Mercurial > hg > sv-dependency-builds
diff src/fftw-3.3.5/configure.ac @ 42:2cd0e3b3e1fd
Current fftw source
| author | Chris Cannam |
|---|---|
| date | Tue, 18 Oct 2016 13:40:26 +0100 |
| parents | |
| children |
line wrap: on
line diff
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/src/fftw-3.3.5/configure.ac Tue Oct 18 13:40:26 2016 +0100 @@ -0,0 +1,841 @@ + + +dnl Process this file with autoconf to produce a configure script. + +dnl Define the fftw version number as M4 macros, so that we can enforce +dnl the invariant that the minor version number in FFTW-X.Y.MINOR is the same +dnl as the revision number in SHARED_VERSION_INFO. +define(FFTW_MAJOR_VERSION, 3.3)dnl +define(FFTW_MINOR_VERSION, 5)dnl + +dnl Version number of the FFTW source package. +AC_INIT(fftw, FFTW_MAJOR_VERSION.FFTW_MINOR_VERSION, fftw@fftw.org) +AC_CONFIG_SRCDIR(kernel/ifftw.h) + +dnl Version number for libtool shared libraries. Libtool wants a string +dnl of the form CURRENT:REVISION:AGE. We adopt the convention that +dnl REVISION is the same as the FFTW minor version number. +dnl fftw-3.1.x was 4:x:1 +dnl fftw-3.2.x was 5:x:2 +dnl fftw-3.3.x was 6:x:3 for x < 4 +dnl fftw-3.3.4 was 7:x:4 +dnl fftw-3.3.5 was 8:x:5 (added planner hooks) +SHARED_VERSION_INFO="8:FFTW_MINOR_VERSION:5" # CURRENT:REVISION:AGE + +AM_INIT_AUTOMAKE(1.7) +AM_CONFIG_HEADER(config.h) +AC_CONFIG_MACRO_DIR([m4]) +AM_MAINTAINER_MODE +AC_SUBST(SHARED_VERSION_INFO) +AC_DISABLE_SHARED dnl to hell with shared libraries +AC_CANONICAL_HOST + +dnl configure options +case "${host_cpu}" in + powerpc*) have_fma=yes;; + ia64*) have_fma=yes;; + hppa*) have_fma=yes;; + mips64*) have_fma=yes;; + *) have_fma=no;; +esac + +AC_ARG_ENABLE(debug, [AC_HELP_STRING([--enable-debug],[compile fftw with extra runtime checks for debugging])], ok=$enableval, ok=no) +if test "$ok" = "yes"; then + AC_DEFINE(FFTW_DEBUG,1,[Define to enable extra FFTW debugging code.]) + debug_malloc=yes +else + debug_malloc=no +fi + +AC_ARG_ENABLE(doc, [AC_HELP_STRING([--disable-doc],[disable building the documentation])], build_doc=$enableval, build_doc=yes) +AM_CONDITIONAL(BUILD_DOC, test x"$build_doc" = xyes) + +AC_ARG_ENABLE(debug-malloc, [AC_HELP_STRING([--enable-debug-malloc],[enable malloc debugging version])], ok=$enableval, ok=$debug_malloc) +if test "$ok" = "yes"; then + AC_DEFINE(FFTW_DEBUG_MALLOC,1,[Define to enable debugging malloc.]) +fi + +AC_ARG_ENABLE(debug-alignment, [AC_HELP_STRING([--enable-debug-alignment],[enable alignment debugging hacks])], ok=$enableval, ok=no) +if test "$ok" = "yes"; then + AC_DEFINE(FFTW_DEBUG_ALIGNMENT,1,[Define to enable alignment debugging hacks.]) +fi + +AC_ARG_ENABLE(random-estimator, [AC_HELP_STRING([--enable-random-estimator],[enable pseudorandom estimator (debugging hack)])], ok=$enableval, ok=no) +if test "$ok" = "yes"; then + AC_DEFINE(FFTW_RANDOM_ESTIMATOR,1,[Define to enable pseudorandom estimate planning for debugging.]) + CHECK_PL_OPTS="--estimate" +fi + +AC_ARG_ENABLE(alloca, [AC_HELP_STRING([--disable-alloca],[disable use of the alloca() function (may be broken on mingw64)])], ok=$enableval, ok=yes) +if test "$ok" = "yes"; then + AC_DEFINE(FFTW_ENABLE_ALLOCA,1,[Define to enable the use of alloca().]) +fi + +AC_ARG_ENABLE(single, [AC_HELP_STRING([--enable-single],[compile fftw in single precision])], ok=$enableval, ok=no) +AC_ARG_ENABLE(float, [AC_HELP_STRING([--enable-float],[synonym for --enable-single])], ok=$enableval) +if test "$ok" = "yes"; then + AC_DEFINE(FFTW_SINGLE,1,[Define to compile in single precision.]) + AC_DEFINE(BENCHFFT_SINGLE,1,[Define to compile in single precision.]) + PRECISION=s +else + PRECISION=d +fi +AM_CONDITIONAL(SINGLE, test "$ok" = "yes") + +AC_ARG_ENABLE(long-double, [AC_HELP_STRING([--enable-long-double],[compile fftw in long-double precision])], ok=$enableval, ok=no) +if test "$ok" = "yes"; then + if test "$PRECISION" = "s"; then + AC_MSG_ERROR([--enable-single/--enable-long-double conflict]) + fi + AC_DEFINE(FFTW_LDOUBLE,1,[Define to compile in long-double precision.]) + AC_DEFINE(BENCHFFT_LDOUBLE,1,[Define to compile in long-double precision.]) + PRECISION=l +fi +AM_CONDITIONAL(LDOUBLE, test "$ok" = "yes") + +AC_ARG_ENABLE(quad-precision, [AC_HELP_STRING([--enable-quad-precision],[compile fftw in quadruple precision if available])], ok=$enableval, ok=no) +if test "$ok" = "yes"; then + if test "$PRECISION" != "d"; then + AC_MSG_ERROR([conflicting precisions specified]) + fi + AC_DEFINE(FFTW_QUAD,1,[Define to compile in quad precision.]) + AC_DEFINE(BENCHFFT_QUAD,1,[Define to compile in quad precision.]) + PRECISION=q +fi +AM_CONDITIONAL(QUAD, test "$ok" = "yes") + +AC_SUBST(PRECISION) +AC_SUBST(CHECK_PL_OPTS) + +dnl SSE/SSE2 theory: +dnl +dnl Historically, you had to supply --enable-sse in single precision and --enable-sse2 +dnl in double precision. +dnl +dnl This behavior is pointless in 2016. --enable-sse2 now works in both precisions, +dnl and is interpreted as --enable-sse in single precision. The old flag --enable--se +dnl is still supported in single-precision only. +AC_ARG_ENABLE(sse, [AC_HELP_STRING([--enable-sse],[enable SSE optimizations])], have_sse=$enableval, have_sse=no) +if test "$have_sse" = "yes"; then + if test "$PRECISION" != "s"; then + AC_MSG_ERROR([SSE requires single precision]) + fi +fi + +AC_ARG_ENABLE(sse2, [AC_HELP_STRING([--enable-sse2],[enable SSE/SSE2 optimizations])], have_sse2=$enableval, have_sse2=no) +if test "$have_sse" = "yes"; then have_sse2=yes; fi +if test "$have_sse2" = "yes"; then + AC_DEFINE(HAVE_SSE2,1,[Define to enable SSE/SSE2 optimizations.]) + if test "$PRECISION" != "d" -a "$PRECISION" != "s"; then + AC_MSG_ERROR([SSE2 requires single or double precision]) + fi +fi +AM_CONDITIONAL(HAVE_SSE2, test "$have_sse2" = "yes") + +AC_ARG_ENABLE(avx, [AC_HELP_STRING([--enable-avx],[enable AVX optimizations])], have_avx=$enableval, have_avx=no) +if test "$have_avx" = "yes"; then + AC_DEFINE(HAVE_AVX,1,[Define to enable AVX optimizations.]) + if test "$PRECISION" != "d" -a "$PRECISION" != "s"; then + AC_MSG_ERROR([AVX requires single or double precision]) + fi +fi +AM_CONDITIONAL(HAVE_AVX, test "$have_avx" = "yes") + +AC_ARG_ENABLE(avx2, [AC_HELP_STRING([--enable-avx2],[enable AVX2 optimizations])], have_avx2=$enableval, have_avx2=no) +if test "$have_avx2" = "yes"; then + AC_DEFINE(HAVE_AVX2,1,[Define to enable AVX2 optimizations.]) + if test "$PRECISION" != "d" -a "$PRECISION" != "s"; then + AC_MSG_ERROR([AVX2 requires single or double precision]) + fi + have_fma="yes" +fi +AM_CONDITIONAL(HAVE_AVX2, test "$have_avx2" = "yes") + +AC_ARG_ENABLE(avx512, [AC_HELP_STRING([--enable-avx512],[enable AVX512 optimizations])], have_avx512=$enableval, have_avx512=no) +if test "$have_avx512" = "yes"; then + AC_DEFINE(HAVE_AVX512,1,[Define to enable AVX512 optimizations.]) + if test "$PRECISION" != "d" -a "$PRECISION" != "s"; then + AC_MSG_ERROR([AVX512 requires single or double precision]) + fi + have_fma="yes" +fi +AM_CONDITIONAL(HAVE_AVX512, test "$have_avx512" = "yes") + +dnl 128-bit AVX is special. There is no reason to use it on Intel processors +dnl since SSE2 is just as fast. However, on AMD processors we can both use +dnl FMA4, and 128-bit SIMD is better than 256-bit since core pairs in a +dnl compute unit can execute two 128-bit instructions independently. +AC_ARG_ENABLE(avx-128-fma, [AC_HELP_STRING([--enable-avx-128-fma],[enable AVX128/FMA optimizations])], have_avx_128_fma=$enableval, have_avx_128_fma=no) +if test "$have_avx_128_fma" = "yes"; then + AC_DEFINE(HAVE_AVX_128_FMA,1,[Define to enable 128-bit FMA AVX optimization]) + AVX_128_FMA_CFLAGS="${AVX_CFLAGS} -mfma4" + AC_SUBST(AVX_128_FMA_CFLAGS) +fi +AM_CONDITIONAL(HAVE_AVX_128_FMA, test "$have_avx_128_fma" = "yes") + +AC_ARG_ENABLE(kcvi, [AC_HELP_STRING([--enable-kcvi],[enable Knights Corner vector instructions optimizations])], have_kcvi=$enableval, have_kcvi=no) +if test "$have_kcvi" = "yes"; then + AC_DEFINE(HAVE_KCVI,1,[Define to enable KCVI optimizations.]) + if test "$PRECISION" != "d" -a "$PRECISION" != "s"; then + AC_MSG_ERROR([Knights Corner vector instructions requires single or double precision]) + fi +fi +AM_CONDITIONAL(HAVE_KCVI, test "$have_kcvi" = "yes") + +AC_ARG_ENABLE(altivec, [AC_HELP_STRING([--enable-altivec],[enable Altivec optimizations])], have_altivec=$enableval, have_altivec=no) +if test "$have_altivec" = "yes"; then + AC_DEFINE(HAVE_ALTIVEC,1,[Define to enable Altivec optimizations.]) + if test "$PRECISION" != "s"; then + AC_MSG_ERROR([Altivec requires single precision]) + fi +fi +AM_CONDITIONAL(HAVE_ALTIVEC, test "$have_altivec" = "yes") + +AC_ARG_ENABLE(vsx, [AC_HELP_STRING([--enable-vsx],[enable IBM VSX optimizations])], have_vsx=$enableval, have_vsx=no) +if test "$have_vsx" = "yes"; then + AC_DEFINE(HAVE_VSX,1,[Define to enable IBM VSX optimizations.]) +fi +AM_CONDITIONAL(HAVE_VSX, test "$have_vsx" = "yes") + +AC_ARG_ENABLE(neon, [AC_HELP_STRING([--enable-neon],[enable ARM NEON optimizations])], have_neon=$enableval, have_neon=no) +if test "$have_neon" = "yes"; then + AC_DEFINE(HAVE_NEON,1,[Define to enable ARM NEON optimizations.]) + case "${host_cpu}" in + aarch64) + ;; + *) + if test "$PRECISION" != "s"; then + AC_MSG_ERROR([NEON requires single precision]) + fi + ;; + esac +fi +AM_CONDITIONAL(HAVE_NEON, test "$have_neon" = "yes") + +AC_ARG_ENABLE(armv8cyclecounter, [AC_HELP_STRING([--enable-armv8cyclecounter],[enable the cycle counter on ARMv8 ; require enabling in kernel mode, see <https://github.com/rdolbeau/enable_arm_pmu>])], have_armv8cyclecounter=$enableval) +if test "$have_armv8cyclecounter"x = "yes"x; then + AC_DEFINE(HAVE_ARMV8CC,1,[Define if you have enabled the cycle counter on ARMv8]) +fi + +AC_ARG_ENABLE(generic-simd128, [AC_HELP_STRING([--enable-generic-simd128],[enable generic (gcc) 128-bit SIMD optimizations])], have_generic_simd128=$enableval, have_generic_simd128=no) +if test "$have_generic_simd128" = "yes"; then + AC_DEFINE(HAVE_GENERIC_SIMD128,1,[Define to enable generic (gcc) 128-bit SIMD optimizations.]) +fi +AM_CONDITIONAL(HAVE_GENERIC_SIMD128, test "$have_generic_simd128" = "yes") + +AC_ARG_ENABLE(generic-simd256, [AC_HELP_STRING([--enable-generic-simd256],[enable generic (gcc) 256-bit SIMD optimizations])], have_generic_simd256=$enableval, have_generic_simd256=no) +if test "$have_generic_simd256" = "yes"; then + AC_DEFINE(HAVE_GENERIC_SIMD256,1,[Define to enable generic (gcc) 256-bit SIMD optimizations.]) +fi +AM_CONDITIONAL(HAVE_GENERIC_SIMD256, test "$have_generic_simd256" = "yes") + + +dnl FIXME: +dnl AC_ARG_ENABLE(mips-ps, [AC_HELP_STRING([--enable-mips-ps],[enable MIPS pair-single optimizations])], have_mips_ps=$enableval, have_mips_ps=no) +dnl if test "$have_mips_ps" = "yes"; then +dnl AC_DEFINE(HAVE_MIPS_PS,1,[Define to enable MIPS paired-single optimizations.]) +dnl if test "$PRECISION" != "s"; then +dnl AC_MSG_ERROR([MIPS paired-single requires single precision]) +dnl fi +dnl fi +dnl AM_CONDITIONAL(HAVE_MIPS_PS, test "$have_mips_ps" = "yes") + +AC_ARG_WITH(slow-timer, [AC_HELP_STRING([--with-slow-timer],[use low-precision timers (SLOW)])], with_slow_timer=$withval, with_slow_timer=no) +if test "$with_slow_timer" = "yes"; then + AC_DEFINE(WITH_SLOW_TIMER,1,[Use low-precision timers, making planner very slow]) +fi + +AC_ARG_ENABLE(mips_zbus_timer, [AC_HELP_STRING([--enable-mips-zbus-timer],[use MIPS ZBus cycle-counter])], have_mips_zbus_timer=$enableval, have_mips_zbus_timer=no) +if test "$have_mips_zbus_timer" = "yes"; then + AC_DEFINE(HAVE_MIPS_ZBUS_TIMER,1,[Define to enable use of MIPS ZBus cycle-counter.]) +fi + +AC_ARG_WITH(our-malloc, [AC_HELP_STRING([--with-our-malloc],[use our aligned malloc (helpful for Win32)])], with_our_malloc=$withval, with_our_malloc=no) +AC_ARG_WITH(our-malloc16, [AC_HELP_STRING([--with-our-malloc16],[Obsolete alias for --with-our-malloc16])], with_our_malloc=$withval) +if test "$with_our_malloc" = "yes"; then + AC_DEFINE(WITH_OUR_MALLOC,1,[Use our own aligned malloc routine; mainly helpful for Windows systems lacking aligned allocation system-library routines.]) +fi + +AC_ARG_WITH(windows-f77-mangling, [AC_HELP_STRING([--with-windows-f77-mangling],[use common Win32 Fortran interface styles])], with_windows_f77_mangling=$withval, with_windows_f77_mangling=no) +if test "$with_windows_f77_mangling" = "yes"; then + AC_DEFINE(WINDOWS_F77_MANGLING,1,[Use common Windows Fortran mangling styles for the Fortran interfaces.]) +fi + +AC_ARG_WITH(incoming-stack-boundary, [AC_HELP_STRING([--with-incoming-stack-boundary=X],[Assume that stack is aligned to (1<<X) bytes])], with_incoming_stack_boundary=$withval, with_incoming_stack_boundary=no) + + +AC_ARG_ENABLE(fma, [AC_HELP_STRING([--enable-fma],[enable optimizations for machines with fused multiply-add])], have_fma=$enableval) +if test "$have_fma"x = "yes"x; then + AC_DEFINE(HAVE_FMA,1,[Define if you have a machine with fused multiply-add]) +fi + +dnl compute library suffix +case "$PRECISION" in + s) PREC_SUFFIX=f;; + d) PREC_SUFFIX=;; + l) PREC_SUFFIX=l;; + q) PREC_SUFFIX=q;; +esac +AC_SUBST(PREC_SUFFIX) + +dnl Checks for programs. +AC_PROG_CC +AM_PROG_CC_C_O +AX_COMPILER_VENDOR +AC_PROG_CC_STDC +AC_PROG_INSTALL +AC_PROG_LN_S +AC_PROG_MAKE_SET +AC_LIBTOOL_WIN32_DLL +AC_PROG_LIBTOOL + +AC_CHECK_PROG(OCAMLBUILD, ocamlbuild, ocamlbuild) + +dnl ----------------------------------------------------------------------- + +AC_ARG_ENABLE(mpi, [AC_HELP_STRING([--enable-mpi],[compile FFTW MPI library])], enable_mpi=$enableval, enable_mpi=no) + +if test "$enable_mpi" = "yes"; then + if test $PRECISION = q; then + AC_MSG_ERROR([quad precision is not supported in MPI]) + fi + ACX_MPI([],[AC_MSG_ERROR([could not find mpi library for --enable-mpi])]) + AC_CHECK_PROG(MPIRUN, mpirun, mpirun) + AC_SUBST(MPIRUN) + + save_CC=$CC + CC=$MPICC + AC_CHECK_SIZEOF(MPI_Fint, [], [#include <mpi.h>]) + CC=$save_CC + if test 0 = $ac_cv_sizeof_MPI_Fint; then + AC_MSG_WARN([sizeof(MPI_Fint) test failed]); + dnl As a backup, assume Fortran integer == C int + AC_CHECK_SIZEOF(int) + if test 0 = $ac_cv_sizeof_int; then AC_MSG_ERROR([sizeof(int) test failed]); fi + ac_cv_sizeof_MPI_Fint=$ac_cv_sizeof_int + fi + C_MPI_FINT=C_INT`expr $ac_cv_sizeof_MPI_Fint \* 8`_T + AC_SUBST(C_MPI_FINT) +fi +AM_CONDITIONAL(MPI, test "$enable_mpi" = "yes") + +dnl ----------------------------------------------------------------------- + +dnl determine CFLAGS first +AX_CC_MAXOPT + +case "${ax_cv_c_compiler_vendor}" in + intel) # Stop icc from defining __GNUC__, except on MacOS where this fails + case "${host_os}" in + *darwin*) ;; # icc -no-gcc fails to compile some system headers + *) + AX_CHECK_COMPILER_FLAGS([-no-gcc], [CC="$CC -no-gcc"]) + ;; + esac + ;; + + hp) # must (sometimes) manually increase cpp limits to handle fftw3.h + AX_CHECK_COMPILER_FLAGS([-Wp,-H128000], + [CC="$CC -Wp,-H128000"]) + ;; + + portland) # -Masmkeyword required for asm("") cycle counters + AX_CHECK_COMPILER_FLAGS([-Masmkeyword], + [CC="$CC -Masmkeyword"]) + ;; +esac + +dnl Determine SIMD CFLAGS at least for gcc and icc +case "${ax_cv_c_compiler_vendor}" in + gnu|intel) + # SSE/SSE2 + if test "$have_sse2" = "yes" -a "x$SSE2_CFLAGS" = x; then + if test "$PRECISION" = d; then flag=msse2; else flag=msse; fi + AX_CHECK_COMPILER_FLAGS(-$flag, [SSE2_CFLAGS="-$flag"], + [AC_MSG_ERROR([Need a version of gcc with -$flag])]) + fi + + # AVX + if test "$have_avx" = "yes" -a "x$AVX_CFLAGS" = x; then + AX_CHECK_COMPILER_FLAGS(-mavx, [AVX_CFLAGS="-mavx"], + [AC_MSG_ERROR([Need a version of gcc with -mavx])]) + fi + + # AVX2 + # gcc-4.8 works with -march=core-avx2, but -mavx2 is not enough. + # Later versions seem to happy with -mavx2, so try the arch one first. + if test "$have_avx2" = "yes" -a "x$AVX2_CFLAGS" = x; then + AX_CHECK_COMPILER_FLAGS(-march=core-avx2, [AVX2_CFLAGS="-march=core-avx2"], + [AX_CHECK_COMPILER_FLAGS(-mavx2, [AVX2_CFLAGS="-mavx2"], + [AC_MSG_ERROR([Need