Closed susilehtola closed 6 years ago
On the first one it looks like they do not support long double
types which we can work around.
For the second one I guess there is no aligned memory calls for these architectures which could be fun. Is there a way to determine x86 architecture in headers? I believe we would need to change all of our pragma's and align
calls.
The mm_malloc issue should be solvable just by switching to using the C11 standard library
void *aligned_alloc( size_t alignment, size_t size );
function.
Psi4 anyway requires C++11, so that shouldn't be a problem, right?
It shouldn't be an issue for Psi4, but several other programs use it as well now. Let me check with them.
Of course, you could try to conditionalize: check which one is available and use that.
It shouldn't be an issue for Psi4, but several other programs use it as well now. Let me check with them.
Which ones?
Yes, adding additional conditionals is the straightforward option.
Avogadro and Chronus are both looking into adding the library AFAIK.
This is still holding up the Psi4 update in Fedora.
I will try to get to it this weekend.
Same in the FreeBSD port: it only succeeds on amd64.
Any news? It'd be nice to get this fixed so I could build gau2grid and update Psi4 from the ancient revision currently on Fedora.
Please try a build from the current master. The linked PR should have fixed this.
But the linked PR hasn't been merged..?
And then there's the long double
issue on some architectures.
Sure merged, long double
should have been fixed in #26.
It builds! Good to go.
Great! Thanks for checking. @loriab Should I mint a new release?
Not for my sake, but Susi and Yuri would probably prefer to pull from a tag than a sha.
Well, I already had to use the github commit hash to build on Fedora, so not for my sake either...
I think a new (minor?) release would be warranted, after all portability goes up from non-existing to pretty good I think?
Also (i) those two PRs are rather on the big side, especially the changes in rsh_coeffs.*
and (ii) now that this issue is closed (but not solved in a released), it is a bit unclear what is going on portability-wise, i.e. I was just about to submit a new issue that gau2grid is not working on any architecture other than x86-64.
The changes should have allowed architectures other than x86-64. Do you have errors on ARM or similar that we can look into.
@loriab Any objections to a release?
No objections -- I was just about to vote for a release.
Well I only tried building 1.2.0 so far cause nothing indicated there might be problems. I've now uploaded git master to Debian so we'll see - still I think a release would be worthwhile if there are no other blockers.
There are testsuite failures on ppc64el, but I'll file those in a separate issue.
Release minted. Please do open a new issue for ppc64el and other architecture issues. Non x86 is new to me, but happy to try to broaden this project's targets.
Current master turned the autobuilders mostly green, in particular the ppc64el tests passed as well: https://buildd.debian.org/status/package.php?p=gau2grid - the remaining issues are in greyed-out non-release architectures, so not a big deal.
gau2grid has been approved in Fedora, but the builds fail on every other architecture than x86_64.
https://koji.fedoraproject.org/koji/taskinfo?taskID=29866555
It appears there are at least two separate issues in the code.
First, on i686 and armv7hl I get
Second, on ppc64le, aarch64 and s390x I get