amdgcn: Remove omp_gcn pass

amdgcn: Remove omp_gcn pass

This pass only had an optimization for obtaining team/thread numbers in it,
and that turns out to be invalid in the presence of nested parallel regions,
so we can simply delete the whole thing.

Of course, it would be nice to apply the optimization where it is valid, but
that will take more effort than I have to spend right now.

gcc/ChangeLog:

	* config/gcn/gcn-tree.c (execute_omp_gcn): Delete.
	(make_pass_omp_gcn): Delete.
	* config/gcn/t-gcn-hsa (PASSES_EXTRA): Delete.
	* config/gcn/gcn-passes.def: Removed.