Avoid overflow when calculating reverse indices

[michaelkirk-signal]

I don't know of any particularly bad consequences to the old way, but the proposed way avoids an overflow, which was triggering the analyzer.

(also it avoids a division and the asserts might prove helpful)