View Issue Details

IDProjectCategoryView StatusLast Update
0000102ardourbugspublic2004-05-07 03:17
Reporterblubbfisch Assigned Topaul  
Status closedResolutionfixed 
Summary0000102: configure error on powerpc
Descriptionconfigure for ardour 0.9-beta7 fails for powerpc-platform, because gcc doesn't know about -march on powerpc.
Additional InformationQUICK FIX:
find -name configure |xargs sed -i "s/-march=\$target_cpu/-mcpu=7450-maltivec -mabi=altivec/"
TagsNo tags attached.



2003-10-28 14:28

administrator   ~0000117

Can we consider this to be a gcc bug? Which version of gcc are you using?


2003-10-28 15:16

administrator   ~0000118

this is totally a gcc bug. -march is listed in the info pages for gcc. it is not listed as a processor specific option. i really don't want to fix this in our configure scripts (though we will if strictly necessary).


2003-10-28 15:25

reporter   ~0000120

Errm, AFAIK gcc _never_ supported the -march flag on powerpc.... I only ever remember using -mcpu=750 or similar.
Bug or not, it is extremely inconsistent and annoying.

I only managed to convince autoconf and friends to behave by hardcoding the flags into myself. Any interest in getting the altivec-and-co-detection code?


2003-10-28 15:36

administrator   ~0000122

you're right and i'm wrong. i looked at gcc info again; -march is x86 specific.
if you can give us some code that sets the flags properly for PPC, that would be great.


2003-10-28 15:46

reporter   ~0000123

Ok, here we go. it isn't exactly elegant, but it worked for me :-)

if test "$target_cpu" = "powerpc"; then
  AC_DEFINE(POWERPC, 1, "Are we running a ppc CPU?")
  altivecLinux=`cat /proc/cpuinfo | grep -i altivec >/dev/null`
  if test "$?" = "0"; then
    AC_DEFINE(HAVE_ALTIVEC_LINUX, 1, "Is there Altivec Support ?")
    if test "$gcc_major_version" = "3"; then
dnl -mcpu=7450 does not reliably work with gcc 3.*
      OPTIM_FLAGS="-O2 -mcpu=7400 -maltivec -mabi=altivec"
      OPTIM_FLAGS="-O3 -mcpu=7400"
    OPTIM_FLAGS="-O3 -mcpu=750 -mmultiple"
  OPTIM_FLAGS="$OPTIM_FLAGS -mhard-float -mpowerpc-gfxopt"
elif echo $target_cpu | grep "i*86" >/dev/null; then
  cat /proc/cpuinfo | grep mmx >/dev/null
  if test $? = 0; then
  cat /proc/cpuinfo | grep sse >/dev/null
  if test $? = 0; then
    sse="-msse -mfpmath=sse"
  cat /proc/cpuinfo | grep 3dnow >/dev/null
  if test $? = 0; then
  AC_DEFINE(INTEL, 1, "Nope its intel")
  if test "$target_cpu" = "i586"; then
    OPTIM_FLAGS="-O3 -march=i586"
  elif test "$target_cpu" = "i686"; then
    OPTIM_FLAGS="-O3 -march=i686"
    if test "$gcc_major_version" = "3"; then
      OPTIM_FLAGS="$OPTIM_FLAGS $mmx $sse $dreidnow"


and then I use $OPTIM_FLAGS to add to the user's CFLAGS...

Keep in mind that gcc _might_ have problems with Optimization higher than -O2 together with -maltivec, although ardour runs fine on my G4 with -O6!


2003-10-28 17:29

administrator   ~0000124

added ppc detection code to's in my tree.


2003-11-05 04:15

administrator   ~0000145

configure script material added to for libs/ardour and gtk_ardour.


2003-11-12 20:09

administrator   ~0000172

Reminder sent to blubbfisch

Could you mark this bug as closed if this has fixed your problem?


2003-11-13 10:12

reporter   ~0000174

Errm, oops, sorry, forgot about this. thanks for reminding me.

configure and make work now on ppc, BUT:

it's a total mess of different options (some parts are compiled with -O6 only, some are compiled without any optimization at all, only -mpowerpc-gfxopt and -mhard-float). Bug or feature?

Shouldn't software acknowledge any set CFLAGS CXXFLAGS in the users's environment and just strip those flags known not to work with the specific code, respectively add flags known to be very helpful in optimizing?

I know this is a lot of sed/awk magic, but may well be worth the effort to yield the best possible compilation flags.

Have fun*


Issue History

Date Modified Username Field Change
2003-10-28 14:19 blubbfisch New Issue
2003-10-28 14:28 taybin Note Added: 0000117
2003-10-28 15:16 paul Note Added: 0000118
2003-10-28 15:17 taybin Status new => acknowledged
2003-10-28 15:17 taybin Description Updated
2003-10-28 15:25 blubbfisch Note Added: 0000120
2003-10-28 15:36 paul Note Added: 0000122
2003-10-28 15:46 blubbfisch Note Added: 0000123
2003-10-28 17:29 paul Note Added: 0000124
2003-10-28 17:30 paul Status acknowledged => assigned
2003-11-05 04:15 paul Status assigned => resolved
2003-11-05 04:15 paul Resolution open => fixed
2003-11-05 04:15 paul Assigned To => paul
2003-11-05 04:15 paul Note Added: 0000145
2003-11-12 20:09 taybin Note Added: 0000172
2003-11-13 10:12 blubbfisch Note Added: 0000174
2004-05-07 03:17 taybin Status resolved => closed