03_Fortran_osx.indd

Size: px

Start display at page:

Download "03_Fortran_osx.indd"

ありさとべ
5 years ago
Views:

1 Fortran OpenMP*

2 Fortran OpenMP* OpenMP* OpenMP* OpenMP* OpenMP* OpenMP* OpenMP* OpenMP* OpenMP* PARALLEL / OpenMP* OpenMP* OpenMP* / 4 1 OpenMP* 2 C/C++ OpenMP* 3 Fortran OpenMP* 4 PC 1

3 bit 64bit Linux* 32bit 64bit Windows* OpenMP* OpenMP* 9.0 OpenMP* 2.5 OpenMP* OpenMP* 2

4 2. OpenMP* OS OS CPU Memory CPU CPU CPU OpenMP*.1 PC OS.2 3

5 3. OpenMP* OpenMP* APIApplication Programming Interface OpenMP* API 1997 Open MP Architecture Review Board Linux*UNIX* Windows* OpenMP* C/C++ Fortran OpenMP* OpenMP* 9.0 OpenMP* OpenMP* OpenMP* OpenMP* OpenMP* 2.5 C/C++ Fortran 1998 OpenMP* C/C OpenMP* C/C OpenMP* Fortran C/C OpenMP* Fortran 1.0 OpenMP* Fortran 1.1 OpenMP* Fortran OpenMP* 3.1 C/C++ Fortran API OpenMP* OpenMP* C/C++ Fortran OpenMP* OpenMP* OpenMP* /Qopenmp Windows*-openmp Linux* OpenMP* OpenMP* OpenMP* OpenMP* 4

6 OpenMP* OpenMP* OpenMP* OpenMP* OpenMP* OpenMP* 32bit 64bit Linux* 32bit 64bit Windows* OpenMP* 3.2 OpenMP* OpenMP* OpenMP* API OpenMP* /.4 OpenMP* OpenMP* API OpenMP* API OpenMP* API OpenMP* API 5

7 OpenMP* OpenMP* OpenMP* DO!$OMP PARALLEL IF (N.GT.LIMIT) DEFAULT(NON)!$OMP& SHARED(n,a,b,c,x,y,z) PRIVATE(f,i,scale) f = 1.0!$OMP DO NOWAIT do i = 1, n z(i) = x(i) + y(i) end do!$omp!$omp DO NOWAIT do i = 1, n a(i) = b(i) + c(i) end do!$omp!$omp BARRIER scale = sum(a, 0, n) + sum(z, 0, n) + f!$omp END PARALLEL Parallel Region.5 OpenMP* 6

8 OpenMP* OpenMP* Linux* $ cat -n pi.f90 1 program omp 2!$ use omp_lib 3 integer num_steps 4 real*8 step,x,pi,sum 5 6 num_steps = step = 1.0D0 / dble(num_steps) 8 sum = 0.0D0 9!$OMP PARALLEL PRIVATE(X) 10 nthread = OMP_GET_NUM_THREADS() 11!$OMP DO REDUCTION(+:SUM) OpenMP* 12 do i = 1, num_steps 13 x = (dble(i)-0.5d0)*step 14 sum = sum + 4.0D0 / (1.0D0 + x * x) OpenMP* 15 end do 16!$OMP 17!$OMP END PARALLEL 18 pi = step * sum 19 write (6,*) nthread,' Threads',' PI = ',pi 20 end program omp 21 $ ifort -O3 -openmp _openmp-report2 pi.f90 ifort: Command line warning: openmp requires C style preprocessing; using fpp to preprocess pi.f90(11) : (col. 6) remark: OpenMP DEFINED LOOP WAS PARALLELIZED. pi.f90(9) : (col. 6) remark: OpenMP DEFINED REGION WAS PARALLELIZED. $ setenv OMP_NUM_THREADS 2 $ time./a.out 2 Threads PI = u 0.003s 0: % 0+0k 0+0io 153pf+0w.6 OpenMP* OpenMP* OpenMP* /Qopenmp Windows*-openmp Linux* OpenMP* OpenMP* OpenMP* OpenMP* 7

9 3.3 OpenMP* OpenMP* OpenMP* OpenMP* C/C++ Fortran while DO I=1, N A[ 1] = = A[ 2]

