... In this case I'm hitting the double spill error, but this is, I expect, tied in with the infinite loop the register spiller hits on some code.
Should be fixed now. Hopefully. - There were 2 bugs in the code WRT calculating life range of spilled regs and the ordering of registers was suboptimal.
It's probably a Leo or Melvin thing (unless anyone's see Angel Faus recently) to thump the register spilling code to have a fairly stupid fallback scheme (after X tries, or when we hit double spill) rather than trying to be optimal, but this one's going to need to get addressed.
Spill just all in one pass?
leo