[cfe-commits] [PATCH] Optimize vec3 loads/stores

[John McCall]

On Jul 18, 2012, at 6:23 PM, Sean Silva wrote:
> I know vectors aren't part of C11/C++11, but doesn't this violate the
> memory model? In particular, doesn't it introduce spurious stores?
> 
> Forgive me if there's some context behind this that I'm missing which
> makes this concern irrelevant.

I think this is mostly obviated by the rule that the storage size of a vector
is always a power of 2.  To the extent that that's true, you're right that this
isn't a valid transformation.

John.