10 4. OpenMP* 4.1 OpenMP* Fortran Fortran 77 f90 f F OpenMP* Fortran OpenMP* OpenMP*!$OMP 1!$OMP directive [clause[[,] clause] ] &!$OMP directive [clause[[,] clause] ]&!$OMP& [clause[[,] clause] ] Fortran77 OpenMP* c$omp directive [clause[[,] clause] ] *$OMP directive [clause[[,] clause] ] c$omp directive [clause[[,] clause] ] c$omp+ [clause[[,] clause] ] 1. clause 9

11 4.2 OpenMP* C C$ IAM = OMP_GET_THREAD_NUM() OpenMP* C$ 2!$!$ IAM = OMP_GET_THREAD_NUM() + &!$ index #ifdef #indef _OPENMP IAM = OMP_GET_THREAD_NUM() #endif 4.3 OpenMP* OpenMP* API OpenMP* call omp_set_num_threads(integer) integer omp_get_num_threads() 1 integer omp_get_max_threads() OMP_NUM_THREADS integer omp_get_thread_num() 0 0 integer omp_get_num_procs() CPU logical omp_in_parallel().false..true. call omp_set_dynamics (logical) logocal omp_get_dynamic() call omp_set_nested(logical) logical omp_get_nested().1 OpenMP* 10

12 OpenMP* OMP OpenMP* API omp_lib program omp!$ use omp_lib.!$omp parallel Iam = omp_get_thread_num()!$omp end parallel. end program omp OpenMP* OpenMP* Parallel Regions Worksharing Data Environment Synchronization / Runtime functions/environment variables Parallel Regions!$OMP PARALLEL Worksharing!$OMP DO!$OMP SECTIONS!$OMP SINGLE!$OMP WORKSHARE Data Environment : THREADPRIVATE : SHAREDPRIVATELASTPRIVATEREDUCTIONCOPYINCOPYPRIVATE Synchronization : CRITICALBARRIERATOMICFLUSHORDEREDMASTER / Runtime functions/environment variables 11

13 4.5 OpenMP* Fortran OpenMP* stop Fortran!$OMP PARALLEL 10 wrk(id) = garbage(id) res(id) = wrk(id)**2 if(conv(res(id)) goto 10!$OMP END PARALLEL print *,id!$omp PARALLEL 10 wrk(id) = garbage(id) 30 res(id)=wrk(id)**2 if(conv(res(id))goto 20 go to 10!$OMP END PARALLEL if(not_done) goto print *, id.7 12

14 5. OpenMP* 5.1 PARALLEL OpenMP* Parallel Region DO PARALLEL GOTO Fortran PARALLEL!$OMP PARALLEL / $OMP END PARALLEL!$OMP PARALLEL [clause [[,] clause]!$omp END PARALLEL clause OpenMP* OpenMP* Fork-Join Fork!$OMP PARALLEL DO [clause[[,]clause]] DO I = 1, N DO!$OMP END PARALLEL DO!$OMP PARALLEL DO [clause[[,]clause]] DO I = 1, N DO!$OMP END PARALLEL DO ForkJoin Join.8 Fork-Join 13

15 1OpenMP* 2 OpenMP*!$OMP PARALLEL Fork 3!$omp end parallel 4 Join Fork-Join PROGRAM OMP!$ USE OMP_LIB REAL*8 A(4,1000) CALL OMP_SET_NUM_THREADS(4) N=1000!$OMP PARALLEL ID = OMP_GET_THREAD_NUM()+1 CALL POOH(ID, A,N)!$OMP END PARALLEL WRITE (*,*) 'all done' END SUBROUTINE POOH(ID,A,N) REAL*8 A(4,*) DO I= 1, N A(ID,J) = 1.0 RETURN END REAL*8 A(4,1000) OMP_SET_NUM_THREADS(4) POOH(0,A,N) WRITE (*,*) 'all done' A DO POOH(1,A,N) POOH(2,A,N) POOH(3,A,N).9 OpenMP* PARALLEL OpenMP* 14

