Make atomic references use correct address spaces.

With CUDA there is no difference in the API for using
atomic operations on global or shared memory. But with
SYCL there is...

device/common/include/traccc/finding/device/impl/find_tracks.ipp

@@ -62,8 +62,9 @@ TRACCC_DEVICE inline void find_tracks(
     vecmem::device_vector<char> out_params_liveness(
         payload.out_params_liveness_view);
     vecmem::device_vector<candidate_link> links(payload.links_view);
-    vecmem::device_atomic_ref<unsigned int> num_total_candidates(
-        *payload.n_total_candidates);
+    vecmem::device_atomic_ref<unsigned int,
+                              vecmem::device_address_space::global>
+        num_total_candidates(*payload.n_total_candidates);
     vecmem::device_vector<const detray::geometry::barcode> barcodes(
         payload.barcodes_view);
     vecmem::device_vector<const unsigned int> upper_bounds(
@@ -158,9 +159,11 @@ TRACCC_DEVICE inline void find_tracks(
         for (; curr_meas < num_meas &&
                shared_payload.shared_candidates_size < thread_id.getBlockDimX();
              curr_meas++) {
-            unsigned int idx = vecmem::device_atomic_ref<unsigned int>(
-                                   shared_payload.shared_candidates_size)
-                                   .fetch_add(1u);
+            unsigned int idx =
+                vecmem::device_atomic_ref<unsigned int,
+                                          vecmem::device_address_space::local>(
+                    shared_payload.shared_candidates_size)
+                    .fetch_add(1u);
 
             /*
              * The buffer elemements are tuples of the measurement index and
@@ -229,7 +232,8 @@ TRACCC_DEVICE inline void find_tracks(
 
                     // Increase the number of candidates (or branches) per input
                     // parameter
-                    vecmem::device_atomic_ref<unsigned int>(
+                    vecmem::device_atomic_ref<
+                        unsigned int, vecmem::device_address_space::local>(
                         shared_payload
                             .shared_num_candidates[owner_local_thread_id])
                         .fetch_add(1u);