benchtests/Makefile

   1 # Copyright (C) 2013-2020 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 expf exp2f logf log2f sincosf sinf \
  27               cosf isnan isinf isfinite hypot logb logbf
  28
  29 bench-pthread := pthread_once thread_create
  30
  31 bench-string := ffs ffsll
  32
  33 ifeq (${BENCHSET},)
  34 bench := $(bench-math) $(bench-pthread) $(bench-string)
  35 else
  36 bench := $(foreach B,$(filter bench-%,${BENCHSET}), ${${B}})
  37 endif
  38
  39 # String function benchmarks.
  40 string-benchset := memccpy memchr memcmp memcpy memmem memmove \
  41                    mempcpy memset rawmemchr stpcpy stpncpy strcasecmp strcasestr \
  42                    strcat strchr strchrnul strcmp strcpy strcspn strlen \
  43                    strncasecmp strncat strncmp strncpy strnlen strpbrk strrchr \
  44                    strspn strstr strcpy_chk stpcpy_chk memrchr strsep strtok \
  45                    strcoll memcpy-large memcpy-random memmove-large memset-large \
  46                    memcpy-walk memset-walk memmove-walk
  47
  48 # Build and run locale-dependent benchmarks only if we're building natively.
  49 ifeq (no,$(cross-compiling))
  50 wcsmbs-benchset := wcslen wcsnlen wcscpy wcpcpy wcsncpy wcpncpy wcscat wcsncat \
  51                    wcscmp wcsncmp wcschr wcschrnul wcsrchr wcsspn wcspbrk wcscspn \
  52                    wmemchr wmemset wmemcmp
  53 else
  54 wcsmbs-benchset :=
  55 endif
  56
  57 string-benchset-all := $(string-benchset) ${wcsmbs-benchset}
  58
  59 ifeq (no,$(cross-compiling))
  60 # We have to generate locales
  61 LOCALES := en_US.UTF-8 tr_TR.UTF-8 cs_CZ.UTF-8 fa_IR.UTF-8 fr_FR.UTF-8 \
  62            ja_JP.UTF-8 si_LK.UTF-8 en_GB.UTF-8 vi_VN.UTF-8 ar_SA.UTF-8 \
  63            da_DK.UTF-8 pl_PL.UTF-8 pt_PT.UTF-8 el_GR.UTF-8 ru_RU.UTF-8 \
  64            he_IL.UTF-8 is_IS.UTF-8 es_ES.UTF-8 hi_IN.UTF-8 sv_SE.UTF-8 \
  65            hu_HU.UTF-8 it_IT.UTF-8 sr_RS.UTF-8 zh_CN.UTF-8
  66 include ../gen-locales.mk
  67 endif
  68
  69 stdlib-benchset := strtod
  70
  71 stdio-common-benchset := sprintf
  72
  73 math-benchset := math-inlines
  74
  75 ifeq (${BENCHSET},)
  76 benchset := $(string-benchset-all) $(stdlib-benchset) $(stdio-common-benchset) \
  77             $(math-benchset)
  78 else
  79 benchset := $(foreach B,$(filter %-benchset,${BENCHSET}), ${${B}})
  80 endif
  81
  82 CFLAGS-bench-ffs.c += -fno-builtin
  83 CFLAGS-bench-ffsll.c += -fno-builtin
  84 CFLAGS-bench-sqrt.c += -fno-builtin
  85 CFLAGS-bench-fmin.c += -fno-builtin
  86 CFLAGS-bench-fminf.c += -fno-builtin
  87 CFLAGS-bench-fmax.c += -fno-builtin
  88 CFLAGS-bench-fmaxf.c += -fno-builtin
  89 CFLAGS-bench-trunc.c += -fno-builtin
  90 CFLAGS-bench-truncf.c += -fno-builtin
  91 CFLAGS-bench-isnan.c += -fsignaling-nans
  92 CFLAGS-bench-isinf.c += -fsignaling-nans
  93 CFLAGS-bench-isfinite.c += -fsignaling-nans
  94
  95 ifeq (${BENCHSET},)
  96 bench-malloc := malloc-thread malloc-simple
  97 else
  98 bench-malloc := $(filter malloc-%,${BENCHSET})
  99 endif
 100
 101 $(addprefix $(objpfx)bench-,$(bench-math)): $(libm)
 102 $(addprefix $(objpfx)bench-,$(math-benchset)): $(libm)
 103 $(addprefix $(objpfx)bench-,$(bench-pthread)): $(shared-thread-library)
 104 $(addprefix $(objpfx)bench-,$(bench-malloc)): $(shared-thread-library)
 105
 106 \f
 107
 108 # Rules to build and execute the benchmarks.  Do not put any benchmark
 109 # parameters beyond this point.
