When running with a GPU backend (CUDA or HIP), Ginkgo direct linear solver will still expect data on the host. Since HiOp sparse module now provides linear solver data on device (when gpu mode is selected), Ginkgo interface should be modified to take data directly from the device. See #650 as a reference.
pelesh
commented
1 year ago
When running with a GPU backend (CUDA or HIP), Ginkgo direct linear solver will still expect data on the host. Since HiOp sparse module now provides linear solver data on device (when
gpumode is selected), Ginkgo interface should be modified to take data directly from the device. See #650 as a reference.