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: [PATCH] Enhance ifcombine to recover non short circuit branches


On Fri, Oct 18, 2013 at 12:06 PM, Zhenqiang Chen
<zhenqiang.chen@linaro.org> wrote:
> On 18 October 2013 17:53, Richard Biener <richard.guenther@gmail.com> wrote:
>> On Fri, Oct 18, 2013 at 11:21 AM, Zhenqiang Chen
>> <zhenqiang.chen@linaro.org> wrote:
>>> On 18 October 2013 00:58, Jeff Law <law@redhat.com> wrote:
>>>> On 10/17/13 05:03, Richard Biener wrote:
>>>>>>>
>>>>>>> Is it OK for trunk?
>>>>>>
>>>>>>
>>>>>> I had a much simpler change which did basically the same from 4.7 (I
>>>>>> can update it if people think this is a better approach).
>>>>>
>>>>>
>>>>> I like that more (note you can now use is_gimple_condexpr as predicate
>>>>> for force_gimple_operand).
>>>>
>>>> The obvious question is whether or not Andrew's simpler change picks up as
>>>> many transformations as Zhenqiang's change.  If not are the things missed
>>>> important.
>>>>
>>>> Zhenqiang, can you do some testing of your change vs Andrew P.'s change?
>>>
>>> Here is a rough compare:
>>>
>>> 1) Andrew P.'s change can not handle ssa-ifcombine-ccmp-3.c (included
>>> in my patch). Root cause is that it does not skip "LABEL". The guard
>>> to do this opt should be the same the bb_has_overhead_p in my patch.
>>
>> I think we want a "proper" predicate in tree-cfg.c for this, like maybe
>> a subset of tree_forwarder_block_p or whatever it will end up looking
>> like (we need "effectively empty BB" elsewhere, for example in vectorization,
>> add a flag to allow a condition ending the BB and the predicate is done).
>>
>>> 2) Andrew P.'s change always generate TRUTH_AND_EXPR, which is not
>>> efficient for "||". e.g. For ssa-ifcombine-ccmp-6.c, it will generate
>>>
>>>   _3 = a_2(D) > 0;
>>>   _5 = b_4(D) > 0;
>>>   _6 = _3 | _5;
>>>   _9 = c_7(D) <= 0;
>>>   _10 = ~_6;
>>>   _11 = _9 & _10;
>>>   if (_11 == 0)
>>>
>>> With my patch, it will generate
>>>
>>>   _3 = a_2(D) > 0;
>>>   _5 = b_4(D) > 0;
>>>   _6 = _3 | _5;
>>>   _9 = c_7(D) > 0;
>>>   _10 = _6 | _9;
>>>   if (_10 != 0)
>>
>> But that seems like a missed simplification in predicate combining
>> which should be fixed more generally.
>>
>>> 3) The good thing of Andrew P.'s change is that "Inverse the order of
>>> the basic block walk" so it can do combine recursively.
>>>
>>> But I think we need some heuristic to control the number of ifs. Move
>>> too much compares from
>>> the inner_bb to outer_bb is not good.
>>
>> True, but that's what fold-const.c does, no?
>
> Based on current fold-const, we can not generate more than "two"
> compares in a basic block.

I think you are wrong as fold is invoked recursively on a && b && c && d.

> But if ifs can be combined recursively in ifcombine, we might generate
> many compares in a basic block.

Yes, we can end up with that for the other simplifications done in
ifcombine as well.  Eventually there needs to be a cost model for this
(use edge probabilities for example, so that if you have profile data
you never combine cold blocks).

Any takers?

Richard.

>>> 4) Another good thing of Andrew P.'s change is that it reuses some
>>> existing functions. So it looks much simple.
>>
>> Indeed - that's what I like about it.
>>
>>>>>
>>>>> With that we should be able to kill the fold-const.c transform?
>>>>
>>>> That would certainly be nice and an excellent follow-up for Zhenqiang.
>>>
>>> That's my final goal to "kill the fold-const.c transform". I think we
>>> may combine the two changes to make a "simple" and "good" patch.
>>
>> Thanks,
>> Richard.
>>
>>> Thanks!
>>> -Zhenqiang


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