16 main.f PROGRAM OMP INTEGER OMP_GET_THREAD_NUM!$OMP PARALLEL CALL WHOAMI!$OMP END PARALLEL WRITE (6,*) 'All Done' CALL WHOAMI!$OMP PARALLEL IF (OMP_GET_THREAD_NUM().EQ.1) &CALL WHOAMI!$OMP END PARALLEL STOP END PROGRAM OMP whoami.f SUBROUTINE WHOAMI EXTERNAL OMP_GET_THREAD_NUM INTEGER IAM, OMP_GET_THREAD_NUM IAM = OMP_GET_THREAD_NUM()!$OMP CRITICAL WRITE (6,*) 'Hello from',iam!$omp END CRITICAL RETURN END m a i n. f 2 WHOAMI whoami.f.10 OpenMP* main.f 2 WHOAMI whoami.f 15

17 5.2 Fortran PARALLEL OpenMP* API 1 do i = 1, n ; a(i) = a(i) + b(i) ; end do 2OpenMP*!$OMP PARALLEL PRIVATE (id,i,istart,iend) id = omp_get_thread_num(); Nthrds = omp_get_num_threads(); istart = id * n / Nthrds; iend = (id+1) * n / Nthrds; do i=istart,iend ; a(i)=a(i)+b(i) ; end do!$omp END PARALLEL 3OpenMP*!$OMP PARALLEL!$OMP DO SCHEDULE(STATIC) do i = 1, n ; a(i) = a(i) + b(i) ; end do!$omp PARALLEL DO SCHEDULE(static) do i = 1, n ; a(i) = a(i) + b(i) ; end do.11 OpenMP* 1 DO 2 OpenMP* 3 DO DO DO!$OMP DO [clause [[,] clause]..] DO [!$OMP [NOWAIT]] 16

18 Fork!$OMP PARALLEL DO [clause[[,]clause]] DO I = 1, N DO!$OMP END PARALLEL DO!$OMP PARALLEL DO [clause[[,]clause]] DO I = 1, N DO!$OMP END PARALLEL DO ForkJoin Join.12 DO OpenMP* PROGRAM OMP!$ USE OMP_LIB REAL*8 A(100) CALL OMP_SET_NUM_THREADS(4)!$OMP PARALLEL DO DO I = 1, 100 CALL POOH(I,A)!$OMP END PARALLEL DO WRITE (*,*) 'all done' END SUBROUTINE POOH(I,A) REAL*8 A(*) A(I) = 1.0 RETURN END / ADO REAL*8 A(100) OMP_SET_NUM_THREADS(4) POOH(I,A) POOH(I,A) POOH(I,A) POOH(I,A) I=1,2,3.. I=26,27,.. I=51,52.. i=76,77.. WRITE (*,*) 'all done'.13 DO DO SCHEDULE!$OMP DO SCHEDULE(type[,chunk]) [clause [[,] clause]..] DO [!$OMP [NOWAIT]] 17

19 TYPE STATIC chunk chunk DYNAMIC chunk chunk1 GUIDED guidedchunk chunk chunk1 RUNTIME typechunk OMP_SCHEDULEOMP_SCHEDULE SCHEDULESTATIC.2 4 SCHEDULESTATIC,6 4 SCHEDULESTATIC, DO SCHEDULESTATIC,6 4 DO WHILE NOWAIT DO NOWAIT DO 18

20 SECTIONS SECTIONS Fortran 90 SELECT SECTIONS SECTION!$OMP SECTIONS [clause [[,] clause]..] [!$OMP SECTION] [!$OMP SECTION!$OMP END SECTIONS [NOWAIT] NOWAIT SECTIONS NOWAIT SECTIONS SECTION SECTIONS/END SECTIONS SECTION SECTION!$OMP PARALLEL [clause[[,]clause]]!$omp SECTIONS!$OMP SECTION idle idle!$omp SECTION.15 SECTIONS SECTION SECTION SINGLE SINGLE SINGLE NOWAIT END SINGLE!$OMP SINGLE [clause [[,] clause]..]!$omp END SINGLE [NOWAIT] 19

21 !$OMP PARALLEL!$OMP SINGLE idle idle idle idle!$omp END SINGLE!$OMP END PARALLEL.16 SINGLE SINGLE SECTIONS SECTION WORKSHARE WORKSHARE Fortran FORALLWHERE MATMUL WORKSHARE!$OMP WORKSHARE!$OMP END WORKSHARE [NOWAIT] WORKSHARE SCHEDULE REAL A(100,200), B(100,200), C(100,200)...!$OMP PARALLEL!$OMP WORKSHARE A=B+C!$OMP END WORKSHARE!$OMP END PARALLEL OpenMP* PARALLEL PARALLEL DO PARALLEL SECTIONS PARALLEL WORKSHARE PARALLEL 20

