[llvm-commits] [llvm] r84158 - in /llvm/trunk: lib/CodeGen/CodePlacementOpt.cpp test/CodeGen/X86/avoid-loop-align-2.ll test/CodeGen/X86/avoid-loop-align.ll

Dan Gohman gohman at apple.com
Thu Oct 15 10:52:20 PDT 2009


On Oct 14, 2009, at 10:33 PM, Evan Cheng wrote:

> 
> On Oct 14, 2009, at 5:36 PM, Dan Gohman wrote:
> 
>> Author: djg
>> Date: Wed Oct 14 19:36:22 2009
>> New Revision: 84158
>> 
>> URL: http://llvm.org/viewvc/llvm-project?rev=84158&view=rev
>> Log:
>> Make CodePlacementOpt align loops, rather than loop headers. The
>> header is just the entry block to the loop, and it needn't be at
>> the top of the loop in the code layout.
> 
> Thanks. This is aligning the destination of the back edge, right?

It aligns the destination of branches that branch to the top of
the loop, from a layout perspective.

> 
> Have you tested the performance and code size impact of this patch? 

Not exhaustively, but 253.perlbmk, which I happened to have been
running for unrelated reasons, saw an 8% speedup, though that's
dependent on it having gotten unlucky before.

Dan




More information about the llvm-commits mailing list