I've used it twice. Once when inverting large matrices, where the built-in dot product algorithm was inaccurate, and once when working with a very small process

I've used it twice. Once when inverting large matrices, where the built-in dot product algorithm was inaccurate, and once when working with a very small process noise with Kalman filters, with 32 bit floats due to platform restrictions, where the process noise could only increase uncertainty to a certain point after which there was underflow. Both times it has worked brilliantly, allowing a targeted increase in accuracy on only the components that needed it, allowing to choose the memory / computational hit desired, similar to using extra bits in certain temporaries when using fixed point.