22 5.3 OpenMP* clauseopenmp* OpenMP* PARALLEL DO SECTIONS WORKSHARE SINGLE PARALLEL DO PARALLEL SECTIONS PARALLEL WORKSHAPE IF SCHEDULE PRIVATE SHARE DEFAULT FIRSTPRIVATE LASTPRIVATE COPYIN COPYPRIVATE REDUCTION ORDERED NOWAIT NUM_THREADS.3 OpenMP* SHARED PRIVATE COMMON OpenMP* MODULE THREADPRIVATE SAVE COMMON MODULE OpenMP* OpenMP* PRIVATE OpenMP* SHARED.4 21

23 main.f PROGRAM OMP COMMON /WORK/ A(10) REAL*8 A INTEGER INDEX(10)!$OMP PARALLEL CALL SORT(INDEX)!$OMP END PARALLEL WRITE (6,*) INDEX(1) STOP END PROGRAM OMP sort.f A INDEX COUNT TEMP A, INDEX, COUNT SUBROUTINE SORT(INDEX) COMMON /WORK/ A(10) REAL*8 A INTEGER INDEX(*) REAL*8 TEMP(10) INTEGER COUNT SAVE COUNT.. RETURN END TEMP A, INDEX, COUNT TEMP TEMP.17 THREADPRIVATE COMMON COMMON COMMON MASTER THREADPRIVATE COMMON COMMON THREADPRIVATE COPYIN PARALLEL COMMON DATA 1 THREADPRIVATE!$OMP THREADPRIVATElist list COMMON / 22

24 COMMON /JUNK/ NX COMMON /STUFF/ A,B,C!$OMP THREADPRIVATE (/JUNK/,/STUFF/) NX = 32 C = !$OMP PARALLEL PRIVATE(NX2,CSQ) COPYIN(/JUNK/,C) NX2 = NX * 2 CSQ = C*C... THREADPRIVATE THREADPRIVATE PRIVATElist DO PRIVATE list PRIVATE Fork DO FORALL OpenMP* PRIVATE SHAREDlist OpenMP* SHARED DEFAULTSHAREDPRIVATENONE DEFAULT PRIVATE SHARED DEFAULTPRIVATETHREADPRIVATE COMMON COMMON PRIVATE DEFAULTSHARED default DEFAULTSHARED DEFAULTNONEPRIVATE SHARED PRIVATESHAREDFIRSTPRIVATE LASTPRIVATE REDUCTION PRIVATESHAREDFIRSTPRIVATELASTPRIVATEREDUCTION 23

25 !$OMP PARALLEL DO DEFAULT(PRIVATE),!$OMP& FIRSTPRIVATE(I), SHARED(X),!$OMP& SHARED(R), LASTPRIVATE(I) FIRSTPRIVATElist FIRSTPRIVATE PRIVATE list PRIVATE PRIVATE LASTPRIVATElist PRIVATE LASTPRIVATE PRIVATE list PRIVATE SECTION SECTION PROGRAM TEST IS = 0 C$OMP PARALLEL DO FIRSTPRIVATE(IS) C$OMP+ LASTPRIVATE(IS) DO J=1,1000 IS = IS + J 1000 CONTINUE print *, IS is=0 print is j=1000 COPYINlist COPYIN THREADPRIVATE FIRSTPRIVATE list THREADPRIVATE 24

26 COPYPRIVATElist COPYPRIVATE PRIVATE!$OMP PARALLEL PRIVATE(A,B)...!$OMP SINGLE READ(24) A!$OMP END SINGLE COPYPRIVATE(A) B = A*A...!$OMP END PARALLEL 25

27 5.4 MASTERCRITICALBARRIERATOMICFLUSHORDERED 6 MASTER MASTER / END MASTER!$OMP MASTER!$OMP END MASTER MASTER CRITICAL 1 CRITICAL!$OMP CRITICAL [(name)]!$omp END CRITICAL!$OMP PARALLEL!$OMP DO DO I = 1, N!$OMP!$OMP CRITICAL!$OMP END CRITICAL!$OMP DO DO I = 1, N!$OMP!$OMP END PARALLEL.18 26

