Clang has a bug processing inline ifuncs with intrinsics[1].
Since ifuncs can't be inlined anyway (they are always dispatched
via a function pointer that is determined based on the CPU
features present), nothing is gained by inlining them. Deinlining
therefore reduces compile time and works around the clang bug.
[1] https://bugs.llvm.org/show_bug.cgi?id=47691Closes#7358
