std :: ml::gpu_rocm
#include <std/ml/gpu_rocm.h> source ยท definition
#include <std/ml/gpu_rocm.h>
int rocm_sync()
wait for all queued work