From: <bar...@t-...> - 2002-06-04 21:26:53
|
hi michel, On Tue, 4 Jun 2002, Michel LESPINASSE wrote: > This is what I added in my acinclude.m4 to do the detection: > > dnl AC_C_ALWAYS_INLINE > dnl Define inline to something appropriate, including the new always_inline > dnl attribute from gcc 3.1 > AC_DEFUN([AC_C_ALWAYS_INLINE], > [AC_C_INLINE > if test x"$GCC" = x"yes" -a x"$ac_cv_c_inline" = x"inline"; then > AC_MSG_CHECKING([for always_inline]) > SAVE_CFLAGS="$CFLAGS" > CFLAGS="$CFLAGS -Wall -Werror" > AC_TRY_COMPILE([],[__attribute__ ((__always_inline__)) void f (void);], > [ac_cv_always_inline=yes],[ac_cv_always_inline=no]) > CFLAGS="$SAVE_CFLAGS" > AC_MSG_RESULT([$ac_cv_always_inline]) > if test x"$ac_cv_always_inline" = x"yes"; then > AC_DEFINE_UNQUOTED([inline],[__attribute__ ((__always_inline__))]) > fi > fi]) daniel - at least to me this looks like that magci stuff you always take care of? ;> > > have you also experimented with all those -funroll... optimization > > directives? without having looked into it any further i could at least > > imagine some of them to be counter-productive with gcc 3.1 > > I havent looked too much... libmpeg2 only uses quite simple > optimization flags though (-O3 -fomit-frame-pointer -mcpu=pentiumpro) humm - i remember i once started with stolen mpeg2dec optimization directives back when it was still maintained by aaron - guess this is really something to work at again. maybe i should try exactly these cflags as well ... but doesn't the gcc have new optimization flags - something related to profiler based optimizations? cheers, guenter -- time is a funny concept |