Re: question about fortran on the new intel Mac

[Erik Schnetter <schnetter at cct dot lsu dot edu>]

On Apr 7, 2006, at 10:17:47, FX Coudert wrote:

> Hi Erik,
>
> > I have gfortran working, starting from the SVN version and  
> > applying some patches.  It's not clean, and I hope that the  
> > official version will work in a few days/weeks.  There are two or  
> > three small problems, solutions for these are known and patches  
> > exist, but they have somehow not yet been officially applied.
>
> I'm also interested in this. Do you have pointers to these patches?  
> Were they proposed on the gcc-patches ml?

Here is my recipe for using gfortran on a MacBook Pro, which has an  
Intel processor.  I use Mac OS X 10.4.6, which runs on Darwin 8.6.1.   
It seems that gfortran is in a much better shape than when I started  
to use it on a Mac -- I think that all gfortran problems that I  
encountered have been solved in the mean time.



0. I started from gcc 112496 (this is the subversion version  
number).  I assume that later versions also work; maybe some of the  
patches that I applied are then superfluous.

1. I installed a newer version of cctools.  I use version 590.36;  
version 590.23 does not work.  This corrects an error with the symbol  
"__dso_handle".

2. I installed the patch http://gcc.gnu.org/ml/gcc-patches/2006-02/ 
msg02041.html, which is supposed to correct an error with  
"__Unwind_GetIPInfo", apparently for g++.  I do not know whether this  
patch is necessary, or whether it has been applied or superseded.

3. I installed a patch that corrects the initialisation of the SSE  
unit in gfortran.  This is bug 26712.  I see that  this bug has been  
closed in the mean time, so that you don't need the patch below if  
you use a current svn version.

Index: libgfortran/config/fpu-387.h
===================================================================
--- libgfortran/config/fpu-387.h	(revision 112496)
+++ libgfortran/config/fpu-387.h	(working copy)
@@ -93,12 +93,13 @@
       /* SSE */
       asm volatile ("stmxcsr %0" : "=m" (cw_sse));
       cw_sse &= 0xFFFF0000;
-      if (options.fpe & GFC_FPE_INVALID) cw_sse |= 1 << 7;
-      if (options.fpe & GFC_FPE_DENORMAL) cw_sse |= 1 << 8;
-      if (options.fpe & GFC_FPE_ZERO) cw_sse |= 1 << 9;
-      if (options.fpe & GFC_FPE_OVERFLOW) cw_sse |= 1 << 10;
-      if (options.fpe & GFC_FPE_UNDERFLOW) cw_sse |= 1 << 11;
-      if (options.fpe & GFC_FPE_PRECISION) cw_sse |= 1 << 12;
+      cw_sse |= (_FPU_MASK_IM | _FPU_MASK_DM | _FPU_MASK_ZM |  
_FPU_MASK_OM | _FPU_MASK_UM | _FPU_MASK_PM ) << 7;
+      if (options.fpe & GFC_FPE_INVALID) cw_sse &= ~(_FPU_MASK_IM <<  
7);
+      if (options.fpe & GFC_FPE_DENORMAL) cw_sse &= ~(_FPU_MASK_DM  
<< 7);
+      if (options.fpe & GFC_FPE_ZERO) cw_sse &= ~(_FPU_MASK_ZM << 7);
+      if (options.fpe & GFC_FPE_OVERFLOW) cw_sse &= ~(_FPU_MASK_OM  
<< 7);
+      if (options.fpe & GFC_FPE_UNDERFLOW) cw_sse &= ~(_FPU_MASK_UM  
<< 7);
+      if (options.fpe & GFC_FPE_PRECISION) cw_sse &= ~(_FPU_MASK_PM  
<< 7);
       asm volatile ("ldmxcsr %0" : : "m" (cw_sse));
     }
}



Summing up, you have to use a new version of cctools to correct a  
problem with Apple's linker, and you may have to apply a patch to  
correct an error in g++.  I assume that the gfortran problem has been  
solved in the mean time.

-erik

--
Erik Schnetter <schnetter@cct.lsu.edu>

My email is as private as my paper mail.  I therefore support encrypting
and signing email messages.  Get my PGP key from www.keyserver.net.

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