Chris@82: dnl @synopsis AX_CC_MAXOPT
Chris@82: dnl @summary turn on optimization flags for the C compiler
Chris@82: dnl @category C
Chris@82: dnl
Chris@82: dnl Try to turn on "good" C optimization flags for various compilers
Chris@82: dnl and architectures, for some definition of "good".  (In our case,
Chris@82: dnl good for FFTW and hopefully for other scientific codes.  Modify 
Chris@82: dnl as needed.)
Chris@82: dnl
Chris@82: dnl The user can override the flags by setting the CFLAGS environment
Chris@82: dnl variable.  
Chris@82: dnl
Chris@82: dnl Note also that the flags assume that ANSI C aliasing rules are
Chris@82: dnl followed by the code (e.g. for gcc's -fstrict-aliasing), and that
Chris@82: dnl floating-point computations can be re-ordered as needed.
Chris@82: dnl
Chris@82: dnl Requires macros: AX_CHECK_COMPILER_FLAGS, AX_COMPILER_VENDOR,
Chris@82: dnl
Chris@82: dnl @version 2011-06-22
Chris@82: dnl @license GPLWithACException
Chris@82: dnl @author Steven G. Johnson <stevenj@alum.mit.edu> and Matteo Frigo.
Chris@82: AC_DEFUN([AX_CC_MAXOPT],
Chris@82: [
Chris@82: AC_REQUIRE([AC_PROG_CC])
Chris@82: AC_REQUIRE([AX_COMPILER_VENDOR])
Chris@82: AC_REQUIRE([AC_CANONICAL_HOST])
Chris@82: 
Chris@82: # Try to determine "good" native compiler flags if none specified via CFLAGS
Chris@82: if test "$ac_test_CFLAGS" != "set"; then
Chris@82:   CFLAGS=""
Chris@82:   case $ax_cv_c_compiler_vendor in
Chris@82:     dec) CFLAGS="-newc -w0 -O5 -ansi_alias -ansi_args -fp_reorder -tune host"
Chris@82:     	 ;;
Chris@82: 
Chris@82:     sun) CFLAGS="-native -fast -xO5 -dalign"
Chris@82:     	 ;;
Chris@82: 
Chris@82:     hp)  CFLAGS="+Oall +Optrs_ansi +DSnative"
Chris@82:     	 ;;
Chris@82: 
Chris@82:     ibm) xlc_opt="-qarch=auto -qtune=auto"
Chris@82:          AX_CHECK_COMPILER_FLAGS($xlc_opt,
Chris@82:          	CFLAGS="-O3 -qalias=ansi -w $xlc_opt",
Chris@82:                [CFLAGS="-O3 -qalias=ansi -w"])
Chris@82:          ;;
Chris@82: 
Chris@82:     intel) CFLAGS="-O3"
Chris@82:         # Intel seems to have changed the spelling of this flag recently
Chris@82:         icc_ansi_alias="unknown"
Chris@82: 	for flag in -ansi-alias -ansi_alias; do
Chris@82: 	  AX_CHECK_COMPILER_FLAGS($flag, [icc_ansi_alias=$flag; break])
Chris@82: 	done
Chris@82:  	if test "x$icc_ansi_alias" != xunknown; then
Chris@82:             CFLAGS="$CFLAGS $icc_ansi_alias"
Chris@82:         fi
Chris@82: 	AX_CHECK_COMPILER_FLAGS(-malign-double, CFLAGS="$CFLAGS -malign-double")
Chris@82: 	# We used to check for architecture flags here, e.g. -xHost etc.,
Chris@82: 	# but these flags are problematic.  On icc-12.0.0, "-mavx -xHost"
Chris@82: 	# overrides -mavx with -xHost, generating SSE2 code instead of AVX
Chris@82: 	# code.  ICC does not seem to support -mtune=host or equivalent
Chris@82: 	# non-ABI changing flag.
