Hi Tobias,
I have submitted the initial proposal through GSoC. Here is also a
publicly viewable version:
https://drive.google.com/file/d/1Ai5qry978vCXdlaASE3ijNfF0CxIOGYE/view?usp=sharing.
Feel free to let me know if there is anything that needs clarification
or requires work.
I've also finished the proof of concept for the 'reverse'
transformation, which you can find here:
https://github.com/sweiglbosker/gcc/commits/omp-stuff/. There is no
Fortran support, no tests, and some formatting issues, but it works as
expected locally. Does the approach look reasonable?
> Check with Overseers, https://urldefense.proofpoint.com/v2/url?
> u=https-3A__sourceware.org_mission.html&d=DwIDaQ&c=slrrB7dE8n7gBJbeO0g-
> IQ&r=FwVQKHF9gmvMUtFOQWGJOg&m=0crDkdcTMfXZTz9yWr7r7BbnTpdMdeq1svcI1P-
> m4o03YUxssN-hDnTrYdwCIMRr&s=XMg5Qf-
> X5s0A5SIjevv1Up6KIg5F3ZOvx5jZxzwICqA&e= – they probably need the error
> message / email header.
Thanks, this worked.
> * New transformations: OpenMP 6.0 added fuse, interchange, reverse,
> split and stride. OpenMP 6.1 will add flatten (see Technical Report 14;
> 6.1 is scheduled for November this year).
Did you mean "stripe" instead of "stride"? I couldn't find this
directive anywhere in TR-14.
Thanks,
Stefan Weigl-Bosker
