Get the latest tech news
Calculate Throughput with LLVM's Scheduling Model
Compiler, uArch, and a little bit of...jigsaw puzzle?
LLVM’s scheduling model – which we’d covered in several posts previously – is a huge database describing the performance characteristics of instructions in a specific processor. This formula is also what MCSchedModel::getReciprocalThroughput uses to calculate the inverse throughput of an instruction – modulo some differences like accounting for number of units of a resource, which we assume it to be one in our discussions. Now, I won’t spend hours wrestling with grammar, go all the way around to just write a blog post about something you can look up from code in a couple of minutes – the things we’ve discussed so far are just a prelude to the fun part:
Or read this on Hacker News