[llvm-dev] DCE in the presence of control flow.

[Daniel Berlin via llvm-dev]

>
>
>
> Note: GCC is at least 3x faster at computing post-dom than LLVM
>
> Why?
>

It has a real edge structure, and so doing things like walking successors
and predecessors in a row is *really* fast and easily predictable cache
behavior.
LLVM has to walk use structures and look at stuff to walk predecessors,
which is both slow and often cache unfriendly :(

When you time it on larger cases, it comes out to about 600ms vs 200ms.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20160129/1fb4ba10/attachment.html>