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