MITgcm |
|
|
|
|||
File indexing completed on 2025-05-01 05:08:01 UTC
view on githubraw file Latest commit 86f38990 on 2025-04-30 15:10:43 UTC86f389907b Jean*0001 #ifndef _CPP_EEOPTIONS_H_ 0002 #define _CPP_EEOPTIONS_H_ 0003 3d054304e0 Jean*0004 CBOP 0005 C !ROUTINE: CPP_EEOPTIONS.h 0006 C !INTERFACE: 0007 C include "CPP_EEOPTIONS.h" 7f58e89433 Jean*0008 C 3d054304e0 Jean*0009 C !DESCRIPTION: 0010 C *==========================================================* 0011 C | CPP\_EEOPTIONS.h | 0012 C *==========================================================* 7f58e89433 Jean*0013 C | C preprocessor "execution environment" supporting | 0014 C | flags. Use this file to set flags controlling the | 0015 C | execution environment in which a model runs - as opposed | 0016 C | to the dynamical problem the model solves. | 0017 C | Note: Many options are implemented with both compile time| 0018 C | and run-time switches. This allows options to be | 0019 C | removed altogether, made optional at run-time or | 0020 C | to be permanently enabled. This convention helps | 0021 C | with the data-dependence analysis performed by the | 0022 C | adjoint model compiler. This data dependency | 0023 C | analysis can be upset by runtime switches that it | 0024 C | is unable to recoginise as being fixed for the | 0025 C | duration of an integration. | 0026 C | A reasonable way to use these flags is to | 0027 C | set all options as selectable at runtime but then | 0028 C | once an experimental configuration has been | 0029 C | identified, rebuild the code with the appropriate | 0030 C | options set at compile time. | 3d054304e0 Jean*0031 C *==========================================================* 0032 CEOP 7f58e89433 Jean*0033 0034 C In general the following convention applies: 0035 C ALLOW - indicates an feature will be included but it may 0036 C CAN have a run-time flag to allow it to be switched 0037 C on and off. 0038 C If ALLOW or CAN directives are "undef'd" this generally 0039 C means that the feature will not be available i.e. it 0040 C will not be included in the compiled code and so no 0041 C run-time option to use the feature will be available. 0042 C 0043 C ALWAYS - indicates the choice will be fixed at compile time 0044 C so no run-time option will be present 0045 86f389907b Jean*0046 C=== Macro related options === 0047 C-- Control storage of floating point operands 0048 C On many systems it improves performance only to use 0049 C 8-byte precision for time stepped variables. 0050 C Constant in time terms ( geometric factors etc.. ) 0051 C can use 4-byte precision, reducing memory utilisation and 0052 C boosting performance because of a smaller working set size. 0053 C However, on vector CRAY systems this degrades performance. 0054 C Enable to switch REAL4_IS_SLOW from genmake2 (with LET_RS_BE_REAL4): 0055 #ifdef LET_RS_BE_REAL4 0056 #undef REAL4_IS_SLOW 0057 #else /* LET_RS_BE_REAL4 */ 0058 #define REAL4_IS_SLOW 0059 #endif /* LET_RS_BE_REAL4 */ 0060 0061 C-- Control use of "double" precision constants. 