Optimizing reflection in C# via dynamic code generation

I have recently been spending a lot of time working on ComputeSharp, a .NET Standard 2.1 library written in C# 8.0 that lets you run code in parallel on the GPU through DX12 and dynamically generated HLSL compute shaders. That is a somewhat unintuitive description for a library that does something conceptually simple: it runs code in parallel in a manner similar to Parallel.