Thanks.  That patch resolved the issue with clear_with_quad and Cinnamon as
well.  

While I was looking at the source, I did notice that many of the functions
assumed that their inputs were valid.  I was speculating that clear_with_quad
may have calculated a zero or even negative number of elements and passed that
along to draw_pt_arrays, which would have passed a NULL array or negative index
to llvm.

You might want to apply that patch to the 17.1, 17.2 and 17.3 branches as well.
 While researching this bug I saw several posts from people having llvm
problems on Radeon hardware with Fedora and Ubuntu versions based on the 17.x
branches.