sysdeps/x86_64/fpu/Makefile

   1 ifeq ($(subdir),mathvec)
   2 libmvec-support += svml_d_cos2_core svml_d_cos4_core_avx \
   3                    svml_d_cos4_core svml_d_cos8_core \
   4                    svml_d_sin2_core svml_d_sin4_core_avx \
   5                    svml_d_sin4_core svml_d_sin8_core svml_d_trig_data \
   6                    svml_s_cosf4_core svml_s_cosf8_core_avx \
   7                    svml_s_cosf8_core svml_s_cosf16_core svml_s_trig_data \
   8                    svml_s_sinf4_core svml_s_sinf8_core_avx \
   9                    svml_s_sinf8_core svml_s_sinf16_core \
  10                    svml_d_sincos2_core svml_d_sincos4_core_avx \
  11                    svml_d_sincos4_core svml_d_sincos8_core \
  12                    svml_d_log2_core svml_d_log4_core_avx svml_d_log4_core \
  13                    svml_d_log8_core svml_d_log_data svml_s_logf4_core \
  14                    svml_s_logf8_core_avx svml_s_logf8_core svml_s_logf16_core \
  15                    svml_s_logf_data svml_d_exp2_core svml_d_exp4_core_avx \
  16                    svml_d_exp4_core svml_d_exp8_core svml_d_exp_data \
  17                    svml_s_expf4_core svml_s_expf8_core_avx svml_s_expf8_core \
  18                    svml_s_expf16_core svml_s_expf_data svml_d_pow2_core \
  19                    svml_d_pow4_core_avx svml_d_pow4_core svml_d_pow8_core \
  20                    svml_d_pow_data svml_s_powf4_core svml_s_powf8_core_avx \
  21                    svml_s_powf8_core svml_s_powf16_core svml_s_powf_data \
  22                    svml_s_sincosf4_core svml_s_sincosf8_core_avx \
  23                    svml_s_sincosf8_core svml_s_sincosf16_core svml_finite_alias
  24
  25 libmvec-static-only-routines = svml_finite_alias
  26 endif
  27
  28 # Variables for libmvec tests.
  29 ifeq ($(subdir),math)
  30 ifeq ($(build-mathvec),yes)
  31 libmvec-tests += double-vlen2 double-vlen4 double-vlen4-avx2 \
  32                  float-vlen4 float-vlen8 float-vlen8-avx2
  33 tests += test-double-libmvec-alias test-double-libmvec-alias-avx \
  34          test-double-libmvec-alias-avx2 test-double-libmvec-alias-main \
  35          test-double-libmvec-alias-avx-main test-double-libmvec-alias-avx2-main \
  36          test-float-libmvec-alias test-float-libmvec-alias-avx \
  37          test-float-libmvec-alias-avx2 test-float-libmvec-alias-main \
  38          test-float-libmvec-alias-avx-main test-float-libmvec-alias-avx2-main \
  39          test-double-libmvec-sincos test-double-libmvec-sincos-avx \
  40          test-double-libmvec-sincos-avx2 test-float-libmvec-sincosf \
  41          test-float-libmvec-sincosf-avx test-float-libmvec-sincosf-avx2
  42 modules-names += test-double-libmvec-alias-mod \
  43                  test-double-libmvec-alias-avx-mod \
  44                  test-double-libmvec-alias-avx2-mod \
  45                  test-float-libmvec-alias-mod \
  46                  test-float-libmvec-alias-avx-mod \
  47                  test-float-libmvec-alias-avx2-mod
  48 modules-names-tests += test-double-libmvec-alias-mod \
  49                  test-double-libmvec-alias-avx-mod \
  50                  test-double-libmvec-alias-avx2-mod \
  51                  test-float-libmvec-alias-mod \
