On 06/28/13 09:01, Jakub Jelinek wrote:
Hi!
The patch grew already too large, while doing something clever to avoid
unnecessary memory loads/stores in vectorized loop is still unimplemented,
the rest is and it seems to work now (at least on the added testcases),
so I've committed it to gomp-4_0-branch now.
If anyone has any issues with this approach, please let me know.
I'm cool with it. Thanks for working on this.
I merged my aldyh/cilk-in-gomp branch with your current gomp-4_0-branch,
and everything's peachy. I also noticed you merged with trunk, so now
we're all squared with Balaji's array notation work. There was some
small fallout, but I've fixed it all.
At this point, I wonder if it would be reasonable to merge my branch
into gomp-4_0-branch, or if you'd rather keep it separate.
Also, are we at a point where we could perhaps merge Cilk Plus' <#pragma
simd> into trunk (albeit, bringing in the corresponding machinery from
gomp-4_0-branch first)? The Cilk Plus work is incremental, so it's
understandable to bring this in piecemeal (array notation, then pragma
simd, then elemental functions, etc etc).
What are your thoughts?
