call magma function from device (NOT host)

[wtan]

Hi when I tried to call a magma function from a cuda kernel I got an error like this:

error: calling a __host__ function("magma_sposv_batched") from a __global__ function("invokeDeviceMagma") is not allowed

Does magma functions have a device version, like cublas has? I really want to invoke magma from device (i.e., inside a kernel) using dynamic parallelism.

Thanks!
Wei

[mgates3]

No, we don't provide device versions. You have to call magma functions from CPU code.
-mark