On certain indirect LLVM 'call' Instructions, you can get an assertion failure from the V9 back-end's graph-coloring register allocator. Essentially, the interference graph builder needs to handle the case where the indirect function pointer has already been assigned a register. I don't think this is a very common case, but bugpoint managed to trigger it for me at least once. The test case is "test/Regression/CodeGen/SparcV9/2004-08-22-CallNull.llx".
I think these patches take care of it. http://mail.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20040823/017460.html http://mail.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20040823/017459.html