Ignore hive id for CPUs when selecting copy paths.

Hive ID is used during copy path selection to locate an optimal
pool of SDMA engines.  However, for CPU-GPU connections we always
want to use the host port facing engines, known generally as the
PCIe optimzed engines.  We want this selection even when the
connection is XGMI hence dropping the hive id for CPUs.

Change-Id: Iffe44174afecfc0bb3272b806fce549c930a49d9


[ROCm/ROCR-Runtime commit: af0f90800d]

projects/rocr-runtime/runtime/hsa-runtime/core/runtime/amd_gpu_agent.cpp

@@ -1564,9 +1564,13 @@ lazy_ptr<core::Blit>& GpuAgent::GetBlitObject(const core::Agent& dst_agent,
     return blits_[BlitDevToDev];
   }
 
-  // Acquire Hive Id of Src and Dst devices
-  uint64_t src_hive_id = src_agent.HiveId();
-  uint64_t dst_hive_id = dst_agent.HiveId();
+  // Acquire Hive Id of Src and Dst devices - ignore hive id for CPU devices.
+  // CPU-GPU connections should always use the host (aka pcie) facing SDMA engines, even if the
+  // connection is XGMI.
+  uint64_t src_hive_id =
+      (src_agent.device_type() == core::Agent::kAmdGpuDevice) ? src_agent.HiveId() : 0;
+  uint64_t dst_hive_id =
+      (dst_agent.device_type() == core::Agent::kAmdGpuDevice) ? dst_agent.HiveId() : 0;
 
   // Bind to a PCIe facing Blit object if the two
   // devices have different Hive Ids. This can occur