LLVM's MLIR Will Allow More Multi-Threading Within Compilers

One of the developers involved with the GCC efforts around more parallelization / multi-threading within the compiler itself has offered his skills to the LLVM team. Though as part of LLVM's growing embrace of the MLIR intermediate representation will also be better multi-threading within compilers like Clang.

Developer Nicholas Krause started a discussion about multi-threading compilers within the LLVM scope following his involvement on the GCC side.

In that discussion it was quickly pointed out by LLVM lead developer Chris Lattner that with the new MLIR representation that its pass manager is implicitly and automatically multi-threaded. So in turn moving more infrastructure to using this "Machine Learning IR" will help in allowing more compiler work to be multi-threaded in exploiting the potential of today's CPUs with increasing core counts.

Another interesting tid-bit brought up by Lattner in the discussion is that he recently gave a presentation on a proposed path forward for more LLVM core / Clang about possibly moving to MLIR over LLVM IR. He hopes in the days ahead to be able to publicly share that presentation in revealing more plans.

In any case, it is great seeing more discussions happening about multi-threading compilers and we'll keep monitoring to see what evolves.