[llvm-commits] [Review request] Instcombine: extractvalue from load --> load from gep

[John McCall]

On Nov 29, 2010, at 2:21 PM, Frits van Bommel wrote:

> On Mon, Nov 29, 2010 at 10:30 PM, Chris Lattner <clattner at apple.com> wrote:
>> I think that Dan is pointing out that it is unwise for an frontend to generate a ton of aggregate values, because that will cause fastisel to be defeated and lead to slower compile times.  However, despite Dan's objections, I think this is a perfectly reasonable xform for instcombine to do.
> 
> I agree that it's probably not a very good thing to do, but apparently
> they do occur "in the wild" even with clang.

Yes, we use them for pointer-to-function-members, and I'm planning to change
our complex generation to use them as well.  Sometimes subverting fastisel
can lead to faster compiler times by substantially simplifying code. :)

John.