It seems that, although we are building a device-enabled MFEM in our hip/cuda configs,
none of our tests/examples are actually exercising setting up the device and running our mfem kernels on device.
We should fix this and check correctness and performance, e.g. in the shaping examples, and perhaps others.
It seems that, although we are building a device-enabled MFEM in our hip/cuda configs,
none of our tests/examples are actually exercising setting up the device and running our mfem kernels on device.
We should fix this and check correctness and performance, e.g. in the shaping examples, and perhaps others.