Use span.Min/Max helpers to speed up SearchValues.Create

[MihaZupan]

No description provided.

[dotnet-policy-service]

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

[MihaZupan]

@EgorBot

using System.Buffers;
using BenchmarkDotNet.Attributes;
using BenchmarkDotNet.Running;

BenchmarkSwitcher.FromAssembly(typeof(Bench).Assembly).Run(args);

public class Bench
{
    // Standard Base64 alphabet (64 chars)
    private const string Base64Chars = "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/";

    [Benchmark]
    public SearchValues<char> Create_Base64Alphabet() => SearchValues.Create(Base64Chars);
}

[Copilot]

Pull request overview

This PR replaces manual min/max loops with ReadOnlySpan<T>.Min() / .Max() calls (from MemoryExtensions.MinMax) in two places, aiming to streamline and potentially speed up min/max computation used during SearchValues.Create range detection.

Changes:

• Update SearchValues.TryGetSingleRange<T> to compute minInclusive/maxInclusive via span Min()/Max() and drop the IMinMaxValue<T> constraint.
• Update regex BDD.SerializeToBytes (debug-only serializer) to validate positivity via Min() and compute the maximum via Max().

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 4 comments.

File	Description
src/libraries/System.Text.RegularExpressions/src/System/Text/RegularExpressions/Symbolic/BDD.cs	Switches debug-only serialization min/max computation to span Min()/Max() helpers.
src/libraries/System.Private.CoreLib/src/System/SearchValues/SearchValues.cs	Uses span Min()/Max() in TryGetSingleRange and removes IMinMaxValue<T> constraint.

[github-actions]

Copilot Code Review

Holistic Assessment

Motivation: Justified. The PR replaces manual scalar min/max loops with the built-in vectorized Span.Min()/Span.Max() helpers, which use SIMD-accelerated implementations for primitive types like byte, char, and long. This directly speeds up SearchValues.Create.

Approach: Correct and minimal. The refactoring uses existing, well-tested APIs (MemoryExtensions.Min/Max) that take the vectorized MinMaxInteger path for all involved types. The private method's IMinMaxValue<T> constraint is correctly relaxed since T.MaxValue/T.MinValue are no longer needed.

Summary: ✅ LGTM. Both changes are semantically equivalent to the original code, callers guarantee non-empty spans (so no InvalidOperationException risk), and the constraint relaxation is safe since TryGetSingleRange is private.

---

Detailed Findings

Detailed Findings

✅ Correctness — Empty span safety verified

Both call sites of TryGetSingleRange guard against empty/single-element spans before the call:

• Create(ReadOnlySpan<byte>): checks values.IsEmpty and values.Length == 1 first
• Create(ReadOnlySpan<char>): same guards

So values.Min() / values.Max() will never throw InvalidOperationException due to an empty span.

✅ Correctness — BDD.cs behavioral equivalence

The old code initialized m = 0 and used Math.Max(serialized[i], m), meaning m was implicitly ≥ 0. The new code calls serialized.AsSpan().Max() which returns the actual maximum. Since the Debug.Assert(serialized.AsSpan().Min() > 0) guarantees all elements are strictly positive, the two are equivalent. The two-pass approach (Min for assertion, then Max) is fine here since this method is [ExcludeFromCodeCoverage] and "used only to generate src data files."

✅ Constraint relaxation — Safe for private method

Removing IMinMaxValue<T> from the private TryGetSingleRange method is safe. The constraint was only needed for T.MaxValue/T.MinValue initialization of the loop variables, which are no longer present. No external callers are affected.

✅ Performance path — Vectorized implementations confirmed

For the types used at call sites (byte, char, long), the Min()/Max() methods dispatch to MinMaxInteger<T, ...> which uses SIMD-accelerated implementations, confirming the expected performance improvement over the previous scalar loops.

Note

This review was created by GitHub Copilot.

Generated by Code Review for issue #129809 · ● 19.2M · ◷

[MihaZupan]

Method	Toolchain	Mean	Error	Ratio
Create_Base64Bytes	PR #129809	109.5 ns	0.13 ns	1.00
Create_Base64Bytes	main	150.5 ns	0.11 ns	1.37
Create_Base64Chars	PR #129809	111.7 ns	0.18 ns	1.00
Create_Base64Chars	main	158.0 ns	0.20 ns	1.42

Not quite as good as #124667, but that makes sense since we're doing two passes for Min/Max instead