lukedays / excel-benchmarks Goto Github PK

You really threw down the gauntlet here :-) I must admit I was little disappointed with xlOil's performance so I added the xloil.FastArray type in xlOil 0.14.1:

@xlo.func
def SumXloilNumpyFast(x: xlo.FastArray) -> float:  # Or use xlo.Array(float, fast=True)
    return np.sum(x)

This narrows the performance gap considerably (to nearly zero)!

There is another trick which can be done with xlOil (and Keith's xlladdins) in some circumstances: rather than pass around giant arrays, use the cache functionality. Creating a cached Excel object using =xloRef(Test1Input) and passing that into your SumXloilNumpy gives a runtime of about 2x SUM. Even better the numpy array can be cached by defining:

@xlo.func
def ToPyObj(x) -> xlo.Cache:
    return x 

Passing this to SumXloilNumpy benchmarks at around 0.5 x SUM.