Number precision in Quantities

[sebastian-peter]

Currently, a lot of Quantity numbers are converted to BigDecimal during operations like unit conversions or multiplications (cf. profiling of a SIMONA run with mobility sim). Since Quantities are constructed with double precision in PSDM, I don't think this is intended for our use case, especially because BigDecimals (which are based on BigInteger) are significantly slower than doubles during standard operations.

All of this is given narrowing is disabeld, which seems to affect execution speed as well.

[t-ober]

I see. Is the conversion taking place in the indriya package? Which would then make it hard to change right?

[sebastian-peter]

No, it happens inside DefaultNumberSystem, which we partly overwrite anyways

[sebastian-peter]

Albeit an being an old article, this gives an idea of how BigDecimal performes vs. double. We should do some testing on our own for Java 17 though, which could be pretty simple (like this).

[sebastian-peter]

This has become obsolete after switching to squants.