On 08/09/16 16:25, Bernhard Reutner-Fischer wrote:
On 8 September 2016 at 10:31, Kyrill Tkachov
<kyrylo.tkac...@foss.arm.com> wrote:
On 07/09/16 20:03, Bernhard Reutner-Fischer wrote:
On September 6, 2016 5:14:47 PM GMT+02:00, Kyrill Tkachov
<kyrylo.tkac...@foss.arm.com> wrote:
Thanks, fixed all the above in my tree (will be retesting).
What about debug statements? ISTM you should skip those.
(Isn't visited reset before entry of a pass?)
Yes, I'll skip debug statements. Comments in gimple.h say that the visited
property is undefined at pass boundaries, so I'd rather initialize it here.
Right.
Maybe I missed the bikeshedding about the name but I'd have used
-fmerge-stores instead.
Wouldn't be hard to change. I can change it any point if there's a
consensus.
Did you consider any relation to any of
https://gcc.gnu.org/PR22141
https://gcc.gnu.org/PR23684
https://gcc.gnu.org/PR47059
https://gcc.gnu.org/PR54422
and their dups
or https://www.mail-archive.com/gcc-patches@gcc.gnu.org/msg77311.html
(the -fmerge-bitfields suggestion from imgtec; maybe the testcases are
of interest)
Thanks for the pointers. I was not aware of PR23684 and have extended the patch
to handle that case as well.
The current version I'm hoping to get in only handles constant stores.
There are tricks we can do for non-constant contiguous stores as well and I hope
to implement some of them in the future, but for now I'm concentrating on
getting
the initial version up to scratch.
I am now re-testing re-benchmarking the patch after your feedback and some
other improvements I've added since the last version and hope to send out
an updated version soon.
Thanks,
Kyrill
thanks,