0062 C Use D0 where it means REAL*8 but not where it means REAL*16 0063 #define D0 d0 0064 0065 C=== IO related options === 3d054304e0 Jean*0066 C-- Flag used to indicate whether Fortran formatted write 7f58e89433 Jean*0067 C and read are threadsafe. On SGI the routines can be thread 0068 C safe, on Sun it is not possible - if you are unsure then 0069 C undef this option. 3d054304e0 Jean*0070 #undef FMTFTN_IO_THREAD_SAFE 0071 0072 C-- Flag used to indicate whether Binary write to Local file (i.e., 0073 C a different file for each tile) and read are thread-safe. 0074 #undef LOCBIN_IO_THREAD_SAFE 0075 0076 C-- Flag to turn off the writing of error message to ioUnit zero 0077 #undef DISABLE_WRITE_TO_UNIT_ZERO 0078 86f389907b Jean*0079 C-- Flag to turn on old default of opening scratch files with the 0080 C STATUS='SCRATCH' option. This method, while perfectly FORTRAN-standard, 0081 C caused filename conflicts on some multi-node/multi-processor platforms 0082 C in the past and has been replace by something (hopefully) more robust. 0083 #undef USE_FORTRAN_SCRATCH_FILES 0084 0085 C-- Flag defined for eeboot_minimal.F, eeset_parms.F and open_copy_data_file.F 0086 C to write STDOUT, STDERR and scratch files from process 0 only. 0087 C WARNING: to use only when absolutely confident that the setup is working 0088 C since any message (error/warning/print) from any proc <> 0 will be lost. 0089 #undef SINGLE_DISK_IO 3d054304e0 Jean*0090 86f389907b Jean*0091 C=== MPI, EXCH and GLOBAL_SUM related options === 3d054304e0 Jean*0092 C-- Flag turns off MPI_SEND ready_to_receive polling in the 0093 C gather_* subroutines to speed up integrations. 0094 #undef DISABLE_MPI_READY_TO_RECEIVE 0095 0096 C-- Control use of communication that might overlap computation. 0097 C Under MPI selects/deselects "non-blocking" sends and receives. 0098 #undef ALLOW_ASYNC_COMMUNICATION 0099 #undef ALWAYS_USE_ASYNC_COMMUNICATION 0100 C-- Control use of communication that is atomic to computation. 0101 C Under MPI selects/deselects "blocking" sends and receives. 0102 #define ALLOW_SYNC_COMMUNICATION 0103 #undef ALWAYS_USE_SYNC_COMMUNICATION 0104 0105 C-- Control XY periodicity in processor to grid mappings 0106 C Note: Model code does not need to know whether a domain is 0107 C periodic because it has overlap regions for every box. 0108 C Model assume that these values have been 0109 C filled in some way. 0110 #undef ALWAYS_PREVENT_X_PERIODICITY 0111 #undef ALWAYS_PREVENT_Y_PERIODICITY 0112 #define CAN_PREVENT_X_PERIODICITY 0113 #define CAN_PREVENT_Y_PERIODICITY 0114 86f389907b Jean*0115 C-- disconnect tiles (no exchange between tiles, just fill-in edges 0116 C assuming locally periodic subdomain) 0117 #undef DISCONNECTED_TILES 0118 0119 C-- Always cumulate tile local-sum in the same order by applying MPI allreduce 0120 C to array of tiles ; can get slower with large number of tiles (big set-up) 0121 #define GLOBAL_SUM_ORDER_TILES 3d054304e0 Jean*0122 0123 C-- Alternative way of doing global sum without MPI allreduce call 86f389907b Jean*0124 C but instead, explicit MPI send & recv calls. Expected to be slower. 