benchtests/Makefile

   1 # Copyright (C) 2013-2019 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 # <http://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
  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
  92 ifeq (${BENCHSET},)
  93 bench-malloc := malloc-thread malloc-simple
  94 else
  95 bench-malloc := $(filter malloc-%,${BENCHSET})
  96 endif
  97
  98 $(addprefix $(objpfx)bench-,$(bench-math)): $(libm)
  99 $(addprefix $(objpfx)bench-,$(math-benchset)): $(libm)
 100 $(addprefix $(objpfx)bench-,$(bench-pthread)): $(shared-thread-library)
 101 $(addprefix $(objpfx)bench-,$(bench-malloc)): $(shared-thread-library)
 102
 103 \f
 104
 105 # Rules to build and execute the benchmarks.  Do not put any benchmark
 106 # parameters beyond this point.
 107
 108 # We don't want the benchmark programs to run in parallel since that could
 109 # affect their performance.
 110 .NOTPARALLEL:
 111
 112 bench-extra-objs = json-lib.o
 113
 114 extra-objs += $(bench-extra-objs)
 115 others-extras = $(bench-extra-objs)
 116
 117 include ../Rules
 118
 119 binaries-bench := $(addprefix $(objpfx)bench-,$(bench))
 120 binaries-benchset := $(addprefix $(objpfx)bench-,$(benchset))
 121 binaries-bench-malloc := $(addprefix $(objpfx)bench-,$(bench-malloc))
 122
 123 # The default duration: 10 seconds.
 124 ifndef BENCH_DURATION
 125 BENCH_DURATION := 10
 126 endif
 127
 128 CPPFLAGS-nonlib += -DDURATION=$(BENCH_DURATION) -D_ISOMAC
 129
 130 # On x86 processors, use RDTSCP, instead of RDTSC, to measure performance
 131 # of functions.  All x86 processors since 2010 support RDTSCP instruction.
 132 ifdef USE_RDTSCP
 133 CPPFLAGS-nonlib += -DUSE_RDTSCP
 134 endif
 135
 136 DETAILED_OPT :=
 137
 138 ifdef DETAILED
 139 DETAILED_OPT := -d
 140 endif
 141
 142 # This makes sure CPPFLAGS-nonlib and CFLAGS-nonlib are passed
 143 # for all these modules.
 144 cpp-srcs-left := $(binaries-benchset:=.c) $(binaries-bench:=.c) \
 145                  $(binaries-bench-malloc:=.c)
 146 lib := nonlib
 147 include $(patsubst %,$(..)libof-iterator.mk,$(cpp-srcs-left))
 148
 149 bench-deps := bench-skeleton.c bench-timing.h Makefile
 150
 151 run-bench = $(test-wrapper-env) \
 152             $(run-program-env) \
 153             $($*-ENV) $(test-via-rtld-prefix) $${run}
 154
 155 timing-type := $(objpfx)bench-timing-type
 156
 157 bench-clean:
 158         rm -f $(binaries-bench) $(addsuffix .o,$(binaries-bench))
 159         rm -f $(binaries-benchset) $(addsuffix .o,$(binaries-benchset))
 160         rm -f $(binaries-bench-malloc) $(addsuffix .o,$(binaries-bench-malloc))
 161         rm -f $(timing-type) $(addsuffix .o,$(timing-type))
 162         rm -f $(addprefix $(objpfx),$(bench-extra-objs))
 163
 164 # Validate the passed in BENCHSET
 165 ifneq ($(strip ${BENCHSET}),)
 166 VALIDBENCHSETNAMES := bench-pthread bench-math bench-string string-benchset \
 167    wcsmbs-benchset stdlib-benchset stdio-common-benchset math-benchset \
 168    malloc-thread malloc-simple
 169 INVALIDBENCHSETNAMES := $(filter-out ${VALIDBENCHSETNAMES},${BENCHSET})
 170 ifneq (${INVALIDBENCHSETNAMES},)
 171 $(info The following values in BENCHSET are invalid: ${INVALIDBENCHSETNAMES})
 172 $(info The valid ones are: ${VALIDBENCHSETNAMES})
 173 $(error Invalid BENCHSET value)
 174 endif
 175 endif
 176
 177 bench: bench-build bench-set bench-func bench-malloc
 178
 179 # Target to only build the benchmark without running it.  We generate locales
 180 # only if we're building natively.
