[llvm-commits] [llvm] r38517 - in /llvm/trunk: include/llvm/LinkAllPasses.h include/llvm/Transforms/Scalar.h lib/Transforms/Scalar/FastDSE.cpp

[Chris Lattner]

On Jul 11, 2007, at 9:37 AM, David A. Greene wrote:

> On Tuesday 10 July 2007 19:46, Owen Anderson wrote:
>> Author: resistor
>> Date: Tue Jul 10 19:46:18 2007
>> New Revision: 38517
>>
>> URL: http://llvm.org/viewvc/llvm-project?rev=38517&view=rev
>> Log:
>> Add FastDSE, a new algorithm for doing dead store elimination.  This
>> algorithm is not as accurate as the current DSE, but it only a  
>> linear scan
>> over each block, rather than quadratic.  Eventually (once it has been
>> improved somewhat), this will replace the current DSE.
>
> I hope by "replace" you mean, "will become the new default."  Some  
> of us
> are very interested in squeezing out every last possible bit of  
> performance
> and are willing to take a compile time hit to do it.  I don't want  
> to lose the
> current DSE implementation.

Right now, the current DSE pass does *very* little, and is extremely  
expensive.  The new FastDSE current misses some cases that the  
existing DSE pass gets, but Owen is working on completing it.  Once  
done, I'd be very surprised if DSE caught any cases that FastDSE  
doesn't.  If that's the case, we should be able to nuke the old DSE  
pass.  If it's not the case, we'll keep it.

-Chris