NetFabric.Hyperlinq VS DistIL

> AVX instructions, which is implemented for quite a few LINQ methods

Are you sure? Any examples of such methods? And does AVX actually helps?

I don’t think that’s possible because IMO AVX and other SIMD can only help for dense inputs. The C# type is ReadOnlySpan, however ReadOnlySpan doesn’t implement IEnumerable and therefore incompatible with LINQ.

There’s even an alternative LINQ to workaround https://github.com/NetFabric/NetFabric.Hyperlinq but that thing is a third-party library most people aren’t using.

I know it C# it doesn't have to make heap allocations, here's a Linq-clone that mostly eliminates them: https://github.com/NetFabric/NetFabric.Hyperlinq

I'm the developer of one of the libraries and you may find in the benchmarks that it performs better than most others.

How is that relevant to the vast majority of the code targeted by LINQ?

The niche scenario you have outlined is partially covered by a recent System.Numerics.Tensors package update (even though I believe it would have been best if there was a community-maintained package with comparable quality that can be decoupled from .NET release cycle and trade-off support/compat guarantees in favour of more aggressive improvements in the future).

The goal of LINQ itself is to offer optimal codepaths when it can within the constraints of the current design (naturally, you could improve it significantly if not for backwards compatibility with the previous 15 or so years of .NET codebases). The argument that it's not good because it's not the tool to do BLAS is just nonsensical.

There is, however, an IL optimizer that can further vectorize certain common patterns and rewrite LINQ calls into open-coded loops: https://github.com/dubiousconst282/DistIL

Repo: https://github.com/dubiousconst282/DistIL

The lifetime problem should not be that complicated, because the class/lambda allocation is contained entirely within the method. There's an open issue on the runtime repo about object stack allocation that would on paper alleviate some of these issues, but it's quite stagnated. In my toy optimizer, I made a buggy pattern matching transform for "concretizing" lambdas invocations using very little code, but implementing something like this in the JIT can be quite challenging, based on my limited experience.