Re: [Fink-devel] -mtune/-mcpu and macintel

[Daniel Johnson]

On Apr 10, 2006, at 2:54 PM, David R. Morrison wrote:

On Apr 10, 2006, at 11:10 AM, Jack Howarth wrote:

Dave,
   For the sparky-py fink packages, I have a sparky-py.patch which
patches the Makefile to have ...

-CXX            = g++
-CXXFLAGS       =
+CXX            = g++-3.3

 CC             = gcc
-CFLAGS         =
+CFLAGS         = -O3 -ffast-math -mtune=970 -fPIC -fno-common

..so I assume I need to handle this in the patchscript with an ifdef

to '-mtune=i686'.

+CXXFLAGS = -O3 -ffast-math -mtune=970 -fPIC -fno-common - Wno-long-double macintel or whatever and have it sed the patch file to change '- mtune=970'

Either that, or pass CXXFLAGS on the command line, if that works.

By the way, it would seem rather silly to use 'i386'
as Macintel will always have at least a Pentium Pro class cpu if not
higher.
                Jack

Sure. The architecture flag %m comes out at i386 because that's what uname -a gives, but for processor optimization you'd want to use something different.

--
Daniel Johnson
[EMAIL PROTECTED]
PGP public key: http://homepage.mac.com/danielj7/publickey.txt

I just wanted to note that that the mtune/mcpu flags for gcc on Intel Macs are set by default to the best settings for Core processors already and if you change them you'll probably end up decreasing performance. Unfortunately, many packages will try to figure out what x86 processor is being used so that they can tune themselves appropriately, but since Core is so new they end up using a poorly chosen default setting like "i386" or "i486". So it's best to make sure that no mtune/mcpu flags get passed on Intel.

PGP.sig
Description: This is a digitally signed message part