From fec8ec1131ab0797e138a6c3c72ab5a441304e12 Mon Sep 17 00:00:00 2001 From: Marshall Lochbaum Date: Sat, 12 Feb 2022 11:27:11 -0500 Subject: =?UTF-8?q?Stop=20using=20=E2=89=A0=C2=A8=E2=88=98=E2=8A=94=20sinc?= =?UTF-8?q?e=20/=E2=81=BC=20does=20it=20now?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- implementation/compile/dynamic.md | 2 +- implementation/primitive/sort.md | 4 ++-- 2 files changed, 3 insertions(+), 3 deletions(-) (limited to 'implementation') diff --git a/implementation/compile/dynamic.md b/implementation/compile/dynamic.md index bd49a98f..a605545e 100644 --- a/implementation/compile/dynamic.md +++ b/implementation/compile/dynamic.md @@ -71,7 +71,7 @@ Like blocks, it can be valuable to optimize derived functions if they are run ma Compound arithmetic functions like `+´`, `` ⌈` ``, or `=⌜` are essential to array programming, and have fast SIMD implementations, so they need to be recognized wherever they are found. -In addition to these, there are patterns like `≠¨∘⊔` that can be implemented faster than their components, and bindings like `l⊸⊐` where a computation (here, a hash table) on the left argument can be saved. These can be handled by inspecting the function. However, it's more robust to convert it to a canonical form, so this possibility should also be considered. +In addition to these, there are patterns like ``∨`∘≠`` that can be implemented faster than their components, and bindings like `l⊸⊐` where a computation (here, a hash table) on the left argument can be saved. These can be handled by inspecting the function. However, it's more robust to convert it to a canonical form, so this possibility should also be considered. Tacit code can be converted to [SSA](https://en.wikipedia.org/wiki/Static_single_assignment_form) form very easily. To translate it into stack-based bytecode it would need a way to reuse values from the stack in multiple places; instructions to duplicate or extract a value from higher in the stack are an obvious candidate. Either of these forms is a natural step on the way to native compilation, and a bytecode representation would make it easier to optimize mixed tacit and explicit code—but it's easier to do the optimizations on SSA-form rather than stack-based code, so perhaps the right path is to convert both bytecode and derived functions to SSA. diff --git a/implementation/primitive/sort.md b/implementation/primitive/sort.md index 0a4c5211..36b9c916 100644 --- a/implementation/primitive/sort.md +++ b/implementation/primitive/sort.md @@ -36,7 +36,7 @@ A branchless binary search is adequate for Bins but in many cases—very small o ### Counting and bucket sort -Both counting and bucket sort are small-range algorithms that begin by counting the number of each possible value. Bucket sort, as used here, means that the counts are then used to place values in the appropriate position in the result in another pass. Counting sort does not read from the initial values again and instead reconstructs them from the counts. It might be written `(/≠¨∘⊔)⌾(-⟜min)` in BQN, with `≠¨∘⊔` as a single efficient operation. +Both counting and bucket sort are small-range algorithms that begin by counting the number of each possible value. Bucket sort, as used here, means that the counts are then used to place values in the appropriate position in the result in another pass. Counting sort does not read from the initial values again and instead reconstructs them from the counts. It might be written `(//⁼)⌾(-⟜min)` in BQN, relying on the extension of `/⁼` to unsorted arguments. Bucket sort can be used for Grade or sort-by (`⍋⊸⊏`), but counting sort only works for sorting itself. It's not-even-unstable: there's no connection between result values and the input values except that they are constructed to be equal. But with [fast Indices](replicate.md#non-booleans-to-indices), Counting sort is vastly more powerful, and is effective with a range four to eight times the argument length. This is large enough that it might pose a memory usage problem, but the memory use can be made arbitrarily low by partitioning. @@ -97,6 +97,6 @@ If `𝕨` is extremely small, use a vector binary search as described in "Sub-na It's cheap and sometimes worthwhile to trim `𝕨` down to the range of `𝕩`. After finding the range of `𝕩`, binary cut `𝕨` to a smaller list that contains the range. Stop when the middle element fits inside the range, and search each half of `𝕨` for the appropriate endpoint of the range. -If `𝕩` is small-range, then a lookup table method is possible. Check the length of `𝕨` because if it's too large then this method is slower! The approach is simply to create a table of the number of elements in `𝕨` with each value, then take a prefix sum. In BQN, ``𝕩⊏+`∾≠¨⊔𝕨``, assuming a minimum of 0. +If `𝕩` is small-range, then a lookup table method is possible. Check the length of `𝕨` because if it's too large then this method is slower—binary search doesn't have to hit every element! The approach is simply to create a table of the number of elements in `𝕨` with each value, then take a prefix sum. In BQN, ``𝕩⊏+`(1+⌈´𝕩)↑/⁼𝕨``, assuming a minimum of 0. [Partitioning](#partitioning) allows one pivot `t` from `𝕨` to be compared with all of `𝕩` at once. Although the comparison `t≤𝕩` can be vectorized, the overhead of partitioning still makes this method a little slower per-comparison than sequential binary search *when* `𝕨` *fits in L1 cache*. For larger `𝕨` (and randomly positioned `𝕩`) cache churn is a huge cost and partitioning can be many times faster. It should be performed recursively, switching to sequential binary search when `𝕨` is small enough. Unlike quicksort there is no difficulty in pivot selection: always take it from the middle of `𝕨` as in a normal binary search. However, there is a potential issue with memory. If `𝕩` is unbalanced with respect to `𝕨`, then the larger part can be nearly the whole length of `𝕩` (if it's all of `𝕩` partitioning isn't actually needed and it doesn't need to be saved). This can require close to `2⋆⁼≠𝕨` saved partitions of length `≠𝕩`, while the expected use would be a total length `≠𝕩`. -- cgit v1.2.3