  52                  test-float-libmvec-alias-avx-mod \
  53                  test-float-libmvec-alias-avx2-mod
  54 extra-test-objs += test-double-libmvec-sincos-avx-main.o \
  55                    test-double-libmvec-sincos-avx2-main.o \
  56                    test-double-libmvec-sincos-main.o \
  57                    test-float-libmvec-sincosf-avx-main.o \
  58                    test-float-libmvec-sincosf-avx2-main.o\
  59                    test-float-libmvec-sincosf-main.o
  60 test-double-libmvec-alias-mod.so-no-z-defs = yes
  61 test-double-libmvec-alias-avx-mod.so-no-z-defs = yes
  62 test-double-libmvec-alias-avx2-mod.so-no-z-defs = yes
  63 test-float-libmvec-alias-mod.so-no-z-defs = yes
  64 test-float-libmvec-alias-avx-mod.so-no-z-defs = yes
  65 test-float-libmvec-alias-avx2-mod.so-no-z-defs = yes
  66
  67 $(objpfx)test-double-libmvec-alias: \
  68   $(objpfx)test-double-libmvec-alias-mod.so
  69 $(objpfx)test-double-libmvec-alias-mod.so: \
  70   $(objpfx)../mathvec/libmvec_nonshared.a $(libmvec)
  71
  72 $(objpfx)test-double-libmvec-alias-avx: \
  73   $(objpfx)test-double-libmvec-alias-avx-mod.so
  74 $(objpfx)test-double-libmvec-alias-avx-mod.so: \
  75   $(objpfx)../mathvec/libmvec_nonshared.a $(libmvec)
  76
  77 $(objpfx)test-double-libmvec-alias-avx2: \
  78   $(objpfx)test-double-libmvec-alias-avx2-mod.so
  79 $(objpfx)test-double-libmvec-alias-avx2-mod.so: \
  80   $(objpfx)../mathvec/libmvec_nonshared.a $(libmvec)
  81
  82 $(objpfx)test-double-libmvec-alias-main: \
  83   $(objpfx)test-double-libmvec-alias-mod.os \
  84   $(objpfx)../mathvec/libmvec_nonshared.a $(libmvec)
  85
  86 $(objpfx)test-double-libmvec-alias-avx-main: \
  87   $(objpfx)test-double-libmvec-alias-avx-mod.os \
  88   $(objpfx)../mathvec/libmvec_nonshared.a $(libmvec)
  89
  90 $(objpfx)test-double-libmvec-alias-avx2-main: \
  91   $(objpfx)test-double-libmvec-alias-avx2-mod.os \
  92   $(objpfx)../mathvec/libmvec_nonshared.a $(libmvec)
  93
  94 $(objpfx)test-float-libmvec-alias: \
  95   $(objpfx)test-float-libmvec-alias-mod.so
  96 $(objpfx)test-float-libmvec-alias-mod.so: \
  97   $(objpfx)../mathvec/libmvec_nonshared.a $(libmvec)
  98
  99 $(objpfx)test-float-libmvec-alias-avx: \
 100   $(objpfx)test-float-libmvec-alias-avx-mod.so
 101 $(objpfx)test-float-libmvec-alias-avx-mod.so: \
 102   $(objpfx)../mathvec/libmvec_nonshared.a $(libmvec)
 103
 104 $(objpfx)test-float-libmvec-alias-avx2: \
 105   $(objpfx)test-float-libmvec-alias-avx2-mod.so
 106 $(objpfx)test-float-libmvec-alias-avx2-mod.so: \
 107   $(objpfx)../mathvec/libmvec_nonshared.a $(libmvec)
 108
 109 $(objpfx)test-float-libmvec-alias-main: \
 110   $(objpfx)test-float-libmvec-alias-mod.os \
 111   $(objpfx)../mathvec/libmvec_nonshared.a $(libmvec)
 112
 113 $(objpfx)test-float-libmvec-alias-avx-main: \
 114   $(objpfx)test-float-libmvec-alias-avx-mod.os \
 115   $(objpfx)../mathvec/libmvec_nonshared.a $(libmvec)
 116
 117 $(objpfx)test-float-libmvec-alias-avx2-main: \
 118   $(objpfx)test-float-libmvec-alias-avx2-mod.os \
 119   $(objpfx)../mathvec/libmvec_nonshared.a $(libmvec)
 120
 121 $(objpfx)test-double-libmvec-sincos: \
 122   $(objpfx)test-double-libmvec-sincos.o \
 123   $(objpfx)test-double-libmvec-sincos-main.o $(libmvec)
 124
 125 $(objpfx)test-double-libmvec-sincos-avx: \
