Lots of condexec hacking today. Talked to wilson and runrau about using branch prediction info during if-conversion. Realized how suboptimial this first cut is likely to be. Need to spend some time looking through papers soon (trimaran and uiuc come to mind).
Hopefully I'll get the first cut on the non-trivial transformations finished enough to be functional tomorrow. If so, I'll have the pleasure of deleting 2100 lines of butt-ugly jump.c code.