[llvm] r211103 - Fixed jump threading going to infinite loop.

[Rafael Espíndola]

LGTM, thanks!

On 19 June 2014 03:11, Dinesh Dwivedi <dinesh.d at samsung.com> wrote:
> How about this
>
>   // Remove unreachable blocks from function as they may result in infinite loop
>   // We do threading if we found something profitable. Jump threading a branch
>   // can create other opportunities. If these opportunities form a cycle i.e. if
>   // any jump treading is undoing previous threading in the path, then we will
>   // forever. We take care of this issue by not jump threading for back edges.
>   // This works for normal cases but not for unreachable blocks as they may have
>   // cycle with no back edge.
>   removeUnreachableBlocks(F);
>
> Regards
> Dinesh Dwivedi
>
> ------- Original Message -------
> Sender : Rafael Espíndola<rafael.espindola at gmail.com>
> Date   : Jun 18, 2014 17:44 (GMT+05:30)
> Title  : Re: Re: [llvm] r211103 - Fixed jump threading going to infinite loop.
>
> On 18 June 2014 02:23, Dinesh Dwivedi <dinesh.d at samsung.com> wrote:
>> Sorry, committed this patch too soon. How about changing comment as follows.
>
> No, that is fine. In llvm the code review process does continue after commit.
>
>>   // Remove unreachable blocks from function as they may result in infinite loop
>>   // We do threading if we found something profitable. But there may be two or
>>   // more conflicting cases where both are profitable and one results in undoing
>>   // other. This issue will not come for reachable blocks as we don't jump
>>   // thread back edges. Unreachable blocks having cycle with no back edge may
>>   // result in infinite loop.
>>   removeUnreachableBlocks(F);
>
> Close. I wouldn't call the optimizations conflicting since they don't
> exist at the same time. It is just that jump threading a branch can
> create another opportunity. If the blocks for a cycle, we would loop
> forever, except that we avoid that by looking at back edges. The
> problem with unreachable code is that there will be a loop, but not
> back edge.
>
> Cheers,
> Rafael