TryAddWithoutValidation for multiple values could be more efficient

[pedrobsaila]

Fixes #64049
perfomance test dotnet/performance#4240

BenchmarkDotNet v0.13.13-nightly.20240311.145, Windows 11 (10.0.22631.3593/23H2/2023Update/SunValley3)
12th Gen Intel Core i7-12700H, 1 CPU, 20 logical and 14 physical cores
.NET SDK 9.0.100-preview.4.24267.66
  [Host]     : .NET 9.0.0 (9.0.24.26619), X64 RyuJIT AVX2
  Job-RIQMYG : .NET 9.0.0 (42.42.42.42424), X64 RyuJIT AVX2
  Job-UYKONU : .NET 9.0.0 (42.42.42.42424), X64 RyuJIT AVX2

PowerPlanMode=00000000-0000-0000-0000-000000000000  IterationTime=250ms  MaxIterationCount=20  
MinIterationCount=15  WarmupCount=1  

Method	Toolchain	data	Mean	Error	Ratio	Allocated	Alloc Ratio
AddHeaders	pr	Array	45.91 ns	1.350 ns	0.54	336 B	0.69
AddHeaders	main	Array	84.59 ns	1.230 ns	1.00	488 B	1.00
AddHeaders	pr	Hashset	86.72 ns	1.129 ns	0.97	496 B	1.00
AddHeaders	main	Hashset	89.18 ns	1.356 ns	1.00	496 B	1.00
AddHeaders	pr	List	45.95 ns	0.730 ns	0.54	336 B	0.68
AddHeaders	main	List	85.28 ns	0.809 ns	1.00	496 B	1.00

[MihaZupan]

Thanks.

@dotnet/ncl could I get a secondary pair of eyes on this one given it's based on my gist from the original issue https://gist.github.com/MihaZupan/800df7a324a66a8e814c3ec91f3d39b0

[rzikm]

LGTM if CI is green (modulo existing comments from MihaZupan). Maybe just add a few comments for those who are not that familiar with the codebase.

[pedrobsaila]

New results after adjusting performance test and TryAddWithoutValidation method :

Toolchain	data	Mean	Error	StdDev	Median	Min	Max	Ratio	Gen0	Allocated	Alloc Ratio
PR	Array	34.59 ns	0.360 ns	0.300 ns	34.54 ns	34.13 ns	35.23 ns	0.47	0.0102	128 B	0.46
master	Array	73.25 ns	0.744 ns	0.696 ns	73.29 ns	72.11 ns	74.30 ns	1.00	0.0222	280 B	1.00
PR	Hashset	73.48 ns	0.945 ns	0.884 ns	73.63 ns	72.30 ns	75.03 ns	0.95	0.0227	288 B	1.00
master	Hashset	77.19 ns	0.902 ns	0.799 ns	77.15 ns	76.24 ns	79.22 ns	1.00	0.0227	288 B	1.00
PR	List	36.77 ns	0.477 ns	0.446 ns	36.84 ns	35.91 ns	37.65 ns	0.48	0.0101	128 B	0.44
master	List	76.21 ns	0.468 ns	0.438 ns	76.24 ns	75.48 ns	77.11 ns	1.00	0.0228	288 B	1.00

[MihaZupan]

I imagine that the most common scenario for this API is a string[] with 1 value for a header that doesn't yet exist in the collection.
In that case, we can save an allocation from avoiding the boxing of the IEnumerator.

I'm not as convinced optimizing cases like an enumerable that isn't an IList is worth the extra maintainability cost in practice.

[pedrobsaila]

The new result after the last changes (it seems the slow path underperformed by rewriting it as foreach loop that calls unitary TryAddWithoutValidation) :

Toolchain	data	Mean	Error	StdDev	Median	Min	Max	Ratio	RatioSD	Gen0	Allocated	Alloc Ratio
PR	Array	52.23 ns	0.452 ns	0.353 ns	52.18 ns	51.69 ns	52.84 ns	0.68	0.01	0.0100	128 B	0.46
master	Array	76.35 ns	1.302 ns	1.218 ns	76.41 ns	74.83 ns	79.53 ns	1.00	0.02	0.0222	280 B	1.00
PR	Hashset	84.33 ns	3.491 ns	3.880 ns	82.85 ns	80.04 ns	92.59 ns	1.10	0.05	0.0229	288 B	1.00
master	Hashset	76.53 ns	0.741 ns	0.693 ns	76.30 ns	75.67 ns	77.71 ns	1.00	0.01	0.0229	288 B	1.00
PR	List	36.12 ns	0.289 ns	0.271 ns	36.05 ns	35.78 ns	36.68 ns	0.47	0.01	0.0102	128 B	0.44
master	List	76.49 ns	0.829 ns	0.692 ns	76.42 ns	75.49 ns	78.12 ns	1.00	0.01	0.0229	288 B	1.00

[pedrobsaila]

New benchmark results :

Toolchain	data	Mean	Error	StdDev	Median	Min	Max	Ratio	RatioSD	Gen0	Allocated	Alloc Ratio
PR	Array	51.95 ns	0.638 ns	0.498 ns	52.16 ns	51.05 ns	52.46 ns	0.68	0.01	0.0101	128 B	0.46
master	Array	75.87 ns	0.727 ns	0.680 ns	75.67 ns	74.51 ns	77.00 ns	1.00	0.01	0.0223	280 B	1.00
PR	Hashset	80.39 ns	0.579 ns	0.513 ns	80.30 ns	79.74 ns	81.31 ns	1.03	0.02	0.0227	288 B	1.00
master	Hashset	77.68 ns	1.145 ns	1.071 ns	77.89 ns	75.62 ns	79.58 ns	1.00	0.02	0.0227	288 B	1.00
PR	List	36.47 ns	0.373 ns	0.349 ns	36.43 ns	35.98 ns	37.20 ns	0.47	0.01	0.0101	128 B	0.44
master	List	77.63 ns	1.135 ns	1.006 ns	77.47 ns	75.94 ns	79.15 ns	1.00	0.02	0.0227	288 B	1.00

[MihaZupan]

/azp run runtime

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).