/[MITgcm]/MITgcm/eesupp/inc/CPP_EEMACROS.h
ViewVC logotype

Contents of /MITgcm/eesupp/inc/CPP_EEMACROS.h

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


Revision 1.14 - (show annotations) (download)
Wed Nov 9 17:22:08 2005 UTC (18 years, 7 months ago) by cnh
Branch: MAIN
CVS Tags: checkpoint58b_post, checkpoint57y_post, checkpoint58, checkpoint58f_post, checkpoint58d_post, checkpoint58a_post, checkpoint57z_post, checkpoint58m_post, checkpoint57y_pre, checkpoint58e_post, checkpoint58k_post, checkpoint58l_post, checkpoint58g_post, checkpoint58h_post, checkpoint58j_post, checkpoint58i_post, checkpoint57x_post, checkpoint58c_post
Changes since 1.13: +10 -1 lines
File MIME type: text/plain
1 - Tidying up multi-threaded stuff to get rid and automate some CPP junk.
2 - Putting in CPP optional mode for exf_interp_read.F that allows it to work
    multi-threaded with an F90 compiler (this mode wont work with g77).

1 C $Header: /u/gcmpack/MITgcm/eesupp/inc/CPP_EEMACROS.h,v 1.13 2005/11/08 15:53:41 cnh Exp $
2 C $Name: $
3
4 CBOP
5 C !ROUTINE: CPP_EEMACROS.h
6 C !INTERFACE:
7 C include "CPP_EEMACROS.h "
8 C !DESCRIPTION:
9 C *==========================================================*
10 C | CPP_EEMACROS.h
11 C *==========================================================*
12 C | C preprocessor "execution environment" supporting
13 C | macros. Use this file to define macros for simplifying
14 C | execution environment in which a model runs - as opposed
15 C | to the dynamical problem the model solves.
16 C *==========================================================*
17 CEOP
18
19 #ifndef _CPP_EEMACROS_H_
20 #define _CPP_EEMACROS_H_
21
22 C In general the following convention applies:
23 C ALLOW - indicates an feature will be included but it may
24 C CAN have a run-time flag to allow it to be switched
25 C on and off.
26 C If ALLOW or CAN directives are "undef'd" this generally
27 C means that the feature will not be available i.e. it
28 C will not be included in the compiled code and so no
29 C run-time option to use the feature will be available.
30 C
31 C ALWAYS - indicates the choice will be fixed at compile time
32 C so no run-time option will be present
33
34 C Flag used to indicate which flavour of multi-threading
35 C compiler directives to use. Only set one of these.
36 C USE_SOLARIS_THREADING - Takes directives for SUN Workshop
37 C compiler.
38 C USE_KAP_THREADING - Takes directives for Kuck and
39 C Associates multi-threading compiler
40 C ( used on Digital platforms ).
41 C USE_IRIX_THREADING - Takes directives for SGI MIPS
42 C Pro Fortran compiler.
43 C USE_EXEMPLAR_THREADING - Takes directives for HP SPP series
44 C compiler.
45 C USE_C90_THREADING - Takes directives for CRAY/SGI C90
46 C system F90 compiler.
47 #ifdef TARGET_SUN
48 #define USE_SOLARIS_THREADING
49 #define USING_THREADS
50 #endif
51
52 #ifdef TARGET_DEC
53 #define USE_KAP_THREADING
54 #define USING_THREADS
55 #endif
56
57 #ifdef TARGET_SGI
58 #define USE_IRIX_THREADING
59 #define USING_THREADS
60 #endif
61
62 #ifdef TARGET_HP
63 #define USE_EXEMPLAR_THREADING
64 #define USING_THREADS
65 #endif
66
67 #ifdef TARGET_CRAY_VECTOR
68 #define USE_C90_THREADING
69 #define USING_THREADS
70 #endif
71
72 #ifdef USE_OMP_THREADING
73 #define USING_THREADS
74 #endif
75
76 C-- Define the mapping for the _BARRIER macro
77 C On some systems low-level hardware support can be accessed through
78 C compiler directives here.
79 #define _BARRIER CALL BARRIER(myThid)
80
81 C-- Define the mapping for the BEGIN_CRIT() and END_CRIT() macros.
82 C On some systems we simply execute this section only using the
83 C master thread i.e. its not really a critical section. We can
84 C do this because we do not use critical sections in any critical
85 C sections of our code!
86 #define _BEGIN_CRIT(a) _BEGIN_MASTER(a)
87 #define _END_CRIT(a) _END_MASTER(a)
88
89 C-- Define the mapping for the BEGIN_MASTER_SECTION() and
90 C END_MASTER_SECTION() macros. These are generally implemented by
91 C simply choosing a particular thread to be "the master" and have
92 C it alone execute the BEGIN_MASTER..., END_MASTER.. sections.
93
94 #define _BEGIN_MASTER(a) IF ( a .EQ. 1 ) THEN
95 #define _END_MASTER(a) ENDIF
96 CcnhDebugStarts
