[llvm-commits] [llvm] r117174 - in /llvm/trunk: include/llvm/CodeGen/LinkAllCodegenComponents.h include/llvm/CodeGen/Passes.h lib/CodeGen/CMakeLists.txt lib/CodeGen/LiveIntervalUnion.cpp lib/CodeGen/LiveIntervalUnion.h lib/CodeGen/RegAllocBase.h lib/CodeGen/RegAllocBasic.cpp lib/CodeGen/SplitKit.h

Evan Cheng evan.cheng at apple.com
Tue Oct 26 17:22:13 PDT 2010


On Oct 26, 2010, at 3:59 PM, Jakob Stoklund Olesen wrote:

> 
> On Oct 26, 2010, at 3:46 PM, Evan Cheng wrote:
> 
>> 
>> On Oct 26, 2010, at 1:19 PM, Jakob Stoklund Olesen wrote:
> 
>>> If the LiveStacks are minimal, it is fine to calculate them on the fly, right? Or du you have a reason for wanting to compute them after allocation?
>> 
>> No particular reason. LiveStacks is a bit of a hack, I just think we should work towards removing the hack. Not having the allocator updating it is a good start.
> 
> What do you have in mind? It seems like an OK design to me.

1) LiveStacks is only used to track live time of spill slots. We would want to track non-spill ones as well.
2) It's using LiveIntervals. Doesn't it seem a bit heavy weight?

> 
> Obviously, StackSlotColoring needs to stop doing register allocation (!!), and it could probably share more code with the register allocator.

Like I said, a bit of a hack. :-)

Evan

> 
> /jakob
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20101026/d59fabaa/attachment.html>


More information about the llvm-commits mailing list