fix: use mpp_gather to improve iceberg restart read times (#1460)

NOAA-GFDL · May 9, 2024 · 42f8506 · 42f8506
1 parent e2fa28a
commit 42f8506
Showing 1 changed file with 3 additions and 10 deletions.
diff --git a/fms2_io/netcdf_io.F90 b/fms2_io/netcdf_io.F90
@@ -838,16 +838,9 @@ subroutine register_unlimited_compressed_axis(fileobj, dimension_name, dimension
   !Gather all local dimension lengths on the I/O root pe.
   allocate(npes_start(size(fileobj%pelist)))
   allocate(npes_count(size(fileobj%pelist)))
-  do i = 1, size(fileobj%pelist)
-    if (fileobj%pelist(i) .eq. mpp_pe()) then
-      npes_count(i) = dimension_length
-    else
-      call mpp_recv(npes_count(i), fileobj%pelist(i), block=.false.)
-      call mpp_send(dimension_length, fileobj%pelist(i))
-    endif
-  enddo
-  call mpp_sync_self(check=event_recv)
-  call mpp_sync_self(check=event_send)
+
+  call mpp_gather((/dimension_length/),npes_count,pelist=fileobj%pelist)
+
   npes_start(1) = 1
   do i = 1, size(fileobj%pelist)-1
      npes_start(i+1) = npes_start(i) + npes_count(i)