97 C Alternate form to the above macros that increments (decrements) a counter each
98 C time a MASTER section is entered (exited). This counter can then be checked in barrier
99 C to try and detect calls to BARRIER within single threaded sections.
100 C Using these macros requires two changes to Makefile - these changes are written
101 C below.
102 C 1 - add a filter to the CPP command to kill off commented _MASTER lines
103 C 2 - add a filter to the CPP output the converts the string N EWLINE to an actual newline.
104 C The N EWLINE needs to be changes to have no space when this macro and Makefile changes
105 C are used. Its in here with a space to stop it getting parsed by the CPP stage in these
106 C comments.
107 C #define _BEGIN_MASTER(a) IF ( a .EQ. 1 ) THEN N EWLINE CALL BARRIER_MS(a)
108 C #define _END_MASTER(a) CALL BARRIER_MU(a) N EWLINE ENDIF
109 C 'CPP = cat $< | $(TOOLSDIR)/set64bitConst.sh | grep -v '^[cC].*_MASTER' | cpp -traditional -P'
110 C .F.f:
111 C $(CPP) $(DEFINES) $(INCLUDES) | sed 's/N EWLINE/\n/' > $@
112 CcnhDebugEnds
113
114 C-- Control storage of floating point operands
115 C On many systems it improves performance only to use
116 C 8-byte precision for time stepped variables.
117 C Constant in time terms ( geometric factors etc.. )
118 C can use 4-byte precision, reducing memory utilisation and
119 C boosting performance because of a smaller working
120 C set size. However, on vector CRAY systems this degrades
121 C performance.
122 #ifdef REAL4_IS_SLOW
123 #define _RS Real*8
124 #define RS_IS_REAL8
125 #define _GLOBAL_SUM_R4(a,b) CALL GLOBAL_SUM_R8 ( a, b)
126 #define _GLOBAL_MAX_R4(a,b) CALL GLOBAL_MAX_R8 ( a, b )
127 #define _MPI_TYPE_RS MPI_DOUBLE_PRECISION
128 #else
129 #define _RS Real*4
130 #define RS_IS_REAL4
131 #define _GLOBAL_SUM_R4(a,b) CALL GLOBAL_SUM_R4 ( a, b )
132 #define _GLOBAL_MAX_R4(a,b) CALL GLOBAL_MAX_R4 ( a, b )
133 #define _MPI_TYPE_RS MPI_REAL
134 #endif
135 #define _EXCH_XY_R4(a,b) CALL EXCH_XY_RL ( a, b )
136 #define _EXCH_XYZ_R4(a,b) CALL EXCH_XYZ_RL ( a, b )
137
138 #define _RL Real*8
139 #define _EXCH_XY_R8(a,b) CALL EXCH_XY_RL ( a, b )
140 #define _EXCH_XYZ_R8(a,b) CALL EXCH_XYZ_RL ( a, b )
141 #define _GLOBAL_SUM_R8(a,b) CALL GLOBAL_SUM_R8 ( a, b )
142 #define _GLOBAL_MAX_R8(a,b) CALL GLOBAL_MAX_R8 ( a, b )
143 #define _MPI_TYPE_RL MPI_DOUBLE_PRECISION
144
145 #define _EXCH_XY_RS(a,b) CALL EXCH_XY_RL ( a, b )
146 #define _EXCH_XYZ_RS(a,b) CALL EXCH_XYZ_RL ( a, b )
147 #define _EXCH_XY_RL(a,b) CALL EXCH_XY_RL ( a, b )
148 #define _EXCH_XYZ_RL(a,b) CALL EXCH_XYZ_RL ( a, b )
149
150 #define _MPI_TYPE_R4 MPI_REAL
151 #if (defined (TARGET_SGI) || defined (TARGET_AIX) || defined (TARGET_LAM))
152 #define _MPI_TYPE_R8 MPI_DOUBLE_PRECISION
153 #else
154 #define _MPI_TYPE_R8 MPI_REAL8
155 #endif
156 #define _R4 Real*4
157 #define _R8 Real*8
158
159 C-- Control use of JAM routines for Artic network
160 C These invoke optimized versions of "exchange" and "sum" that
161 C utilize the programmable aspect of Artic cards.
162 #ifdef LETS_MAKE_JAM
163 #define _GLOBAL_SUM_R4(a,b) CALL GLOBAL_SUM_R8_JAM ( a, b)
164 #define _EXCH_XY_R4(a,b) CALL EXCH_XY_R8_JAM ( a, b )
165 #define _EXCH_XYZ_R4(a,b) CALL EXCH_XYZ_R8_JAM ( a, b )
166 #define _EXCH_XY_R8(a,b) CALL EXCH_XY_R8_JAM ( a, b )
167 #define _EXCH_XYZ_R8(a,b) CALL EXCH_XYZ_R8_JAM ( a, b )
168 #define _GLOBAL_SUM_R8(a,b) CALL GLOBAL_SUM_R8_JAM ( a, b )
169
170 #define _EXCH_XY_RS(a,b) CALL EXCH_XY_R8_JAM ( a, b )
171 #define _EXCH_XYZ_RS(a,b) CALL EXCH_XYZ_R8_JAM ( a, b )
172 #define _EXCH_XY_RL(a,b) CALL EXCH_XY_R8_JAM ( a, b )
173 #define _EXCH_XYZ_RL(a,b) CALL EXCH_XYZ_R8_JAM ( a, b )
174 #endif
175
176 C-- Control use of "double" precision constants.
177 C Use D0 where it means REAL*8 but not where it means REAL*16
178 #ifdef REAL_D0_IS_16BYTES
179 #define D0
180 #endif
181
182 C-- Substitue for 1.D variables
183 C Sun compilers do not use 8-byte precision for literals
184 C unless .Dnn is specified. CRAY vector machines use 16-byte
185 C precision when they see .Dnn which runs very slowly!
186 #ifdef REAL_D0_IS_16BYTES
187 #define _d E
188 #define _F64( a ) a
189 #endif
190 #ifndef REAL_D0_IS_16BYTES
191 #define _d D
192 #define _F64( a ) DFLOAT( a )
193 #endif
194
195 #endif /* _CPP_EEMACROS_H_ */

  ViewVC Help
Powered by ViewVC 1.1.22