oxidecomputer/omicron#10887 will, imminently, enable ThinLTO in Omicron. we'll save a CD of memory per sled and TUFs will be smaller and things might be faster and everyone will rejoyce.
but recall that LLVM is not actively tested on illumos and ThinLTO is an opt-in feature. before calling it a done deal and shipping new software to customers, we ought to make sure the ThinLTO tests (afaik this is them, anyway) all seem ... reasonable. if they do not, we might want to not ship software with lto=thin. or we might find it's fine in practice and make some judgement call about R23 and later? but right now we don't know how bad it'll look, and we should know that.
I don't really know how to run LLVM tests so I have no better advice than "I see an llvm/test/TestRunner.sh that seems promising ..." - sorry to you, dear reader, and probably me if I'm trying to run these in a few weeks! I do plan to take notes in this issue once I've gotten somewhere with this.
oxidecomputer/omicron#10887 will, imminently, enable ThinLTO in Omicron. we'll save a CD of memory per sled and TUFs will be smaller and things might be faster and everyone will rejoyce.
but recall that LLVM is not actively tested on illumos and ThinLTO is an opt-in feature. before calling it a done deal and shipping new software to customers, we ought to make sure the ThinLTO tests (afaik this is them, anyway) all seem ... reasonable. if they do not, we might want to not ship software with
lto=thin. or we might find it's fine in practice and make some judgement call about R23 and later? but right now we don't know how bad it'll look, and we should know that.I don't really know how to run LLVM tests so I have no better advice than "I see an
llvm/test/TestRunner.shthat seems promising ..." - sorry to you, dear reader, and probably me if I'm trying to run these in a few weeks! I do plan to take notes in this issue once I've gotten somewhere with this.