Re: Installing GCC-4.x in userspace / Compiling Octave from CVS

[Sergei Steshenko]

--- "John W. Eaton" <address@hidden> wrote:

> On  1-Oct-2007, Sergei Steshenko wrote:
> 
> | Everything is compiled with gcc-4.2.1, which means it is compiled with
> | gfortran, but while building 'lapack' I also gave path to gcc-3.4.6 
> libraries
> | in order to satisfy the linker.
> | 
> | I think that was needed for built-in self-tests.
> | 
> | Here is the relevant part of 'make.inc' which comes with 'lapack':
> | 
> |     19  FORTRAN = 
> /maxtor5/sergei/AppsFromScratchWD/install/gcc-4.2.1/bin/gfortran
> |     20  OPTS     = -funroll-all-loops -O3
> |     21  DRVOPTS  = $(OPTS)
> |     22  NOOPT    =
> |     23  LOADER = 
> /maxtor5/sergei/AppsFromScratchWD/install/gcc-4.2.1/bin/gfortran
> |     24  LOADOPTS = -L 
> /maxtor5/sergei/AppsFromScratchWD/install/gcc-4.2.1/lib -L
> | 
> /maxtor5/sergei/AppsFromScratchWD/install/gcc-4.2.1/libexec/gcc/i686-pc-linux-gnu/4.2.1
>  -L
> | /maxtor5/sergei/AppsFromScratchWD/install/gcc-3.4.6/lib -lgfortran -l g2c#
> | .
> 
> If you compiled everything with gfortran, then this doesn't look right
> to me.  Does it also work if you remove the -lg2c and the -L option
> that includes the gcc-3.4.6 directory?  Where did the -lg2c come from?
> Compiling a simple program with "gfortran -v" doesn't show any -lg2c
> in the output, so I don't think compiling something with gfortran
> should require -lg2c.
> 
> jwe
> 

'lapack' linking fails without "-l g2c" - that is why I had to complicate 
matters in the
above manner in the first place.

I didn't dig into this, I just wanted to build the whole thing which I did.
And I'm not the only one who had this problem.

In the URL I posted :

http://archives.free.net.ph/message/20070902.203315.12c42b92.en.html

one can read this:

"
But now, there seems to be something screwy, possibly 
specific to cygwin. I downloaded and installed the 
high performance math libraries from AMD's website. 
When I built the examples using the original gcc 
compiler collection, it compiled perfectly and all 
tests using the high performance library passed. Now, 
doing exactly the same thing, make dies half way 
through at link time for one of the tests half way 
through make complaining that -lg2c doesn't exist. 
But it does, obviously, since make using gcc v 3.4.4 
found it. Where do I look for a fix to this? gcc v 
3.4.4 must have had something hard coded into it to 
tell it where this library is, because I didn't tell 
it where to look. I didn't even know this library 
existed the first time I built the AMD acml examples. 
My guess, from what I saw today, is that my current 
build of gcc v 4.2.1 doesn't have this information 
built in. I assume that is this information was in 
either environment variables or in the GnuMakefile, my 
attempt at compiling the examples using gcc v 4.2.1 
would have succeeded. 
".

I contacted the guy and gave him my solution.

I first resolved the problem and then started looking for a more elegant
solution - all I found was the above post.

I'd be glad to further test 'octave'; built-in tests pass - just tell me
what else to try.

At all, I am trying to always run 'make test' and its moral equivalents -
WRT 'octave' everything looks fine, i.e. I had no problems with the libraries
I've built.

By the way, I'm using 'atlas' library which uses 'lapack' - don't know
whether this matters or not regarding the issue we're discussing.

Regards,
  Sergei.

Applications From Scratch: http://appsfromscratch.berlios.de/


       
____________________________________________________________________________________
Boardwalk for $500? In 2007? Ha! Play Monopoly Here and Now (it's updated for 
today's economy) at Yahoo! Games.
http://get.games.yahoo.com/proddesc?gamekey=monopolyherenow