[llvm] r185393 - Remove floating point computations form SpillPlacement.cpp.

Stepan Dyatkovskiy stpworld at narod.ru
Tue Jul 2 11:17:33 PDT 2013


Hi Jakob, Benjamin
Currently I'm investigating this issue. Hope, will present first patch soon.
I just wanted to explain reasons, why it fires sometimes and sometimes not.
I suppose, nobody did the tests for case when stack part of byval 
structure is bigger than 4Kbytes. It seems, that your patch got llvm to 
emit stack part bigger than 4Kb and it causes llvm to generate wrong 
code (does it affect r0-r3 regs allocation?). See more details in PR16393.
So if you need to apply your patch ASAP, you may just try to avoid that 
case. Suppose it will make lencode green.

Anyway, me and Manman are working on it.

-Stepan.

Jakob Stoklund Olesen wrote:
>
> On Jul 2, 2013, at 5:48 AM, Renato Golin <renato.golin at linaro.org> wrote:
>
>> Hi Jakob, Benjamin,
>>
>> Just FYI, this commit broke Lencod on ARM again, though, if this is the same problem as before, it seems that Manman and Stepan already have a solution in mind.
>>
>> I'm not sure the current problem is, again, due to spilling the wrong offsets on the ByVal stack, but just to let you know, in case it doesn't, we might have to investigate this commit further as well. ;)
>
> Thanks, Renato.
>
> I’ve reverted the patch in r185447 while we investigate. I talked to Manman about the byval problem, and it does sound like a likely culprit. I’ll reapply the patch once the byval problem has been fixed.
>
> Thanks,
> /jakob
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>




More information about the llvm-commits mailing list