NegateValue vs ScaleValue

[SeungheonOh]

compares NegateValue and ScaleValue builtin.

I benchmarked two implementation of NegateValue. First, with (-1) * quantity other with negate quantity and two results differ significantly.

This is result is from having (-1) * quantity

This one is with negate quantity

where

negateQuantity :: Quantity -> Maybe Quantity
negateQuantity (UnsafeQuantity x) =
  if x == -170141183460469231731687303715884105728
  then Nothing
  else Just $ UnsafeQuantity (negate x)

I didn't know negate would perform so much better than (-1) *

[SeungheonOh]

This is with slower negateValue:

negateQuantity :: Quantity -> Maybe Quantity
negateQuantity (UnsafeQuantity x) = quantity (negate x)
{-# INLINE negateQuantity #-}

[zliu41]

Does any of these graphs compare two implementations that are otherwise exactly the same, except one uses negateQuantity, and the other uses scaleQuantity?

You seem to be comparing negateValue and scaleValue, but there might be other differences between them other than negation vs. multiplication?

[zliu41]

It would also be nice to verify whether 0 - x is indeed that much faster than (-1) * x (a simple criterion benchmark would do). Because according to their cost models, the differences aren't that big.

[zliu41]

In any case, I think we should still add scaleValue instead of negateValue. We can always optimize for special cases like 1 or -1 later.

[SeungheonOh]

Does any of these graphs compare two implementations that are otherwise exactly the same, except one uses negateQuantity, and the other uses scaleQuantity?

Yes!

[SeungheonOh]

We got all we needed from benchmark!