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

[Quentin Colombet]

Hi Renato,

Le 11 juin 2013 à 00:53, Renato Golin <renato.golin at linaro.org> a écrit :

> On 7 June 2013 01:00, Quentin Colombet <qcolombet at apple.com> wrote:
>> - For ARM backend expert: Confirm that the value is not required. Or put differently, what was the purpose checking if  a value was set?
> 
> Hi Quentin,
> 
> What I think that line is checking is to make sure the value from the operand is addressable, ie. there's something at the other end.
> 
> The comment above that condition is wrong, it should be below when it's actually checking for the alignment, and that put me off for a while, but now it's clear that this is just a sanity check, possibly introduced to fix a bug somewhere where the load was being created when the dest was not available, possibly breaking validation further down.
> 
> I don't claim to understand that part of the code, but would be good to know why having a non-addressable "Value" makes your case work. It feels as though the fix should be somewhere else...
Well, like I said, the information is destroyed by the stack coloring pass. We may be able to do something there but at least, Nadav confirms that it does what it is expected in that case.
I am not familiar with the use of value. What do you mean by the value may not be addressable?
If you can access it through single loads, what prevent you to do it through one multiple load?

Thanks,

Quentin

> 
> cheers,
> --renato
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20130611/afe8caa0/attachment.html>