 181 ifeq (no,$(cross-compiling))
 182 bench-build: $(gen-locales) $(timing-type) $(binaries-bench) \
 183         $(binaries-benchset) $(binaries-bench-malloc)
 184 else
 185 bench-build: $(timing-type) $(binaries-bench) $(binaries-benchset) \
 186         $(binaries-bench-malloc)
 187 endif
 188
 189 bench-set: $(binaries-benchset)
 190         for run in $^; do \
 191           echo "Running $${run}"; \
 192           $(run-bench) > $${run}.out; \
 193         done
 194
 195 bench-malloc: $(binaries-bench-malloc)
 196         for run in $^; do \
 197           echo "$${run}"; \
 198           if [ `basename $${run}` = "bench-malloc-thread" ]; then \
 199                 for thr in 1 8 16 32; do \
 200                         echo "Running $${run} $${thr}"; \
 201                         $(run-bench) $${thr} > $${run}-$${thr}.out; \
 202                 done;\
 203           else \
 204                 for thr in 8 16 32 64 128 256 512 1024 2048 4096; do \
 205                   echo "Running $${run} $${thr}"; \
 206                   $(run-bench) $${thr} > $${run}-$${thr}.out; \
 207                 done;\
 208           fi;\
 209         done
 210
 211 # Build and execute the benchmark functions.  This target generates JSON
 212 # formatted bench.out.  Each of the programs produce independent JSON output,
 213 # so one could even execute them individually and process it using any JSON
 214 # capable language or tool.
 215 bench-func: $(binaries-bench)
 216         if [ -n '$^' ] ; then \
 217         { timing_type=$$($(timing-type)); \
 218           echo "{\"timing_type\": \"$${timing_type}\","; \
 219           echo " \"functions\": {"; \
 220           for run in $^; do \
 221             if ! [ "x$${run}" = "x$<" ]; then \
 222               echo ","; \
 223             fi; \
 224             echo "Running $${run}" >&2; \
 225             $(run-bench) $(DETAILED_OPT); \
 226           done; \
 227           echo; \
 228           echo " }"; \
 229           echo "}"; \
 230           } > $(objpfx)bench.out-tmp; \
 231           if [ -f $(objpfx)bench.out ]; then \
 232             mv -f $(objpfx)bench.out $(objpfx)bench.out.old; \
 233           fi; \
 234           mv -f $(objpfx)bench.out-tmp $(objpfx)bench.out; \
 235           $(PYTHON) scripts/validate_benchout.py $(objpfx)bench.out \
 236           scripts/benchout.schema.json; \
 237         fi
 238
 239 $(timing-type) $(binaries-bench) $(binaries-benchset) \
 240         $(binaries-bench-malloc): %: %.o $(objpfx)json-lib.o \
 241         $(link-extra-libs-tests) \
 242   $(sort $(filter $(common-objpfx)lib%,$(link-libc))) \
 243   $(addprefix $(csu-objpfx),start.o) $(+preinit) $(+postinit)
 244         $(+link-tests)
 245
 246 $(objpfx)bench-%.c: %-inputs $(bench-deps)
 247         { if [ -n "$($*-INCLUDE)" ]; then \
 248           cat $($*-INCLUDE); \
 249         fi; \
 250         $(PYTHON) scripts/bench.py $(patsubst %-inputs,%,$<); } > $@-tmp
 251         mv -f $@-tmp $@