ax_ext.m4 15.3 KB
Newer Older
Pedro Gonnet's avatar
Pedro Gonnet committed
1
# ===========================================================================
Josh Borrow's avatar
Josh Borrow committed
2
#          https://www.gnu.org/software/autoconf-archive/ax_ext.html
Pedro Gonnet's avatar
Pedro Gonnet committed
3
4
5
6
7
8
9
10
# ===========================================================================
#
# SYNOPSIS
#
#   AX_EXT
#
# DESCRIPTION
#
11
#   Find supported SIMD extensions by requesting cpuid. When a SIMD
12
#   extension is found, the -m"simdextensionname" is added to SIMD_FLAGS if
13
#   compiler supports it. For example, if "sse2" is available then "-msse2"
14
#   is added to SIMD_FLAGS.
Pedro Gonnet's avatar
Pedro Gonnet committed
15
#
16
17
18
19
20
#   Find other supported CPU extensions by requesting cpuid. When a
#   processor extension is found, the -m"extensionname" is added to
#   CPUEXT_FLAGS if compiler supports it. For example, if "bmi2" is
#   available then "-mbmi2" is added to CPUEXT_FLAGS.
#
Pedro Gonnet's avatar
Pedro Gonnet committed
21
22
23
#   This macro calls:
#
#     AC_SUBST(SIMD_FLAGS)
24
#     AC_SUBST(CPUEXT_FLAGS)
Pedro Gonnet's avatar
Pedro Gonnet committed
25
26
27
#
#   And defines:
#
28
29
30
31
32
33
#     HAVE_RDRND / HAVE_BMI1 / HAVE_BMI2 / HAVE_ADX / HAVE_MPX
#     HAVE_PREFETCHWT1 / HAVE_ABM / HAVE_MMX / HAVE_SSE / HAVE_SSE2
#     HAVE_SSE3 / HAVE_SSSE3 / HAVE_SSE4_1 / HAVE_SSE4_2 / HAVE_SSE4a
#     HAVE_SHA / HAVE_AES / HAVE_AVX / HAVE_FMA3 / HAVE_FMA4 / HAVE_XOP
#     HAVE_AVX2 / HAVE_AVX512_F / HAVE_AVX512_CD / HAVE_AVX512_PF
#     HAVE_AVX512_ER / HAVE_AVX512_VL / HAVE_AVX512_BW / HAVE_AVX512_DQ
Josh Borrow's avatar
Josh Borrow committed
34
#     HAVE_AVX512_IFMA / HAVE_AVX512_VBMI / HAVE_ALTIVEC / HAVE_VSX
Pedro Gonnet's avatar
Pedro Gonnet committed
35
36
37
#
# LICENSE
#
38
#   Copyright (c) 2007 Christophe Tournayre <turn3r@users.sourceforge.net>
39
#   Copyright (c) 2013,2015 Michael Petch <mpetch@capp-sysware.com>
Josh Borrow's avatar
Josh Borrow committed
40
#   Copyright (c) 2017 Rafael de Lucena Valle <rafaeldelucena@gmail.com>
Pedro Gonnet's avatar
Pedro Gonnet committed
41
42
43
44
45
46
#
#   Copying and distribution of this file, with or without modification, are
#   permitted in any medium without royalty provided the copyright notice
#   and this notice are preserved. This file is offered as-is, without any
#   warranty.

47
48
49
50
#   SWIFT modifications. Intel compilers accept these flags, but warn about
#   the order of the flags when more than one is used. Given that we just
#   set SIMD_FLAGS to the most specific value, rather than all accepted ones.

