Read on Twitter
quite impressed with how well clang++ is coping with this 126 MB, 841000 line long C++ file i am feeding it, by which i mean it still hasn't eaten all of my RAM and that's already more than what i'd expect
i'm not entirely sure but i think there should be at least 1-2 million template function instantiations in that file
just in case you are wondering, almost the entire file consists of a single function. it is called "eval()". my editor is still trying to open it
i bet the linker will -love- it
i bet the linker will -love- it
you know what's really fascinating? first clang pegged a core and my laptop tried to take off, which is what i'd expect
but now clang is still pegging a core (which sits in C0) and my laptop is barely warm
is it -that- memory bottlenecked?
but now clang is still pegging a core (which sits in C0) and my laptop is barely warm
is it -that- memory bottlenecked?
seems to be deeply stuck calculating dominance frontier
i've never seen such a profoundly memory latency limited workload before. a lot of the time (around ½, i'd say?) clang spends >95% of time in C1. then goes back to >95% of time in C0. repeat. appears to be doing nothing but calculating dominance frontier in supposedly linear time
i left it to run overnight, it ended up allocating 14 GB and running out of memory. now i know what happens if i try to compile several million template instantiations in one function. who could have possibly predicted it?
