Changes between Version 2 and Version 3 of ticket/0871/Review

Timestamp:

2011-10-19T16:58:16+02:00 (13 years ago)

Author:

rblod

Comment:

--

ticket/0871/Review

@@ -48 +48 @@
 Various tests have been performed on intel with ifort compiler and IBM with xlf:
 
-- first without openmp : works fine for different kind of files (restart, outputs), if the size of the variable to rebuild doesn't exceed half of the RAM memory. For instance in the case of
+- first without OpenMP : works fine for different kind of files (restart, outputs), if the size of the variable to rebuild doesn't exceed half of the RAM memory. For instance it fails in the case of
+a 3D variable of dimension 300*300*50 with daily outputs during on year on a 8Gb computer. The variable size is around 4Gb, and has to be allocated twice, for the global array to be rebuilt and the sum of the local arrays.[[BR]]
+Note that the old rebuild is using a time loop to write the fileds record by record and save memory.  [[BR]]
+- with OpenMP and ifort : it fails at first and i had to declare indlimlens as SHARED. After that change, it runs, but the rebuilt fields are not correct. I had it working in commenting the nested OpenMP loops and using OpenMP only on the number of input files. [[BR]]
 
-a 3D variable of dimension 300*300*50 with daily outputs during 
+- with OpenMP and xlf: it runs fine with 3D variables, 2D variables are not correct (to be re-tested by me). It also fails with big arrays
 
 '''Code Review'''
@@ -64 +67 @@
 === Review Summary ===
 Add summary here
+I don't thing OpenMP problems are major. For ifort, it may be due to the support of nested OpenMP directives n the given implementation, the 2D variables problem has tobe re-checked.
+I've got more concern with the memory issue, even if it's strongly related to the configuration in use.
 
 === Approval for the trunk ===