FlattenSequence/distance(from:to:) benchmarks. #71786
Conversation
|
@swift-ci please benchmark |
|
@swift-ci please smoke test |
| let minor = repeatElement(00000, count: inner) | ||
| let major = repeatElement(minor, count: outer) | ||
| let flattened: FlattenSequence = major.joined() |
There was a problem hiding this comment.
These could be at the top level or, alternatively, in a setUpFunction, so that creating them isn't part of the measurement.
There was a problem hiding this comment.
Are you sure you want to move these? It would be a lot of ceremony for 3 integers. Note that distance(from:to:) is called 2 * outer^2 * inner^2 * iterations times. I'm asking just in case you thought they were arrays.
▿ Swift.FlattenSequence<Swift.Repeated<Swift.Repeated<Swift.Int>>>
▿ _base: Swift.Repeated<Swift.Repeated<Swift.Int>>
- count: 32
▿ repeatedValue: Swift.Repeated<Swift.Int>
- count: 32
- repeatedValue: 0
There was a problem hiding this comment.
Nevermind. I found a solution I don't hate, then added a few sequential benchmarks for good measure. It's boilerplate-y but symmetrical. You could make it generic, but perhaps manual specialization is preferred in benchmarks?
| @@ -89,6 +89,7 @@ import Exclusivity | |||
| import ExistentialPerformance | |||
| import Fibonacci | |||
| import FindStringNaive | |||
| import FlattenDistanceFromTo | |||
There was a problem hiding this comment.
Please also add register(FlattenDistanceFromTo.benchmarks) at line 286 below.
With the PR as-is the benchmark runner still does not know about this addition.
There was a problem hiding this comment.
Sure. The script could use some love too:
The script will automatically:
- Edit main.swift by importing and registering* your new Swift module.
*The
cakeauto-registration is a lie.
1. Registered the new benchmarks in benchmark/utils/main.swift.
1. Parameterized benchmark dependencies, per code review.
1. Added some sequential tests with String as the inner collection.
|
I've updated the PR and added a few string-based tests. Here's some results using XCTest:
Swift 5.9After (#71648)So if these tests do not "run in a few milliseconds for N = 1", they might after applying the patch. |
1. Replaced occurrences of Repeated<T> with Array<T>.
|
I suppose you may write a better algorithm if all inner collections have the same number of elements, so I replaced occurrences of |
|
@swift-ci please benchmark |
1. The auto-generated [BenchmarkInfo] was given the wrong name.
|
I looked at the logs and saw that the auto-generated [BenchmarkInfo] was given the wrong name. I have changed the name in the most recent commit. |
|
@swift-ci please benchmark |
|
|
It looks like they'd still be a bit long-running even after the performance improvement. |
The benchmarks took too long to perform, so I reduced the size of each payload.
|
I have cut the payload dimensions into halves and quarters in the most recent commit. |
This commit addresses some trials and tribulations I encountered while working on (swiftlang#71786). It: 1. fixes the auto-registration regex 2. fixes the auto-generated array's name 3. generates the current year for the license header 4. generates some dashes for the license header
|
@swift-ci please benchmark |
|
✅ | Benchmark Check Report |
|
Two mysterious issues remaining in the benchmarks:
|
|
@swift-ci please smoke test |
This commit addresses some trials and tribulations I encountered while working on (swiftlang#71786). It: 1. fixes the auto-registration regex 2. fixes the auto-generated array's name 3. generates the current year for the license header 4. generates some dashes for the license header
I wrote some benchmarks for (#71648). I'm new to
/benchmark, but I read the README and ran the script.