Add BMP fast path to StringBuilder Rune overloads

[prozolic]

Add a fast path that calls StringBuilder.Append/Insert/Replace(char) when Rune.IsBmp holds.
Move the existing span path into a private method so that the stackalloc stays out of the fast path.
This follows the pattern used in #127939.

Benchmark

Method	Job	Toolchain	Mean	Error	StdDev	Median	Min	Max	Ratio	RatioSD	Allocated	Alloc Ratio
Append_Rune_Bmp	Job-QIDXWP	\main\corerun.exe	2.818 ns	0.0166 ns	0.0130 ns	2.819 ns	2.790 ns	2.844 ns	1.00	0.00	-	NA
Append_Rune_Bmp	Job-APEIBT	\pr\corerun.exe	2.378 ns	0.0672 ns	0.0773 ns	2.358 ns	2.266 ns	2.514 ns	0.84	0.03	-	NA
Append_Rune_Supplementary	Job-QIDXWP	\main\corerun.exe	3.299 ns	0.1453 ns	0.1674 ns	3.237 ns	3.075 ns	3.648 ns	1.00	0.00	-	NA
Append_Rune_Supplementary	Job-APEIBT	\pr\corerun.exe	3.092 ns	0.0594 ns	0.0526 ns	3.088 ns	3.012 ns	3.201 ns	0.94	0.05	-	NA
Insert_Rune_Bmp	Job-QIDXWP	\main\corerun.exe	5.885 ns	0.1921 ns	0.2212 ns	5.800 ns	5.629 ns	6.302 ns	1.00	0.00	-	NA
Insert_Rune_Bmp	Job-APEIBT	\pr\corerun.exe	5.465 ns	0.0894 ns	0.0747 ns	5.454 ns	5.372 ns	5.605 ns	0.93	0.04	-	NA
Insert_Rune_Supplementary	Job-QIDXWP	\main\corerun.exe	7.564 ns	0.1035 ns	0.0864 ns	7.530 ns	7.421 ns	7.704 ns	1.00	0.00	-	NA
Insert_Rune_Supplementary	Job-APEIBT	\pr\corerun.exe	7.802 ns	0.1499 ns	0.1539 ns	7.796 ns	7.499 ns	8.132 ns	1.03	0.02	-	NA
Replace_Rune_Bmp	Job-QIDXWP	\main\corerun.exe	19.057 ns	0.2855 ns	0.2384 ns	18.997 ns	18.794 ns	19.642 ns	1.00	0.00	-	NA
Replace_Rune_Bmp	Job-APEIBT	\pr\corerun.exe	15.674 ns	0.3009 ns	0.2955 ns	15.719 ns	15.130 ns	16.245 ns	0.82	0.02	-	NA
Replace_Rune_Supplementary	Job-QIDXWP	\main\corerun.exe	21.403 ns	0.3263 ns	0.2548 ns	21.416 ns	20.926 ns	21.709 ns	1.00	0.00	-	NA
Replace_Rune_Supplementary	Job-APEIBT	\pr\corerun.exe	21.162 ns	0.4119 ns	0.3440 ns	21.135 ns	20.591 ns	21.723 ns	0.99	0.02	-	NA

Benchmark source

    [BenchmarkCategory(Categories.Libraries, Categories.Runtime)]
    public class Perf_RuneForStringBuilder
    {
        private Rune _bmp;
        private Rune _supplementary;

        private StringBuilder _appendTarget;
        private StringBuilder _insertTarget;
        private StringBuilder _replaceBmp;
        private StringBuilder _replaceSupplementary;

        [GlobalSetup]
        public void Setup()
        {
            _bmp = new Rune('A');
            _supplementary = new Rune(0x1F600); // 😀

            _appendTarget = new StringBuilder(16);
            _insertTarget = new StringBuilder(16);

            var source = new string('x', 100);
            _replaceBmp = new StringBuilder(source + _bmp.ToString(), source.Length);
            _replaceSupplementary = new StringBuilder(source + _supplementary.ToString(), source.Length);
        }

        [Benchmark]
        public StringBuilder Append_Rune_Bmp()
        {
            _appendTarget.Clear();
            return _appendTarget.Append(_bmp);
        }

        [Benchmark]
        public StringBuilder Append_Rune_Supplementary()
        {
            _appendTarget.Clear();
            return _appendTarget.Append(_supplementary);
        }

