[llvm-commits] [PATCH] Adding x32 ABI triple and separate StackSlotSize in MCAsmInfo

[Jim Grosbach]

On Jan 22, 2013, at 8:22 PM, Evan Cheng <evan.cheng at apple.com> wrote:

> 
> On Jan 22, 2013, at 2:35 PM, Eli Bendersky <eliben at google.com> wrote:
> 
>>>> 
>>>> Evan, I'm not sure what you're referring to here. Could you please clarify?
>>> 
>>> CalleeSaveStackSlotSize now has special meaning when it is zero, perhaps something like CalleeSaveStackSlotSizeOverride is more appropriate? I personally really don't like this, I think it optimizing for out-of-tree targets is not the right approach. It's making the code harder to understand.
>>> 
>> 
>> I'm fine with reverting to the originally proposed approach of simply
>> setting this value in the target MCAsmInfo constructors and losing the
>> "if 0 then use default" behavior.
> 
> That would be my preference but I'm certainly not trying to dictate it. It's your call.


Sorry for coming in late to the discussion.

I disagree with this and would very much prefer if we keep the "default to PointerSize" behavior. This isn't optimizing for out of tree targets (which Evan is quite right that isn't something we do), but rather putting the burden of special complexity only on those targets which exercise it. Targets which do a "normal" thing don't have to worry about it and get sensible behavior. Those that need something more complex have the flexibility to do so.

-Jim