28 BARRIER BARRIER!$OMP BARRIER ATOMIC ATOMIC!$OMP ATOMIC Expression-statement 1 x = x op expr, x = expr op x, x = intr (x, expr), x = intr(expr, x) x expr x INTR MAX, MIN, IAND, IOR, IEOR op +, *, -, /,.AND.,.OR.,.EQV.,.NEQV.!$OMP PARALLEL!$OMP DO DO I = 1, N!$OMP!$OMP ATOMIC!$OMP DO DO I = 1, N!$OMP!$OMP END PARALLEL x ATOMIC ATOMIC CRITICAL ATOMIC CRITICAL.19 ATOMIC 27

29 C$OMP PARALLEL PRIVATE(B) B = DOIT(I) C$OMP ATOMIC X = X + FOO(B) C$OMP END PARALLEL!$OMP PARALLEL PRIVATE(B, tmp) B = DOIT(I) TMP = FOO(B)!$OMP CRITICAL X = X + tmp!$omp END CRITICAL!$OMP END PARALLEL 2 X ATOMIC CRITICAL foo 2 FLUSH flush!$omp FLUSH [(list)] list, COMMON SAVE SAVE SHARED list,list ORDERD ORDERD / END ORDERED!$OMP ORDERED!$OMP END ORDERED ORDERED DO PARALLEL DO DO ORDERED 28

30 REDUCTION REDUCTION DO REDUCTION REDUCTION( {op intr }: list ) list SHARED list intr MAX, MIN, IAND, IOR, IEOR op +, *, -, /,.AND.,.OR.,.EQV.,.NEQV. REDUCTION op REDUCTION NOWAIT REDUCTION BARRIER REDUCTION ASUM = 0.0 APROD = 1.0!$OMP PARALLEL DO REDUCTION(+:ASUM) REDUCTION(*:APROD) DO I = 1, n ASUM = ASUM + A(I) APROD = APROD * A(I)!$OMP END PARALLEL DO + 0.OR. 0 * 1 MAX 1-0 MIN 0.AND. 1 // 1.5 REDUCTION REDUCTION REDUCTION!$OMP DO REDUCTION(+: A, Y) REDUCTION(.OR.: AM) 29

31 5.5 IF IF PARALLEL IF TRUE NUM_THREADS NUM_THREADS PARALLEL NOWAIT OpenMP* END SECTIONSEND SINGLE NOWAIT!$OMP PARALLEL!$OMP DO DO I = 1, N!$OMP!$OMP CRITICAL!$OMP END CRITICAL!$OMP DO DO I = 1, N!$OMP!$OMP END PARALLEL.20 NOWAIT NOWAIT!$OMP DO DO J=1,N A(J) = B(J) + C(J)!$OMP DO DO J=1,N D(J) = E(J) * f!$omp DO DO J=1,n Z(J) = (A(J)+A(J+1)) *

32 5.6 / OpenMP* OMP_NUM_THREADS OMP_SCHEDULE OMP_DYNAMIC OMP_NESTED setenv OMP_NUM_THREADS 4Linux* set OMP_NUM_THREADS=4Windows* type[,chunk] type static/dynamic/guided chunk chunk static 1 static chunk setenv OMP_SCHEDULE "DYNAMIC"Linux* set OMP_SCHEDULE=DYNAMICWindows* true false setenv OMP_DYNAMIC TRUELinux* set OMP_DYNAMIC=TRUEWindows* true false false setenv OMP_NESTED TRUELinux* set OMP_NESTED=TRUEWindows*.6 OpenMP* 31

33 6. OpenMP* OpenMP* OpenMP* OpenMP* OpenMP* OpenMP* OpenMP*.21 OpenMP* OpenMP* 1. VTune TM 2. OpenMP* 3. MKL OpenMP* 32

34 7. OpenMP* 33

35 OpenMP* OpenMP* OpenMP* md.f OpenMP* OpenMP* SGI Altix / Itanium GHz $ IFORT -O3 -PARALLEL -PAR_THRESHOLD0 md.f md.f(35) : (COL. 6) REMARK: LOOP WAS AUTO-PARALLELIZED. md.f(98) : (COL. 8) REMARK: LOOP WAS AUTO-PARALLELIZED. md.f(161) : (COL. 6) REMARK: LOOP WAS AUTO-PARALLELIZED. md.f(181) : (COL. 6) REMARK: LOOP WAS AUTO-PARALLELIZED. md.f(212) : (COL. 6) REMARK: LOOP WAS AUTO-PARALLELIZED. dist 34

