add rocshmem_barrier() (#61)

* add team-barrier implementation

add a team-barrier API and implementation in the IPC and RO conduit.
Clean up some of the logic in the RO Conduit to distinguish between
sync, sync_all, barrier, and barrier_all.

* add team_barrier_tests to functional tests

src/ipc/context_ipc_device.hpp

@@ -61,6 +61,8 @@ class IPCContext : public Context {
 
   __device__ void barrier_all();
 
+  __device__ void barrier(rocshmem_team_t team);
+
   __device__ void sync_all();
 
   __device__ void sync(rocshmem_team_t team);

src/ipc/context_ipc_device_coll.cpp

@@ -122,4 +122,20 @@ __device__ void IPCContext::barrier_all() {
   __syncthreads();
 }
 
+__device__ void IPCContext::barrier(rocshmem_team_t team) {
+  IPCTeam *team_obj = reinterpret_cast<IPCTeam *>(team);
+
+  int pe = team_obj->my_pe_in_world;
+  int pe_start = team_obj->tinfo_wrt_world->pe_start;
+  int pe_stride = team_obj->tinfo_wrt_world->stride;
+  int pe_size = team_obj->num_pes;
+  long *p_sync = team_obj->barrier_pSync;
+
+  if (is_thread_zero_in_block()) {
+    quiet();
+  }
+  internal_sync(pe, pe_start, pe_stride, pe_size, p_sync);
+  __syncthreads();
+}
+
 }  // namespace rocshmem