Improve TimeSpan.ToString/TryFormat throughput for default format

[stephentoub]

Improves the throughput of the null/"c"/"t"/"T" default format for TimeSpan.ToString/TryFormat, porting over the approach/specialization from Utf8Formatter.

Contributes to https://github.com/dotnet/corefx/issues/30612
cc: @jkotas, @danmosemsft, @ahsonkhan

Benchmark:

using System;
using BenchmarkDotNet.Attributes;
using BenchmarkDotNet.Attributes.Jobs;
using BenchmarkDotNet.Running;

[MemoryDiagnoser]
[InProcess]
public class Benchmark
{
    private static void Main() => BenchmarkRunner.Run<Benchmark>();

    private TimeSpan _simple = new TimeSpan(1, 2, 3);
    private TimeSpan _maxValue = new TimeSpan(long.MaxValue);
    private char[] _tmp = new char[100];

    [Benchmark] public string SimpleToString() => _simple.ToString();
    [Benchmark] public bool SimpleTryFormat() => _simple.TryFormat(_tmp, out int charsWritten);

    [Benchmark] public string MaxToString() => _maxValue.ToString();
    [Benchmark] public bool MaxTryFormat() => _maxValue.TryFormat(_tmp, out int charsWritten);
}

Before/After:

Benchmark	Before (ns)	After (ns)	Improvement
SimpleToString	126.8	44.76	2.83x
SimpleTryFormat	147.2	30.25	4.87x
MaxToString	391.9	111.56	3.51x
MaxTryFormat	282.7	106.23	2.66x

[ahsonkhan]

I am curious, why re-order these lines?

[stephentoub]

No great reason, just that since it's trivial to do either way, seemed better to only write the number of chars once they'd actually already been written. It doesn't really matter.

[ahsonkhan]

Debug.Assert?

[stephentoub]

We throw below. That seemed sufficient.

[ahsonkhan]

Yep. Makes sense.

[ahsonkhan]

We use mod here and division within the loop. Would using DivRem be faster?

[stephentoub]

Would using DivRem be faster?

It would require measuring. I didn't do any substantive changes to this G-formatting code, though, only minor cleanup.

[Spaier]

the second if (ticks < 0) is always false, isn't it?

if ticks is less than a zero then ticks = -ticks make it more than zero.

[stephentoub]

the second if (ticks < 0) is always false, isn't it?

No. -long.MinValue == long.MinValue.

[Spaier]

Thanks for an answer. Pretty interesting that the same applies to int.MinValue == -int.MinValue, but not to sbyte and short.

[jakobbotsch]

That is because all arithmetic operations on types smaller than int are automatically promoted to int beforehand, see here.