 110
 111 # We don't want the benchmark programs to run in parallel since that could
 112 # affect their performance.
 113 .NOTPARALLEL:
 114
 115 bench-extra-objs = json-lib.o
 116
 117 extra-objs += $(bench-extra-objs)
 118 others-extras = $(bench-extra-objs)
 119
 120 include ../Rules
 121
 122 binaries-bench := $(addprefix $(objpfx)bench-,$(bench))
 123 binaries-benchset := $(addprefix $(objpfx)bench-,$(benchset))
 124 binaries-bench-malloc := $(addprefix $(objpfx)bench-,$(bench-malloc))
 125
 126 # The default duration: 1 seconds.
 127 ifndef BENCH_DURATION
 128 BENCH_DURATION := 1
 129 endif
 130
 131 CPPFLAGS-nonlib += -DDURATION=$(BENCH_DURATION) -D_ISOMAC
 132
 133 # On x86 processors, use RDTSCP, instead of RDTSC, to measure performance
 134 # of functions.  All x86 processors since 2010 support RDTSCP instruction.
 135 ifdef USE_RDTSCP
 136 CPPFLAGS-nonlib += -DUSE_RDTSCP
 137 endif
 138
 139 DETAILED_OPT :=
 140
 141 ifdef DETAILED
 142 DETAILED_OPT := -d
 143 endif
 144
 145 # This makes sure CPPFLAGS-nonlib and CFLAGS-nonlib are passed
 146 # for all these modules.
 147 cpp-srcs-left := $(binaries-benchset:=.c) $(binaries-bench:=.c) \
 148                  $(binaries-bench-malloc:=.c)
 149 lib := nonlib
 150 include $(patsubst %,$(..)libof-iterator.mk,$(cpp-srcs-left))
 151
 152 bench-deps := bench-skeleton.c bench-timing.h Makefile
 153
 154 run-bench = $(test-wrapper-env) \
 155             $(run-program-env) \
 156             $($*-ENV) $(test-via-rtld-prefix) $${run}
 157
 158 timing-type := $(objpfx)bench-timing-type
 159
 160 bench-clean:
 161         rm -f $(binaries-bench) $(addsuffix .o,$(binaries-bench))
 162         rm -f $(binaries-benchset) $(addsuffix .o,$(binaries-benchset))
 163         rm -f $(binaries-bench-malloc) $(addsuffix .o,$(binaries-bench-malloc))
 164         rm -f $(timing-type) $(addsuffix .o,$(timing-type))
 165         rm -f $(addprefix $(objpfx),$(bench-extra-objs))
 166
 167 # Validate the passed in BENCHSET
 168 ifneq ($(strip ${BENCHSET}),)
 169 VALIDBENCHSETNAMES := bench-pthread bench-math bench-string string-benchset \
 170    wcsmbs-benchset stdlib-benchset stdio-common-benchset math-benchset \
 171    malloc-thread malloc-simple
 172 INVALIDBENCHSETNAMES := $(filter-out ${VALIDBENCHSETNAMES},${BENCHSET})
 173 ifneq (${INVALIDBENCHSETNAMES},)
 174 $(info The following values in BENCHSET are invalid: ${INVALIDBENCHSETNAMES})
 175 $(info The valid ones are: ${VALIDBENCHSETNAMES})
 176 $(error Invalid BENCHSET value)
 177 endif
 178 endif
 179
 180 bench: bench-build bench-set bench-func bench-malloc
 181
 182 # Target to only build the benchmark without running it.  We generate locales
 183 # only if we're building natively.
