Tracking Down Floating Point Corruption
Recently a long-time customer of mine emailed me with a question.
They were having anomalies when computing floating point values.
Specifically, hardware would be read – values would be processed into a
float, and that value was then used as a metric.
Every now and then the metric would be off – it would jump by
several orders of magnitude. The next computation would be just fine,
though, exactly where it should have been.