blev grid size with oversubscription factor

NVIDIA · Dec 23, 2022 · f321135 · f321135
1 parent a16d905
commit f321135
Showing 1 changed file with 2 additions and 1 deletion.
diff --git a/cub/device/dispatch/dispatch_batch_memcpy.cuh b/cub/device/dispatch/dispatch_batch_memcpy.cuh
@@ -540,7 +540,8 @@ struct DispatchBatchMemcpy : SelectedPolicy
       return error;
     }
 
-    int batch_memcpy_blev_grid_size = static_cast<int>(sm_count * batch_memcpy_blev_occupancy);
+    int batch_memcpy_blev_grid_size =
+      static_cast<int>(sm_count * batch_memcpy_blev_occupancy * CUB_SUBSCRIPTION_FACTOR(0));
 
     // Construct the tile status for the buffer prefix sum
     BLevBufferOffsetTileState buffer_scan_tile_state;