36 96 DO I=1,NP 97! COMPUTE POTENTIAL ENERGY AND FORCES 98 F(1:ND,I) = DO J=1,NP 100 IF (I.NE. J) THEN 101 CALL DIST(ND,BOX,POS(1,I),POS(1,J),RIJ,D) 102! ATTRIBUTE HALF OF THE POTENTIAL ENERGY TO PARTICLE 'J' 103 POT = POT + 0.5*V(D) 104 DO K=1,ND 105 F(K,I) = F(K,I) - RIJ(K)*DV(D)/D 106 ENDDO 107 ENDIF 108 ENDDO 109! COMPUTE KINETIC ENERGY 110 KIN = KIN + DOTR8(ND,VEL(1,I),VEL(1,I)) 111 ENDDO 148 SUBROUTINE DIST(ND,BOX,R1,R2,DR,D) 160 D = DO I=1,ND 162 DR(I) = R1(I) - R2(I) 163 D = D + DR(i)** ENDDO 165 D = SQRT(D) RETURN 168 END 35

37 OpenMP*!$OMP PARALLEL DO!$OMP& DEFAULT(SHARED)!$OMP& PRIVATE(I,J,K,RIJ,D)!$OMP& REDUCTION(+ : POT, KIN) DO I=1,NP! COMPUTE POTENTIAL ENERGY AND FORCES F(1:ND,I) = 0.0 DO J=1,NP IF (I.NE. J) THEN CALL DIST(ND,BOX,POS(1,I),POS(1,J),RIJ,D)! ATTRIBUTE HALF OF THE POTENTIAL ENERGY TO PARTICLE 'J' POT = POT + 0.5*V(D) DO K=1,ND F(K,I) = F(K,I) - RIJ(K)*DV(D)/D ENDDO ENDIF ENDDO! COMPUTE KINETIC ENERGY KIN = KIN + DOTR8(ND,VEL(1,I),VEL(1,I)) ENDDO!$OMP END PARALLEL DO KIN = KIN*0.5*MASS OpenMP* 36

38 OpenMP* PROGRAM OMP INTEGER NUM_STEPS REAL*8 STEP,X,PI,SUM NUM_STEPS = STEP = 1.0D0 / DBLE(NUM_STEPS) SUM = 0.0D0!$OMP PARALLEL DO REDUCTION(+:SUM) PRIVATE(X) DO I = 1, NUM_STEPS X = (DBLE(I)-0.5D0)*STEP SUM = SUM + 4.0D0 / (1.0D0 + X * X) PI = STEP * SUM WRITE (6,*) 'PI = ',PI END PROGRAM OMP PROGRAM OMP!$ USE OMP_LIB PARAMETER (NUM_THREADS=4) INTEGER NUM_STEPS REAL*8 STEP,X,PI,SUM(0:NUM_THREADS-1) NUM_STEPS = STEP = 1.0D0 / DBLE(NUM_STEPS) CALL OMP_SET_NUM_THREADS(NUM_THREADS)!$OMP PARALLEL PRIVATE(X) ID = OMP_GET_THREAD_NUM() SUM(ID) = 0.0D0 ISTART = ID * NUM_STEPS / NUM_THREADS + 1 IEND = (ID+1) * NUM_STEPS / NUM_THREADS DO I = ISTART, IEND X = (DBLE(I)-0.5D0)*STEP SUM(ID) = SUM(ID) + 4.0D0 / (1.0D0 + X * X)!$OMP END PARALLEL PI = 0.0D0 DO I = 0, NUM_THREADS-1 PI = PI + STEP * SUM(I) WRITE (6,*) 'PI = ',PI END PROGRAM OMP 37

39 HPC IntelIntel PentiumXeonItaniumVTune Intel Corporation * 2006 Intel Corporation J-001 JPN/0605/PDF/SE/DEG/KS

02_C-C++_osx.indd

02_C-C++_osx.indd C/C++ OpenMP* / 2 C/C++ OpenMP* OpenMP* 9.0 1... 2 2... 3 3OpenMP*... 5 3.1... 5 3.2 OpenMP*... 6 3.3 OpenMP*... 8 4OpenMP*... 9 4.1... 9 4.2 OpenMP*... 9 4.3 OpenMP*... 10 4.4... 10 5OpenMP*... 11 5.1