Thomas Tsou | 3eaae80 | 2013-08-20 19:31:14 -0400 | [diff] [blame] | 1 | # =========================================================================== |
| 2 | # http://www.gnu.org/software/autoconf-archive/ax_ext.html |
| 3 | # =========================================================================== |
| 4 | # |
| 5 | # SYNOPSIS |
| 6 | # |
| 7 | # AX_EXT |
| 8 | # |
| 9 | # DESCRIPTION |
| 10 | # |
| 11 | # Find supported SIMD extensions by requesting cpuid. When an 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. |
| 15 | # |
| 16 | # This macro calls: |
| 17 | # |
| 18 | # AC_SUBST(SIMD_FLAGS) |
| 19 | # |
| 20 | # And defines: |
| 21 | # |
| 22 | # HAVE_MMX / HAVE_SSE / HAVE_SSE2 / HAVE_SSE3 / HAVE_SSSE3 / HAVE_SSE4.1 / HAVE_SSE4.2 / HAVE_AVX |
| 23 | # |
| 24 | # LICENSE |
| 25 | # |
| 26 | # Copyright (c) 2007 Christophe Tournayre <turn3r@users.sourceforge.net> |
| 27 | # Copyright (c) 2013 Michael Petch <mpetch@capp-sysware.com> |
| 28 | # |
| 29 | # Copying and distribution of this file, with or without modification, are |
| 30 | # permitted in any medium without royalty provided the copyright notice |
| 31 | # and this notice are preserved. This file is offered as-is, without any |
| 32 | # warranty. |
| 33 | |
| 34 | #serial 12 |
| 35 | |
| 36 | AC_DEFUN([AX_EXT], |
| 37 | [ |
| 38 | AC_REQUIRE([AC_CANONICAL_HOST]) |
| 39 | |
| 40 | case $host_cpu in |
| 41 | i[[3456]]86*|x86_64*|amd64*) |
| 42 | |
| 43 | AC_REQUIRE([AX_GCC_X86_CPUID]) |
| 44 | AC_REQUIRE([AX_GCC_X86_AVX_XGETBV]) |
| 45 | |
| 46 | AX_GCC_X86_CPUID(0x00000001) |
| 47 | ecx=`echo $ax_cv_gcc_x86_cpuid_0x00000001 | cut -d ":" -f 3` |
| 48 | edx=`echo $ax_cv_gcc_x86_cpuid_0x00000001 | cut -d ":" -f 4` |
| 49 | |
| 50 | AC_CACHE_CHECK([whether mmx is supported], [ax_cv_have_mmx_ext], |
| 51 | [ |
| 52 | ax_cv_have_mmx_ext=no |
| 53 | if test "$((0x$edx>>23&0x01))" = 1; then |
| 54 | ax_cv_have_mmx_ext=yes |
| 55 | fi |
| 56 | ]) |
| 57 | |
| 58 | AC_CACHE_CHECK([whether sse is supported], [ax_cv_have_sse_ext], |
| 59 | [ |
| 60 | ax_cv_have_sse_ext=no |
| 61 | if test "$((0x$edx>>25&0x01))" = 1; then |
| 62 | ax_cv_have_sse_ext=yes |
| 63 | fi |
| 64 | ]) |
| 65 | |
| 66 | AC_CACHE_CHECK([whether sse2 is supported], [ax_cv_have_sse2_ext], |
| 67 | [ |
| 68 | ax_cv_have_sse2_ext=no |
| 69 | if test "$((0x$edx>>26&0x01))" = 1; then |
| 70 | ax_cv_have_sse2_ext=yes |
| 71 | fi |
| 72 | ]) |
| 73 | |
| 74 | AC_CACHE_CHECK([whether sse3 is supported], [ax_cv_have_sse3_ext], |
| 75 | [ |
| 76 | ax_cv_have_sse3_ext=no |
| 77 | if test "$((0x$ecx&0x01))" = 1; then |
| 78 | ax_cv_have_sse3_ext=yes |
| 79 | fi |
| 80 | ]) |
| 81 | |
| 82 | AC_CACHE_CHECK([whether ssse3 is supported], [ax_cv_have_ssse3_ext], |
| 83 | [ |
| 84 | ax_cv_have_ssse3_ext=no |
| 85 | if test "$((0x$ecx>>9&0x01))" = 1; then |
