matmul intrinsic procedure

[FX Coudert]

> I will get the CVS version of both 4.0 and 4.1, and compile it using gcc 
> 4.0.0 (which came with my operating system), and will report back on the 
> result.

Please note that there are binaries for linux available from 
gfortran.org. They are based on 4.1, and will save you the trouble of 
compiling gfortran yourself.

> The source code for "faim" is part of my research and I am not allowed to 
> send it out. I will, however, come up with equivalent code shortly.

I would advise testing it with an up-to-date before taking the time to 
reduce it. But if the bug still exists in 4.1, then we'll definetely 
need a reduced testcase.

FX