https://github.com/vgvassilev commented: I'd feel more comfortable if @lhames takes a look, too. Can you run some benchmarks and compare in-process to out-of-process to out-of-process with shared memory (same architecture)? https://github.com/llvm/llvm-project/pull/110418