I've committed this patch to gomp4 branch.  It changes the auto partitioning 
logic to allocate the outermost loop to the outermost available partitioning. 
For instance, gang partitioning will be used for the outermost loop of a 
parallel region.   Innermost loops remain partitioned at the  innermost 
available level.

This means that if we run out of available partitions, we've parallelized the 
outer loop and the innermost loops, rather than just parallelized the inner loops.

nathan