[SR-7869] String.init(decoding:sourceEncoding:) Suffers Lack of Speed #50404
Comments
|
Cc @milseman |
|
Comment by Caleb Kleveter (JIRA) I thought this might be relevant. The String(bytes:encoding: ) initializer is faster than converting to data and then a string. Using the same measuring as before, I got 0.333 sec. func testBytesToStringSpeed() {
let bytes: [UInt8] = [49, 50, 51, 52, 53, 54, 55, 56, 57, 48, 113, 119, 101, 114, 116, 121, 117, 105, 111, 112, 97, 115, 100, 102, 103, 104, 106, 107, 108, 122, 120, 99, 118, 98, 110, 109]
measure {
for _ in 0...1_000_000 {
guard let _ = String(bytes: bytes, encoding: .utf8) else {
XCTFail()
return
}
}
}
}Test Suite 'Selected tests' started at 2018-06-04 16:22:52.128
Test Suite 'CSVTests.xctest' started at 2018-06-04 16:22:52.130
Test Suite 'CSVTests' started at 2018-06-04 16:22:52.131
Test Case '-[CSVTests.CSVTests testBytesToStringSpeed]' started.
/Users/calebkleveter/Development/CSV/Tests/CSVTests/CSVTests.swift:117: Test Case '-[CSVTests.CSVTests testBytesToStringSpeed]' measured [Time, seconds] average: 0.333, relative standard deviation: 1.154%, values: [0.337403, 0.328976, 0.340650, 0.337263, 0.329202, 0.329619, 0.333868, 0.332411, 0.334991, 0.330322], performanceMetricID:com.apple.XCTPerformanceMetric_WallClockTime, baselineName: "", baselineAverage: , maxPercentRegression: 10.000%, maxPercentRelativeStandardDeviation: 10.000%, maxRegression: 0.100, maxStandardDeviation: 0.100
Test Case '-[CSVTests.CSVTests testBytesToStringSpeed]' passed (3.901 seconds).
Test Suite 'CSVTests' passed at 2018-06-04 16:22:56.033.
Executed 1 test, with 0 failures (0 unexpected) in 3.901 (3.902) seconds
Test Suite 'CSVTests.xctest' passed at 2018-06-04 16:22:56.033.
Executed 1 test, with 0 failures (0 unexpected) in 3.901 (3.903) seconds
Test Suite 'Selected tests' passed at 2018-06-04 16:22:56.034.
Executed 1 test, with 0 failures (0 unexpected) in 3.901 (3.905) seconds
Program ended with exit code: 0 |
|
Thanks for providing all of these |
|
Opened #17213 for the benchmarks. |
|
This PR speeds up the example code here: #17244 Unfortunately, we still have some poor performance when it comes to transcoding/parsing the UTF-8 into UTF-16. These won't be fixed at this moment, but we're planning on tackling it from another direction soon. I'll close this bug when that PR is merged; feel free to open any followup reports. |
|
@milseman You wanted to close this already, right? |
|
Yes, this is tracked in the UTF8Decode.swift benchmarks, and currently String.init(decoding:as:)is around 8x faster than the init from Foundation. |
|
In progress at #21959 edit: To clarify, this bug was originally that the stdlib's was slow. Then we made the stdlib's way way faster than Foundation's. That PR is now making Foundation's faster and more in line with the stdlib's by producing fast native Strings. All related general perf goodness. |
Environment
macOS 10.13.4 (17E202)
Xcode 9.4 (9F1027a)
Apple Swift version 4.1.1 (swiftlang-902.0.53 clang-902.0.39.2)
Target: x86_64-apple-darwin17.5.0
Additional Detail from JIRA
md5: 779e5c587cc70109119e81ca1b157c0d
Issue Description:
When converting an array of UInt8 to String, there are couple ways to do it. You can convert the bytes to data and use the data initializer:
Or decoding it as UTF-8:
Although the second initializer has much cleaner syntax, it suffers from lack of speed. I ran some benchmarks, doing the operation 1,000,000 (one million) times, and got the following results:
String(data:encoding) 0.902sec:
String(decoding:as: ) 18.517sec:
This is the test method I used:
The text was updated successfully, but these errors were encountered: