109 $(addprefix $(objpfx)bench-,$(math-benchset)): $(libm)
 110 $(addprefix $(objpfx)bench-,$(bench-pthread)): $(shared-thread-library)
 111 $(addprefix $(objpfx)bench-,$(bench-malloc)): $(shared-thread-library)
 112 $(addprefix $(objpfx)bench-,pthread-locks): $(libm)
 113
 114 \f
 115
 116 # Rules to build and execute the benchmarks.  Do not put any benchmark
 117 # parameters beyond this point.
 118
 119 # We don't want the benchmark programs to run in parallel since that could
 120 # affect their performance.
 121 .NOTPARALLEL:
 122
 123 bench-extra-objs = json-lib.o
 124
 125 extra-objs += $(bench-extra-objs)
 126 others-extras = $(bench-extra-objs)
 127
 128 include ../Rules
 129
 130 binaries-bench := $(addprefix $(objpfx)bench-,$(bench))
 131 binaries-benchset := $(addprefix $(objpfx)bench-,$(benchset))
 132 binaries-bench-malloc := $(addprefix $(objpfx)bench-,$(bench-malloc))
 133
 134 # The default duration: 1 seconds.
 135 ifndef BENCH_DURATION
 136 BENCH_DURATION := 1
 137 endif
 138
 139 CPPFLAGS-nonlib += -DDURATION=$(BENCH_DURATION) -D_ISOMAC
 140
 141 # Use clock_gettime to measure performance of functions.  The default is
 142 # to use the architecture-specific high precision timing instructions.
 143 ifdef USE_CLOCK_GETTIME
 144 CPPFLAGS-nonlib += -DUSE_CLOCK_GETTIME
 145 else
 146 # On x86 processors, use RDTSCP, instead of RDTSC, to measure performance
 147 # of functions.  All x86 processors since 2010 support RDTSCP instruction.
 148 ifdef USE_RDTSCP
 149 CPPFLAGS-nonlib += -DUSE_RDTSCP
 150 endif
 151 endif
 152
 153 DETAILED_OPT :=
 154
 155 ifdef DETAILED
 156 DETAILED_OPT := -d
 157 endif
 158
 159 # This makes sure CPPFLAGS-nonlib and CFLAGS-nonlib are passed
 160 # for all these modules.
 161 cpp-srcs-left := $(binaries-benchset:=.c) $(binaries-bench:=.c) \
 162                  $(binaries-bench-malloc:=.c)
 163 lib := nonlib
 164 include $(patsubst %,$(..)libof-iterator.mk,$(cpp-srcs-left))
 165
 166 bench-deps := bench-skeleton.c bench-timing.h Makefile
 167
 168 run-bench = $(test-wrapper-env) \
 169             $(run-program-env) \
 170             $($*-ENV) $(test-via-rtld-prefix) $${run}
 171
 172 timing-type := $(objpfx)bench-timing-type
 173
 174 bench-clean:
 175         rm -f $(binaries-bench) $(addsuffix .o,$(binaries-bench))
 176         rm -f $(binaries-benchset) $(addsuffix .o,$(binaries-benchset))
 177         rm -f $(binaries-bench-malloc) $(addsuffix .o,$(binaries-bench-malloc))
 178         rm -f $(timing-type) $(addsuffix .o,$(timing-type))
 179         rm -f $(addprefix $(objpfx),$(bench-extra-objs))
 180
 181 # Validate the passed in BENCHSET
 182 ifneq ($(strip ${BENCHSET}),)
 183 VALIDBENCHSETNAMES := bench-pthread bench-math bench-string string-benchset \
 184    wcsmbs-benchset stdlib-benchset stdio-common-benchset math-benchset \
 185    malloc-thread malloc-simple
 186 INVALIDBENCHSETNAMES := $(filter-out ${VALIDBENCHSETNAMES},${BENCHSET})
 187 ifneq (${INVALIDBENCHSETNAMES},)
 188 $(info The following values in BENCHSET are invalid: ${INVALIDBENCHSETNAMES})
 189 $(info The valid ones are: ${VALIDBENCHSETNAMES})
 190 $(error Invalid BENCHSET value)
 191 endif
 192 endif
 193
 194 bench: bench-build bench-set bench-func bench-malloc
 195
 196 # Target to only build the benchmark without running it.  We generate locales
 197 # only if we're building natively.