| 86 | ax_cv_have_ssse3_ext=yes |
| 87 | fi |
| 88 | ]) |
| 89 | |
| 90 | AC_CACHE_CHECK([whether sse4.1 is supported], [ax_cv_have_sse41_ext], |
| 91 | [ |
| 92 | ax_cv_have_sse41_ext=no |
| 93 | if test "$((0x$ecx>>19&0x01))" = 1; then |
| 94 | ax_cv_have_sse41_ext=yes |
| 95 | fi |
| 96 | ]) |
| 97 | |
| 98 | AC_CACHE_CHECK([whether sse4.2 is supported], [ax_cv_have_sse42_ext], |
| 99 | [ |
| 100 | ax_cv_have_sse42_ext=no |
| 101 | if test "$((0x$ecx>>20&0x01))" = 1; then |
| 102 | ax_cv_have_sse42_ext=yes |
| 103 | fi |
| 104 | ]) |
| 105 | |
| 106 | AC_CACHE_CHECK([whether avx is supported by processor], [ax_cv_have_avx_cpu_ext], |
| 107 | [ |
| 108 | ax_cv_have_avx_cpu_ext=no |
| 109 | if test "$((0x$ecx>>28&0x01))" = 1; then |
| 110 | ax_cv_have_avx_cpu_ext=yes |
| 111 | fi |
| 112 | ]) |
| 113 | |
| 114 | if test x"$ax_cv_have_avx_cpu_ext" = x"yes"; then |
| 115 | AX_GCC_X86_AVX_XGETBV(0x00000000) |
| 116 | |
| 117 | xgetbv_eax="0" |
| 118 | if test x"$ax_cv_gcc_x86_avx_xgetbv_0x00000000" != x"unknown"; then |
| 119 | xgetbv_eax=`echo $ax_cv_gcc_x86_avx_xgetbv_0x00000000 | cut -d ":" -f 1` |
| 120 | fi |
| 121 | |
| 122 | AC_CACHE_CHECK([whether avx is supported by operating system], [ax_cv_have_avx_ext], |
| 123 | [ |
| 124 | ax_cv_have_avx_ext=no |
| 125 | |
| 126 | if test "$((0x$ecx>>27&0x01))" = 1; then |
| 127 | if test "$((0x$xgetbv_eax&0x6))" = 6; then |
| 128 | ax_cv_have_avx_ext=yes |
| 129 | fi |
| 130 | fi |
| 131 | ]) |
| 132 | if test x"$ax_cv_have_avx_ext" = x"no"; then |
| 133 | AC_MSG_WARN([Your processor supports AVX, but your operating system doesn't]) |
| 134 | fi |
| 135 | fi |
| 136 | |
| 137 | if test "$ax_cv_have_mmx_ext" = yes; then |
| 138 | AX_CHECK_COMPILE_FLAG(-mmmx, ax_cv_support_mmx_ext=yes, []) |
| 139 | if test x"$ax_cv_support_mmx_ext" = x"yes"; then |
| 140 | SIMD_FLAGS="$SIMD_FLAGS -mmmx" |
| 141 | AC_DEFINE(HAVE_MMX,,[Support mmx instructions]) |
| 142 | else |
| 143 | AC_MSG_WARN([Your processor supports mmx instructions but not your compiler, can you try another compiler?]) |
| 144 | fi |
| 145 | fi |
| 146 | |
| 147 | if test "$ax_cv_have_sse_ext" = yes; then |
| 148 | AX_CHECK_COMPILE_FLAG(-msse, ax_cv_support_sse_ext=yes, []) |
| 149 | if test x"$ax_cv_support_sse_ext" = x"yes"; then |
| 150 | SIMD_FLAGS="$SIMD_FLAGS -msse" |
| 151 | AC_DEFINE(HAVE_SSE,,[Support SSE (Streaming SIMD Extensions) instructions]) |
| 152 | else |
| 153 | AC_MSG_WARN([Your processor supports sse instructions but not your compiler, can you try another compiler?]) |
| 154 | fi |
| 155 | fi |
| 156 | |
| 157 | if test "$ax_cv_have_sse2_ext" = yes; then |
| 158 | AX_CHECK_COMPILE_FLAG(-msse2, ax_cv_support_sse2_ext=yes, []) |
| 159 | if test x"$ax_cv_support_sse2_ext" = x"yes"; then |
| 160 | SIMD_FLAGS="$SIMD_FLAGS -msse2" |