a version of gcc with either -march=core-avx2 or -mavx2])])]) + AX_CHECK_COMPILER_FLAGS(-mfma, [AVX2_CFLAGS="$AVX2_CFLAGS -mfma"], + [AC_MSG_WARN([Need a version of gcc with -mfma (harmless for icc)])]) + fi + + # AVX512 + if test "$have_avx512" = "yes" -a "x$AVX512_CFLAGS" = x; then + AX_CHECK_COMPILER_FLAGS(-mavx512f, [AVX512_CFLAGS="-mavx512f"], + [AC_MSG_ERROR([Need a version of gcc with -mavx512f])]) + fi + + if test "$host_vendor" = "apple"; then + # We need to tell gcc to use an external assembler to get AVX/AVX2 with gcc on OS X + AX_CHECK_COMPILER_FLAGS([-Wa,-q], [CFLAGS="$CFLAGS -Wa,-q"]) + # Disable the new compact unwinding format so we avoid warnings/potential errors. + AX_CHECK_COMPILER_FLAGS([-Wl,-no_compact_unwind], [CFLAGS="$CFLAGS -Wl,-no_compact_unwind"]) + fi + + # KCVI + if test "$have_kcvi" = "yes" -a "x$KCVI_CFLAGS" = x; then + AX_CHECK_COMPILER_FLAGS(-mmic, [KCVI_CFLAGS="-mmic"], + [AC_MSG_ERROR([Need a version of icc with -mmic])]) + fi + + if test "$have_altivec" = "yes" -a "x$ALTIVEC_CFLAGS" = x; then + # -DFAKE__VEC__ is a workaround because gcc-3.3 does not + # #define __VEC__ with -maltivec. + AX_CHECK_COMPILER_FLAGS(-faltivec, [ALTIVEC_CFLAGS="-faltivec"], + [AX_CHECK_COMPILER_FLAGS(-maltivec -mabi=altivec, + [ALTIVEC_CFLAGS="-maltivec -mabi=altivec -DFAKE__VEC__"], + [AX_CHECK_COMPILER_FLAGS(-fvec, [ALTIVEC_CFLAGS="-fvec"], + [AC_MSG_ERROR([Need a version of gcc with -maltivec])])])]) + fi + + case "${host_cpu}" in + aarch64) + ;; + *) + if test "$have_neon" = "yes" -a "x$NEON_CFLAGS" = x; then + AX_CHECK_COMPILER_FLAGS(-mfpu=neon, [NEON_CFLAGS="-mfpu=neon"], + [AC_MSG_ERROR([Need a version of gcc with -mfpu=neon])]) + fi + ;; + esac + + if test "$have_vsx" = "yes" -a "x$VSX_CFLAGS" = x; then + AX_CHECK_COMPILER_FLAGS(-mvsx, [VSX_CFLAGS="-mvsx"], + [AC_MSG_ERROR([Need a version of gcc with -mvsx])]) + fi + + dnl FIXME: + dnl elif test "$have_mips_ps" = "yes"; then + dnl # Just punt here and use only new 4.2 compiler :( + dnl # Should add section for older compilers... + dnl AX_CHECK_COMPILER_FLAGS(-mpaired-single, + dnl [SIMD_CFLAGS="-mpaired-single"], + dnl #[AC_MSG_ERROR([Need a version of gcc with -mpaired-single])]) + dnl [AX_CHECK_COMPILER_FLAGS(-march=mips64, + dnl [SIMD_CFLAGS="-march=mips64"], + dnl [AC_MSG_ERROR( + dnl [Need a version of gcc with -mpaired-single or -march=mips64]) + dnl ])]) + dnl fi + ;; + + clang) + + if test "$have_avx" = "yes" -a "x$AVX_CFLAGS" = x; then + AX_CHECK_COMPILER_FLAGS(-mavx, [AVX_CFLAGS="-mavx"], + [AC_MSG_ERROR([Need a version of clang with -mavx])]) + fi + + if test "$have_avx2" = "yes" -a "x$AVX2_CFLAGS" = x; then + AX_CHECK_COMPILER_FLAGS(-mavx2, [AVX2_CFLAGS="-mavx2"], + [AC_MSG_ERROR([Need a version of clang with -mavx2])]) + AX_CHECK_COMPILER_FLAGS(-mfma, [AVX2_CFLAGS="$AVX2_CFLAGS -mfma"]) + fi + + if test "$have_vsx" = "yes" -a "x$VSX_CFLAGS" = x; then + # clang appears to need both -mvsx and -maltivec for VSX + AX_CHECK_COMPILER_FLAGS(-maltivec, [VSX_CFLAGS="-maltivec"], + [AC_MSG_ERROR([Need a version of gcc with -maltivec])]) + AX_CHECK_COMPILER_FLAGS(-mvsx, [VSX_CFLAGS="-mvsx $VSX_CFLAGS"], + [AC_MSG_ERROR([Need a version of gcc with -mvsx])]) + fi + ;; + + ibm) + if test "$have_vsx" = "yes" -a "x$VSX_CFLAGS" = x; then + # Note that IBM xlC uses -qaltivec for VSX too. + AX_CHECK_COMPILER_FLAGS(-qaltivec, [VSX_CFLAGS="-qaltivec"], + [AC_MSG_ERROR([Need a version of gcc with -qaltivec])]) + fi + ;; +esac + +AC_SUBST(SSE2_CFLAGS) +AC_SUBST(AVX_CFLAGS) +AC_SUBST(AVX2_CFLAGS) +AC_SUBST(AVX512_CFLAGS) +AC_SUBST(KCVI_CFLAGS) +AC_SUBST(ALTIVEC_CFLAGS) +AC_SUBST(VSX_CFLAGS) +AC_SUBST(NEON_CFLAGS) + +dnl add stack alignment CFLAGS if so requested +if test "$with_incoming_stack_boundary"x != "no"x; then + case "${ax_cv_c_compiler_vendor}" in + gnu) + tentative_flags="-mincoming-stack-boundary=$with_incoming_stack_boundary"; + AX_CHECK_COMPILER_FLAGS($tentative_flags, + [STACK_ALIGN_CFLAGS=$tentative_flags]) + ;; + esac +fi +AC_SUBST(STACK_ALIGN_CFLAGS) + +dnl Checks for header files. +AC_HEADER_STDC +AC_CHECK_HEADERS([libintl.h malloc.h stddef.h stdlib.h string.h strings.h sys/time.h unistd.h limits.h c_asm.h intrinsics.h stdint.h mach/mach_time.h sys/sysctl.h]) +dnl c_asm.h: Header file for enabling asm() on Digital Unix +dnl intrinsics.h: cray unicos +dnl sys/sysctl.h: MacOS X altivec detection + +dnl altivec.h requires $ALTIVEC_CFLAGS (we use this for VSX too, which uses the same header) +save_CFLAGS="$CFLAGS" +save_CPPFLAGS="$CPPFLAGS" +CFLAGS="$CFLAGS $ALTIVEC_CFLAGS $VSX_CFLAGS" +CPPFLAGS="$CPPFLAGS $ALTIVEC_CFLAGS $VSX_CFLAGS" +AC_CHECK_HEADERS([altivec.h]) +CFLAGS="$save_CFLAGS" +CPPFLAGS="$save_CPPFLAGS" + +dnl Checks for typedefs, structures, and compiler characteristics. +AC_C_CONST +AC_C_INLINE +AC_TYPE_SIZE_T +AC_HEADER_TIME +AC_CHECK_TYPE([long double], + [AC_DEFINE(HAVE_LONG_DOUBLE, 1, [Define to 1 if the compiler supports `long double'])], +[ +if test $PRECISION = l; then + AC_MSG_ERROR([long double is not a supported type with your compiler.]) +fi +]) +AC_CHECK_TYPE([hrtime_t],[AC_DEFINE(HAVE_HRTIME_T, 1, [Define to 1 if hrtime_t is defined in <sys/time.h>])],, +[ +#if HAVE_SYS_TIME_H +#include <sys/time.h> +#endif +]) + +AC_CHECK_SIZEOF(int) +AC_CHECK_SIZEOF(unsigned int) +AC_CHECK_SIZEOF(long) +AC_CHECK_SIZEOF(unsigned long) +AC_CHECK_SIZEOF(long long) +AC_CHECK_SIZEOF(unsigned long long) +AC_CHECK_SIZEOF(size_t) +AC_CHECK_SIZEOF(ptrdiff_t) + +AC_CHECK_TYPES(uintptr_t, [], [AC_CHECK_SIZEOF(void *)], [$ac_includes_default +#ifdef HAVE_STDINT_H +# include <stdint.h> +#endif]) + +AC_CHECK_SIZEOF(float) +AC_CHECK_SIZEOF(double) + +dnl Check sizeof fftw_r2r_kind for Fortran interface [it has == sizeof(int) +dnl for years, but being paranoid]. Note: the definition here must match +dnl the one in api/fftw3.h! +AC_CHECK_SIZEOF(fftw_r2r_kind, [], [typedef enum { + FFTW_R2HC=0, FFTW_HC2R=1, FFTW_DHT=2, + FFTW_REDFT00=3, FFTW_REDFT01=4, FFTW_REDFT10=5, FFTW_REDFT11=6, + FFTW_RODFT00=7, FFTW_RODFT01=8, FFTW_RODFT10=9, FFTW_RODFT11=10 +} fftw_r2r_kind;]) +if test 0 = $ac_cv_sizeof_fftw_r2r_kind; then AC_MSG_ERROR([sizeof(fftw_r2r_kind) test failed]); fi +C_FFTW_R2R_KIND=C_INT`expr $ac_cv_sizeof_fftw_r2r_kind \* 8`_T +AC_SUBST(C_FFTW_R2R_KIND) + +dnl Checks for library functions. +AC_FUNC_ALLOCA +AC_FUNC_STRTOD +AC_FUNC_VPRINTF +AC_CHECK_LIB(m, sin) + +if test $PRECISION = q; then + AX_GCC_VERSION(4,6,0,[],[AC_MSG_ERROR([gcc 4.6 or later required for quad precision support])]) + AC_CHECK_LIB(quadmath, sinq, [], [AC_MSG_ERROR([quad precision requires libquadmath for quad-precision trigonometric routines])]) + LIBQUADMATH=-lquadmath +fi +AC_SUBST(LIBQUADMATH) + +AC_CHECK_FUNCS([BSDgettimeofday gettimeofday gethrtime read_real_time time_base_to_time drand48 sqrt memset posix_memalign memalign _mm_malloc _mm_free clock_gettime mach_absolute_time sysctl abort sinl cosl snprintf]) +AC_CHECK_DECLS([sinl, cosl, sinq, cosq],,,[#include <math.h>]) +AC_CHECK_DECLS([memalign],,,[ +#ifdef HAVE_MALLOC_H +#include <malloc.h> +#endif]) +AC_CHECK_DECLS([drand48, srand48, posix_memalign]) dnl in stdlib.h + +dnl Cray UNICOS _rtc() (real-time clock) intrinsic +AC_MSG_CHECKING([for _rtc intrinsic]) +rtc_ok=yes +AC_TRY_LINK([#ifdef HAVE_INTRINSICS_H +#include <intrinsics.h> +#endif], [_rtc()], [AC_DEFINE(HAVE__RTC,1,[Define if you have the UNICOS _rtc() intrinsic.])], [rtc_ok=no]) +AC_MSG_RESULT($rtc_ok) + +if test "$PRECISION" = "l"; then + AC_CHECK_FUNCS([cosl sinl tanl], [], [AC_MSG_ERROR([long-double precision requires long-double trigonometric routines])]) +fi + +AC_MSG_CHECKING([for isnan]) +AC_TRY_LINK([#include <math.h> +], if (!isnan(3.14159)) isnan(2.7183);, ok=yes, ok=no) +if test "$ok" = "yes"; then + AC_DEFINE(HAVE_ISNAN,1,[Define if the isnan() function/macro is available.]) +fi +AC_MSG_RESULT(${ok}) + +dnl TODO +AX_GCC_ALIGNS_STACK() + +dnl override CFLAGS selection when debugging +if test "${enable_debug}" = "yes"; then + CFLAGS="-g" +fi + +dnl add gcc warnings, in debug/maintainer mode only +if test "$enable_debug" = yes || test "$USE_MAINTAINER_MODE" = yes; then +if test "$ac_test_CFLAGS" != "set"; then + if test $ac_cv_prog_gcc = yes; then + CFLAGS="$CFLAGS -Wall -W -Wcast-qual -Wpointer-arith -Wcast-align -pedantic -Wno-long-long -Wshadow -Wbad-function-cast -Wwrite-strings -Wstrict-prototypes -Wredundant-decls -Wnested-externs" # -Wundef -Wconversion -Wmissing-prototypes -Wmissing-declarations + fi +fi +fi + +dnl check for a proper indent in maintainer mode +if test "$USE_MAINTAINER_MODE" = yes; then + AC_PATH_PROG(INDENT, indent, indent) + # if INDENT is set to 'indent' then we didn't find indent + if test "$INDENT" != indent ; then + AC_MSG_CHECKING(if $INDENT is GNU indent) + if $INDENT --version 2>/dev/null | head -n 1|grep "GNU indent" > /dev/null ; then + AC_MSG_RESULT(yes) + INDENT="$INDENT -kr -cs -i5 -l800 -fca -nfc1 -sc -sob -cli4 -TR -Tplanner -TV" + else + AC_MSG_RESULT(no) + AC_MSG_WARN($INDENT does not appear to be GNU indent.) + fi + else + AC_MSG_WARN(no indent program found: codelets will be ugly) + INDENT=cat + fi +fi + +dnl ----------------------------------------------------------------------- + +AC_ARG_ENABLE(fortran, [AC_HELP_STRING([--disable-fortran],[don't include Fortran-callable wrappers])], enable_fortran=$enableval, enable_fortran=yes) + +if test "$enable_fortran" = "yes"; then + AC_PROG_F77 + if test -z "$F77"; then + enable_fortran=no + AC_MSG_WARN([*** Couldn't find f77 compiler; using default Fortran wrappers.]) + else + AC_F77_DUMMY_MAIN([], [enable_fortran=no + AC_MSG_WARN([*** Couldn't figure out how to link C and Fortran; using default Fortran wrappers.])]) + fi +else + AC_DEFINE([DISABLE_FORTRAN], 1, [Define to disable Fortran wrappers.]) +fi + +if test "x$enable_fortran" = xyes; then + AC_F77_WRAPPERS + AC_F77_FUNC(f77foo) + AC_F77_FUNC(f77_foo) + f77_foo2=`echo $f77foo | sed 's/77/77_/'` + if test "$f77_foo" = "$f77_foo2"; then + AC_DEFINE(F77_FUNC_EQUIV, 1, [Define if F77_FUNC and F77_FUNC_ are equivalent.]) + + # Include g77 wrappers by default for GNU systems or gfortran + with_g77_wrappers=$ac_cv_f77_compiler_gnu + case $host_os in *gnu*) with_g77_wrappers=yes ;; esac + fi +else + with_g77_wrappers=no +fi + +AC_ARG_WITH(g77-wrappers, [AC_HELP_STRING([--with-g77-wrappers],[force inclusion of g77-compatible wrappers in addition to any other Fortran compiler that is detected])], with_g77_wrappers=$withval) +if test "x$with_g77_wrappers" = "xyes"; then + AC_DEFINE(WITH_G77_WRAPPERS,1,[Include g77-compatible wrappers in addition to any other Fortran wrappers.]) +fi + +dnl ----------------------------------------------------------------------- +have_smp="no" +AC_ARG_ENABLE(openmp, [AC_HELP_STRING([--enable-openmp],[use OpenMP directives for parallelism])], enable_openmp=$enableval, enable_openmp=no) + +if test "$enable_openmp" = "yes"; then + AC_DEFINE(HAVE_OPENMP,1,[Define to enable OpenMP]) + AX_OPENMP([], [AC_MSG_ERROR([don't know how to enable OpenMP])]) +fi + +AC_ARG_ENABLE(threads, [AC_HELP_STRING([--enable-threads],[compile FFTW SMP threads library])], enable_threads=$enableval, enable_threads=no) + +if test "$enable_threads" = "yes"; then + AC_DEFINE(HAVE_THREADS,1,[Define to enable SMP threads]) +fi + +AC_ARG_WITH(combined-threads, [AC_HELP_STRING([--with-combined-threads],[combine threads into main libfftw3])], with_combined_threads=$withval, with_combined_threads=no) + +if test "$with_combined_threads" = yes; then + if test "$enable_openmp" = "yes"; then + AC_MSG_ERROR([--with-combined-threads incompatible with --enable-openmp]) + fi + if test "$enable_threads" != "yes"; then + AC_MSG_ERROR([--with-combined-threads requires --enable-threads]) + fi +fi + +dnl Check for threads library... +THREADLIBS="" +if test "$enable_threads" = "yes"; then + # Win32 threads are the default on Windows: + if test -z "$THREADLIBS"; then + AC_MSG_CHECKING([for