[llvm-commits] Patch for review: Speeding up ScheduleDAG computations

David Greene dag at cray.com
Fri Feb 29 13:33:39 PST 2008

On Friday 29 February 2008 15:00, Evan Cheng wrote:

> Please resubmit it again. I'll evaluate the compile time penalty today.

Ok, thanks.

> There is something I am not grasping. I would think the problem can
> still occur even if the special tie breaker is removed.  Take any two
> nodes A, B, suppose all of the conditions evaluates to false the
> function will returns false regardless if it's comparing A <= B or B
> <= A, right? I would think the right is just to change the default from
> return false;
> to
> return left->NodeNum < right->NodeNum;
> This forces a strict ordering between two nodes, no?

Yes, I think that's right.  I glossed over that bit of code.


More information about the llvm-commits mailing list