[PATCH] ARM: Form paired load after stack coloring.

[Quentin Colombet]

Le 11 juin 2013 à 10:11, Renato Golin <renato.golin at linaro.org> a écrit :

> On 11 June 2013 17:21, Quentin Colombet <qcolombet at apple.com> wrote:
>> I am not familiar with the use of value. What do you mean by the value may not be addressable?
> 
> Me neither, I'm basing my answer on the comments around the code, so I could be totally wrong...
> 
>  
>> If you can access it through single loads, what prevent you to do it through one multiple load?
> 
> getValue() seems to return the base address of the memory operand, which by that time we know it had only one. Not having getValue() (or the return of that being zero) means, I assume, that LLVM doesn't know what the address of that operand is or where it can be reached. Maybe not...
The thing is that the virtual registers and the offsets are used for this purpose (the base virtual register has to be the same).

> 
> I just think that removing a validation on the memory operand will make a multiple load work. I'm not even sure what that validation means in the first place... :/
I do not know either and that is my point :). I wanted to get more information on why value is tested and never used.
To me it is not relevant to use the value here. 
> 
> Would be good for someone with more knowledge on that area to chime in, as I'm unsure that this is the correct fix for the problem...
Agree, have you someone in mind that I could ping about that?

> 
> cheers,
> --renato

Thanks,

-Quentin
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20130611/bc88fc67/attachment.html>