Win32 threads]) + AC_TRY_LINK([#include <windows.h>], + [_beginthreadex(0,0,0,0,0,0);], + [THREADLIBS=" "; AC_MSG_RESULT(yes)], + [AC_MSG_RESULT(no)]) + fi + + # POSIX threads, the default choice everywhere else: + if test -z "$THREADLIBS"; then + ACX_PTHREAD([THREADLIBS="$PTHREAD_LIBS " + CC="$PTHREAD_CC" + AC_DEFINE(USING_POSIX_THREADS, 1, [Define if we have and are using POSIX threads.])]) + fi + + if test -z "$THREADLIBS"; then + AC_MSG_ERROR([couldn't find threads library for --enable-threads]) + fi + AC_DEFINE(HAVE_THREADS, 1, [Define if we have a threads library.]) +fi +AC_SUBST(THREADLIBS) +AM_CONDITIONAL(THREADS, test "$enable_threads" = "yes") +AM_CONDITIONAL(OPENMP, test "$enable_openmp" = "yes") +AM_CONDITIONAL(SMP, test "$enable_threads" = "yes" -o "$enable_openmp" = "yes") +AM_CONDITIONAL(COMBINED_THREADS, test x"$with_combined_threads" = xyes) + + +dnl ----------------------------------------------------------------------- +dnl Check for not-always-available (not quite) cycle counters +case "${host_cpu}" in + armv7*) + AC_MSG_CHECKING([armv7a has 64 bits readable CNTVCT]) + AC_RUN_IFELSE( + [AC_LANG_PROGRAM([[#include <stdint.h>]], + [[uint32_t Rt, Rt2 = 0;asm volatile("mrrc p15, 1, %0, %1, c14" : "=r"(Rt), "=r"(Rt2));]] + )], + [AC_DEFINE(ARMV7A_HAS_CNTCVT,1,[Define if CNTVCT is 64 bits readable on armv7a ]) + AC_MSG_RESULT([yes])], + [AC_MSG_RESULT([no])] + ) + ;; + aarch64) + AC_MSG_CHECKING([armv8 has 64 bits readable CNTVCT_EL0]) + AC_RUN_IFELSE( + [AC_LANG_PROGRAM([[#include <stdint.h>]], + [[uint64_t Rt;asm volatile("mrs %0, CNTVCT_EL0" : "=r" (Rt));]] + )], + [AC_DEFINE(ARMV8_HAS_CNTCVT_EL0,1,[Define if CNTVCT_EL0 is 64 bits readable on armv8 ]) + AC_MSG_RESULT([yes])], + [AC_MSG_RESULT([no])] + ) + ;; + *) + ;; +esac + +dnl ----------------------------------------------------------------------- + +AC_MSG_CHECKING([whether a cycle counter is available]) +save_CPPFLAGS=$CPPFLAGS +CPPFLAGS="$CPPFLAGS -I$srcdir/kernel" +AC_TRY_CPP([#include "cycle.h" +#ifndef HAVE_TICK_COUNTER +# error No cycle counter +#endif], [ok=yes], [ok=no]) +CPPFLAGS=$save_CPPFLAGS +AC_MSG_RESULT($ok) +if test $ok = no && test "x$with_slow_timer" = xno; then + echo "***************************************************************" + echo "WARNING: No cycle counter found. FFTW will use ESTIMATE mode " + echo " for all plans. See the manual for more information." + echo "***************************************************************" +fi + +dnl ----------------------------------------------------------------------- + +AC_DEFINE_UNQUOTED(FFTW_CC, "$CC $CFLAGS", [C compiler name and flags]) + +AC_CONFIG_FILES([ + Makefile + support/Makefile + genfft/Makefile + kernel/Makefile + simd-support/Makefile + + dft/Makefile + dft/scalar/Makefile + dft/scalar/codelets/Makefile + dft/simd/Makefile + dft/simd/common/Makefile + dft/simd/sse2/Makefile + dft/simd/avx/Makefile + dft/simd/avx-128-fma/Makefile + dft/simd/avx2/Makefile + dft/simd/avx2-128/Makefile + dft/simd/avx512/Makefile + dft/simd/kcvi/Makefile + dft/simd/altivec/Makefile + dft/simd/vsx/Makefile + dft/simd/neon/Makefile + dft/simd/generic-simd128/Makefile + dft/simd/generic-simd256/Makefile + + rdft/Makefile + rdft/scalar/Makefile + rdft/scalar/r2cf/Makefile + rdft/scalar/r2cb/Makefile + rdft/scalar/r2r/Makefile + rdft/simd/Makefile + rdft/simd/common/Makefile + rdft/simd/sse2/Makefile + rdft/simd/avx/Makefile + rdft/simd/avx-128-fma/Makefile + rdft/simd/avx2/Makefile + rdft/simd/avx2-128/Makefile + rdft/simd/avx512/Makefile + rdft/simd/kcvi/Makefile + rdft/simd/altivec/Makefile + rdft/simd/vsx/Makefile + rdft/simd/neon/Makefile + rdft/simd/generic-simd128/Makefile + rdft/simd/generic-simd256/Makefile + + reodft/Makefile + + threads/Makefile + + api/Makefile + + mpi/Makefile + + libbench2/Makefile + tests/Makefile + doc/Makefile + doc/FAQ/Makefile + + tools/Makefile + tools/fftw_wisdom.1 + tools/fftw-wisdom-to-conf + + m4/Makefile + + fftw.pc +]) + +AC_OUTPUT