        [Benchmark]
        public StringBuilder Insert_Rune_Bmp()
        {
            _insertTarget.Clear();
            return _insertTarget.Insert(0, _bmp);
        }

        [Benchmark]
        public StringBuilder Insert_Rune_Supplementary()
        {
            _insertTarget.Clear();
            return _insertTarget.Insert(0, _supplementary);
        }

        [Benchmark]
        public StringBuilder Replace_Rune_Bmp() => _replaceBmp.Replace(_bmp, _bmp);

        [Benchmark]
        public StringBuilder Replace_Rune_Supplementary() => _replaceSupplementary.Replace(_supplementary, _supplementary);

    }

[azure-pipelines]

Azure Pipelines:
Successfully started running 3 pipeline(s).
13 pipeline(s) were filtered out due to trigger conditions.
There may be pipelines that require an authorized user to comment /azp run to run.

[azure-pipelines]

Azure Pipelines:
Successfully started running 3 pipeline(s).
13 pipeline(s) were filtered out due to trigger conditions.
There may be pipelines that require an authorized user to comment /azp run to run.

[dotnet-policy-service]

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

[Copilot]

Pull request overview

This PR updates StringBuilder’s Rune overloads to add a BMP fast path that routes to the existing char-based APIs, with the non-BMP path moved into private helpers.

Changes:

• Add value.IsBmp / oldRune.IsBmp && newRune.IsBmp fast paths that call Append(char), Insert(char), and Replace(char, char, ...).
• Move the existing span + stackalloc conversion logic into private *Rune helper methods.
• Add Debug.Assert(!value.IsBmp) in the new slow-path helpers for Append / Insert.

Suppressed comments (2)

src/libraries/System.Private.CoreLib/src/System/Text/StringBuilder.cs:1397

• Same concern as AppendRune: this helper contains a tiny fixed-size stackalloc that the JIT may optimize and still inline, reintroducing stackalloc / GS-cookie overhead into the public fast path. Marking the helper NoInlining keeps the BMP path reliably lean.

        private unsafe StringBuilder InsertRune(int index, Rune value)

src/libraries/System.Private.CoreLib/src/System/Text/StringBuilder.cs:2357

• This helper uses small fixed-size stackallocs; if it gets inlined, the public entrypoint could inherit stackalloc-related prolog costs. To keep the fast-path benefit reliable, consider marking this helper NoInlining.

        private unsafe StringBuilder ReplaceRune(Rune oldRune, Rune newRune, int startIndex, int count)

[prozolic]

I'm not sure whether to apply [MethodImpl(MethodImplOptions.NoInlining)].
Should I add it to make it explicit that these private methods should not be inlined?

[Copilot]

🔵 Needs a closer look

Mark the three extracted span helpers NoInlining to keep stackalloc work out of the BMP fast paths.

Review details

Suppressed comments (3)

src/libraries/System.Private.CoreLib/src/System/Text/StringBuilder.cs:1070

• This split does not by itself guarantee that the stackalloc stays out of the BMP fast path: the JIT can inline a small helper after promoting a fixed-size localloc, which would put the span stack setup back into Append(Rune). The existing CoreLib span helpers use NoInlining for this exact reason (for example, Encoding.Internal.cs:225-230); please mark this helper as NoInlining, and apply the same treatment to the other two Rune helpers below.

        private unsafe StringBuilder AppendRune(Rune value)

src/libraries/System.Private.CoreLib/src/System/Text/StringBuilder.cs:1397

• The same inlining risk applies here: without NoInlining, the JIT may inline this fixed-size-stackalloc helper into Insert(int, Rune), defeating the stated goal of keeping stack setup out of the fast path. Please add the attribute here as well, matching the existing span-helper pattern in Encoding.Internal.cs.

        private unsafe StringBuilder InsertRune(int index, Rune value)

src/libraries/System.Private.CoreLib/src/System/Text/StringBuilder.cs:2357

• Without NoInlining, this small helper can be inlined and its two fixed-size stackallocs can become part of the caller's prolog, so extracting it does not reliably keep the slow path out of Replace(Rune, Rune, ...). Please mark this helper NoInlining too, consistent with the existing CoreLib span-helper pattern.

        private unsafe StringBuilder ReplaceRune(Rune oldRune, Rune newRune, int startIndex, int count)

• Files reviewed: 1/1 changed files
• Comments generated: 0 new
• Review effort level: Lite