Use Span<Char>.Fill in String(Char, Int32) constructor

[jfd16]

This would most likely be a significant performance improvement, especially with the vectorized span fill introduced in #51365.

I couldn't figure out the best area label to add to this issue. If you have write-permissions please help me learn by adding exactly one area label.

Tagging subscribers to this area: @dotnet/area-system-runtime
See info in area-owners.md if you want to be subscribed.

Issue Details

---

This would most likely be a significant performance improvement, especially with the vectorized span fill introduced in #51365.

Author:	jfd16
Assignees:	-
Labels:	area-System.Runtime, up-for-grabs
Milestone:	-

[GrabYourPitchforks]

@jfd16 Thanks for the suggestion! I've marked it up for grabs so that anybody can tackle it if they're interested. Or feel free to submit a PR if you're feeling up to it.

For whoever submits a PR: please also submit benchmarks demonstrating that this is a perf win for common use cases. Maybe try lengths of 1, 2, 5, and 10 as typical use cases (to mimic code which emits padding); or 4, 8, 12, and 16 (to mimic code which emits spaces as a tab char replacement).

[ahydrax]

@GrabYourPitchforks I can look into it. Could you assign this to me?

[SteveDunn]

I was curious about this one, so I did the change and ran the benchmarks. The results were surprising though; it was slower!

Here's the results:

BenchmarkDotNet=v0.13.1, OS=Windows 10.0.22000
Intel Core i7-7700K CPU 4.20GHz (Kaby Lake), 1 CPU, 8 logical and 4 physical cores
.NET SDK=7.0.100-alpha.1.21467.19
  [Host]     : .NET 5.0.10 (5.0.1021.41214), X64 RyuJIT
  Job-KJFVRU : .NET 7.0.0 (42.42.42.42424), X64 RyuJIT
  Job-MZHKWO : .NET 7.0.0 (42.42.42.42424), X64 RyuJIT

IterationCount=3  LaunchCount=1  WarmupCount=3  

Method	Job	Toolchain	Size	Mean	Error	StdDev	Ratio	RatioSD	Gen 0	Allocated
Medium	Job-KJFVRU	main	1	6.512 ns	2.859 ns	0.1567 ns	1.00	0.00	0.0057	24 B
Medium	Job-MZHKWO	PR	1	13.985 ns	2.461 ns	0.1349 ns	2.15	0.07	0.0134	56 B
Medium	Job-KJFVRU	main	2	6.360 ns	1.075 ns	0.0589 ns	1.00	0.00	0.0076	32 B
Medium	Job-MZHKWO	PR	2	14.040 ns	3.786 ns	0.2075 ns	2.21	0.03	0.0153	64 B
Medium	Job-KJFVRU	main	4	6.149 ns	3.932 ns	0.2155 ns	1.00	0.00	0.0076	32 B
Medium	Job-MZHKWO	PR	4	14.773 ns	12.972 ns	0.7110 ns	2.40	0.10	0.0153	64 B
Medium	Job-KJFVRU	main	5	6.680 ns	1.670 ns	0.0916 ns	1.00	0.00	0.0076	32 B
Medium	Job-MZHKWO	PR	5	13.978 ns	12.688 ns	0.6955 ns	2.09	0.08	0.0172	72 B
Medium	Job-KJFVRU	main	8	7.167 ns	3.092 ns	0.1695 ns	1.00	0.00	0.0096	40 B
Medium	Job-MZHKWO	PR	8	15.836 ns	3.509 ns	0.1924 ns	2.21	0.05	0.0191	80 B
Medium	Job-KJFVRU	main	10	7.764 ns	5.893 ns	0.3230 ns	1.00	0.00	0.0115	48 B
Medium	Job-MZHKWO	PR	10	15.461 ns	9.475 ns	0.5194 ns	1.99	0.07	0.0229	96 B
Medium	Job-KJFVRU	main	12	7.629 ns	3.200 ns	0.1754 ns	1.00	0.00	0.0115	48 B
Medium	Job-MZHKWO	PR	12	16.313 ns	12.362 ns	0.6776 ns	2.14	0.12	0.0229	96 B
Medium	Job-KJFVRU	main	16	8.485 ns	4.583 ns	0.2512 ns	1.00	0.00	0.0134	56 B
Medium	Job-MZHKWO	PR	16	16.567 ns	14.305 ns	0.7841 ns	1.96	0.14	0.0268	112 B
Medium	Job-KJFVRU	main	32	10.847 ns	2.463 ns	0.1350 ns	1.00	0.00	0.0210	88 B
Medium	Job-MZHKWO	PR	32	17.667 ns	6.613 ns	0.3625 ns	1.63	0.03	0.0421	176 B
Medium	Job-KJFVRU	main	64	16.476 ns	6.452 ns	0.3537 ns	1.00	0.00	0.0363	152 B
Medium	Job-MZHKWO	PR	64	24.305 ns	5.120 ns	0.2806 ns	1.48	0.02	0.0727	304 B
Medium	Job-KJFVRU	main	128	29.586 ns	14.826 ns	0.8127 ns	1.00	0.00	0.0669	280 B
Medium	Job-MZHKWO	PR	128	35.916 ns	33.364 ns	1.8288 ns	1.22	0.09	0.1339	560 B
Medium	Job-KJFVRU	main	256	54.535 ns	22.270 ns	1.2207 ns	1.00	0.00	0.1281	536 B
Medium	Job-MZHKWO	PR	256	61.755 ns	20.355 ns	1.1157 ns	1.13	0.02	0.2563	1,072 B

[jkoritzinsky]

@SteveDunn if the implementation for the "PR" case is the code in your PR, I think there's some improvements we can do to make it perform much better (and possibly faster than main)

[ahydrax]

I've been playing around finding conditions where SpanHelpers.Fill could perform better than current implementation. It looks like adding nested if on string length adds a performance regression.
I stucked with this:
https://github.com/ahydrax/runtime/blob/string-ctor-span-fill/src/libraries/System.Private.CoreLib/src/System/String.cs#L329

[ahydrax]

My benchmark: https://github.com/ahydrax/string-ctor-span-fill

I would be thankful for any comments on that