https://github.com/vgvassilev commented: Overall this looks quite promising to me. Have you run that patch on bigger workflows? Do we have some performance numbers to compare? I will run some tests on our infrastructure and report back. https://github.com/llvm/llvm-project/pull/76774