>>105560913
yeah, but it's also kind of complicated. hence why being able to arbitrarily use a GENERIC COMPILER FRAMEWORK that already has robust target for spir-v would be a huge boon.
https://github.com/KhronosGroup/SPIRV-Guide
unfortunately, llvm ain't it chief.
>opencl
based. i really wish somebody would knock some sense into vendors and force them to knock off all of this cuda and rocm bullshit. there's no reason we should have to use these dogshit vendor lock-in memes when we have perfectly good APIs for compute that are isomorphic with graphics pipelines.

>>105560977
i'd rather use chez scheme than either desu. codegen is easier and debugging it is easier. it's basically the one usecase i've found sexprs tolerable for.
debugging stuff that uses a c as an il is a royal pain in the ass.
debugging stuff on the gpu is already annoying enough without also having to work with machine-generated c and mapping it to another language on top of that.

>>105561401
>c's terrible memory model
has nothing to do with what makes c a mid choice for an il
oh no, arithmetic over addresses? in my abstract machine code? heavens to betsy, save me MIT! aiieee!!!