pelesh
pelesh
I think this is relevant for other GPU solvers in HiOp. @kswirydo @nychiang
> The only downside is that now all CI platforms/developers that use this config will by default now build with all the architectures, rather than only on Marianas as would...
CC @maksud
In PR #548 tests are still failing on Marianas, but now with different error messages ([PNNL Pipeline](https://gitlab.pnnl.gov/exasgd/solvers/hiop-mirror/-/jobs/102551)): NlpSparse1_6 (Click to show logs) ```shell Start 22: NlpSparse1_6 22: Test command: /share/apps/openmpi/4.1.0/gcc/10.2.0/bin/mpirun...
@maksud, please check if #548 fixes the issue you reported. Ginkgo tests pass on ascent.
> Am I correct in assuming that no matter if we are in `cpu` or `hybrid` mode, the data for the linear solver will in the beginning be on the...
> @pelesh do you remember the context of this feature? I believe this was about implemnting compliance with [R8](https://github.com/xsdk-project/xsdk-community-policies/blob/master/package_policies/R8.md) xSDK policy.
> `hiop` 0.7.1, `ginkgo` 1.5.0, `gcc` 12.2.0 @barracuda156, you need to use Ginkgo from the head of `glu_experimental` branch. Direct solvers in Ginkgo are under heavy development right now and...
@barracuda156, also see #585.
> Has this been addressed btw? Almost there :). See #669. Direct solvers from `glu_experimental` branch have been released in Ginkgo 1.7.