Josh Borrow's avatar
Josh Borrow committed
51
#serial 18
Pedro Gonnet's avatar
Pedro Gonnet committed
52
53
54

AC_DEFUN([AX_EXT],
[
55
  AC_REQUIRE([AC_CANONICAL_HOST])
56
57
58
59
  AC_REQUIRE([AC_PROG_CC])

  CPUEXT_FLAGS=""
  SIMD_FLAGS=""
60
61
62

  case $host_cpu in
    powerpc*)
Josh Borrow's avatar
Josh Borrow committed
63
      AC_CACHE_CHECK([whether altivec is supported for old distros], [ax_cv_have_altivec_old_ext],
64
65
66
          [
            if test `/usr/sbin/sysctl -a 2>/dev/null| grep -c hw.optional.altivec` != 0; then
                if test `/usr/sbin/sysctl -n hw.optional.altivec` = 1; then
Josh Borrow's avatar
Josh Borrow committed
67
                  ax_cv_have_altivec_old_ext=yes
68
69
70
71
                fi
            fi
          ])

Josh Borrow's avatar
Josh Borrow committed
72
          if test "$ax_cv_have_altivec_old_ext" = yes; then
73
74
75
            AC_DEFINE(HAVE_ALTIVEC,,[Support Altivec instructions])
            AX_CHECK_COMPILE_FLAG(-faltivec, SIMD_FLAGS="$SIMD_FLAGS -faltivec", [])
          fi
Josh Borrow's avatar
Josh Borrow committed
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99

      AC_CACHE_CHECK([whether altivec is supported], [ax_cv_have_altivec_ext],
          [
            if test `LD_SHOW_AUXV=1 /bin/true 2>/dev/null|grep -c altivec` != 0; then
              ax_cv_have_altivec_ext=yes
            fi
          ])

          if test "$ax_cv_have_altivec_ext" = yes; then
            AC_DEFINE(HAVE_ALTIVEC,,[Support Altivec instructions])
            AX_CHECK_COMPILE_FLAG(-maltivec, SIMD_FLAGS="$SIMD_FLAGS -maltivec", [])
          fi

      AC_CACHE_CHECK([whether vsx is supported], [ax_cv_have_vsx_ext],
          [
            if test `LD_SHOW_AUXV=1 /bin/true 2>/dev/null|grep -c vsx` != 0; then
                ax_cv_have_vsx_ext=yes
            fi
          ])

          if test "$ax_cv_have_vsx_ext" = yes; then
            AC_DEFINE(HAVE_VSX,,[Support VSX instructions])
            AX_CHECK_COMPILE_FLAG(-mvsx, SIMD_FLAGS="$SIMD_FLAGS -mvsx", [])
          fi
100
101
102
103
104
    ;;

    i[[3456]]86*|x86_64*|amd64*)

      AC_REQUIRE([AX_GCC_X86_CPUID])
105
      AC_REQUIRE([AX_GCC_X86_CPUID_COUNT])
106
107
      AC_REQUIRE([AX_GCC_X86_AVX_XGETBV])

108
109
110
      eax_cpuid0=0
      AX_GCC_X86_CPUID(0x00000000)
      if test "$ax_cv_gcc_x86_cpuid_0x00000000" != "unknown";
111
      then
112
        eax_cpuid0=`echo $ax_cv_gcc_x86_cpuid_0x00000000 | cut -d ":" -f 1`
113
114
      fi

115
116
117
118
119
120
      eax_cpuid80000000=0
      AX_GCC_X86_CPUID(0x80000000)
      if test "$ax_cv_gcc_x86_cpuid_0x80000000" != "unknown";
      then
        eax_cpuid80000000=`echo $ax_cv_gcc_x86_cpuid_0x80000000 | cut -d ":" -f 1`
      fi
121

122
123
124
125
126
127
128
129
      ecx_cpuid1=0
      edx_cpuid1=0
      if test "$((0x$eax_cpuid0))" -ge 1 ; then
        AX_GCC_X86_CPUID(0x00000001)
        if test "$ax_cv_gcc_x86_cpuid_0x00000001" != "unknown";
        then
          ecx_cpuid1=`echo $ax_cv_gcc_x86_cpuid_0x00000001 | cut -d ":" -f 3`
          edx_cpuid1=`echo $ax_cv_gcc_x86_cpuid_0x00000001 | cut -d ":" -f 4`
130
        fi
131
      fi
132

