[llvm-commits] [llvm] r159969 - /llvm/trunk/lib/Analysis/ScalarEvolution.cpp

Dan Gohman gohman at apple.com
Tue Jul 10 11:22:31 PDT 2012


On Jul 9, 2012, at 5:18 PM, Chris Lattner <clattner at apple.com> wrote:

> 
> On Jul 9, 2012, at 4:51 PM, Dan Gohman wrote:
> 
>> Author: djg
>> Date: Mon Jul  9 18:51:20 2012
>> New Revision: 159969
>> 
>> URL: http://llvm.org/viewvc/llvm-project?rev=159969&view=rev
>> Log:
>> Delete code for folding undefs in ScalarEvolution. It's invalid in
>> obscure ways, and it isn't actually important in the real world.
> 
> Can't it just fold to zero?

ScalarEvolution can't modify the IR, so it can't pick a definite value
for a particular undef use and guarantee that other code which also
looks at the IR will pick the same definite value for that use. I
can see room for disagreement about this, but I also claim that it
isn't that important in the real world.

Dan




More information about the llvm-commits mailing list