This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [graphite] instancewise dependence analysis functions and structures
On Tue, Aug 19, 2008 at 4:05 PM, Jack Howarth <howarth@bromo.msbb.uc.edu> wrote:
> On Tue, Aug 19, 2008 at 03:30:10PM -0500, Sebastian Pop wrote:
>>
>> Huh? Who said so? The fact that you see no performance improvement
>> does not mean that the code is not transformed. It is an inefficiency
>> of the cloog-ppl port of domain_simplify function that messes up the
>> code right now. If you want to help porting a 2000 lines code
>> function please step up ;-)
>>
>
> Sebastian,
> I guess I misunderstood Tobias Grosser's previous comments that...
>
>> The problem seems to be that cloog-ppl does not remove all unused
>> conditions. So there remain conditions in the cloog AST and, as our
>> backend at the moment is not able to build conditions, we do not
>> generate any code.
>
Oh, sorry! It's my misunderstanding here: yes, when there are
conditions in the transformed loop nest the graphite code generator
does not produce the transformed code.
> ...to imply that you might have the intention to fix this issue by
> adding support for the backend to build conditions rather than to
> fix cloog-ppl to remove all the unused conditions.
Yes I am trying to address as well the domain_simplify to not produce
inefficient code.
I will also include an option in the configure of GCC to make it
possible to build GCC with polylib that does not have these problems.
Sebastian