1 # ===========================================================================
2 # https://www.gnu.org/software/autoconf-archive/ax_cc_maxopt.html
3 # ===========================================================================
11 # Try to turn on "good" C optimization flags for various compilers and
12 # architectures, for some definition of "good". (In our case, good for
13 # FFTW and hopefully for other scientific codes. Modify as needed.)
15 # The user can override the flags by setting the CFLAGS environment
16 # variable. The user can also specify --enable-portable-binary in order to
17 # disable any optimization flags that might result in a binary that only
18 # runs on the host architecture.
20 # Note also that the flags assume that ANSI C aliasing rules are followed
21 # by the code (e.g. for gcc's -fstrict-aliasing), and that floating-point
22 # computations can be re-ordered as needed.
24 # Requires macros: AX_CHECK_COMPILE_FLAG, AX_COMPILER_VENDOR,
25 # AX_GCC_ARCHFLAG, AX_GCC_X86_CPUID.
29 # Copyright (c) 2008 Steven G. Johnson <stevenj@alum.mit.edu>
30 # Copyright (c) 2008 Matteo Frigo
32 # This program is free software: you can redistribute it and/or modify it
33 # under the terms of the GNU General Public License as published by the
34 # Free Software Foundation, either version 3 of the License, or (at your
35 # option) any later version.
37 # This program is distributed in the hope that it will be useful, but
38 # WITHOUT ANY WARRANTY; without even the implied warranty of
39 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General
40 # Public License for more details.
42 # You should have received a copy of the GNU General Public License along
43 # with this program. If not, see <https://www.gnu.org/licenses/>.
45 # As a special exception, the respective Autoconf Macro's copyright owner
46 # gives unlimited permission to copy, distribute and modify the configure
47 # scripts that are the output of Autoconf when processing the Macro. You
48 # need not follow the terms of the GNU General Public License when using
49 # or distributing such scripts, even though portions of the text of the
50 # Macro appear in them. The GNU General Public License (GPL) does govern
51 # all other use of the material that constitutes the Autoconf Macro.
53 # This special exception to the GPL applies to versions of the Autoconf
54 # Macro released by the Autoconf Archive. When you make and distribute a
55 # modified version of the Autoconf Macro, you may extend this special
56 # exception to the GPL to apply to your modified version as well.
60 AC_DEFUN([AX_CC_MAXOPT],
62 AC_REQUIRE([AC_PROG_CC])
63 AC_REQUIRE([AX_COMPILER_VENDOR])
64 AC_REQUIRE([AC_CANONICAL_HOST])
66 AC_ARG_ENABLE(portable-binary, [AS_HELP_STRING([--enable-portable-binary], [disable compiler optimizations that would produce unportable binaries])],
67 acx_maxopt_portable=$enableval, acx_maxopt_portable=no)
69 # Try to determine "good" native compiler flags if none specified via CFLAGS
70 if test "x$ac_test_CFLAGS" = "x"; then
71 case $ax_cv_c_compiler_vendor in
72 dec) CFLAGS="$CFLAGS -newc -w0 -O5 -ansi_alias -ansi_args -fp_reorder -tune host"
73 if test "x$acx_maxopt_portable" = xno; then
74 CFLAGS="$CFLAGS -arch host"
77 sun) CFLAGS="$CFLAGS -native -fast -xO5 -dalign"
78 if test "x$acx_maxopt_portable" = xyes; then
79 CFLAGS="$CFLAGS -xarch=generic"
82 hp) CFLAGS="$CFLAGS +Oall +Optrs_ansi +DSnative"
83 if test "x$acx_maxopt_portable" = xyes; then
84 CFLAGS="$CFLAGS +DAportable"
87 ibm) if test "x$acx_maxopt_portable" = xno; then
88 xlc_opt="-qarch=auto -qtune=auto"
