Significant Reverse(ReadOnlySpan<byte>) performance regression for certain sizes

[jkotas]

Repro

Run Reverse(ReadOnlySpan<byte>) micro-benchmark (https://github.com/dotnet/performance/blob/c599f4c65e60ffe74c4c959ee02d1dddf551968e/src/benchmarks/micro/libraries/System.Memory/Span.cs#L44) for array size 33.

Result

Current .NET 8 is about 1.6x slower compared to .NET 7 (AVX2 machine)

Introduced by #70944.

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

Issue Details

---

Run Reverse(ReadOnlySpan<byte>) micro-benchmark (https://github.com/dotnet/performance/blob/c599f4c65e60ffe74c4c959ee02d1dddf551968e/src/benchmarks/micro/libraries/System.Memory/Span.cs#L44) for array size 33.

Result:

Introduced by #70944. The

Author:	jkotas
Assignees:	-
Labels:	area-System.Memory, untriaged
Milestone:	-