| 161 | AC_DEFINE(HAVE_SSE2,,[Support SSE2 (Streaming SIMD Extensions 2) instructions]) |
| 162 | else |
| 163 | AC_MSG_WARN([Your processor supports sse2 instructions but not your compiler, can you try another compiler?]) |
| 164 | fi |
| 165 | fi |
| 166 | |
| 167 | if test "$ax_cv_have_sse3_ext" = yes; then |
| 168 | AX_CHECK_COMPILE_FLAG(-msse3, ax_cv_support_sse3_ext=yes, []) |
| 169 | if test x"$ax_cv_support_sse3_ext" = x"yes"; then |
| 170 | SIMD_FLAGS="$SIMD_FLAGS -msse3" |
| 171 | AC_DEFINE(HAVE_SSE3,,[Support SSE3 (Streaming SIMD Extensions 3) instructions]) |
| 172 | else |
| 173 | AC_MSG_WARN([Your processor supports sse3 instructions but not your compiler, can you try another compiler?]) |
| 174 | fi |
| 175 | fi |
| 176 | |
| 177 | if test "$ax_cv_have_ssse3_ext" = yes; then |
| 178 | AX_CHECK_COMPILE_FLAG(-mssse3, ax_cv_support_ssse3_ext=yes, []) |
| 179 | if test x"$ax_cv_support_ssse3_ext" = x"yes"; then |
| 180 | SIMD_FLAGS="$SIMD_FLAGS -mssse3" |
| 181 | AC_DEFINE(HAVE_SSSE3,,[Support SSSE3 (Supplemental Streaming SIMD Extensions 3) instructions]) |
| 182 | else |
| 183 | AC_MSG_WARN([Your processor supports ssse3 instructions but not your compiler, can you try another compiler?]) |
| 184 | fi |
| 185 | fi |
| 186 | |
| 187 | if test "$ax_cv_have_sse41_ext" = yes; then |
| 188 | AX_CHECK_COMPILE_FLAG(-msse4.1, ax_cv_support_sse41_ext=yes, []) |
| 189 | if test x"$ax_cv_support_sse41_ext" = x"yes"; then |
| 190 | SIMD_FLAGS="$SIMD_FLAGS -msse4.1" |
| 191 | AC_DEFINE(HAVE_SSE4_1,,[Support SSSE4.1 (Streaming SIMD Extensions 4.1) instructions]) |
| 192 | else |
| 193 | AC_MSG_WARN([Your processor supports sse4.1 instructions but not your compiler, can you try another compiler?]) |
| 194 | fi |
| 195 | fi |
| 196 | |
| 197 | if test "$ax_cv_have_sse42_ext" = yes; then |
| 198 | AX_CHECK_COMPILE_FLAG(-msse4.2, ax_cv_support_sse42_ext=yes, []) |
| 199 | if test x"$ax_cv_support_sse42_ext" = x"yes"; then |
| 200 | SIMD_FLAGS="$SIMD_FLAGS -msse4.2" |
| 201 | AC_DEFINE(HAVE_SSE4_2,,[Support SSSE4.2 (Streaming SIMD Extensions 4.2) instructions]) |
| 202 | else |
| 203 | AC_MSG_WARN([Your processor supports sse4.2 instructions but not your compiler, can you try another compiler?]) |
| 204 | fi |
| 205 | fi |
| 206 | |
| 207 | if test "$ax_cv_have_avx_ext" = yes; then |
| 208 | AX_CHECK_COMPILE_FLAG(-mavx, ax_cv_support_avx_ext=yes, []) |
| 209 | if test x"$ax_cv_support_avx_ext" = x"yes"; then |
| 210 | SIMD_FLAGS="$SIMD_FLAGS -mavx" |
| 211 | AC_DEFINE(HAVE_AVX,,[Support AVX (Advanced Vector Extensions) instructions]) |
| 212 | else |
| 213 | AC_MSG_WARN([Your processor supports avx instructions but not your compiler, can you try another compiler?]) |
| 214 | fi |
| 215 | fi |
| 216 | |
| 217 | ;; |
| 218 | esac |
| 219 | |
| 220 | AC_SUBST(SIMD_FLAGS) |
| 221 | ]) |