3d054304e0 Jean*0125 #undef GLOBAL_SUM_SEND_RECV 0126 0127 C-- Alternative way of doing global sum on a single CPU 86f389907b Jean*0128 C to eliminate tiling-dependent roundoff errors. Note: This is slow. 0129 #undef CG2D_SINGLECPU_SUM 0130 0131 C=== Other options (to add/remove pieces of code) === 0132 C-- Flag to turn on checking for errors from all threads and procs 0133 C (calling S/R STOP_IF_ERROR) before stopping. 0134 #define USE_ERROR_STOP 0135 0136 C-- Control use of communication with other component: 0137 C allow to import and export from/to Coupler interface. 0138 #undef COMPONENT_MODULE 3d054304e0 Jean*0139 86f389907b Jean*0140 C-- Activate some pieces of code for coupling to GEOS AGCM 0141 #undef HACK_FOR_GMAO_CPL 0142 0143 C=== And define Macros === 0144 c#include "CPP_EEMACROS.h" 3d054304e0 Jean*0145 C--- Taken from file "CPP_EEMACROS.h": 7f58e89433 Jean*0146 0147 C Flag used to indicate which flavour of multi-threading 0148 C compiler directives to use. Only set one of these. 0149 C USE_SOLARIS_THREADING - Takes directives for SUN Workshop 0150 C compiler. 9274434acc Jean*0151 C USE_KAP_THREADING - Takes directives for Kuck and 7f58e89433 Jean*0152 C Associates multi-threading compiler 0153 C ( used on Digital platforms ). 0154 C USE_IRIX_THREADING - Takes directives for SGI MIPS 0155 C Pro Fortran compiler. 0156 C USE_EXEMPLAR_THREADING - Takes directives for HP SPP series 0157 C compiler. 0158 C USE_C90_THREADING - Takes directives for CRAY/SGI C90 0159 C system F90 compiler. 0160 #ifdef TARGET_SUN 0161 #define USE_SOLARIS_THREADING 0162 #endif 0163 0164 #ifdef TARGET_DEC 0165 #define USE_KAP_THREADING 0166 #endif 0167 0168 #ifdef TARGET_SGI 0169 #define USE_IRIX_THREADING 0170 #endif 0171 0172 #ifdef TARGET_HP 0173 #define USE_EXEMPLAR_THREADING 0174 #endif 0175 0176 #ifdef TARGET_CRAY_VECTOR 0177 #define USE_C90_THREADING 0178 #endif 0179 86f389907b Jean*0180 #ifdef USE_OMP_THREADING 0181 #define USING_THREADS 0182 #endif 0183 7f58e89433 Jean*0184 C-- Define the mapping for the _BARRIER macro 0185 C On some systems low-level hardware support can be accessed through 0186 C compiler directives here. 0187 #define _BARRIER CALL BARRIER(myThid) 0188 9274434acc Jean*0189 C-- Define the mapping for the BEGIN_CRIT() and END_CRIT() macros. 7f58e89433 Jean*0190 C On some systems we simply execute this section only using the 0191 C master thread i.e. its not really a critical section. We can 0192 C do this because we do not use critical sections in any critical 0193 C sections of our code! 0194 #define _BEGIN_CRIT(a) _BEGIN_MASTER(a) 0195 #define _END_CRIT(a) _END_MASTER(a) 0196 0197 C-- Define the mapping for the BEGIN_MASTER_SECTION() and 0198 C END_MASTER_SECTION() macros. These are generally implemented by 0199 C simply choosing a particular thread to be "the master" and have 0200 C it alone execute the BEGIN_MASTER..., END_MASTER.. sections. 