133
134
135
136
137
138
139
140
      ebx_cpuid7=0
      ecx_cpuid7=0
      if test "$((0x$eax_cpuid0))" -ge 7 ; then
        AX_GCC_X86_CPUID_COUNT(0x00000007, 0x00)
        if test "$ax_cv_gcc_x86_cpuid_0x00000007" != "unknown";
        then
          ebx_cpuid7=`echo $ax_cv_gcc_x86_cpuid_0x00000007 | cut -d ":" -f 2`
          ecx_cpuid7=`echo $ax_cv_gcc_x86_cpuid_0x00000007 | cut -d ":" -f 3`
141
        fi
142
      fi
143

144
145
146
147
148
149
150
151
      ecx_cpuid80000001=0
      edx_cpuid80000001=0
      if test "$((0x$eax_cpuid80000000))" -ge "$((0x80000001))" ; then
        AX_GCC_X86_CPUID(0x80000001)
        if test "$ax_cv_gcc_x86_cpuid_0x80000001" != "unknown";
        then
          ecx_cpuid80000001=`echo $ax_cv_gcc_x86_cpuid_0x80000001 | cut -d ":" -f 3`
          edx_cpuid80000001=`echo $ax_cv_gcc_x86_cpuid_0x80000001 | cut -d ":" -f 4`
152
        fi
153
      fi
154

155
      AC_CACHE_VAL([ax_cv_have_mmx_os_support_ext],
156
      [
157
        ax_cv_have_mmx_os_support_ext=yes
158
159
      ])

160
      ax_cv_have_none_os_support_ext=yes
161

162
      AC_CACHE_VAL([ax_cv_have_sse_os_support_ext],
163
      [
164
165
166
        ax_cv_have_sse_os_support_ext=no,
        if test "$((0x$edx_cpuid1>>25&0x01))" = 1; then
          AC_LANG_PUSH([C])
Josh Borrow's avatar
Josh Borrow committed
167
          AC_RUN_IFELSE([AC_LANG_SOURCE([[
168
169
170
171
172
173
174
175
176
177
178
#include <signal.h>
#include <stdlib.h>
            /* No way at ring1 to ring3 in protected mode to check the CR0 and CR4
               control registers directly. Execute an SSE instruction.
               If it raises SIGILL then OS doesn't support SSE based instructions */
            void sig_handler(int signum){ exit(1); }
            int main(){
              signal(SIGILL, sig_handler);
              /* SSE instruction xorps  %xmm0,%xmm0 */
              __asm__ __volatile__ (".byte 0x0f, 0x57, 0xc0");
              return 0;
Josh Borrow's avatar
Josh Borrow committed
179
180
181
182
            }]])],
            [ax_cv_have_sse_os_support_ext=yes],
            [ax_cv_have_sse_os_support_ext=no],
            [ax_cv_have_sse_os_support_ext=no])
183
          AC_LANG_POP([C])
184
185
186
        fi
      ])

187
188
      xgetbv_eax=0
      if test "$((0x$ecx_cpuid1>>28&0x01))" = 1; then
189
190
191
192
193
194
        AX_GCC_X86_AVX_XGETBV(0x00000000)

        if test x"$ax_cv_gcc_x86_avx_xgetbv_0x00000000" != x"unknown"; then
          xgetbv_eax=`echo $ax_cv_gcc_x86_avx_xgetbv_0x00000000 | cut -d ":" -f 1`
        fi

195
        AC_CACHE_VAL([ax_cv_have_avx_os_support_ext],
196
        [
197
198
          ax_cv_have_avx_os_support_ext=no
          if test "$((0x$ecx_cpuid1>>27&0x01))" = 1; then
199
            if test "$((0x$xgetbv_eax&0x6))" = 6; then
200
              ax_cv_have_avx_os_support_ext=yes
201
202
203
204
205
            fi
          fi
        ])
      fi

206
207
208
209
210
211
212
      AC_CACHE_VAL([ax_cv_have_avx512_os_support_ext],
      [
        ax_cv_have_avx512_os_support_ext=no
        if test "$ax_cv_have_avx_os_support_ext" = yes; then
          if test "$((0x$xgetbv_eax&0xe6))" = "$((0xe6))"; then
            ax_cv_have_avx512_os_support_ext=yes
          fi
213
        fi
214
      ])
215

