[llvm-bugs] [Bug 30373] Regression(281162): clang miscompiles some of chrome's tests

via llvm-bugs llvm-bugs at lists.llvm.org
Mon Sep 19 01:26:11 PDT 2016


https://llvm.org/bugs/show_bug.cgi?id=30373

James Molloy <james.molloy at arm.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|---                         |FIXED

--- Comment #14 from James Molloy <james.molloy at arm.com> ---
Hi Nico,

Thanks so much for taking so much effort to find a reduced testcase! As I
suspected, r281162 wasn't actually the problem commit, an earlier one was and
code permutation at r281162 exposed it.

The issue was that when picking a candidate for CSE, I was combining metadata
correctly but not AND-ing IR flags which in this testcase caused an "exact"
flag to erroneously propagate to a right shift when it wasn't there originally.

This is fixed in r281889.

I'm working on removing this whole slew of code in SimplifyCFG by creating a
new GVN-sinking pass - it causes far too much code churn and the number of
defects it's raised (in canonical form expectations, SROA, ...) is massive.

Thanks again,

James

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20160919/d50a6078/attachment.html>


More information about the llvm-bugs mailing list