Improve the performance of BigInteger.{Max|Min}Magnitude/Abs & fix a edge case of TryParse

[karakasa]

Close #105324

The PR is to:

• Improve worst-case performance of BigInteger.{Max|Min}Magnitude by changing multiple comparison operators to a single CompareTo call
• Improve perf of BigInteger.Abs() by refactoring code and making the method branchless.
• Fix a bug where BigInteger.TryParse still throws instead of returning false.
  • PS: I wrote an OuterLoop test covering this but it would take 1.5~2GB which may unstablize the testing environment. Feel free to notify me if you think the test should be removed. The bug is on a very rare path.

Method	Toolchain	Length	Mean	Error	StdDev	Ratio	RatioSD
Abs	main	2	2.902 ns	0.0061 ns	0.0048 ns	1.00	0.00
Abs	pr	2	1.367 ns	0.0071 ns	0.0059 ns	0.47	0.00
MaxMagnitude_WorstPath	main	2	11.603 ns	0.0660 ns	0.0585 ns	1.00	0.00
MaxMagnitude_WorstPath	pr	2	3.553 ns	0.0092 ns	0.0081 ns	0.31	0.00
Abs	main	16	3.183 ns	0.0133 ns	0.0118 ns	1.00	0.00
Abs	pr	16	1.173 ns	0.0433 ns	0.0405 ns	0.37	0.01
MaxMagnitude_WorstPath	main	16	14.485 ns	0.3156 ns	0.5690 ns	1.00	0.00
MaxMagnitude_WorstPath	pr	16	6.758 ns	0.0230 ns	0.0204 ns	0.46	0.02
Abs	main	256	2.912 ns	0.0106 ns	0.0099 ns	1.00	0.00
Abs	pr	256	1.138 ns	0.0032 ns	0.0027 ns	0.39	0.00
MaxMagnitude_WorstPath	main	256	56.795 ns	0.2398 ns	0.2243 ns	1.00	0.00
MaxMagnitude_WorstPath	pr	256	35.709 ns	0.0808 ns	0.0675 ns	0.63	0.00
Abs	main	4096	3.007 ns	0.0355 ns	0.0277 ns	1.00	0.00
Abs	pr	4096	1.146 ns	0.0181 ns	0.0161 ns	0.38	0.01
MaxMagnitude_WorstPath	main	4096	547.978 ns	0.7053 ns	0.5889 ns	1.00	0.00
MaxMagnitude_WorstPath	pr	4096	479.951 ns	0.8376 ns	0.7425 ns	0.88	0.00

[LEI-Hongfaan]

Interesting.

        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        private BigInteger(int n, uint[]? rgu, [ConstantExpected] bool noLengthCheck)
        {
            if (noLengthCheck && (rgu is not null) && (rgu.Length > MaxLength))
            {
                ThrowHelper.ThrowOverflowException();
            }

            _sign = n;
            _bits = rgu;

            AssertValid();
        }

https://github.com/LEI-Hongfaan/dotnet-runtime/blob/f49abbbd68a222c0ce77f80aa19ecd53630505e7/src/libraries/System.Runtime.Numerics/src/System/Numerics/BigInteger.cs#L492

And

                    return new BigInteger(-1, buffer, noLengthCheck: true);

https://github.com/LEI-Hongfaan/dotnet-runtime/blob/f49abbbd68a222c0ce77f80aa19ecd53630505e7/src/libraries/System.Runtime.Numerics/src/System/Numerics/BigInteger.cs#L1675

Maybe we can have...

        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        internal BigInteger(int n, uint[]? rgu, [ConstantExpected] bool noLengthCheck = false)

[karakasa]

        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        internal BigInteger(int n, uint[]? rgu, [ConstantExpected] bool noLengthCheck = false)

I tend not to break any existing method, as long as it's not private. ConstantExpected is a good idea.

[huoyaoyuan]

I tend not to break any existing method, as long as it's not private.

There should not be difference between private and internal in this case. The internal accessors would be dedicated utility types for BigInteger, like the BigIntegerCalculator.

[jeffhandley]

/azp run runtime-libraries-coreclr outerloop

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[jeffhandley]

@tannergooding This is a more targeted change that seems worthwhile to bring in to .NET 11.