216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
      for ac_instr_info dnl
      in "none;rdrnd;RDRND;ecx_cpuid1,30;-mrdrnd;HAVE_RDRND;CPUEXT_FLAGS" dnl
         "none;bmi1;BMI1;ebx_cpuid7,3;-mbmi;HAVE_BMI1;CPUEXT_FLAGS" dnl
         "none;bmi2;BMI2;ebx_cpuid7,8;-mbmi2;HAVE_BMI2;CPUEXT_FLAGS" dnl
         "none;adx;ADX;ebx_cpuid7,19;-madx;HAVE_ADX;CPUEXT_FLAGS" dnl
         "none;mpx;MPX;ebx_cpuid7,14;-mmpx;HAVE_MPX;CPUEXT_FLAGS" dnl
         "none;prefetchwt1;PREFETCHWT1;ecx_cpuid7,0;-mprefetchwt1;HAVE_PREFETCHWT1;CPUEXT_FLAGS" dnl
         "none;abm;ABM;ecx_cpuid80000001,5;-mabm;HAVE_ABM;CPUEXT_FLAGS" dnl
         "mmx;mmx;MMX;edx_cpuid1,23;-mmmx;HAVE_MMX;SIMD_FLAGS" dnl
         "sse;sse;SSE;edx_cpuid1,25;-msse;HAVE_SSE;SIMD_FLAGS" dnl
         "sse;sse2;SSE2;edx_cpuid1,26;-msse2;HAVE_SSE2;SIMD_FLAGS" dnl
         "sse;sse3;SSE3;ecx_cpuid1,1;-msse3;HAVE_SSE3;SIMD_FLAGS" dnl
         "sse;ssse3;SSSE3;ecx_cpuid1,9;-mssse3;HAVE_SSSE3;SIMD_FLAGS" dnl
         "sse;sse41;SSE4.1;ecx_cpuid1,19;-msse4.1;HAVE_SSE4_1;SIMD_FLAGS" dnl
         "sse;sse42;SSE4.2;ecx_cpuid1,20;-msse4.2;HAVE_SSE4_2;SIMD_FLAGS" dnl
         "sse;sse4a;SSE4a;ecx_cpuid80000001,6;-msse4a;HAVE_SSE4a;SIMD_FLAGS" dnl
         "sse;sha;SHA;ebx_cpuid7,29;-msha;HAVE_SHA;SIMD_FLAGS" dnl
         "sse;aes;AES;ecx_cpuid1,25;-maes;HAVE_AES;SIMD_FLAGS" dnl
         "avx;avx;AVX;ecx_cpuid1,28;-mavx;HAVE_AVX;SIMD_FLAGS" dnl
         "avx;fma3;FMA3;ecx_cpuid1,12;-mfma;HAVE_FMA3;SIMD_FLAGS" dnl
         "avx;fma4;FMA4;ecx_cpuid80000001,16;-mfma4;HAVE_FMA4;SIMD_FLAGS" dnl
         "avx;xop;XOP;ecx_cpuid80000001,11;-mxop;HAVE_XOP;SIMD_FLAGS" dnl
         "avx;avx2;AVX2;ebx_cpuid7,5;-mavx2;HAVE_AVX2;SIMD_FLAGS" dnl
         "avx512;avx512f;AVX512-F;ebx_cpuid7,16;-mavx512f;HAVE_AVX512_F;SIMD_FLAGS" dnl
         "avx512;avx512cd;AVX512-CD;ebx_cpuid7,28;-mavx512cd;HAVE_AVX512_CD;SIMD_FLAGS" dnl
         "avx512;avx512pf;AVX512-PF;ebx_cpuid7,26;-mavx512pf;HAVE_AVX512_PF;SIMD_FLAGS" dnl
         "avx512;avx512er;AVX512-ER;ebx_cpuid7,27;-mavx512er;HAVE_AVX512_ER;SIMD_FLAGS" dnl
         "avx512;avx512vl;AVX512-VL;ebx_cpuid7,31;-mavx512vl;HAVE_AVX512_VL;SIMD_FLAGS" dnl
         "avx512;avx512bw;AVX512-BW;ebx_cpuid7,30;-mavx512bw;HAVE_AVX512_BW;SIMD_FLAGS" dnl
         "avx512;avx512dq;AVX512-DQ;ebx_cpuid7,17;-mavx512dq;HAVE_AVX512_DQ;SIMD_FLAGS" dnl
         "avx512;avx512ifma;AVX512-IFMA;ebx_cpuid7,21;-mavx512ifma;HAVE_AVX512_IFMA;SIMD_FLAGS" dnl
         "avx512;avx512vbmi;AVX512-VBMI;ecx_cpuid7,1;-mavx512vbmi;HAVE_AVX512_VBMI;SIMD_FLAGS" dnl
         #
      do ac_instr_os_support=$(eval echo \$ax_cv_have_$(echo $ac_instr_info | cut -d ";" -f 1)_os_support_ext)
         ac_instr_acvar=$(echo $ac_instr_info | cut -d ";" -f 2)
         ac_instr_shortname=$(echo $ac_instr_info | cut -d ";" -f 3)
         ac_instr_chk_loc=$(echo $ac_instr_info | cut -d ";" -f 4)
         ac_instr_chk_reg=0x$(eval echo \$$(echo $ac_instr_chk_loc | cut -d "," -f 1))
         ac_instr_chk_bit=$(echo $ac_instr_chk_loc | cut -d "," -f 2)
         ac_instr_compiler_flags=$(echo $ac_instr_info | cut -d ";" -f 5)
         ac_instr_have_define=$(echo $ac_instr_info | cut -d ";" -f 6)
         ac_instr_flag_type=$(echo $ac_instr_info | cut -d ";" -f 7)

