PERF: Optimize Reader.TypeDeserializerCache.cs

[perf-cat]

Hi @bdlane, I've tried to make some performance optimizations to this repo. Following are the benchmark results before and after the changes:

Benchmarks before changes:

Method	NumberOfRecords	Mean	Error	StdDev	Median	Ratio	RatioSD	Rank	Gen 0	Gen 1	Gen 2	Allocated
XmlSerializer	1000	3.843 ms	0.2893 ms	0.8529 ms	3.636 ms	1.00	0.00	2	70.3125	35.1563	-	310.18 KB
LinqToXml	1000	5.057 ms	0.3136 ms	0.9247 ms	5.369 ms	1.39	0.40	4	93.7500	46.8750	-	527.9 KB
XmlReader	1000	2.755 ms	0.2076 ms	0.6120 ms	2.692 ms	0.75	0.22	1	62.5000	27.3438	-	277.52 KB
XmlGridReader_Properties	1000	4.110 ms	0.2972 ms	0.8762 ms	4.069 ms	1.12	0.35	3	74.2188	31.2500	-	332.53 KB
XmlGridReader_Constructor	1000	4.080 ms	0.2659 ms	0.7840 ms	4.170 ms	1.11	0.33	3	74.2188	31.2500	-	332.53 KB

Benchmarks after changes:

Method	NumberOfRecords	Mean	Error	StdDev	Median	Ratio	RatioSD	Rank	Gen 0	Gen 1	Gen 2	Allocated
XmlSerializer	1000	2.724 ms	0.0842 ms	0.2319 ms	2.632 ms	1.00	0.00	4	70.3125	31.2500	-	310.18 KB
LinqToXml	1000	3.018 ms	0.0578 ms	0.1351 ms	2.978 ms	1.10	0.10	5	93.7500	46.8750	-	527.9 KB
XmlReader	1000	1.682 ms	0.0266 ms	0.0222 ms	1.681 ms	0.57	0.06	1	58.5938	27.3438	-	277.54 KB
XmlGridReader_Properties	1000	2.494 ms	0.0489 ms	0.0523 ms	2.476 ms	0.85	0.07	2	78.1250	31.2500	-	332.49 KB
XmlGridReader_Constructor	1000	2.600 ms	0.0516 ms	0.1322 ms	2.564 ms	0.95	0.09	3	70.3125	27.3438	-	332.49 KB

As you can see the duration and allocations (Properties and Constructor benchmarks) have gone down. I've also made sure that all the unit tests pass following these changes as well. Could you please confirm whether or not you think my changes are valid?

Thank you!

[bdlane]

Hi @perf-cat

Thanks for the contribution.

I see you've made a number of similar contributions to several other repos in the past few days, all of which are small edits around performance, and are measured by BenchmarkDotNET. Just wondering what the deal is with this?

[perf-cat]

Hi @bdlane, thank you for your response!
You are absolutely right. This is actually part of our research project for automatically finding performance optimizations. We rely on BenchmarkDotNet and unit tests to validate our fixes. I hope this answers your question :)