[PATCH] Propagate DAG node ordering during legalization and instruction selection

[Justin Holewinski]

On Tue, Mar 19, 2013 at 2:26 AM, Evan Cheng <evan.cheng at apple.com> wrote:

>
>
> Sent from my iPad
>
> On Mar 18, 2013, at 2:02 PM, Justin Holewinski <
> justin.holewinski at gmail.com> wrote:
>
> Compile-time impact is negligible for a release build on the unit tests.
>  There is about an 8% impact with assertions enabled.
>
>
> Unit tests are much too small for measuring compile time. 8% for assertion
> build is massive. Why are there such large discrepancy?
>

I've been trying to get measurements from LNT, but I'm getting too much
run-to-run variation.  A few benchmarks show significant changes (both
positive and negative), but the affected benchmarks are different from run
to run.  The vast majority of benchmarks show no change.  Do you have any
suggestions for getting concrete compile time data?


>
> Evan
>
>
> I'm not sure how CSE factors in here.  This propagation happens during
> legalization.  I agree that it should be addressed, but I see it as a
> separate issue.  Unless I'm missing some instance of CSE being performed
> during legalization?
>
>
> On Mon, Mar 18, 2013 at 4:04 PM, Jakob Stoklund Olesen <stoklund at 2pi.dk>wrote:
>
>>
>> On Mar 15, 2013, at 10:51 AM, Justin Holewinski <
>> justin.holewinski at gmail.com> wrote:
>>
>> Thanks Jakob!
>>
>> Modified patch attached.  Anyone adverse to this change?
>>
>>
>> +
>> +  // Propagate ordering
>> +  unsigned Order = DAG.GetOrdering(Op.getNode());
>> +  DAG.AssignOrdering(Lo.getNode(), Order);
>> +  DAG.AssignOrdering(Hi.getNode(), Order);
>>
>> Did you consider what happens when CSE kicks in? The new nodes may not be
>> new, they could have been CSE'ed with an existing expression.
>>
>> When that happens, the ordering should be the earlier of the two.
>>
>> I think it would make sense to teach the SDNodeOrdering class about this.
>>
>> /jakob
>>
>>
>
>
> --
>
> Thanks,
>
> Justin Holewinski
>
> _______________________________________________
>
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>
>


-- 

Thanks,

Justin Holewinski
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20130319/b3a32d83/attachment.html>