rjmccall added a comment.

In https://reviews.llvm.org/D27680#629182, @ahatanak wrote:

> In https://reviews.llvm.org/D27680#628272, @rjmccall wrote:
> > Wouldn't it be simpler to just record an insertion point for the beginning 
> > of the current lexical scope and insert the lifetime.begin there instead of 
> > at the current IP?
> I'm not sure I understood your comment, but it seems to me that simply moving 
> the lifetime.start intrinsics to the current lexical scope wouldn't work in a 
> case like this:

I'm suggesting that, instead of moving instructions retroactively when you see 
a goto, you just insert the lifetime.start intrinsics at the start of the 
current lexical scope when you're emitting the variable.  That more exactly 
models the C language rule, I think, and shouldn't have any significant 
negative impact on optimization.


cfe-commits mailing list

Reply via email to