Use case for the Math.fma() method?

[axkr]

What is the use case for the Math.fma() method?
Is it only available to improve accuracy for calculating (a * b) + c or can it sometimes also help speeding up computation?

For example in the vector dot product this operation can be modified to use the fma method:

• hipparchus/hipparchus-core/src/main/java/org/hipparchus/linear/ArrayRealVector.java

  Line 462 in 3ab8c5e

  dot += data[i] * vData[i];

I assume the fma method call is "in general" slower than calculating (a * b) + c directly?

[maisonobe]

Out of Hipparchus, I would say fma is probably intended for better accuracy (see for example Jean-Michel Muller books on floating point arithmetic and on elementary functions). It may be faster too than computing separately because fma is handled in hardware in many modern processors.
In Hippachus, we do have linearCombination which basically is a generalization of fma (and existed well before fma was added to Java Math). As it is more general and can take care of cancellations in dot products, I personally only use linearCombination.