[SWDEV-560681] Allowed GPU enumeration to continue with non-contiguous render nodes (#1609)

* Fix uninitialized variable in GPU enumeration loop (#1643)
* Initialize node_to_gpu_id to prevent undefined behavior

---------

Signed-off-by: gabrpham <Gabriel.Pham@amd.com>
Co-authored-by: Allan Xavier <axavier@digitalocean.com>

projects/rocm-smi-lib/src/rocm_smi_kfd.cc

@@ -685,7 +685,7 @@ KFDNode::Initialize(void) {
 
   std::map<uint32_t, std::shared_ptr<IOLink>>::iterator it;
   uint32_t node_to;
-  uint64_t node_to_gpu_id;
+  uint64_t node_to_gpu_id = 0;
   std::shared_ptr<IOLink> link;
   bool numa_node_found = false;
   for (it = io_link_map_tmp.begin(); it != io_link_map_tmp.end(); it++) {
@@ -693,7 +693,7 @@ KFDNode::Initialize(void) {
     node_to = it->first;
     link = it->second;
     ret = ReadKFDGpuId(node_to, &node_to_gpu_id);
-    if (ret) {return ret;}
+    if (ret) {continue;}
     if (node_to_gpu_id == 0) {  //  CPU node
       if (numa_node_found) {
         if (numa_node_weight_ > link->weight()) {