 126   $(objpfx)test-double-libmvec-sincos-avx.o \
 127   $(objpfx)test-double-libmvec-sincos-avx-main.o $(libmvec)
 128
 129 $(objpfx)test-double-libmvec-sincos-avx2: \
 130   $(objpfx)test-double-libmvec-sincos-avx2.o \
 131   $(objpfx)test-double-libmvec-sincos-avx2-main.o $(libmvec)
 132
 133 $(objpfx)test-float-libmvec-sincosf: \
 134   $(objpfx)test-float-libmvec-sincosf.o \
 135   $(objpfx)test-float-libmvec-sincosf-main.o $(libmvec)
 136
 137 $(objpfx)test-float-libmvec-sincosf-avx: \
 138   $(objpfx)test-float-libmvec-sincosf-avx.o \
 139   $(objpfx)test-float-libmvec-sincosf-avx-main.o $(libmvec)
 140
 141 $(objpfx)test-float-libmvec-sincosf-avx2: \
 142   $(objpfx)test-float-libmvec-sincosf-avx2.o \
 143   $(objpfx)test-float-libmvec-sincosf-avx2-main.o $(libmvec)
 144
 145 ifeq (yes,$(config-cflags-avx512))
 146 libmvec-tests += double-vlen8 float-vlen16
 147 tests += test-double-libmvec-alias-avx512 \
 148          test-float-libmvec-alias-avx512 \
 149          test-double-libmvec-alias-avx512-main \
 150          test-float-libmvec-alias-avx512-main \
 151          test-double-libmvec-sincos-avx512 \
 152          test-float-libmvec-sincosf-avx512
 153 modules-names += test-double-libmvec-alias-avx512-mod \
 154                  test-float-libmvec-alias-avx512-mod
 155 modules-names-tests += test-double-libmvec-alias-avx512-mod \
 156                  test-float-libmvec-alias-avx512-mod
 157 extra-test-objs += test-double-libmvec-sincos-avx512-main.o \
 158                    test-float-libmvec-sincosf-avx512-main.o
 159 test-double-libmvec-alias-avx512-mod.so-no-z-defs = yes
 160 test-float-libmvec-alias-avx512-mod.so-no-z-defs = yes
 161
 162 $(objpfx)test-double-libmvec-alias-avx512: \
 163   $(objpfx)test-double-libmvec-alias-avx512-mod.so
 164 $(objpfx)test-double-libmvec-alias-avx512-mod.so: \
 165   $(objpfx)../mathvec/libmvec_nonshared.a $(libmvec)
 166
 167 $(objpfx)test-double-libmvec-alias-avx512-main: \
 168   $(objpfx)test-double-libmvec-alias-avx512-mod.os \
 169   $(objpfx)../mathvec/libmvec_nonshared.a $(libmvec)
 170
 171 $(objpfx)test-float-libmvec-alias-avx512: \
 172   $(objpfx)test-float-libmvec-alias-avx512-mod.so
 173 $(objpfx)test-float-libmvec-alias-avx512-mod.so: \
 174   $(objpfx)../mathvec/libmvec_nonshared.a $(libmvec)
 175
 176 $(objpfx)test-float-libmvec-alias-avx512-main: \
 177   $(objpfx)test-float-libmvec-alias-avx512-mod.os \
 178   $(objpfx)../mathvec/libmvec_nonshared.a $(libmvec)
 179
 180 $(objpfx)test-double-libmvec-sincos-avx512: \
 181   $(objpfx)test-double-libmvec-sincos-avx512.o \
 182   $(objpfx)test-double-libmvec-sincos-avx512-main.o $(libmvec)
 183
 184 $(objpfx)test-float-libmvec-sincosf-avx512: \
 185   $(objpfx)test-float-libmvec-sincosf-avx512.o \
 186   $(objpfx)test-float-libmvec-sincosf-avx512-main.o $(libmvec)
 187 endif
 188
 189 double-vlen2-funcs = cos exp log pow sin sincos
 190 double-vlen4-funcs = cos exp log pow sin sincos
 191 double-vlen4-avx2-funcs = cos exp log pow sin sincos
 192 double-vlen8-funcs = cos exp log pow sin sincos
 193 float-vlen4-funcs = cos exp log pow sin sincos
 194 float-vlen8-funcs = cos exp log pow sin sincos
 195 float-vlen8-avx2-funcs = cos exp log pow sin sincos
 196 float-vlen16-funcs = cos exp log pow sin sincos
