[llvm-commits] [llvm] r164408 - in /llvm/trunk/lib/CodeGen: LiveStackAnalysis.cpp RegAllocGreedy.cpp StackSlotColoring.cpp VirtRegMap.cpp

[Evan Cheng]

On Sep 26, 2012, at 11:37 AM, Jakob Stoklund Olesen <stoklund at 2pi.dk> wrote:

> 
> On Sep 21, 2012, at 1:14 PM, Benjamin Kramer <benny.kra at gmail.com> wrote:
> 
>> 
>> On 21.09.2012, at 22:04, Evan Cheng <evan.cheng at apple.com> wrote:
>> 
>>> Author: evancheng
>>> Date: Fri Sep 21 15:04:28 2012
>>> New Revision: 164408
>>> 
>>> URL: http://llvm.org/viewvc/llvm-project?rev=164408&view=rev
>>> Log:
>>> Fix a significant recent(?) regression. StackSlotColoring no longer did anything
>>> because LiveStackAnalysis was not preserved by VirtRegWriter. This caused
>>> big stack usage regression in some cases.
>> 
>> Ugh, is there a way to add a test to the test-suite or regression tests so the regression can't come up again?
> 
> The stack slot coloring pass could assert if a spill slot doesn't have a live range.

If the livestack analysis is empty, then it could ever mean no spills or the analysis was cleared. 

I don't see an easy way to add a non-small test case for this. 

Evan

> 
> /jakob
>