Hi!

On Tue, 13 Sep 2016 16:27:39 +0200, Bernd Schmidt <bschmidt@redhat.com> wrote:
> On 09/13/2016 04:24 PM, Jason Merrill wrote:
> 
> > But we could define TARGET_ABSOLUTE_BIGGEST_ALIGNMENT on nvptx instead
> > of on x86; is this OK?
> 
> That's what I had in mind. It would be good if Thomas or Nathan could 
> give this patch a spin, I'm not currently really set up for it. But it 
> looks like a reasonable try to me.

I'm happy to report that this patch doesn't cause any changes in test
results both for nvptx target testing, and for nvptx offloading testing.
But I have not examined in detail what it actually does ;-) -- currently
occupied with too much other work already.

> > I'm still not sure why you need an alignment cap on nvptx, but I'm not
> > going to worry about it anymore.  :)
> 
> I think it was the cfgexpand machinery that uses dynamic allocations 
> when a variable has a bigger alignment than the stack, and you really 
> don't want these on ptx.

It will be good to document that, next to the definition in
gcc/config/nvptx/nvptx.h maybe?


Grüße
 Thomas