| 1 |
C $Header$ |
C $Header$ |
| 2 |
C $Name$ |
C $Name$ |
| 3 |
C |
CBOP |
| 4 |
C /==========================================================\ |
C !ROUTINE: BARRIER.h |
| 5 |
C | FBAR.h | |
C !INTERFACE: |
| 6 |
C | o Globals used by Fortran barrier routine. | |
C include "BARRIER.h" |
| 7 |
C |==========================================================| |
C !DESCRIPTION: |
| 8 |
C | These variables support a simple Fortran shared memory | |
C *==========================================================* |
| 9 |
C | barrier routine. They do busy waiting, that is the | |
C | BARRIER.h |
| 10 |
C | thread that is waiting sits in a tight loop reading | |
C | o Globals used by Fortran barrier routine. |
| 11 |
C | some memory location and waiting for all other threads. | |
C *==========================================================* |
| 12 |
C | On some systems this is sometimes a good method to use. | |
C | These variables support a simple Fortran shared memory |
| 13 |
C | On the T3E memory is not shared so the routine should | |
C | barrier routine. They do busy waiting, that is the |
| 14 |
C | use the T3E "eureka" barriers. On CRAY and NEC there are | |
C | thread that is waiting sits in a tight loop reading |
| 15 |
C | hardware barriers that are accessed through a compiler | |
C | some memory location and waiting for all other threads. |
| 16 |
C | directives. Finally proper multi-threading compilers | |
C | On some systems this is sometimes a good method to use. |
| 17 |
C | support barrier compile directives - sometimes these | |
C | On the T3E memory is not shared so the routine should |
| 18 |
C | are good, sometimes they are lousy. | |
C | use the T3E "eureka" barriers. On CRAY and NEC there are |
| 19 |
C | The barrier mechanism is used as follows | |
C | hardware barriers that are accessed through a compiler |
| 20 |
C | 1. In the single-threaded part of the code | |
C | directives. Finally proper multi-threading compilers |
| 21 |
C | CALL FBAR_INIT | |
C | support barrier compile directives - sometimes these |
| 22 |
C | on CRAY, NEC this routine does nothing | |
C | are good, sometimes they are lousy. |
| 23 |
C | on T3E there is no single-threaded code | |
C | The barrier mechanism is used as follows |
| 24 |
C | but there may be barrier initialisation - | |
C | 1. In the single-threaded part of the code |
| 25 |
C | need to check. | |
C | CALL FBAR_INIT |
| 26 |
C | 2. When we need to synchronize everybody just | |
C | on CRAY, NEC this routine does nothing |
| 27 |
C | CALL FBAR( myThid ) | |
C | on T3E there is no single-threaded code |
| 28 |
C | where myThid is myThreadId | |
C | but there may be barrier initialisation - |
| 29 |
C | on CRAY, NEC FBAR will just do C$DIR BARRIER | |
C | need to check. |
| 30 |
C | or the like. | |
C | 2. When we need to synchronize everybody just |
| 31 |
C | on T3E FBAR does CALL BARRIER(...) or something | |
C | CALL FBAR( myThid ) |
| 32 |
C | need to check this. | |
C | where myThid is myThreadId |
| 33 |
C \==========================================================/ |
C | on CRAY, NEC FBAR will just do C$DIR BARRIER |
| 34 |
|
C | or the like. |
| 35 |
|
C | on T3E FBAR does CALL BARRIER(...) or something |
| 36 |
|
C | need to check this. |
| 37 |
|
C *==========================================================* |
| 38 |
|
CEOP |
| 39 |
COMMON / BARRIER_COMMON / key1, key2, key3, |
COMMON / BARRIER_COMMON / key1, key2, key3, |
| 40 |
& door1, door2, door3 |
& door1, door2, door3 |
| 41 |
INTEGER key1(lShare4,MAX_NO_THREADS) |
INTEGER key1(lShare4,MAX_NO_THREADS) |
Parent Directory
| 
Revision Log
| 
Revision Graph
| 
Patch