Chris@82: 	;;
Chris@82:     
Chris@82:     clang)
Chris@82:         CFLAGS="-O3 -fomit-frame-pointer"
Chris@82:         AX_CHECK_COMPILER_FLAGS(-mtune=native, CFLAGS="$CFLAGS -mtune=native")
Chris@82:         AX_CHECK_COMPILER_FLAGS(-fstrict-aliasing,CFLAGS="$CFLAGS -fstrict-aliasing")
Chris@82:         ;;
Chris@82: 
Chris@82:     gnu) 
Chris@82:      # Default optimization flags for gcc on all systems.
Chris@82:      # Somehow -O3 does not imply -fomit-frame-pointer on ia32
Chris@82:      CFLAGS="-O3 -fomit-frame-pointer"
Chris@82: 
Chris@82:      # tune for the host by default
Chris@82:      AX_CHECK_COMPILER_FLAGS(-mtune=native, CFLAGS="$CFLAGS -mtune=native")
Chris@82: 
Chris@82:      # -malign-double for x86 systems
Chris@82:      AX_CHECK_COMPILER_FLAGS(-malign-double, CFLAGS="$CFLAGS -malign-double")
Chris@82: 
Chris@82:      #  -fstrict-aliasing for gcc-2.95+
Chris@82:      AX_CHECK_COMPILER_FLAGS(-fstrict-aliasing,
Chris@82: 	CFLAGS="$CFLAGS -fstrict-aliasing")
Chris@82: 
Chris@82:      # -fno-schedule-insns is pretty much required on all risc
Chris@82:      # processors.
Chris@82:      # 
Chris@82:      # gcc performs one pass of instruction scheduling, then a pass of
Chris@82:      # register allocation, then another pass of instruction
Chris@82:      # scheduling.  The first pass reorders instructions in a way that
Chris@82:      # is pretty much the worst possible for the purposes of register
Chris@82:      # allocation.  We disable the first pass.
Chris@82:      AX_CHECK_COMPILER_FLAGS(-fno-schedule-insns, CFLAGS="$CFLAGS -fno-schedule-insns")
Chris@82: 
Chris@82:      # flags to enable power ISA 2.07 instructions with gcc (always true with vsx)
Chris@82:      if test "$have_vsx" = "yes"; then
Chris@82:          AX_CHECK_COMPILER_FLAGS(-mcpu=power8, CFLAGS="$CFLAGS -mcpu=power8")
Chris@82:          AX_CHECK_COMPILER_FLAGS(-mpower8-fusion, CFLAGS="$CFLAGS -mpower8-fusion")
Chris@82:          AX_CHECK_COMPILER_FLAGS(-mpower8-vector, CFLAGS="$CFLAGS -mpower8-vector")
Chris@82:          AX_CHECK_COMPILER_FLAGS(-mdirect-move, CFLAGS="$CFLAGS -mdirect-move")
Chris@82:      fi
Chris@82:      ;;
Chris@82:   esac
Chris@82: 
Chris@82:   if test -z "$CFLAGS"; then
Chris@82: 	echo ""
Chris@82: 	echo "********************************************************"
Chris@82:         echo "* WARNING: Don't know the best CFLAGS for this system  *"
Chris@82:         echo "* Use ./configure CFLAGS=... to specify your own flags *"
Chris@82: 	echo "* (otherwise, a default of CFLAGS=-O3 will be used)    *"
Chris@82: 	echo "********************************************************"
Chris@82: 	echo ""
Chris@82:         CFLAGS="-O3"
Chris@82:   fi
Chris@82: 
Chris@82:   AX_CHECK_COMPILER_FLAGS($CFLAGS, [], [
Chris@82: 	echo ""
Chris@82:         echo "********************************************************"
Chris@82:         echo "* WARNING: The guessed CFLAGS don't seem to work with  *"
Chris@82:         echo "* your compiler.                                       *"
Chris@82:         echo "* Use ./configure CFLAGS=... to specify your own flags *"
Chris@82:         echo "********************************************************"
Chris@82:         echo ""
Chris@82:         CFLAGS=""
Chris@82:   ])
Chris@82: 
Chris@82: fi
Chris@82: ])