92 AX_CHECK_COMPILE_FLAG($xlc_opt,
93 CFLAGS="$CFLAGS -O3 -qansialias -w $xlc_opt",
94 [CFLAGS="$CFLAGS -O3 -qansialias -w"
95 echo "******************************************************"
96 echo "* You seem to have the IBM C compiler. It is *"
97 echo "* recommended for best performance that you use: *"
99 echo "* CFLAGS=-O3 -qarch=xxx -qtune=xxx -qansialias -w *"
101 echo "* where xxx is pwr2, pwr3, 604, or whatever kind of *"
102 echo "* CPU you have. (Set the CFLAGS environment var. *"
103 echo "* and re-run configure.) For more info, man cc. *"
104 echo "******************************************************"])
107 intel) CFLAGS="$CFLAGS -O3 -ansi_alias"
108 if test "x$acx_maxopt_portable" = xno; then
113 # icc accepts gcc assembly syntax, so these should work:
116 case $ax_cv_gcc_x86_cpuid_0 in # see AX_GCC_ARCHFLAG
117 *:756e6547:6c65746e:49656e69) # Intel
118 case $ax_cv_gcc_x86_cpuid_1 in
119 *0?6[[78ab]]?:*:*:*|?6[[78ab]]?:*:*:*|6[[78ab]]?:*:*:*) icc_flags="-xK" ;;
120 *0?6[[9d]]?:*:*:*|?6[[9d]]?:*:*:*|6[[9d]]?:*:*:*|*1?65?:*:*:*) icc_flags="-xSSE2 -xB -xK" ;;
121 *0?6e?:*:*:*|?6e?:*:*:*|6e?:*:*:*) icc_flags="-xSSE3 -xP -xO -xB -xK" ;;
122 *0?6f?:*:*:*|?6f?:*:*:*|6f?:*:*:*|*1?66?:*:*:*) icc_flags="-xSSSE3 -xT -xB -xK" ;;
123 *1?6[[7d]]?:*:*:*) icc_flags="-xSSE4.1 -xS -xT -xB -xK" ;;
124 *1?6[[aef]]?:*:*:*|*2?6[[5cef]]?:*:*:*) icc_flags="-xSSE4.2 -xS -xT -xB -xK" ;;
125 *2?6[[ad]]?:*:*:*) icc_flags="-xAVX -SSE4.2 -xS -xT -xB -xK" ;;
126 *3?6[[ae]]?:*:*:*) icc_flags="-xCORE-AVX-I -xAVX -SSE4.2 -xS -xT -xB -xK" ;;
127 *3?6[[cf]]?:*:*:*|*4?6[[56]]?:*:*:*) icc_flags="-xCORE-AVX2 -xCORE-AVX-I -xAVX -SSE4.2 -xS -xT -xB -xK" ;;
128 *000?f[[346]]?:*:*:*|?f[[346]]?:*:*:*|f[[346]]?:*:*:*) icc_flags="-xSSE3 -xP -xO -xN -xW -xK" ;;
129 *00??f??:*:*:*|??f??:*:*:*|?f??:*:*:*|f??:*:*:*) icc_flags="-xSSE2 -xN -xW -xK" ;;
133 if test "x$icc_flags" != x; then
134 for flag in $icc_flags; do
135 AX_CHECK_COMPILE_FLAG($flag, [icc_archflag=$flag; break])
138 AC_MSG_CHECKING([for icc architecture flag])
139 AC_MSG_RESULT($icc_archflag)
140 if test "x$icc_archflag" != xunknown; then
141 CFLAGS="$CFLAGS $icc_archflag"
147 # default optimization flags for nvhpc
152 # default optimization flags for gcc on all systems
153 CFLAGS="$CFLAGS -O3 -fomit-frame-pointer"
155 # -malign-double for x86 systems
156 AX_CHECK_COMPILE_FLAG(-malign-double, CFLAGS="$CFLAGS -malign-double")
158 # -fstrict-aliasing for gcc-2.95+
159 AX_CHECK_COMPILE_FLAG(-fstrict-aliasing,
160 CFLAGS="$CFLAGS -fstrict-aliasing")
162 # note that we enable "unsafe" fp optimization with other compilers, too
163 AX_CHECK_COMPILE_FLAG(-ffast-math, CFLAGS="$CFLAGS -ffast-math")
165 AX_GCC_ARCHFLAG($acx_maxopt_portable)
167 # drop to -O1 for gcc 4.2
169 sed -e 's/.* \(@<:@0-9@:>@@<:@0-9@:>@*\)\.\(@<:@0-9@:>@@<:@0-9@:>@*\).*/\1 \2/' |
171 if test $major -eq 4 -a $minor -eq 2; then
179 # default optimization flags for MSVC opt builds
184 if test -z "$CFLAGS"; then
186 echo "********************************************************"
187 echo "* WARNING: Don't know the best CFLAGS for this system *"
188 echo "* Use ./configure CFLAGS=... to specify your own flags *"
189 echo "* (otherwise, a default of CFLAGS=-O3 will be used) *"
190 echo "********************************************************"
195 AX_CHECK_COMPILE_FLAG($CFLAGS, [], [
197 echo "********************************************************"
198 echo "* WARNING: The guessed CFLAGS don't seem to work with *"
199 echo "* your compiler. *"
200 echo "* Use ./configure CFLAGS=... to specify your own flags *"
201 echo "********************************************************"