Tweak code gen for Guid parsing

[stephentoub]

Following-up on @GrabYourPitchforks' suggestion in #55792 (comment)

Method	Toolchain	Mean	Ratio	Code Size
D	\main\CoreRun.exe	38.43 ns	1.00	2,951 B
D	\pr\CoreRun.exe	37.52 ns	0.98	2,900 B
N	\main\CoreRun.exe	37.52 ns	1.00	2,302 B
N	\pr\CoreRun.exe	37.08 ns	0.99	2,249 B
B	\main\CoreRun.exe	39.30 ns	1.00	3,050 B
B	\pr\CoreRun.exe	38.07 ns	0.97	2,999 B
P	\main\CoreRun.exe	39.41 ns	1.00	3,055 B
P	\pr\CoreRun.exe	38.85 ns	0.99	3,004 B
InvalidD	\main\CoreRun.exe	42.55 ns	1.00	3,139 B
InvalidD	\pr\CoreRun.exe	41.01 ns	0.96	3,088 B
WeirdD	\main\CoreRun.exe	102.16 ns	1.00	3,418 B
WeirdD	\pr\CoreRun.exe	100.44 ns	0.98	3,367 B

[dotnet-issue-labeler]

I couldn't figure out the best area label to add to this PR. If you have write-permissions please help me learn by adding exactly one area label.

[ghost]

Tagging subscribers to this area: @dotnet/area-system-runtime
See info in area-owners.md if you want to be subscribed.

Issue Details

---

Following-up on @GrabYourPitchforks' suggestion in #55792 (comment)

Method	Toolchain	Mean	Ratio	Code Size
D	\main\CoreRun.exe	38.43 ns	1.00	2,951 B
D	\pr\CoreRun.exe	37.52 ns	0.98	2,900 B
N	\main\CoreRun.exe	37.52 ns	1.00	2,302 B
N	\pr\CoreRun.exe	37.08 ns	0.99	2,249 B
B	\main\CoreRun.exe	39.30 ns	1.00	3,050 B
B	\pr\CoreRun.exe	38.07 ns	0.97	2,999 B
P	\main\CoreRun.exe	39.41 ns	1.00	3,055 B
P	\pr\CoreRun.exe	38.85 ns	0.99	3,004 B
InvalidD	\main\CoreRun.exe	42.55 ns	1.00	3,139 B
InvalidD	\pr\CoreRun.exe	41.01 ns	0.96	3,088 B
WeirdD	\main\CoreRun.exe	102.16 ns	1.00	3,418 B
WeirdD	\pr\CoreRun.exe	100.44 ns	0.98	3,367 B

Author:	stephentoub
Assignees:	-
Labels:	area-System.Runtime
Milestone:	-

[GrabYourPitchforks]

Was hoping for a slightly better clock time measurement. But hey, I'll take the 50-byte codegen win. :)

[GrabYourPitchforks]

Here's another experiment which may be worth running. Changing the definition of DecodeByte to use nuint instead of char results in a 93-byte savings in the TryParse method (per sharplab). I'm not currently set up to measure how it might affect clock time.

private static byte DecodeByte(nuint ch1, nuint ch2, ref int invalidIfNegative)
{
    // TODO https://github.com/dotnet/runtime/issues/13464:
    // Replace the Unsafe.Add with HexConverter.FromChar once the bounds checks are eliminated.

    ReadOnlySpan<byte> lookup = CharToHexLookup;

    int h1 = -1;
    if (ch1 < (nuint)lookup.Length)
    {
        h1 = (sbyte)Unsafe.Add(ref MemoryMarshal.GetReference(lookup), (nint)ch1);
    }
    h1 <<= 4;

    int h2 = -1;
    if (ch2 < (nuint)lookup.Length)
    {
        h2 = (sbyte)Unsafe.Add(ref MemoryMarshal.GetReference(lookup), (nint)ch2);
    }

    int result = h1 | h2;
    invalidIfNegative |= result;
    return (byte)result;
}

[jeffhandley]

@GrabYourPitchforks This PR is assigned to you for follow-up/decision before the RC1 snap.

[stephentoub]

Here's another experiment

Thanks. Gives a small bump in throughput.