std :: ml::gpu_rocm
Struct std::RocmNativeStream
struct RocmNativeStream { }The hipStream_t every op above is queued on (MIOpen and hipFFT queue their work on it too); null if there is no GPU. Work another library queues on it is ordered with this backend's ops and recorded by graph capture. Opaque: never constructed, only passed back to HIP / MIOpen / hipFFT (std's own bindings of those APIs take their stream as one).