Improve Character method performance

[lahma]

• add SkipLocalsInit attribute
• use pre-calculated answers for chars, generate using a test case using old methods
• use HexConverter borrowed from .NET sources which has superior check performance for common case

This does change the Character.HexValue to return 0xff instead of 0 like before. But as it's invalid value I would consider it a corner case and better handle the way .NET libraries do it?

Reasoning and more fine-grained results can be found here:

• Add SkipLocalsInit attribute esprima-dotnet#339
• Improve Character method performance esprima-dotnet#340

BenchmarkDotNet=v0.13.1, OS=Windows 10.0.22621
AMD Ryzen 9 5950X, 1 CPU, 32 logical and 16 physical cores
.NET SDK=7.0.100-rc.1.22431.12
  [Host]   : .NET 6.0.9 (6.0.922.41905), X64 RyuJIT
  ShortRun : .NET 6.0.9 (6.0.922.41905), X64 RyuJIT

Job=ShortRun  IterationCount=3  LaunchCount=1  
WarmupCount=3  

Parlot.Benchmarks.ParlotBenchmarks

Diff	Method	Mean	Error	Gen 0	Gen 1	Gen 2	Allocated
Old	CreateCompiledSmallParser	56,338.023 ns	1,968.3448 ns	2.0142	0.9766	-	34,640 B
New		55,003.465 ns (-2%)	1,201.7026 ns	2.0142 (0%)	0.9766 (0%)	-	34,640 B (0%)
Old	CreateCompiledExpressionParser	10,915.005 ns	1,398.7247 ns	0.3357	0.1678	0.0153	5,688 B
New		10,330.240 ns (-5%)	187.2058 ns	0.3357 (0%)	0.1678 (0%)	0.0153 (0%)	5,688 B (0%)
Old	CursorMatchHello	30.261 ns	1.6033 ns	0.0076	-	-	128 B
New		30.814 ns (+2%)	0.3755 ns	0.0076 (0%)	-	-	128 B (0%)
Old	CursorMatchGoodbye	32.938 ns	5.4822 ns	0.0076	-	-	128 B
New		30.666 ns (-7%)	7.1631 ns	0.0076 (0%)	-	-	128 B (0%)
Old	CursorMatchNone	24.658 ns	1.9902 ns	0.0076	-	-	128 B
New		25.442 ns (+3%)	0.3252 ns	0.0076 (0%)	-	-	128 B (0%)
Old	Lookup	32.622 ns	1.2212 ns	0.0076	-	-	128 B
New		30.357 ns (-7%)	0.6192 ns	0.0076 (0%)	-	-	128 B (0%)
Old	SkipWhiteSpace_1	27.005 ns	0.7584 ns	0.0076	-	-	128 B
New		22.965 ns (-15%)	1.5206 ns	0.0076 (0%)	-	-	128 B (0%)
Old	SkipWhiteSpace_10	40.532 ns	1.4653 ns	0.0076	-	-	128 B
New		29.865 ns (-26%)	12.5532 ns	0.0076 (0%)	-	-	128 B (0%)
Old	DecodeStringWithoutEscapes	6.904 ns	0.1219 ns	-	-	-	0 B
New		7.712 ns (+12%)	0.0565 ns	-	-	-	0 B
Old	DecodeStringWithEscapes	70.397 ns	0.6172 ns	0.0072	-	-	120 B
New		68.612 ns (-3%)	2.9249 ns	0.0072 (0%)	-	-	120 B (0%)
Old	ExpressionRawSmall	229.810 ns	18.4341 ns	0.0181	-	-	304 B
New		217.793 ns (-5%)	36.7999 ns	0.0181 (0%)	-	-	304 B (0%)
Old	ExpressionCompiledSmall	428.956 ns	21.7107 ns	0.0391	-	-	656 B
New		386.041 ns (-10%)	14.2656 ns	0.0391 (0%)	-	-	656 B (0%)
Old	ExpressionFluentSmall	583.263 ns	23.7512 ns	0.0391	-	-	656 B
New		530.177 ns (-9%)	18.2044 ns	0.0391 (0%)	-	-	656 B (0%)
Old	ExpressionRawBig	1,156.974 ns	14.1275 ns	0.0706	-	-	1,200 B
New		1,129.001 ns (-2%)	48.1586 ns	0.0706 (0%)	-	-	1,200 B (0%)
Old	ExpressionCompiledBig	2,376.195 ns	201.3357 ns	0.1717	-	-	2,888 B
New		2,124.498 ns (-11%)	38.6111 ns	0.1717 (0%)	-	-	2,888 B (0%)
Old	ExpressionFluentBig	3,237.802 ns	54.4850 ns	0.1717	-	-	2,888 B
New		2,928.574 ns (-10%)	230.3049 ns	0.1717 (0%)	-	-	2,888 B (0%)
Old	BigJson	92,982.792 ns	14,716.9926 ns	5.9814	1.7090	-	101,672 B
New		92,182.898 ns (-1%)	4,901.0568 ns	5.9814 (0%)	1.7090 (0%)	-	101,672 B (0%)
Old	BigJsonCompiled	73,206.250 ns	6,353.9512 ns	5.9814	1.9531	-	101,672 B
New		75,104.378 ns (+3%)	2,620.1724 ns	5.9814 (0%)	1.9531 (0%)	-	101,672 B (0%)
Old	DeepJson	64,070.520 ns	4,004.4522 ns	6.7139	1.5869	-	112,976 B
New		61,565.955 ns (-4%)	6,416.8976 ns	6.7139 (0%)	1.5869 (0%)	-	112,976 B (0%)
Old	DeepJsonCompiled	41,217.159 ns	2,123.3436 ns	6.7139	2.1362	-	112,976 B
New		41,193.115 ns (0%)	1,414.2750 ns	6.7139 (0%)	2.1362 (0%)	-	112,976 B (0%)
Old	LongJson	78,860.205 ns	4,896.0293 ns	8.0566	2.6855	-	135,512 B
New		78,844.609 ns (0%)	13,264.2833 ns	8.0566 (0%)	2.6855 (0%)	-	135,512 B (0%)
Old	LongJsonCompiled	65,387.073 ns	9,289.1072 ns	8.0566	2.6855	-	135,512 B
New		61,799.943 ns (-5%)	7,839.5657 ns	8.0566 (0%)	2.6855 (0%)	-	135,512 B (0%)
Old	WideJson	42,593.520 ns	1,019.5747 ns	2.4414	0.4272	-	41,584 B
New		42,646.167 ns (0%)	107.7185 ns	2.4414 (0%)	0.4272 (0%)	-	41,584 B (0%)
Old	WideJsonCompiled	35,827.002 ns	8,399.4675 ns	2.4414	0.4272	-	41,584 B
New		34,939.266 ns (-2%)	3,123.3980 ns	2.4414 (0%)	0.4272 (0%)	-	41,584 B (0%)

[lahma]

@sebastienros you no longer want to go fast? 🤔

[sebastienros]

Didn't notice this one ;)

[lahma]

If this lands as a release and consumed by Fluid, might be interesting to update the Fluid benchmarks 😉

[sebastienros]

I just published Fluid knowing that Parlot is on its way, I'll do a separate one with just Parlot updated. I didn't want to delay it. Checking if I need to add a net7.0 target, then will run the benchmarks.

[lahma]

Sound good, I'm not in a hurry, Parlot has been fast enough 🙂