Generalize using the additional ways to get the CUDA hwloc location

src/core/perfmodel/perfmodel_bus.c

@@ -2961,8 +2961,8 @@ static void write_bus_platform_file_content(int version)
 			/* Record RAM/CUDA bandwidths */
 			if (!nvlinkhost[i])
 			{
-				find_platform_forward_path(hwloc_cuda_get_device_osdev_by_index(topology, i), 1000000. / search_bus_best_timing(i, "CUDA", 0));
-				find_platform_backward_path(hwloc_cuda_get_device_osdev_by_index(topology, i), 1000000. / search_bus_best_timing(i, "CUDA", 1));
+				find_platform_forward_path(get_hwloc_cuda_obj(topology, i), 1000000. / search_bus_best_timing(i, "CUDA", 0));
+				find_platform_backward_path(get_hwloc_cuda_obj(topology, i), 1000000. / search_bus_best_timing(i, "CUDA", 1));
 			}
 		}
 
@@ -2986,8 +2986,8 @@ static void write_bus_platform_file_content(int version)
 							fprintf(f, "    <link_ctn id=\"Host\"/>\n");
 						else
 							emit_platform_path_up(f,
-									hwloc_cuda_get_device_osdev_by_index(topology, i),
-									hwloc_cuda_get_device_osdev_by_index(topology, j));
+									get_hwloc_cuda_obj(topology, i),
+									get_hwloc_cuda_obj(topology, j));
 					}
 					fprintf(f, "   </route>\n");
 				}
@@ -2998,7 +2998,7 @@ static void write_bus_platform_file_content(int version)
 				/* TODO: NUMA affinity */
 				fprintf(f, "    <link_ctn id=\"Host\"/>\n");
 			else
-				emit_platform_forward_path(f, hwloc_cuda_get_device_osdev_by_index(topology, i));
+				emit_platform_forward_path(f, get_hwloc_cuda_obj(topology, i));
 			fprintf(f, "   </route>\n");
 
 			fprintf(f, "   <route src=\"RAM\" dst=\"CUDA%u\" symmetrical=\"NO\">\n", i);
@@ -3007,7 +3007,7 @@ static void write_bus_platform_file_content(int version)
 				/* TODO: NUMA affinity */
 				fprintf(f, "    <link_ctn id=\"Host\"/>\n");
 			else
-				emit_platform_backward_path(f, hwloc_cuda_get_device_osdev_by_index(topology, i));
+				emit_platform_backward_path(f, get_hwloc_cuda_obj(topology, i));
 			fprintf(f, "   </route>\n");
 		}