[LLVMdev] How to avoid loopverify failures after replacing the backedge with an edge(latchBB to exitBB) in a looppass?

zhiyuan yang sjtu.yzy at gmail.com
Wed May 8 05:06:40 PDT 2013

Sorry for forgetting to reply all...

It works! Thank you very much!

But I also wonder how do you know this function will work while there are
no documents noticing that. I try learning LLVM by reading its code, but
soon feel lost in so many functions. Just like I have many tools, but don't
know which to use and what difference it could make. Is there some better
ways to learn LLVM?

On Wed, May 8, 2013 at 1:02 PM, Andrew Trick <atrick at apple.com> wrote:

> On May 7, 2013, at 8:26 PM, zhiyuan yang <sjtu.yzy at gmail.com> wrote:
> Hi,
> I am writing a loop pass to replace the backedge with an edge from latch
> to exit.
> Now I just replace the terminator of latch with another BranchInst, and
> the loop will not be a loop after my pass. However, it turns out a failure
> of loopverify after executing my pass:
> opt: ~/llvm/llvm-trunk/include/llvm/Analysis/LoopInfoImpl.h:297: void
> llvm::LoopBase<N, M>::verifyLoop() const [with BlockT = llvm::BasicBlock,
> LoopT = llvm::Loop]: Assertion `HasInsideLoopPreds && "Loop block has no
> in-loop predecessors!"' failed.
> Does this mean I should keep it still a loop after my pass? If so, how
> could I bypass it?
> You might need to call LPPassManager::deleteLoopFromQueue().
> -Andy
> --
> *Thank you && Best Regards,*
> *Zhiyuan Yang*
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev


*Thank you && Best Regards,*

*Zhiyuan Yang*
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20130508/db3fbbee/attachment.html>

More information about the llvm-dev mailing list