added missing GPU timer logging to Loki field driver

Co-authored-by: Balthasar Reuter <[email protected]>
ecmwf-ifs · Dec 19, 2024 · b2815a3 · b2815a3
1 parent 3786554
commit b2815a3
Showing 1 changed file with 6 additions and 0 deletions.
diff --git a/src/cloudsc_loki/cloudsc_driver_field_loki_mod.F90 b/src/cloudsc_loki/cloudsc_driver_field_loki_mod.F90
@@ -131,6 +131,12 @@ SUBROUTINE CLOUDSC_DRIVER_FIELD( &
 
       CALL TIMER%END()
 
+#ifdef CLOUDSC_GPU_TIMING
+    ! On GPUs, adding block-level column totals is cumbersome and
+    ! error prone, and of little value due to the large number of
+    ! processing "thread teams". Instead we register the total here.
+    CALL TIMER % THREAD_LOG(TID=TID, IGPC=NGPTOT)
+#endif
       CALL TIMER%PRINT_PERFORMANCE(NPROMA, NGPBLKS, NGPTOT)