0201 #define _BEGIN_MASTER(a) IF ( a .EQ. 1 ) THEN 0202 #define _END_MASTER(a) ENDIF 0203 3d054304e0 Jean*0204 C- Note: global_sum/max macros were used to switch to JAM routines (obsolete); 0205 C in addition, since only the R4 & R8 S/R are coded, GLOBAL RS & RL macros 0206 C enable to call the corresponding R4 or R8 S/R. 7f58e89433 Jean*0207 #ifdef REAL4_IS_SLOW 3d054304e0 Jean*0208 7f58e89433 Jean*0209 #define _RS Real*8 0210 #define RS_IS_REAL8 3d054304e0 Jean*0211 #define _GLOBAL_SUM_RS(a,b) CALL GLOBAL_SUM_R8( a, b ) 0212 #define _GLOBAL_MAX_RS(a,b) CALL GLOBAL_MAX_R8( a, b ) 0213 #define _MPI_TYPE_RS MPI_DOUBLE_PRECISION 7f58e89433 Jean*0214 3d054304e0 Jean*0215 #else /* REAL4_IS_SLOW */ 7f58e89433 Jean*0216 0217 #define _RS Real*4 0218 #define RS_IS_REAL4 3d054304e0 Jean*0219 #define _GLOBAL_SUM_RS(a,b) CALL GLOBAL_SUM_R4( a, b ) 0220 #define _GLOBAL_MAX_RS(a,b) CALL GLOBAL_MAX_R4( a, b ) 0221 #define _MPI_TYPE_RS MPI_REAL 7f58e89433 Jean*0222 3d054304e0 Jean*0223 #endif /* REAL4_IS_SLOW */ 7f58e89433 Jean*0224 3d054304e0 Jean*0225 #define _RL Real*8 0226 #define RL_IS_REAL8 0227 #define _GLOBAL_SUM_RL(a,b) CALL GLOBAL_SUM_R8( a, b ) 0228 #define _GLOBAL_MAX_RL(a,b) CALL GLOBAL_MAX_R8( a, b ) 0229 #define _MPI_TYPE_RL MPI_DOUBLE_PRECISION 0230 86f389907b Jean*0231 #define _MPI_TYPE_R4 MPI_REAL 0232 #if (defined (TARGET_SGI) || defined (TARGET_AIX) || defined (TARGET_LAM)) 0233 #define _MPI_TYPE_R8 MPI_DOUBLE_PRECISION 0234 #else 0235 #define _MPI_TYPE_R8 MPI_REAL8 0236 #endif 0237 #define _R4 Real*4 0238 #define _R8 Real*8 0239 3d054304e0 Jean*0240 C- Note: a) exch macros were used to switch to JAM routines (obsolete) 0241 C b) exch R4 & R8 macros are not practically used ; if needed, 0242 C will directly call the corrresponding S/R. 0243 #define _EXCH_XY_RS(a,b) CALL EXCH_XY_RS ( a, b ) 0244 #define _EXCH_XY_RL(a,b) CALL EXCH_XY_RL ( a, b ) 0245 #define _EXCH_XYZ_RS(a,b) CALL EXCH_XYZ_RS ( a, b ) 0246 #define _EXCH_XYZ_RL(a,b) CALL EXCH_XYZ_RL ( a, b ) 7f58e89433 Jean*0247 0248 C-- Control use of "double" precision constants. 0249 C Use D0 where it means REAL*8 but not where it means REAL*16 0250 #ifdef REAL_D0_IS_16BYTES 0251 #define D0 0252 #endif 0253 0254 C-- Substitue for 1.D variables 0255 C Sun compilers do not use 8-byte precision for literals 0256 C unless .Dnn is specified. CRAY vector machines use 16-byte 0257 C precision when they see .Dnn which runs very slowly! 0258 #ifdef REAL_D0_IS_16BYTES 0259 #define _F64( a ) a 0260 #endif 0261 #ifndef REAL_D0_IS_16BYTES 0262 #define _F64( a ) DFLOAT( a ) 0263 #endif 0264 86f389907b Jean*0265 C-- Set the format for writing processor IDs, e.g. in S/R eeset_parms 0266 C and S/R open_copy_data_file. The default of I9.9 should work for 0267 C a long time (until we will use 10e10 processors and more) 0268 #define FMT_PROC_ID 'I9.9' 0269 0270 C-- Set the format for writing ensemble task IDs in S/R eeset_parms 0271 C and S/R open_copy_data_file. 0272 #define FMT_TSK_ID 'I6.6' 0273 0274 C-- Set ACTION= in OPEN instruction for input file (before doing IO) 0275 C leave it empty (if EXCLUDE_OPEN_ACTION) or set it to proper value 0276 #ifdef EXCLUDE_OPEN_ACTION 0277 # define _READONLY_ACTION 0278 #else 0279 # define _READONLY_ACTION ACTION='read', 0280 #endif 3d054304e0 Jean*0281 86f389907b Jean*0282 #endif /* _CPP_EEOPTIONS_H_ */
| [ Source navigation ] | [ Diff markup ] | [ Identifier search ] | [ general search ] |
|
This page was automatically generated from https://github.com/MITgcm/MITgcm by the 2.2.1-MITgcm-0.1 LXR engine. The LXR team |
|