[LLVMdev] PR5373

[Dan Gohman]

Hello Jakub,

This patch isn't right. The underlying problem that loop
unswitching was having was that it wasn't considering an
infinite-loop CFG pattern to be a "side effect". It was
assuming that the loop would exit eventually, and that
by eliminating the infinite loop, it thought it was just
accelerating that :-).

I went ahead and fixed this on trunk.

Dan

On Aug 11, 2010, at 12:07 AM, Jakub Staszak wrote:

> Hello,
> 
> Fixed patch attached. Can anyone test it?
> 
> Regards
> -- 
> Jakub Staszak
> <pr5373.patch>
>