/[MITgcm]/MITgcm/eesupp/inc/CPP_EEOPTIONS.h

Diff of /MITgcm/eesupp/inc/CPP_EEOPTIONS.h

Parent Directory | Revision Log | View Revision Graph Revision Graph | View Patch Patch

-revision 1.32 by heimbach,
Mon May 25 04:33:15 2009 UTC
+revision 1.40 by jmc,
Tue Aug 25 20:29:33 2015 UTC
 Line 46 
 C
  C     ALWAYS - indicates the choice will be fixed at compile time
  C              so no run-time option will be present
+ C=== Macro related options ===
+ C--   Control storage of floating point operands
+ C     On many systems it improves performance only to use
+ C     8-byte precision for time stepped variables.
+ C     Constant in time terms ( geometric factors etc.. )
+ C     can use 4-byte precision, reducing memory utilisation and
+ C     boosting performance because of a smaller working set size.
+ C     However, on vector CRAY systems this degrades performance.
+ C     Enable to switch REAL4_IS_SLOW from genmake2 (with LET_RS_BE_REAL4):
+ #ifdef LET_RS_BE_REAL4
+ #undef REAL4_IS_SLOW
+ #else /* LET_RS_BE_REAL4 */
+ #define REAL4_IS_SLOW
+ #endif /* LET_RS_BE_REAL4 */
+ C--   Control use of "double" precision constants.
+ C     Use D0 where it means REAL*8 but not where it means REAL*16
+ #define D0 d0
+ C--   Enable some old macro conventions for backward compatibility
+ #undef USE_OLD_MACROS_R4R8toRSRL
+ C=== IO related options ===
  C--   Flag used to indicate whether Fortran formatted write
  C     and read are threadsafe. On SGI the routines can be thread
  C     safe, on Sun it is not possible - if you are unsure then
-Line 59 
 C     a different file for each tile) an
+Line 82 
 C     a different file for each tile) an
  C--   Flag to turn off the writing of error message to ioUnit zero
  #undef DISABLE_WRITE_TO_UNIT_ZERO
- C--   Flag to turn on checking for errors from all threads and procs
+ C--   Alternative formulation of BYTESWAP, faster than
- C     (calling S/R STOP_IF_ERROR) before stopping.
+ C     compiler flag -byteswapio on the Altix.
- #define USE_ERROR_STOP
+ #undef FAST_BYTESWAP
+ C--   Flag defined for eeboot_minimal.F, eeset_parms.F and open_copy_data_file.F
+ C     to write STDOUT, STDERR and scratch files from process 0 only.
+ C WARNING: to use only when absolutely confident that the setup is working
+ C     since any message (error/warning/print) from any proc <> 0 will be lost.
+ #undef SINGLE_DISK_IO
+ C=== MPI, EXCH and GLOBAL_SUM related options ===
  C--   Flag turns off MPI_SEND ready_to_receive polling in the
  C     gather_* subroutines to speed up integrations.
  #undef DISABLE_MPI_READY_TO_RECEIVE
-Line 72 
 CXXX We no longer select the use of MPI
+Line 102 
 CXXX We no longer select the use of MPI
  CXXX To use MPI, use an appropriate genmake2 options file or use
  CXXX genmake2 -mpi .
  CXXX #undef  ALLOW_USE_MPI
- CXXX #undef  ALWAYS_USE_MPI
  C--   Control use of communication that might overlap computation.
  C     Under MPI selects/deselects "non-blocking" sends and receives.
-Line 84 
 C     Under MPI selects/deselects "block
+Line 113 
 C     Under MPI selects/deselects "block
  #define ALLOW_SYNC_COMMUNICATION
  #undef  ALWAYS_USE_SYNC_COMMUNICATION
- C--   Control use of JAM routines for Artic network
- C     These invoke optimized versions of "exchange" and "sum" that
- C     utilize the programmable aspect of Artic cards.
- #undef  LETS_MAKE_JAM
- #undef  JAM_WITH_TWO_PROCS_PER_NODE
- C--   Control storage of floating point operands
- C     On many systems it improves performance only to use
- C     8-byte precision for time stepped variables.
- C     Constant in time terms ( geometric factors etc.. )
- C     can use 4-byte precision, reducing memory utilisation and
- C     boosting performance because of a smaller working
- C     set size. However, on vector CRAY systems this degrades
- C     performance.
- #define REAL4_IS_SLOW
- C--   Control use of "double" precision constants.
- C     Use D0 where it means REAL*8 but not where it means REAL*16
- #define D0 d0
  C--   Control XY periodicity in processor to grid mappings
  C     Note: Model code does not need to know whether a domain is
  C           periodic because it has overlap regions for every box.
-Line 114 
 C           filled in some way.
+Line 123 
 C           filled in some way.
  #define CAN_PREVENT_X_PERIODICITY
  #define CAN_PREVENT_Y_PERIODICITY
- C--   Alternative formulation of BYTESWAP, faster than
+ C--   disconnect tiles (no exchange between tiles, just fill-in edges
- C     compiler flag -byteswapio on the Altix.
+ C     assuming locally periodic subdomain)
- #undef FAST_BYTESWAP
+ #undef DISCONNECTED_TILES
+ C--   Always cumulate tile local-sum in the same order by applying MPI allreduce
+ C     to array of tiles ; can get slower with large number of tiles (big set-up)
+ #undef GLOBAL_SUM_ORDER_TILES
  C--   Alternative way of doing global sum without MPI allreduce call
- C     but instead, explicit MPI send & recv calls.
+ C     but instead, explicit MPI send & recv calls. Expected to be slower.
  #undef GLOBAL_SUM_SEND_RECV
  C--   Alternative way of doing global sum on a single CPU
- C     to eliminate tiling-dependent roundoff errors.
+ C     to eliminate tiling-dependent roundoff errors. Note: This is slow.
- C     Note: This is slow.
  #undef  CG2D_SINGLECPU_SUM
- C--   Enablbe some old macro conventions for
+ C=== Other options (to add/remove pieces of code) ===
- C     backward compatibility
+ C--   Flag to turn on checking for errors from all threads and procs
- #undef USE_OLD_MACROS_R4R8toRSRL
+ C     (calling S/R STOP_IF_ERROR) before stopping.
+ #define USE_ERROR_STOP
+ C--   Control use of communication with other component:
+ C     allow to import and export from/to Coupler interface.
+ #undef COMPONENT_MODULE
  #endif /* _CPP_EEOPTIONS_H_ */

 Legend:



Removed from v.1.32
 


changed lines


 
Added in v.1.40
 Legend:



Removed from v.1.32
 


changed lines


 
Added in v.1.40
-Removed from v.1.32
+Added in v.1.40

	ViewVC Help
Powered by ViewVC 1.1.22