Use safe Span.Slice loop pattern in Enumerable.SumSignedIntegersVectorized

[EgorBo]

Note

This PR was authored with assistance from GitHub Copilot (AI-generated content).

This PR replaces unsafe code with safe one. Also, slightly changes the code to check overflow status in the end so we don't pay for overflow check every iteration and assume overflow happens rarely and when it does - the cost of the exception is so big that it doesn't make sense to quit earlier.

Perf

Benchmark: @EgorBot request sweeping int[].Sum() and long[].Sum() over N ∈ {4, 8, 16, 32, 50, 100, 256, 1024, 4096, 16384, 65536, 1_000_000}. Full results: EgorBot/Benchmarks#155.

TL;DR: neutral on most sizes, with notable wins around the small-vectorized regime (N=32-256). No regressions.

Highlights (main time vs PR time, ratio = main/PR; values >1.00 mean PR is faster):

Hardware	Workload	N	main	PR	main/PR
Apple M4	SumInt	32	2.73 ns	1.82 ns	1.51x
Apple M4	SumInt	16	1.11 ns	1.06 ns	1.05x
Apple M4	SumInt	65536	7.89 us	7.69 us	1.03x
Intel Xeon	SumInt	32	4.53 ns	3.71 ns	1.22x
Intel Xeon	SumLong	32	5.63 ns	4.60 ns	1.22x
Intel Xeon	SumInt	50	6.04 ns	5.18 ns	1.17x
Intel Xeon	SumInt	256	21.14 ns	19.26 ns	1.10x
Intel Xeon	SumLong	1M	272.7 us	248.4 us	1.10x

All other size/workload combinations are within run-to-run noise (ratio in [0.99, 1.03]).

The wins at N=32-256 are mostly the deferred-overflow-check paying off (the test+branch is a meaningful fraction of cycles when the loop body is small and only runs a few times). The 1.51x at N=32 on M4 is amplified because Apple Silicon also benefits from the simpler control flow (one fewer branch per main-loop iter).

[Copilot]

Pull request overview

Copilot reviewed 5 out of 5 changed files in this pull request and generated no new comments.

[EgorBo]

@EgorBot -amd -intel -arm

using System.Linq;
using BenchmarkDotNet.Attributes;

public class SumBench
{
    // Cover edge cases (below Vector<T>.Count*4 dispatch threshold), main-loop boundaries
    // around Vector<T>.Count multiples on V128/V256/V512 hosts, and L1/L2/RAM-resident sizes.
    [Params(4, 8, 16, 32, 50, 100, 256, 1024, 4096, 16384, 65536, 1_000_000)]
    public int N;

    private int[] _ints = null!;
    private long[] _longs = null!;

    [GlobalSetup]
    public void Setup()
    {
        _ints = Enumerable.Range(0, N).ToArray();
        _longs = Enumerable.Range(0, N).Select(i => (long)i).ToArray();
    }

    [Benchmark]
    public int SumInt() => _ints.Sum();

    [Benchmark]
    public long SumLong() => _longs.Sum();
}

[Copilot]

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated no new comments.

[Copilot]

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated 1 comment.