1 |
C $Header: $ |
2 |
C $Name: $ |
3 |
|
4 |
#include "CPP_OPTIONS.h" |
5 |
|
6 |
SUBROUTINE EXCH2_RECV_RX2( |
7 |
I tIlo, tIhi, tiStride, |
8 |
I tJlo, tJhi, tjStride, |
9 |
I tKlo, tKhi, tkStride, |
10 |
I thisTile, thisI, nN, |
11 |
I e2Bufr1_RX, e2Bufr2_RX, e2BufrRecSize, |
12 |
I mnb, nt, |
13 |
U array1, |
14 |
I i1Lo, i1Hi, j1Lo, j1Hi, k1Lo, k1Hi, |
15 |
U array2, |
16 |
I i2Lo, i2Hi, j2Lo, j2Hi, k2Lo, k2Hi, |
17 |
U e2_msgHandles, myTiles, |
18 |
I commSetting, |
19 |
I myThid ) |
20 |
|
21 |
IMPLICIT NONE |
22 |
|
23 |
C |
24 |
#include "W2_OPTIONS.h" |
25 |
#include "W2_EXCH2_TOPOLOGY.h" |
26 |
|
27 |
#include "EEPARAMS.h" |
28 |
CHARACTER*(MAX_LEN_MBUF) messageBuffer |
29 |
C |
30 |
C === Routine arguments === |
31 |
C tIlo, tIhi, tIstride :: index range in I that will be filled in target "array" |
32 |
C tJlo, tJhi, tJstride :: index range in J that will be filled in target "array" |
33 |
C tKlo, tKhi, tKstride :: index range in K that will be filled in target "array" |
34 |
C thisTile :: Rank of the receiveing tile |
35 |
C thisI :: Index of the receiving tile within this process (used |
36 |
C :: to select buffer slots that are allowed). |
37 |
C nN :: Neighbour entry that we are processing |
38 |
C e2Bufr1_RX :: Data transport buffer array. This array is used in one of |
39 |
C :: two ways. For PUT communication the entry in the buffer |
40 |
C :: associated with the source for this receive (determined |
41 |
C :: from the opposing_send index) is read. For MSG communication |
42 |
C :: the entry in the buffer associated with this neighbor of this |
43 |
C :: tile is used as a receive location for loading a linear |
44 |
C :: stream of bytes. |
45 |
C e2BufrRecSize :: Number of elements in each entry of e2Bufr1_RX |
46 |
C mnb :: Second dimension of e2Bufr1_RX |
47 |
C nt :: Third dimension of e2Bufr1_RX |
48 |
C array :: Target array that this receive writes to. |
49 |
C i1Lo, i1Hi :: I coordinate bounds of target array |
50 |
C j1Lo, j1Hi :: J coordinate bounds of target array |
51 |
C k1Lo, k1Hi :: K coordinate bounds of target array |
52 |
C e2_msgHandles :: Synchronization and coordination data structure used to coordinate access |
53 |
C :: to e2Bufr1_RX or to regulate message buffering. In PUT communication |
54 |
C :: sender will increment handle entry once data is ready in buffer. |
55 |
C :: Receiver will decrement handle once data is consumed from buffer. For |
56 |
C :: MPI MSG communication MPI_Wait uses hanlde to check Isend has cleared. |
57 |
C :: This is done in routine after receives. |
58 |
C myTiles :: List of nt tiles that this process owns. |
59 |
C commSetting :: Mode of communication used to exchnage with this neighbor |
60 |
C myThid :: Thread number of this instance of EXCH2_RECV_RX1 |
61 |
C |
62 |
INTEGER tILo, tIHi, tiStride |
63 |
INTEGER tJLo, tJHi, tjStride |
64 |
INTEGER tKLo, tKHi, tkStride |
65 |
INTEGER i1Lo, i1Hi, j1Lo, j1Hi, k1Lo, k1Hi |
66 |