 198 ifeq (no,$(cross-compiling))
 199 bench-build: $(gen-locales) $(timing-type) $(binaries-bench) \
 200         $(binaries-benchset) $(binaries-bench-malloc)
 201 else
 202 bench-build: $(timing-type) $(binaries-bench) $(binaries-benchset) \
 203         $(binaries-bench-malloc)
 204 endif
 205
 206 bench-set: $(binaries-benchset)
 207         for run in $^; do \
 208           echo "Running $${run}"; \
 209           $(run-bench) > $${run}.out; \
 210         done
 211
 212 bench-malloc: $(binaries-bench-malloc)
 213         for run in $^; do \
 214           echo "$${run}"; \
 215           if [ `basename $${run}` = "bench-malloc-thread" ]; then \
 216                 for thr in 1 8 16 32; do \
 217                         echo "Running $${run} $${thr}"; \
 218                         $(run-bench) $${thr} > $${run}-$${thr}.out; \
 219                 done;\
 220           else \
 221                 for thr in 8 16 32 64 128 256 512 1024 2048 4096; do \
 222                   echo "Running $${run} $${thr}"; \
 223                   $(run-bench) $${thr} > $${run}-$${thr}.out; \
 224                 done;\
 225           fi;\
 226         done
 227
 228 # Build and execute the benchmark functions.  This target generates JSON
 229 # formatted bench.out.  Each of the programs produce independent JSON output,
 230 # so one could even execute them individually and process it using any JSON
 231 # capable language or tool.
 232 bench-func: $(binaries-bench)
 233         if [ -n '$^' ] ; then \
 234         { timing_type=$$($(test-wrapper-env) \
 235                          $(run-program-env) \
 236                          $(test-via-rtld-prefix) \
 237                          $(timing-type)); \
 238           echo "{\"timing_type\": \"$${timing_type}\","; \
 239           echo " \"functions\": {"; \
 240           for run in $^; do \
 241             if ! [ "x$${run}" = "x$<" ]; then \
 242               echo ","; \
 243             fi; \
 244             echo "Running $${run}" >&2; \
 245             $(run-bench) $(DETAILED_OPT); \
 246           done; \
 247           echo; \
 248           echo " }"; \
 249           echo "}"; \
 250           } > $(objpfx)bench.out-tmp; \
 251           if [ -f $(objpfx)bench.out ]; then \
 252             mv -f $(objpfx)bench.out $(objpfx)bench.out.old; \
 253           fi; \
 254           mv -f $(objpfx)bench.out-tmp $(objpfx)bench.out; \
 255           $(PYTHON) scripts/validate_benchout.py $(objpfx)bench.out \
 256           scripts/benchout.schema.json; \
 257         fi
 258
 259 ifeq ($(bind-now),yes)
 260 link-bench-bind-now = -Wl,-z,now
 261 endif
 262
 263 bench-link-targets = $(timing-type) $(binaries-bench) $(binaries-benchset) \
 264         $(binaries-bench-malloc)
 265
 266 $(bench-link-targets): %: %.o $(objpfx)json-lib.o \
 267         $(link-extra-libs-tests) \
 268   $(sort $(filter $(common-objpfx)lib%,$(link-libc))) \
 269   $(addprefix $(csu-objpfx),start.o) $(+preinit) $(+postinit)
 270         $(+link-tests)
 271
 272 $(bench-link-targets): LDFLAGS += $(link-bench-bind-now)
 273
 274 $(objpfx)bench-%.c: %-inputs $(bench-deps)
 275         { if [ -n "$($*-INCLUDE)" ]; then \
 276           cat $($*-INCLUDE); \
 277         fi; \
 278         $(PYTHON) scripts/bench.py $(patsubst %-inputs,%,$<); } > $@-tmp
 279         mv -f $@-tmp $@