Teach alias analysis to reason about inline assembly nodes and
associated glued CopyToReg/CopyFromReg nodes.
This fixes PR9517.
I'm not sure I like the algorithm here. It seems like it would be more straightforward to handle the INLINEASM and all the glued nodes together, in a separate inner loop.
I don't think there's an comment anywhere explaining why you're putting chain operands into the visited list? Thinking about it a bit, I guess you're trying to avoid computing "useless" aliases? Does that actually work reliably, given the visitation order of the operands of a TokenFactor is sort of arbitrary?