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: [lno] fix division by zero


On Saturday 03 January 2004 7:24 pm, Toon Moene wrote:
> Pop Sébastian wrote:
> > 	* tree-vectorizer.c (vect_analyze_operations): Test vectorization_factor
> > 	before computing modulo.
>
> Well, I tried, and I tried, but I can't get no ... vectorization:
>
> $ cat vector.f95

Probably because g95 does "interesting" things with arrays. Basically we 
access flatten all array references and access everyhting though a 
one-dimensional array pointer. I'm guessing this might loose whatever 
alignment guarantees we need.

Look at the output of -fdump-tree-optimized and you'll see what I mean.

This is fairly high on the list of things to fix this year. I'd like to do it 
properly by making array accesses use proper multi-dimensional/nested 
ARRAY_TYPE nodes. We could hack something together to make particular simple 
cases work, but I'd prefer to switch everything over. That way we only have 
one system, and everything "just works".

The first part of this is (me) figuring out exactly what additions we need to 
make to ARRAY_TYPE. I don't think there is anything fundamentally missing, it 
just needs a bit of tweaking.

Paul


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