 184 ifeq (no,$(cross-compiling))
 185 bench-build: $(gen-locales) $(timing-type) $(binaries-bench) \
 186         $(binaries-benchset) $(binaries-bench-malloc)
 187 else
 188 bench-build: $(timing-type) $(binaries-bench) $(binaries-benchset) \
 189         $(binaries-bench-malloc)
 190 endif
 191
 192 bench-set: $(binaries-benchset)
 193         for run in $^; do \
 194           echo "Running $${run}"; \
 195           $(run-bench) > $${run}.out; \
 196         done
 197
 198 bench-malloc: $(binaries-bench-malloc)
 199         for run in $^; do \
 200           echo "$${run}"; \
 201           if [ `basename $${run}` = "bench-malloc-thread" ]; then \
 202                 for thr in 1 8 16 32; do \
 203                         echo "Running $${run} $${thr}"; \
 204                         $(run-bench) $${thr} > $${run}-$${thr}.out; \
 205                 done;\
 206           else \
 207                 for thr in 8 16 32 64 128 256 512 1024 2048 4096; do \
 208                   echo "Running $${run} $${thr}"; \
 209                   $(run-bench) $${thr} > $${run}-$${thr}.out; \
 210                 done;\
 211           fi;\
 212         done
 213
 214 # Build and execute the benchmark functions.  This target generates JSON
 215 # formatted bench.out.  Each of the programs produce independent JSON output,
 216 # so one could even execute them individually and process it using any JSON
 217 # capable language or tool.
 218 bench-func: $(binaries-bench)
 219         if [ -n '$^' ] ; then \
 220         { timing_type=$$($(test-wrapper-env) \
 221                          $(run-program-env) \
 222                          $(test-via-rtld-prefix) \
 223                          $(timing-type)); \
 224           echo "{\"timing_type\": \"$${timing_type}\","; \
 225           echo " \"functions\": {"; \
 226           for run in $^; do \
 227             if ! [ "x$${run}" = "x$<" ]; then \
 228               echo ","; \
 229             fi; \
 230             echo "Running $${run}" >&2; \
 231             $(run-bench) $(DETAILED_OPT); \
 232           done; \
 233           echo; \
 234           echo " }"; \
 235           echo "}"; \
 236           } > $(objpfx)bench.out-tmp; \
 237           if [ -f $(objpfx)bench.out ]; then \
 238             mv -f $(objpfx)bench.out $(objpfx)bench.out.old; \
 239           fi; \
 240           mv -f $(objpfx)bench.out-tmp $(objpfx)bench.out; \
 241           $(PYTHON) scripts/validate_benchout.py $(objpfx)bench.out \
 242           scripts/benchout.schema.json; \
 243         fi
 244
 245 ifeq ($(bind-now),yes)
 246 link-bench-bind-now = -Wl,-z,now
 247 endif
 248
 249 bench-link-targets = $(timing-type) $(binaries-bench) $(binaries-benchset) \
 250         $(binaries-bench-malloc)
 251
 252 $(bench-link-targets): %: %.o $(objpfx)json-lib.o \
 253         $(link-extra-libs-tests) \
 254   $(sort $(filter $(common-objpfx)lib%,$(link-libc))) \
 255   $(addprefix $(csu-objpfx),start.o) $(+preinit) $(+postinit)
 256         $(+link-tests)
 257
 258 $(bench-link-targets): LDFLAGS += $(link-bench-bind-now)
 259
 260 $(objpfx)bench-%.c: %-inputs $(bench-deps)
 261         { if [ -n "$($*-INCLUDE)" ]; then \
 262           cat $($*-INCLUDE); \
 263         fi; \
 264         $(PYTHON) scripts/bench.py $(patsubst %-inputs,%,$<); } > $@-tmp
 265         mv -f $@-tmp $@