INTEGER i2Lo, i2Hi, j2Lo, j2Hi, k2Lo, k2Hi |
67 |
INTEGER thisTile, nN, thisI |
68 |
INTEGER e2BufrRecSize |
69 |
INTEGER mnb, nt |
70 |
_RX e2Bufr1_RX( e2BufrRecSize, mnb, nt, 2 ) |
71 |
_RX e2Bufr2_RX( e2BufrRecSize, mnb, nt, 2 ) |
72 |
_RX array1(i1Lo:i1Hi,j1Lo:j1Hi,k1Lo:k1Hi) |
73 |
_RX array2(i2Lo:i2Hi,j2Lo:j2Hi,k2Lo:k2Hi) |
74 |
INTEGER e2_msgHandles(2, mnb, nt) |
75 |
INTEGER myThid |
76 |
INTEGER myTiles(nt) |
77 |
CHARACTER commSetting |
78 |
|
79 |
C == Local variables == |
80 |
C itl, jtl, ktl :: Loop counters |
81 |
C :: itl etc... target local |
82 |
C :: itc etc... target canonical |
83 |
C :: isl etc... source local |
84 |
C :: isc etc... source canonical |
85 |
INTEGER itl, jtl, ktl |
86 |
INTEGER itc, jtc, ktc |
87 |
INTEGER isc, jsc, ksc |
88 |
INTEGER isl, jsl, ksl |
89 |
C tt :: Target tile |
90 |
C iBufr1 :: Buffer counter |
91 |
C iBufr2 :: |
92 |
INTEGER tt |
93 |
INTEGER iBufr1, iBufr2 |
94 |
C mb, nb :: Selects e2Bufr, msgHandle record to use |
95 |
C ir :: |
96 |
INTEGER mb, nb, ir |
97 |
C oN :: Opposing send record number |
98 |
INTEGER oN |
99 |
C Loop counters |
100 |
INTEGER I, nri1, nrj1, nrk1 |
101 |
INTEGER nri2, nrj2, nrk2 |
102 |
INTEGER itl1reduce, jtl1reduce |
103 |
INTEGER itl2reduce, jtl2reduce |
104 |
|
105 |
C MPI setup |
106 |
#include "SIZE.h" |
107 |
#include "EESUPPORT.h" |
108 |
INTEGER theTag1, theSize1, theType |
109 |
INTEGER theTag2, theSize2 |
110 |
INTEGER sProc, tProc |
111 |
#ifdef ALLOW_USE_MPI |
112 |
INTEGER mpiStatus(MPI_STATUS_SIZE), mpiRc |
113 |
#endif |
114 |
|
115 |
tt=exch2_neighbourId(nN, thisTile ) |
116 |
oN=exch2_opposingSend_record(nN, thisTile ) |
117 |
itl1reduce=0 |
118 |
jtl1reduce=0 |
119 |
itl2reduce=0 |
120 |
jtl2reduce=0 |
121 |
IF ( exch2_pi(1,oN,tt) .EQ. -1 ) itl1reduce=1 |
122 |
IF ( exch2_pj(1,oN,tt) .EQ. -1 ) itl1reduce=1 |
123 |
IF ( exch2_pi(2,oN,tt) .EQ. -1 ) jtl2reduce=1 |
124 |
IF ( exch2_pj(2,oN,tt) .EQ. -1 ) jtl2reduce=1 |
125 |
|
126 |
C Handle receive end data transport according to communication mechanism between |
127 |
C source and target tile |
128 |
IF ( commSetting .EQ. 'P' ) THEN |
129 |
C 1 Need to check and spin on data ready assertion for multithreaded mode, for now do nothing i.e. |
130 |
C assume only one thread per process. |
131 |
|
132 |
C 2 Need to set e2Bufr to use put buffer from opposing send. |
133 |
oN = exch2_opposingSend_record(nN, thisTile ) |
134 |
mb = oN |
135 |
ir = 1 |
136 |
DO I=1,nt |
137 |
IF ( myTiles(I) .EQ. tt ) THEN |
138 |
nb = I |
139 |
ENDIF |
140 |
ENDDO |
141 |
C Get data from e2Bufr(1,mb,nb) |
142 |
ELSEIF ( commSetting .EQ. 'M' ) THEN |
143 |
#ifdef ALLOW_USE_MPI |
144 |
C Setup MPI stuff here |
145 |
nb = thisI |
146 |
mb = nN |
147 |
ir = 2 |
148 |
theTag1 = (tt-1)*MAX_NEIGHBOURS*2 + oN-1 |
149 |
& + 10000*( |
150 |
& (thisTile-1)*MAX_NEIGHBOURS*2 + oN-1 |
151 |
& ) |
152 |
theTag2 = (tt-1)*MAX_NEIGHBOURS*2 + MAX_NEIGHBOURS + oN-1 |
153 |
& + 10000*( |
154 |
