On 11/26/13 03:52, Bin.Cheng wrote:
On Tue, Nov 26, 2013 at 6:06 AM, Jeff Law <l...@redhat.com> wrote:
On 11/25/13 02:11, Bin.Cheng wrote:
Slightly tune to make iv cand choosing algorithm more accurate:
http://gcc.gnu.org/ml/gcc-patches/2013-11/msg01574.html
It would help if you had some sample codes where this patch was useful. I
can kind-of see what's going on, but I'm way too unfamiliar with the tree-IV
code to review this change.
Jeff, Thanks to your review.
As for example, consider below code on ARM/cortex-m4, (the code itself
may be non-sense):
[ snip ]
So in iv_ca_narrow you have the following change:
@@ -5705,13 +5709,15 @@ iv_ca_narrow (struct ivopts_data *data, struct iv_
if (!cp)
continue;
- if (!iv_ca_has_deps (ivs, cp))
- continue;
+ iv_ca_set_cp (data, ivs, use, cp);
+ acost = iv_ca_cost (ivs);
+ iv_ca_set_cp (data, ivs, use, old_cp);
- if (!cheaper_cost_pair (cp, new_cp))
- continue;
-
- new_cp = cp;
+ if (compare_costs (acost, best_cost) < 0)
+ {
+ best_cost = acost;
+ new_cp = cp;
+ }
}
}
else
You're set a new cost pair (cp), compute the cost into acost, the
restore the old cost pair (oldcp). Then you see if you got a cheaper
cost and if so, record the best cost and set new_cp.
What's the point behind restoring the old cost pair within the loop?
Can't you just restore it outside the loop if nothing cheaper was found?
AFAICT, the calls to iv_ca_has_deps aren't there for correctness, they
were there to prune out IVs that were not useful in this loop, correct?
Jeff
