While working on SSE support for C1 I noticed that the interpreter doesn't handle float return values properly after the adapter changes. It's easily noticeable if you run runThese with the math intrinsics disabled since they are the only built in natives that return floats as far as I can see or at least the only ones we test the return value of. The problem is that the interpreter conflates double and float returns which is fine if SSE is disabled but can't work one we begin using SSE.