 197
 198 double-vlen4-arch-ext-cflags = -mavx
 199 double-vlen4-arch-ext2-cflags = -mavx2
 200 double-vlen8-arch-ext-cflags = -mavx512f
 201
 202 float-vlen8-arch-ext-cflags = -mavx
 203 float-vlen8-arch-ext2-cflags = -mavx2
 204 float-vlen16-arch-ext-cflags = -mavx512f
 205
 206 libmvec-sincos-cflags = $(libm-test-fast-math-cflags) -fno-inline -fopenmp -Wno-unknown-pragmas
 207 libmvec-alias-cflags = $(libmvec-sincos-cflags) -ffloat-store -ffinite-math-only
 208
 209 CFLAGS-test-double-libmvec-alias-mod.c = $(libmvec-alias-cflags)
 210 CFLAGS-test-double-libmvec-alias-avx-mod.c = $(double-vlen4-arch-ext-cflags) $(libmvec-alias-cflags) -DREQUIRE_AVX
 211 CFLAGS-test-double-libmvec-alias-avx2-mod.c = $(double-vlen4-arch-ext2-cflags) $(libmvec-alias-cflags) -DREQUIRE_AVX2
 212 CFLAGS-test-double-libmvec-alias-avx512-mod.c = $(double-vlen8-arch-ext-cflags) $(libmvec-alias-cflags) -DREQUIRE_AVX512F
 213
 214 CFLAGS-test-float-libmvec-alias-mod.c = $(libmvec-alias-cflags)
 215 CFLAGS-test-float-libmvec-alias-avx-mod.c = $(double-vlen4-arch-ext-cflags) $(libmvec-alias-cflags) -DREQUIRE_AVX
 216 CFLAGS-test-float-libmvec-alias-avx2-mod.c = $(double-vlen4-arch-ext2-cflags) $(libmvec-alias-cflags) -DREQUIRE_AVX2
 217 CFLAGS-test-float-libmvec-alias-avx512-mod.c = $(double-vlen8-arch-ext-cflags) $(libmvec-alias-cflags) -DREQUIRE_AVX512F
 218
 219 CFLAGS-test-double-vlen4-avx2-wrappers.c = $(double-vlen4-arch-ext2-cflags)
 220
 221 CFLAGS-test-float-vlen8-avx2-wrappers.c = $(float-vlen8-arch-ext2-cflags)
 222
 223 CFLAGS-test-double-libmvec-sincos-main.c = $(libmvec-sincos-cflags)
 224 CFLAGS-test-double-libmvec-sincos-avx.c = -DREQUIRE_AVX
 225 CFLAGS-test-double-libmvec-sincos-avx-main.c = $(libmvec-sincos-cflags) $(double-vlen4-arch-ext-cflags)
 226 CFLAGS-test-double-libmvec-sincos-avx2.c = -DREQUIRE_AVX2
 227 CFLAGS-test-double-libmvec-sincos-avx2-main.c = $(libmvec-sincos-cflags) $(double-vlen4-arch-ext2-cflags)
 228 CFLAGS-test-double-libmvec-sincos-avx512.c = -DREQUIRE_AVX512F
 229 CFLAGS-test-double-libmvec-sincos-avx512-main.c = $(libmvec-sincos-cflags) $(double-vlen8-arch-ext-cflags)
 230
 231 CFLAGS-test-float-libmvec-sincosf-main.c = $(libmvec-sincos-cflags)
 232 CFLAGS-test-float-libmvec-sincosf-avx.c = -DREQUIRE_AVX
 233 CFLAGS-test-float-libmvec-sincosf-avx-main.c = $(libmvec-sincos-cflags) $(float-vlen8-arch-ext-cflags)
 234 CFLAGS-test-float-libmvec-sincosf-avx2.c = -DREQUIRE_AVX2
 235 CFLAGS-test-float-libmvec-sincosf-avx2-main.c = $(libmvec-sincos-cflags) $(float-vlen8-arch-ext2-cflags)
 236 CFLAGS-test-float-libmvec-sincosf-avx512.c = -DREQUIRE_AVX512F
 237 CFLAGS-test-float-libmvec-sincosf-avx512-main.c = $(libmvec-sincos-cflags) $(float-vlen16-arch-ext-cflags)
 238 endif
 239 endif
 240
 241 ifeq ($(subdir)$(config-cflags-mprefer-vector-width),mathyes)
 242 # When compiled with -O3 -march=skylake, GCC 8 and 9 optimize some loops
 243 # in branred.c with 256-bit vector instructions, which leads to store
 244 # forward stall:
 245 #
 246 # https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90579
 247 #
 248 # Limit vector width to 128 bits to work around this issue.  It improves
 249 # performance of sin and cos by more than 40% on Skylake.
 250 CFLAGS-branred.c = -mprefer-vector-width=128
 251 endif