         AC_CACHE_CHECK([whether ${ac_instr_shortname} is supported by the processor], [ax_cv_have_${ac_instr_acvar}_cpu_ext],
         [
           eval ax_cv_have_${ac_instr_acvar}_cpu_ext=no
           if test "$((${ac_instr_chk_reg}>>${ac_instr_chk_bit}&0x01))" = 1 ; then
             eval ax_cv_have_${ac_instr_acvar}_cpu_ext=yes
           fi
         ])

         if test x"$(eval echo \$ax_cv_have_${ac_instr_acvar}_cpu_ext)" = x"yes"; then
           AC_CACHE_CHECK([whether ${ac_instr_shortname} is supported by the processor and OS], [ax_cv_have_${ac_instr_acvar}_ext],
           [
             eval ax_cv_have_${ac_instr_acvar}_ext=no
             if test x"${ac_instr_os_support}" = x"yes"; then
               eval ax_cv_have_${ac_instr_acvar}_ext=yes
             fi
           ])

           if test "$(eval echo \$ax_cv_have_${ac_instr_acvar}_ext)" = yes; then
             AX_CHECK_COMPILE_FLAG(${ac_instr_compiler_flags}, eval ax_cv_support_${ac_instr_acvar}_ext=yes,
                                                               eval ax_cv_support_${ac_instr_acvar}_ext=no)
             if test x"$(eval echo \$ax_cv_support_${ac_instr_acvar}_ext)" = x"yes"; then
280
               eval ${ac_instr_flag_type}=\"${ac_instr_compiler_flags}\"
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
               AC_DEFINE_UNQUOTED([${ac_instr_have_define}])
             else
               AC_MSG_WARN([Your processor and OS supports ${ac_instr_shortname} instructions but not your compiler, can you try another compiler?])
             fi
           else
             if test x"${ac_instr_os_support}" = x"no"; then
               AC_CACHE_VAL(ax_cv_support_${ac_instr_acvar}_ext, eval ax_cv_support_${ac_instr_acvar}_ext=no)
               AC_MSG_WARN([Your processor supports ${ac_instr_shortname}, but your OS doesn't])
             fi
           fi
         else
           AC_CACHE_VAL(ax_cv_have_${ac_instr_acvar}_ext, eval ax_cv_have_${ac_instr_acvar}_ext=no)
           AC_CACHE_VAL(ax_cv_support_${ac_instr_acvar}_ext, eval ax_cv_support_${ac_instr_acvar}_ext=no)
         fi
      done
296
297
  ;;
  esac
Pedro Gonnet's avatar
Pedro Gonnet committed
298

