[Bug tree-optimization/60172] [4.9/5/6/7 Regression] ARM performance regression from trunk@207239

[thopre01 at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=60172

Thomas Preud'homme <thopre01 at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |thopre01 at gcc dot gnu.org

--- Comment #29 from Thomas Preud'homme <thopre01 at gcc dot gnu.org> ---
(In reply to Richard Biener from comment #28)
> For GCC 6 we now have a DCE right before RTL expansion so what's the status
> of this bug for GCC6/7?

The RTL expansion looks better with:


20: r127=r132<<0x2

21: r137=r125+r127
22: r126=r131+r137  // r126=r131+r125+r127

29: r140=r131+r125
30: r141=r140+r127  // r141=r131+r125+r127
31: r142=r141+1000
32: [r142+20]=r123

So there should be a clear CSE opportunity to replace r141 by r126. Yet this
does not happen at cse1. Beside, gimple before expand is different so I cannot
really confirmed that the changes at expand actually fixed the issue.

I think the ticket can be closed and we'll reopen it if needed.

Best regards,

Thomas