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

[SteveDunn]

Fix for Use Span.Fill in String(Char, Int32) constructor #57993

[dotnet-issue-labeler]

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

[ghost]

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

Issue Details

---

Fix for Use Span.Fill in String(Char, Int32) constructor #57993

Author:	SteveDunn
Assignees:	-
Labels:	area-System.Runtime, community-contribution
Milestone:	-

[adamsitnik]

I've benchmarked the initial version of this PR (it was always allocating an array), against @jkoritzinsky suggestion to use Span.Fill and @CyberAndrii suggestion to call SpanHelpers.Fill directly and what we have in main.

[Benchmark]
[Arguments(1)]
[Arguments(10)]
[Arguments(20)]
[Arguments(50)]
[Arguments(100)]
[Arguments(1000)]
public string CtorCharCount(int size) => new string('a', size);

Here are the results:

BenchmarkDotNet=v0.13.1.1616-nightly, OS=Windows 10.0.22000                        
AMD Ryzen Threadripper PRO 3945WX 12-Cores, 1 CPU, 24 logical and 12 physical cores

Method	Job	size	Mean	Ratio	Allocated
CtorCharCount	initial	1	19.487 ns	3.45	56 B
CtorCharCount	Jeremy	1	6.976 ns	1.24	24 B
CtorCharCount	Andrii	1	6.857 ns	1.21	24 B
CtorCharCount	main	1	5.650 ns	1.00	24 B
CtorCharCount	initial	10	16.537 ns	2.50	96 B
CtorCharCount	Jeremy	10	7.931 ns	1.20	48 B
CtorCharCount	Andrii	10	7.924 ns	1.20	48 B
CtorCharCount	main	10	6.638 ns	1.00	48 B
CtorCharCount	initial	20	16.553 ns	2.20	128 B
CtorCharCount	Jeremy	20	7.923 ns	1.03	64 B
CtorCharCount	Andrii	20	7.769 ns	1.03	64 B
CtorCharCount	main	20	7.575 ns	1.00	64 B
CtorCharCount	initial	50	22.185 ns	1.70	256 B
CtorCharCount	Jeremy	50	9.777 ns	0.75	128 B
CtorCharCount	Andrii	50	10.152 ns	0.78	128 B
CtorCharCount	main	50	13.052 ns	1.00	128 B
CtorCharCount	initial	100	28.198 ns	1.36	448 B
CtorCharCount	Jeremy	100	13.447 ns	0.65	224 B
CtorCharCount	Andrii	100	13.445 ns	0.65	224 B
CtorCharCount	main	100	20.748 ns	1.00	224 B
CtorCharCount	initial	1000	172.225 ns	0.97	4,048 B
CtorCharCount	Jeremy	1000	80.894 ns	0.46	2,024 B
CtorCharCount	Andrii	1000	79.703 ns	0.45	2,024 B
CtorCharCount	main	1000	177.664 ns	1.00	2,024 B

For small inputs (count < 20), we can see a 1 ns regression. For larger inputs, we can see an up to 2x improvement.

@tannergooding @GrabYourPitchforks What is the typical usage of new string(char, count)? Is it the right tradeoff to be made?

[tannergooding]

or small inputs (count < 20), we can see a 1 ns regression

The 3945WX has a base speed of 4.0GHz, so 1ns represents about 4 cycles. Even if < 20 is common, I think that's an acceptable tradeoff here as it represents less noise than a typical memory read can have.

[danmoseley]

Thanks for this improvement @SteveDunn !