This simplifies the static analysis, since phi nodes must only consider the incoming edges, and never the instructions in the block
Open question: should a phi node be allowed to mention a variable defined by a phi-node in the same block?
x = phi ...
y = phi x ...
I think the answer should be NO, since there is no incoming edge, with which a phi-bound value in the current block can be associated. That breaks the "if I came form here then take this value, if I came from there then take the other value" semantics of the phi node.
This simplifies the static analysis, since phi nodes must only consider the incoming edges, and never the instructions in the block
Open question: should a phi node be allowed to mention a variable defined by a phi-node in the same block?
I think the answer should be NO, since there is no incoming edge, with which a phi-bound value in the current block can be associated. That breaks the "if I came form here then take this value, if I came from there then take the other value" semantics of the phi node.