& (thisTile-1)*MAX_NEIGHBOURS*2 + MAX_NEIGHBOURS + oN-1 |
155 |
& ) |
156 |
tProc = exch2_tProc(thisTile)-1 |
157 |
sProc = exch2_tProc(tt)-1 |
158 |
theType = MPI_REAL8 |
159 |
nri1 = (tIhi-tIlo+1-itl1reduce)/tiStride |
160 |
nrj1 = (tJhi-tJlo+1-jtl1reduce)/tjStride |
161 |
nrk1 = (tKhi-tKlo+1)/tkStride |
162 |
iBufr1 = nri1*nrj1*nrk1 |
163 |
nri2 = (tIhi-tIlo+1-itl2reduce)/tiStride |
164 |
nrj2 = (tJhi-tJlo+1-jtl2reduce)/tjStride |
165 |
nrk2 = (tKhi-tKlo+1)/tkStride |
166 |
iBufr2 = nri2*nrj2*nrk2 |
167 |
CALL MPI_Recv( e2Bufr1_RX(1,mb,nb,ir), iBufr1, theType, sProc, |
168 |
& theTag1, MPI_COMM_MODEL, mpiStatus, mpiRc ) |
169 |
CALL MPI_Recv( e2Bufr2_RX(1,mb,nb,ir), iBufr2, theType, sProc, |
170 |
& theTag2, MPI_COMM_MODEL, mpiStatus, mpiRc ) |
171 |
#ifdef W2_E2_DEBUG_ON |
172 |
WRITE(messageBuffer,'(A,I4,A,I4,A)') ' RECV FROM TILE=', tt, |
173 |
& ' (proc = ',sProc,')' |
174 |
CALL PRINT_MESSAGE(messageBuffer, |
175 |
I standardMessageUnit,SQUEEZE_RIGHT, |
176 |
I myThid) |
177 |
WRITE(messageBuffer,'(A,I4,A,I4,A)') ' INTO TILE=', thisTile, |
178 |
& ' (proc = ',tProc,')' |
179 |
CALL PRINT_MESSAGE(messageBuffer, |
180 |
I standardMessageUnit,SQUEEZE_RIGHT, |
181 |
I myThid) |
182 |
WRITE(messageBuffer,'(A,I10)') ' TAG1=', theTag1 |
183 |
CALL PRINT_MESSAGE(messageBuffer, |
184 |
I standardMessageUnit,SQUEEZE_RIGHT, |
185 |
I myThid) |
186 |
WRITE(messageBuffer,'(A,I4)') ' NEL1=', iBufr1 |
187 |
CALL PRINT_MESSAGE(messageBuffer, |
188 |
I standardMessageUnit,SQUEEZE_RIGHT, |
189 |
I myThid) |
190 |
WRITE(messageBuffer,'(A,I10)') ' TAG2=', theTag2 |
191 |
CALL PRINT_MESSAGE(messageBuffer, |
192 |
I standardMessageUnit,SQUEEZE_RIGHT, |
193 |
I myThid) |
194 |
WRITE(messageBuffer,'(A,I4)') ' NEL2=', iBufr2 |
195 |
CALL PRINT_MESSAGE(messageBuffer, |
196 |
I standardMessageUnit,SQUEEZE_RIGHT, |
197 |
I myThid) |
198 |
#endif /* W2_E2_DEBUG_ON */ |
199 |
C Set mb to neighbour entry |
200 |
C Set nt to this tiles rank |
201 |
mb = nN |
202 |
#endif |
203 |
ELSE |
204 |
STOP 'EXCH2_RECV_RX2:: commSetting VALUE IS INVALID' |
205 |
ENDIF |
206 |
|
207 |
iBufr1=0 |
208 |
DO ktl=tKlo,tKhi,tKStride |
209 |
DO jtl=tJLo+jtl1reduce, tJHi, tjStride |
210 |
DO itl=tILo+itl1reduce, tIHi, tiStride |
211 |
C Read from e2Bufr1_RX(iBufr,mb,nb) |
212 |
iBufr1=iBufr1+1 |
213 |
array1(itl,jtl,ktl)=e2Bufr1_RX(iBufr1,mb,nb,ir) |
214 |
ENDDO |
215 |
ENDDO |
216 |
ENDDO |
217 |
|
218 |
iBufr2=0 |
219 |
DO ktl=tKlo,tKhi,tKStride |
220 |
DO jtl=tJLo+jtl2reduce, tJHi, tjStride |
221 |
DO itl=tILo+itl2reduce, tIHi, tiStride |
222 |
C Read from e2Bufr1_RX(iBufr,mb,nb) |
223 |
iBufr2=iBufr2+1 |
224 |
array2(itl,jtl,ktl)=e2Bufr2_RX(iBufr2,mb,nb,ir) |
225 |
ENDDO |
226 |
ENDDO |
227 |
ENDDO |
228 |
|
229 |
RETURN |
230 |
END |
231 |
|
232 |
C---+----1----+----2----+----3----+----4----+----5----+----6----+----7-|--+----| |
233 |
|
234 |
CEH3 ;;; Local Variables: *** |
235 |
CEH3 ;;; mode:fortran *** |
236 |
CEH3 ;;; End: *** |