On 04/28/2015 01:15 AM, Marek Polacek wrote:
On Mon, Apr 27, 2015 at 10:07:46PM -0600, Jeff Law wrote:
Thanks for the nice summary in the first part of the mail. Two typos:
+ /* Most of the time tree_swap_operands_p does what we want. But there's
Shouldn't that be "there are"?
+ cas
On Mon, Apr 27, 2015 at 10:07:46PM -0600, Jeff Law wrote:
Thanks for the nice summary in the first part of the mail. Two typos:
> + /* Most of the time tree_swap_operands_p does what we want. But there's
Shouldn't that be "there are"?
> + cases where we we know one operand is better for
Richi recently changed tree-ssa-dom.c::record_equality to use
tree_swap_operands_p to canonicalize the implied copy we record for
equality comparisons. This is a good thing.
However, there is a case where tree_swap_operands_p gives us operands in
an undesirable order for this routine. Spec
