This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [tree-ssa] Virtual operands in tree-tailcall


Hello,

> >while playing with new loop invariant motion pass,
> 
> 
> Just to note, you need to be careful with invariant motion. It's not 
> always good to hoist everything we can (as it can increase register 
> pressure).
> It's almost always good to move invariant memory loads/sink invariant 
> memory stores, though.
> I'm just talking about general loop invariants.

yes, I am aware of this.  In fact I would like to move only the
"obviously beneficial" invariants on tree level (so that we may get
rid of libcall notes), and leave the rest on rtl level pass that may
much better estimate things like costs of the computations and register
presure.

Zdenek


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]