Optimize DateTime decomposition in CoreLib

[ts2do]

Add uses of DateTime.GetDate (renamed from GetDatePart) where appropriate.
Add internal DateTime.GetTime method to eliminate duplicate computations.

Note that the change to TimeZoneInfo.TransitionTimeToDateTime, which adds TimeOfDay directly in ticks precision, can't break anything because the TransitionTime constructor ensures that TimeOfDay doesn't have sub-millisecond time.

Here are the relevant microbenchmarks in dotnet/performance, of which the rows with the bold text being the build with the changes:

System.Tests.Perf_DateTime

Method	format	Mean	Error	StdDev	Median	Min	Max	Ratio	Gen 0	Gen 1	Gen 2	Allocated
ToString	?	292.30 ns	3.399 ns	2.838 ns	292.33 ns	286.58 ns	296.96 ns	0.99	0.0144	-	-	64 B
ToString	?	295.86 ns	2.520 ns	2.104 ns	296.08 ns	291.85 ns	300.16 ns	1.00	0.0143	-	-	64 B
ToString	G	293.25 ns	3.067 ns	2.869 ns	293.66 ns	288.34 ns	297.03 ns	0.99	0.0143	-	-	64 B
ToString	G	295.89 ns	3.830 ns	3.583 ns	295.99 ns	289.15 ns	303.13 ns	1.00	0.0143	-	-	64 B
ToString	o	66.41 ns	0.943 ns	0.882 ns	66.37 ns	64.78 ns	67.96 ns	0.72	0.0190	-	-	80 B
ToString	o	92.34 ns	0.826 ns	0.773 ns	92.28 ns	91.13 ns	93.41 ns	1.00	0.0191	-	-	80 B
ToString	r	52.63 ns	0.433 ns	0.384 ns	52.64 ns	51.90 ns	53.41 ns	0.92	0.0190	-	-	80 B
ToString	r	57.13 ns	0.514 ns	0.481 ns	56.93 ns	56.59 ns	58.31 ns	1.00	Base	0.0189	-	-
ToString	s	289.34 ns	3.105 ns	2.904 ns	289.99 ns	283.88 ns	292.82 ns	1.02	0.0150	-	-	64 B
ToString	s	284.92 ns	1.898 ns	1.683 ns	285.08 ns	281.30 ns	287.85 ns	1.00	0.0151	-	-	64 B

System.Tests.Perf_DateTimeOffset

Method	format	value	Mean	Error	StdDev	Median	Min	Max	Ratio	Gen 0	Gen 1	Gen 2	Allocated
ToString	?	?	666.74 ns	6.687 ns	6.255 ns	664.85 ns	658.04 ns	681.22 ns	1.01	0.0296	-	-	128 B
ToString	?	?	662.49 ns	5.139 ns	4.555 ns	663.86 ns	653.89 ns	670.28 ns	1.00	0.0294	-	-	128 B
ToString	?	12/30/2017 3:45:22 AM -08:00	658.87 ns	5.386 ns	5.038 ns	658.13 ns	651.09 ns	670.00 ns	0.98	0.0286	-	-	128 B
ToString	?	12/30/2017 3:45:22 AM -08:00	671.86 ns	5.994 ns	5.607 ns	673.26 ns	663.39 ns	683.24 ns	1.00	0.0294	-	-	128 B
ToString	G	?	308.34 ns	4.298 ns	3.810 ns	308.64 ns	300.68 ns	315.02 ns	1.01	0.0148	-	-	64 B
ToString	G	?	304.10 ns	2.380 ns	2.226 ns	304.11 ns	300.51 ns	307.61 ns	1.00	0.0150	-	-	64 B
ToString	o	?	76.22 ns	0.368 ns	0.308 ns	76.17 ns	75.78 ns	76.92 ns	0.69	0.0209	-	-	88 B
ToString	o	?	110.96 ns	1.202 ns	1.066 ns	110.72 ns	109.07 ns	113.46 ns	1.00	0.0206	-	-	88 B
ToString	r	?	63.59 ns	0.742 ns	0.620 ns	63.76 ns	62.62 ns	64.42 ns	0.97	0.0190	-	-	80 B
ToString	r	?	65.52 ns	0.940 ns	0.880 ns	65.33 ns	63.77 ns	67.45 ns	1.00	0.0190	-	-	80 B
ToString	s	?	304.85 ns	3.213 ns	3.006 ns	305.61 ns	298.44 ns	309.04 ns	1.00	0.0147	-	-	64 B
ToString	s	?	305.75 ns	3.682 ns	3.444 ns	304.98 ns	300.63 ns	310.69 ns	1.00	0.0148	-	-	64 B

System.Buffers.Text.Tests.Utf8FormatterTests.FormatterDateTimeOffsetNow

Method	Mean	Error	StdDev	Median	Min	Max	Ratio	Gen 0	Gen 1	Gen 2	Allocated
FormatterDateTimeOffsetNow	44.22 ns	0.473 ns	0.442 ns	44.35 ns	43.59 ns	45.01 ns	0.62	-	-	-	-
FormatterDateTimeOffsetNow	71.83 ns	0.662 ns	0.587 ns	71.79 ns	71.00 ns	73.06 ns	1.00	-	-	-	-

System.Text.Json.Tests.Perf_DateTimes

Method	Formatted	SkipValidation	Mean	Error	StdDev	Median	Min	Max	Ratio	Gen0	Gen1	Gen2	Allocated
WriteDateTimes	False	False	6.700 ms	0.0729 ms	0.0647 ms	6.700 ms	6.584 ms	6.801 ms	0.72	-	-	-	140 B
WriteDateTimes	False	False	9.313 ms	0.1085 ms	0.0962 ms	9.310 ms	9.159 ms	9.498 ms	1.00	-	-	-	120 B
WriteDateTimes	False	True	6.513 ms	0.0543 ms	0.0453 ms	6.529 ms	6.428 ms	6.578 ms	0.71	-	-	-	120 B
WriteDateTimes	False	True	9.149 ms	0.0635 ms	0.0563 ms	9.149 ms	9.025 ms	9.231 ms	1.00	-	-	-	120 B
WriteDateTimes	True	False	7.284 ms	0.0855 ms	0.0758 ms	7.270 ms	7.145 ms	7.394 ms	0.73	-	-	-	120 B
WriteDateTimes	True	False	9.926 ms	0.1158 ms	0.1083 ms	9.918 ms	9.735 ms	10.091 ms	1.00	-	-	-	120 B
WriteDateTimes	True	True	7.343 ms	0.0763 ms	0.0714 ms	7.328 ms	7.250 ms	7.479 ms	0.73	-	-	-	141 B
WriteDateTimes	True	True	10.120 ms	0.1492 ms	0.1323 ms	10.077 ms	9.938 ms	10.373 ms	1.00	-	-	-	120 B

[GrabYourPitchforks]

One minor question, but otherwise LGTM!

Please double-check that the 1-2% variance in the benchmarks is just noise and not something that requires further investigation.

[ts2do]

Based on what I'm seeing in additional runs and taking into account the nature of the changes, the 1-3% variance in either direction is noise.

[danmoseley]

Thanks for this optimization!