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] Fix PR 33512 Folding of x & ((~x) | y) into x & y on the tree level


On Mon, Apr 23, 2012 at 2:41 AM, Richard Guenther
<richard.guenther@gmail.com> wrote:
> On Sat, Apr 21, 2012 at 6:06 AM, Andrew Pinski
> <andrew.pinski@caviumnetworks.com> wrote:
>> This time with the patch and describing what the bug was. ÂThe problem
>> was defcodefor_name does not always set arg1 and arg2. ÂThis fixes it
>> so it is always set to NULL if they don't exist.
>
> Ok with ...
>
> + Âif (code1 == SSA_NAME)
> + Â Â{
> + Â Â Âdef = SSA_NAME_DEF_STMT (name);
> +
> + Â Â Âif (def && is_gimple_assign (def)
> + Â Â Â Â && can_propagate_from (def))
> + Â Â Â {
> + Â Â Â Â code1 = gimple_assign_rhs_code (def);
> + Â Â Â Â arg11 = gimple_assign_rhs1 (def);
> + Â Â Â Â Âarg21 = gimple_assign_rhs2 (def);
> + Â Â Â Â Âarg31 = gimple_assign_rhs2 (def);
> + Â Â Â }
> + Â Â}
>
> ... recursing here instead.


Recursing how?  Or do you mean when code1 is a SSA_NAME do a recursive
call so that we can get some simple copy-prop happening?
The current code does not do that though.

Thanks,
Andrew


>
> Thanks,
> Richard.
>
>
>> Thanks,
>> Andrew
>>
>> On Fri, Apr 20, 2012 at 9:05 PM, Andrew Pinski
>> <andrew.pinski@caviumnetworks.com> wrote:
>>> On Thu, Jan 19, 2012 at 3:13 AM, Richard Guenther
>>> <richard.guenther@gmail.com> wrote:
>>>> On Thu, Jan 19, 2012 at 10:00 AM, Andrew Pinski
>>>> <andrew.pinski@caviumnetworks.com> wrote:
>>>>> On Tue, Jan 17, 2012 at 1:38 AM, Richard Guenther
>>>>> <richard.guenther@gmail.com> wrote:
>>>>>> On Tue, Jan 17, 2012 at 8:06 AM, Andrew Pinski
>>>>>> <andrew.pinski@caviumnetworks.com> wrote:
>>>>>>> Hi,
>>>>>>> ÂThis adds the folding of x & ((~x) | y)) into x & y on the tree
>>>>>>> level via fold-const.c
>>>>>>> There is already partly done on the RTL level but it would be a good
>>>>>>> thing for the tree level also.
>>>>>>>
>>>>>>>
>>>>>>> OK for 4.8 (yes I know we have not branched yet but I thought I would
>>>>>>> send it out so I don't forget about it)?
>>>>>>> Bootstrapped and tested on x86_64-linux-gnu with no regressions.
>>>>>>
>>>>>> Can you instead patch tree-ssa-forwprop.c:simplify_bitwise_binary?
>>>>>
>>>>> Yes and here is a new patch which also adds optimizing x | ((~x) & y)) to x | y.
>>>>> Also it adds the optimizing x & (x | y) to x and x | (x & y) to x to
>>>>> tree-ssa-forwprop.c:simplify_bitwise_binary
>>>>> since it was an easy extension on top of the ~x case (well I
>>>>> implemented the one without the ~ first). ÂI did not remove those
>>>>> folding from fold-const.c though.
>>>>>
>>>>> Also I was thinking maybe this belongs in reassociate though I don't
>>>>> see how to do it.
>>>>
>>>> I still have plans to create that piecewise gimple_fold (see my proposal
>>>> from early last year) that would be the container for this kind of pattern
>>>> matching. ÂIt would then be usable from reassoc as well (but reassoc
>>>> has the issue of only collecting one kind of op, so its simplification
>>>> wouldn't trigger reliably on these).
>>>>
>>>> http://gcc.gnu.org/ml/gcc-patches/2011-03/msg01099.html
>>>>
>>>>> OK for 4.8, once in stage 1? Again bootstrapped and tested on
>>>>> x86_64-linux-gnu with no regressions.
>>>>
>>>> Ok.
>>>
>>> Here is an updated patch which fixes a bug which I found while doing
>>> the treecombine branch. ÂI rewrote defcodefor_name so more than
>>> SSA_NAMEs can be passed to it.
>>>
>>> OK? Bootstrapped and tested on x86_64-linux-gnu with no regressions.
>>>
>>> Thanks,
>>> Andrew Pinski
>>>
>>> ChangeLog:
>>>
>>> ChangeLog:
>>> * tree-ssa-forwprop.c (defcodefor_name): New function.
>>> (simplify_bitwise_binary): Use defcodefor_name instead of manually
>>> Simplify "( X | Y) & X" to X and "( X & Y) | X" to X.
>>> Simplify "(~X | Y) & X" to "X & Y" and
>>> "(~X & Y) | X" to "X | Y".
>>>
>>> testsuite/ChangeLog:
>>> * gcc.dg/tree-ssa/andor-3.c: New testcase.
>>> * gcc.dg/tree-ssa/andor-4.c: New testcase.
>>> * gcc.dg/tree-ssa/andor-5.c: New testcase.
>>>
>>>>
>>>> Thanks,
>>>> Richard.
>>>>
>>>>> Thanks,
>>>>> Andrew Pinski
>>>>>
>>>>> ChangeLog:
>>>>> * tree-ssa-forwprop.c (defcodefor_name): New function.
>>>>> (simplify_bitwise_binary): Use defcodefor_name.
>>>>> Simplify "( X | Y) & X" to X and "( X & Y) | X" to X.
>>>>> Simplify "(~X | Y) & X" to "X & Y" and
>>>>> "(~X & Y) | X" to "X | Y".
>>>>>
>>>>> testsuite/ChangeLog:
>>>>> * gcc.dg/tree-ssa/andor-3.c: New testcase.
>>>>> * gcc.dg/tree-ssa/andor-4.c: New testcase.
>>>>> * gcc.dg/tree-ssa/andor-5.c: New testcase.
>>>>>
>>>>>
>>>>>>
>>>>>> Thanks,
>>>>>> Richard.
>>>>>>
>>>>>>> Thanks,
>>>>>>> Andrew Pinski
>>>>>>>
>>>>>>> ChangeLog:
>>>>>>> * fold-const.c (fold_binary_loc <case BIT_AND_EXPR>): Add folding of x
>>>>>>> & (~x | y) into x & y.
>>>>>>>
>>>>>>> testsuite/ChangeLog:
>>>>>>> * gcc.dg/tree-ssa/andor-3.c: New testcase.


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