299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
  AH_TEMPLATE([HAVE_RDRND],[Define to 1 to support Digital Random Number Generator])
  AH_TEMPLATE([HAVE_BMI1],[Define to 1 to support Bit Manipulation Instruction Set 1])
  AH_TEMPLATE([HAVE_BMI2],[Define to 1 to support Bit Manipulation Instruction Set 2])
  AH_TEMPLATE([HAVE_ADX],[Define to 1 to support Multi-Precision Add-Carry Instruction Extensions])
  AH_TEMPLATE([HAVE_MPX],[Define to 1 to support Memory Protection Extensions])
  AH_TEMPLATE([HAVE_PREFETCHWT1],[Define to 1 to support Prefetch Vector Data Into Caches WT1])
  AH_TEMPLATE([HAVE_ABM],[Define to 1 to support Advanced Bit Manipulation])
  AH_TEMPLATE([HAVE_MMX],[Define to 1 to support Multimedia Extensions])
  AH_TEMPLATE([HAVE_SSE],[Define to 1 to support Streaming SIMD Extensions])
  AH_TEMPLATE([HAVE_SSE2],[Define to 1 to support Streaming SIMD Extensions])
  AH_TEMPLATE([HAVE_SSE3],[Define to 1 to support Streaming SIMD Extensions 3])
  AH_TEMPLATE([HAVE_SSSE3],[Define to 1 to support Supplemental Streaming SIMD Extensions 3])
  AH_TEMPLATE([HAVE_SSE4_1],[Define to 1 to support Streaming SIMD Extensions 4.1])
  AH_TEMPLATE([HAVE_SSE4_2],[Define to 1 to support Streaming SIMD Extensions 4.2])
  AH_TEMPLATE([HAVE_SSE4a],[Define to 1 to support AMD Streaming SIMD Extensions 4a])
  AH_TEMPLATE([HAVE_SHA],[Define to 1 to support Secure Hash Algorithm Extension])
  AH_TEMPLATE([HAVE_AES],[Define to 1 to support Advanced Encryption Standard New Instruction Set (AES-NI)])
  AH_TEMPLATE([HAVE_AVX],[Define to 1 to support Advanced Vector Extensions])
  AH_TEMPLATE([HAVE_FMA3],[Define to 1 to support  Fused Multiply-Add Extensions 3])
  AH_TEMPLATE([HAVE_FMA4],[Define to 1 to support Fused Multiply-Add Extensions 4])
  AH_TEMPLATE([HAVE_XOP],[Define to 1 to support eXtended Operations Extensions])
  AH_TEMPLATE([HAVE_AVX2],[Define to 1 to support Advanced Vector Extensions 2])
  AH_TEMPLATE([HAVE_AVX512_F],[Define to 1 to support AVX-512 Foundation Extensions])
  AH_TEMPLATE([HAVE_AVX512_CD],[Define to 1 to support AVX-512 Conflict Detection Instructions])
  AH_TEMPLATE([HAVE_AVX512_PF],[Define to 1 to support AVX-512 Conflict Prefetch Instructions])
  AH_TEMPLATE([HAVE_AVX512_ER],[Define to 1 to support AVX-512 Exponential & Reciprocal Instructions])
  AH_TEMPLATE([HAVE_AVX512_VL],[Define to 1 to support AVX-512 Vector Length Extensions])
  AH_TEMPLATE([HAVE_AVX512_BW],[Define to 1 to support AVX-512 Byte and Word Instructions])
  AH_TEMPLATE([HAVE_AVX512_DQ],[Define to 1 to support AVX-512 Doubleword and Quadword Instructions])
  AH_TEMPLATE([HAVE_AVX512_IFMA],[Define to 1 to support AVX-512 Integer Fused Multiply Add Instructions])
  AH_TEMPLATE([HAVE_AVX512_VBMI],[Define to 1 to support AVX-512 Vector Byte Manipulation Instructions])
Pedro Gonnet's avatar
Pedro Gonnet committed
330
  AC_SUBST(SIMD_FLAGS)
331
  AC_SUBST(CPUEXT_FLAGS)
Pedro Gonnet's avatar
Pedro Gonnet committed
332
])