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] [Graphite] Using SESE regions as black-boxes


On Mon, May 24, 2010 at 10:50, Antoniu Pop
<antoniu.pop@mines-paristech.fr> wrote:
>> The else part in copy_gbb_and_scalar_dependences seems to not be
>> taken for the moment: how was that code tested? ?Would there be
>
> It was tested using a separate feature that abstracts information from
> OpenMP annotations and allows to build SESE black boxes.
>
>> another way to enable the use of SESE regions in the GBBs before
>> we commit this patch? ?For example, we would have to improve the
>> scop detection to create SESE black boxes.
>
> Yes, this is one of the goals. The current implementation will not
> check any properties within black boxes, so it is sufficient to build
> them prior to SCoP detection.

I don't think that the black boxes should be built before SCoP
detection, but instead during SCoP detection: if a statement is
considered too difficult (during the SCoP detection), then right now
we cut out that statement and create new scops around it, so it could
be good to avoid this by enclosing the statement and the other control
flow constructs around that stmt in a black box.

I think that for testing purposes we should have a driver that does
trigger that code, otherwise it may just become unstable with time.

Sebastian


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