[PATCH] D74185: Revert the revert of vectorization commits

Tue Feb 11 08:36:04 PST 2020

ABataev added a comment.

In D74185#1868457 <https://reviews.llvm.org/D74185#1868457>, @ABataev wrote:

> In D74185#1868423 <https://reviews.llvm.org/D74185#1868423>, @george.karpenkov wrote:
>
> > @ABataev Thanks for looking into this! You are right: when I try the `opt` tool I also get the same result. Yet somehow when inside TensorFlow, rolling back this revision does fix the miscompile.
> >  Do you think you could give any advice what I can do to narrow it down? Maybe places to print before and after in vectorization passes?
>
>
> There is an option in LLVM, which allows to limit the number of passes and find the pass, which causes the miscompilation. Can't tell you this option right now, will tell you tomorrow.

Yep, found it. Here it is https://llvm.org/docs/OptBisect.html. You can try to bisect optimization passes and find the wrong one.

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D74185/new/

https://reviews.llvm.org/D74185