Vectorize Guid.ToString

[EgorBo]

I tried to keep the implementation compact - around +66 lines to handle D, N, P and B formats (except X basically, which is rarely-used). For both x64 and Arm64.

public static IEnumerable<Guid> TestData()
{
    yield return new Guid("{483BC244-A481-4D29-986F-8C7D647AA20D}");
}

[Benchmark]
[ArgumentsSource(nameof(TestData))]
public string ToStringD(Guid guid) => guid.ToString(); // Default - "D"

[Benchmark]
[ArgumentsSource(nameof(TestData))]
public string ToStringN(Guid guid) => guid.ToString("N");

[Benchmark]
[ArgumentsSource(nameof(TestData))]
public string ToStringB(Guid guid) => guid.ToString("B");

[Benchmark]
[ArgumentsSource(nameof(TestData))]
public string ToStringP(Guid guid) => guid.ToString("P");

Method	Toolchain	guid	Mean	Error	StdDev	Ratio
ToStringD	\Core_Root\corerun.exe	483bc(...)aa20d [36]	13.26 ns	0.035 ns	0.033 ns	0.57
ToStringD	\Core_Root_base\corerun.exe	483bc(...)aa20d [36]	23.27 ns	0.081 ns	0.071 ns	1.00
ToStringN	\Core_Root\corerun.exe	483bc(...)aa20d [36]	12.69 ns	0.072 ns	0.056 ns	0.58
ToStringN	\Core_Root_base\corerun.exe	483bc(...)aa20d [36]	22.00 ns	0.048 ns	0.040 ns	1.00
ToStringB	\Core_Root\corerun.exe	483bc(...)aa20d [36]	14.45 ns	0.083 ns	0.065 ns	0.59
ToStringB	\Core_Root_base\corerun.exe	483bc(...)aa20d [36]	24.29 ns	0.167 ns	0.139 ns	1.00
ToStringP	\Core_Root\corerun.exe	483bc(...)aa20d [36]	15.11 ns	0.247 ns	0.231 ns	0.61
ToStringP	\Core_Root_base\corerun.exe	483bc(...)aa20d [36]	24.66 ns	0.065 ns	0.051 ns	1.00

The main motivation was to slightly speed up JSON serialization for Guid fields, e.g.:

record Entity(Guid Id, string Name);

string json = JsonSerializer.Serialize(entity, EntityJsonContext.Default.Entity);

Around 5% improvement here.

[ghost]

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.

[EgorBo]

@stephentoub @MihaZupan anything else?
I'll simplify this impl even further by sharing code with HexConverter (and, probably, Utf8Formatter.Guid) and will move hand-made Vector128 crossplat helpers to Vector128 itself (clean up) in my next PR

[ghost]

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

Issue Details

---

I tried to keep the implementation compact - around +66 lines to handle D, N, P and B formats (except X basically, which is rarely-used). For both x64 and Arm64.

public static IEnumerable<Guid> TestData()
{
    yield return new Guid("{483BC244-A481-4D29-986F-8C7D647AA20D}");
}

[Benchmark]
[ArgumentsSource(nameof(TestData))]
public string ToStringD(Guid guid) => guid.ToString(); // Default - "D"

[Benchmark]
[ArgumentsSource(nameof(TestData))]
public string ToStringN(Guid guid) => guid.ToString("N");

[Benchmark]
[ArgumentsSource(nameof(TestData))]
public string ToStringB(Guid guid) => guid.ToString("B");

[Benchmark]
[ArgumentsSource(nameof(TestData))]
public string ToStringP(Guid guid) => guid.ToString("P");

Method	Toolchain	guid	Mean	Error	StdDev	Ratio
ToStringD	\Core_Root\corerun.exe	483bc(...)aa20d [36]	13.26 ns	0.035 ns	0.033 ns	0.57
ToStringD	\Core_Root_base\corerun.exe	483bc(...)aa20d [36]	23.27 ns	0.081 ns	0.071 ns	1.00
ToStringN	\Core_Root\corerun.exe	483bc(...)aa20d [36]	12.69 ns	0.072 ns	0.056 ns	0.58
ToStringN	\Core_Root_base\corerun.exe	483bc(...)aa20d [36]	22.00 ns	0.048 ns	0.040 ns	1.00
ToStringB	\Core_Root\corerun.exe	483bc(...)aa20d [36]	14.45 ns	0.083 ns	0.065 ns	0.59
ToStringB	\Core_Root_base\corerun.exe	483bc(...)aa20d [36]	24.29 ns	0.167 ns	0.139 ns	1.00
ToStringP	\Core_Root\corerun.exe	483bc(...)aa20d [36]	15.11 ns	0.247 ns	0.231 ns	0.61
ToStringP	\Core_Root_base\corerun.exe	483bc(...)aa20d [36]	24.66 ns	0.065 ns	0.051 ns	1.00

The main motivation was to slightly speed up JSON serialization for Guid fields, e.g.:

record Entity(Guid Id, string Name);

string json = JsonSerializer.Serialize(entity, EntityJsonContext.Default.Entity);

Around 5% improvement here.

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

[MihaZupan]

will move hand-made Vector128 crossplat helpers to Vector128 itself (clean up) in my next PR

You can skip the Shuffle (#80963 already does it) -- I'll rebase once your PR is merged.

[EgorBo]