Far better than the unfortunately somewhat generic-rant title implies. It’s really about compute shaders, and the trouble with handling control flow on high-width SIMD systems. Turns out that despite nice shiny VM languages like SPIR-V and Webassembly going “we should really just ditch goto and have structured control everywhere, it’ll be easier to analyze”… that really might not be the case. Do we actually know what we’re doing, when it comes to writing compilers? Obviously not yet!