Tolerated Comparison, Part 2
This post might be more difficult than our usual fare. You won’t find any interesting APL coding insights here—instead we’ll be focused on the tricky topic of floating-point error analysis. If that’s not your thing, feel free to skip this one! Although if you plan to use tolerated comparison in a real application, you really […]
Speed versus Accuracy: the User's Choice
At Dyalog we have long striven for both correctness and high performance in our implementation. However, our views on this matter have recently undergone an historic shift in paradigm which we are excited to share with our users. We now intend to provide the best experience to the user of Dyalog APL not by providing […]
Tolerated Comparison, Part 1
Warning: The formula for “tolerated” comparison in this post is not correct for actual floating-point numbers. Do not use it! The follow-up post discusses how to compute the required value precisely in the presence of floating-point error. Tolerant comparison is an important feature in making APL usable in the real world. But it’s also a […]
Expanding Bits in Shrinking Time
The chart above compares the performance of ∘.∧ in Dyalog versions 16.0 and 17.0. The improvement is a factor of 3-4 across the whole range (except at multiple-of-8 lengths, where 16.0 spikes up). That’s not easy to get, especially for a function like ∘.∧ which was already the target of a fair amount of special […]
Stackless Traversal
Enlist (∊) is twice as fast in Dyalog 16.0 as it was in Dyalog 15.0. Pretty much across the board: ∊⍳100 is not going to be any faster, but whenever the argument is a nested array and the simple arrays it contains are reasonably small, there are huge performance improvements. How did we achieve the […]
