matmul intrinsic procedure

[Jan Stanisław Owoc <jsowoc at ucalgary dot ca>]

Hi,


I'm a student at the University of Calgary and I'm a big fan of Fortran 
for numerical methods calculations. I've been using ifort for some time 
and am thrilled to find out that there is a gcc replacement now.

Unfortunately, many of my programs deal with large matrices (I do 
finite-element analysis a lot), including multiplying them. To the best of 
my knowledge, there is an intrinsic method in the Fortran-95 standard which 
is called "matmul" and takes two matrices as arguments and returns the 
matrix product of them. Both ifort (Intel's compiler) and g95 (I am aware 
that this is not a GPL-based product) support this intrinsic method, but 
according to the gcc documentation, gfortran does not. It generates no 
error at compile time, but an assert fails when the program runs.

I use this method a lot and have implemented it before (in C, C++, and 
Fortran), and wouldn't mind helping out the community by submitting an 
implementation. My implementations have always been specific to my needs 
of the function, frequently just in-lining it.

If there was some way I could get an exact specification for the intrinsic 
procedure "matmul", as well as an implementation guideline (language, 
formatting, comments etc.), I wouldn't mind contributing this function. 
As I am not a software engineer, I cannot guarantee I would be able to 
come up with something exactly matching the specification, but I can give 
it a try.

Thank you very much for your time; gcc is an awesome compiler collection.


Jan Owoc