Clean up SDMA ganging

We don't need to keep track of specific blit engines in gang for
submission anymore as ganging early exits on pending bytes.
So tidy up the fluff.

Change-Id: I77e80bf1ad8f561a03fff77bce33aa09d02760c6


[ROCm/ROCR-Runtime commit: 132815bcfb]

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

@@ -902,7 +902,6 @@ hsa_status_t GpuAgent::DmaCopy(void* dst, core::Agent& dst_agent,
   }
 
   int gang_factor = 0;
-  uint32_t gang_mask = 0;
   // Use non-D2D (auxillary) SDMA engines in the event of xGMI D2D support
   // when xGMI SDMA context is not available.
   bool has_aux_gang = tmp_gang_factor >= properties_.NumSdmaEngines && !!!properties_.NumSdmaXgmiEngines;
@@ -925,7 +924,6 @@ hsa_status_t GpuAgent::DmaCopy(void* dst, core::Agent& dst_agent,
       }
     }
 
-    gang_mask |= 1 << i;
     gang_factor++;
   }
 
@@ -961,9 +959,6 @@ hsa_status_t GpuAgent::DmaCopy(void* dst, core::Agent& dst_agent,
   bool gang_leader_set = false;
   int gang_sig_count = 0;
   for (int i = 0; i < gang_factor; i++) {
-    if (gang_factor > 1 && !!!(gang_mask & (1 << i)))
-      continue;
-
     // Set leader and gang status to blit
     SetCopyRequestRefCount(true);
     lazy_ptr<core::Blit>& blit = has_aux_gang ? blits_[i + 1] :

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

@@ -376,7 +376,6 @@ void BuildTopology() {
         continue;
 
       auto linfo = core::Runtime::runtime_singleton_->GetLinkInfo(src_id, dst_id);
-      GpuAgent *gpu = (AMD::GpuAgent*)src_gpu;
       // xGMI link type cannot determine bandwidth keep it fixed for ganging
       bool has_fixed_gang = linfo.info.link_type == HSA_AMD_LINK_INFO_TYPE_XGMI &&
                             linfo.info.numa_distance != 15;
@@ -392,7 +391,7 @@ void BuildTopology() {
       uint32_t gang_factor = has_fixed_gang ? 2 : (linfo.info.min_bandwidth ?
                              linfo.info.max_bandwidth/linfo.info.min_bandwidth : 0);
 
-      gpu->RegisterGangPeer(*dst_gpu, gang_factor);
+      ((AMD::GpuAgent*)src_gpu)->RegisterGangPeer(*dst_gpu, gang_factor);
     }
   }
 }