2011, Oracle / U.S. GOVERNMENT END USERS: Oracle programs, including any operating system, integrated software, any programs installed on the hardware

Size: px

Start display at page:

Download "2011, Oracle / U.S. GOVERNMENT END USERS: Oracle programs, including any operating system, integrated software, any programs installed on the hardware"

顺毕
5 years ago
Views:

1 Oracle Solaris Studio 12.3 OpenMP API E E

2 2011, Oracle / U.S. GOVERNMENT END USERS: Oracle programs, including any operating system, integrated software, any programs installed on the hardware, and/or documentation, delivered to U.S. Government end users are commercial computer software pursuant to the applicable Federal Acquisition Regulation and agency-specific supplemental regulations. As such, use, duplication, disclosure, modification, and adaptation of the programs, including any operating system, integrated software, any programs installed on the hardware, and/or documentation, shall be subject to license terms and license restrictions applicable to the programs. No other rights are granted to the U.S. Government. Oracle Corporation Oracle Java Oracle / Intel Intel Xeon Intel Corporation SPARC SPARC International, Inc AMD Opteron AMD AMD Opteron Advanced Micro Devices UNIX The Open Group Oracle Corporation Oracle Corporation @25097

3 OpenMP API OpenMP OpenMP OpenMP OpenMP Oracle Solaris Studio ID SUNW_MP_PROCBIND OS OpenMP SECTIONS SINGLE

4 3.6.3 ATOMIC omp_set_num_threads() omp_set_schedule() omp_set_max_active_levels() omp_get_max_active_levels() Fortran THREADPRIVATE SHARED OMP_NESTED OMP_THREAD_LIMIT OMP_MAX_ACTIVE_LEVELS OpenMP THREADPRIVATE auto default( auto) Oracle Solaris Studio 12.3 OpenMP API E

5 Oracle Solaris OS A

6 6

7 Oracle Solaris Studio 12.3 C C++ Fortran OpenMP API Oracle Solaris Studio Oracle Solaris SPARC Oracle Solaris Linux x86 x86 "x86" x86 "x64" 64 x86 CPU 32 x86 x86 32 SPARC x86 Linux Linux x86 Oracle Solaris Oracle Solaris Oracle Solaris Studio 12.3 Oracle Solaris Studio Oracle Solaris Studio Oracle Solaris Studio Web dbxtool DLight IDE "Help" F1 "Help" 7

8 Web URL Oracle Web Oracle Oracle Oracle Oracle Solaris Studio Oracle Web Oracle My Oracle Support pls/topic/lookup?ctx=acc&id=trs P 1 AaBbCc123.login ls -a machine_name% you have mail. AaBbCc123 machine_name% su Password: 8 Oracle Solaris Studio 12.3 OpenMP API E

9 P 1 aabbcc123 rm filename AaBbCc123 Class 6 shell Oracle Solaris OS UNIX shell Oracle Solaris P 2 shell shell Bash shell Korn shell Bourne shell $ Bash shell Korn shell Bourne shell # C shell C shell machine_name% machine_name# 9

10 10

11 1 1 OpenMP API OpenMP (Application Program Interface, API) OpenMP OpenMP API Oracle Solaris Oracle Solaris Studio 1.1 OpenMP OpenMP 3.1 API Oracle Solaris Studio OpenMP Web OpenMP compunity Web Oracle Solaris Studio OpenMP API Oracle Solaris Studio solarisstudio 1.2 Fortran Fortran C/C++ [...] "Fortran" Fortran 95 Oracle Solaris Studio f95(1) 11

12 1.2 "pragma" OpenMP API C C++ Fortran 12 Oracle Solaris Studio 12.3 OpenMP API E

13 2 OpenMP OpenMP API Oracle Solaris Studio 12.3 OpenMP 2 1 OMP_NUM_THREADS omp_set_num_threads() PARALLEL num_threads 2.1 OpenMP cc CC f95 -xopenmp f95 -xopenmp -openmp -xopenmp -xopenmp=parallel -xopenmp=noopt OpenMP pragma -xopenmp=parallel -xo3 -xo3 OpenMP pragma -xo3 -xo3 -xo2 -openmp=noopt -openmp=noopt OpenMP pragma 13

14 2.1 -xopenmp=stubs -xopenmp=none OpenMP OpenMP OpenMP OpenMP pragma -xopenmp libompstubs.a % cc omp_ignore.c -lompstubs libompstubs.a OpenMP libmtsk.so OpenMP pragma Oracle Solaris OpenMP libmtsk.so xopenmp xopenmp=none OpenMP pragma xopenmp xopenmp=parallel -xopenmp=parallel noopt _OPENMP YYYYMM C/C L Fortran dbx OpenMP -xopenmp=noopt -g -xopenmp Fortran 95 -xopenmp -xopenmp=parallel -xopenmp=noopt -stackvar OpenMP -xopenmp xopenmp, xopenmp=parallel xopenmp=noopt xvpara OpenMP 6 7 OpenMP OpenMP OpenMP shared private 14 Oracle Solaris Studio 12.3 OpenMP API E

15 2.2 OpenMP 2.2 OpenMP OpenMP OpenMP OpenMP API 3.1 Oracle Solaris Studio OpenMP Oracle Solaris Studio OpenMP Oracle Solaris Studio OpenMP OpenMP OMP_SCHEDULE STATIC OMP_NUM_THREADS OMP_DYNAMIC OMP_NESTED OMP_STACKSIZE OMP_WAIT_POLICY OMP_MAX_ACTIVE_LEVELS OMP_THREAD_LIMIT OMP_PROC_BIND setenv OMP_SCHEDULE GUIDED,4 2 setenv OMP_NUM_THREADS 16 TRUE setenv OMP_DYNAMIC FALSE FALSE setenv OMP_NESTED FALSE 32 4MB 64 8MB setenv OMP_STACKSIZE 10M PASSIVE FALSE Oracle Solaris Studio OpenMP OpenMP 2 OpenMP 15

16 2.2 OpenMP PARALLEL PARALLEL OMP_NUM_THREADS PARALLEL OMP_NUM_THREADS SUNW_MP_WARN OpenMP SUNW_MP_WARN TRUE stderr FALSE FALSE setenv SUNW_MP_WARN TRUE int sunw_mp_register_warn (void (*func)(void *)); sunw_mp_register_warn() 0 1 OpenMP OpenMP SUNW_MP_WARN TRUE OpenMP SUNW_MP_THR_IDLE OpenMP SPIN SLEEP SLEEP(times) SLEEP(timems) SLEEP(timemc) time s ms mc SPIN SLEEP SLEEP SLEEP SLEEP(0) SLEEP(0s) SLEEP(0ms) SLEEP(0mc) 16 Oracle Solaris Studio 12.3 OpenMP API E

17 2.2 OpenMP SUNW_MP_THR_IDLE OMP_WAIT_POLICY OMP_WAIT_POLICY setenv SUNW_MP_THR_IDLE SPIN setenv SUNW_MP_THR_IDLE SLEEP setenv SUNW_MP_THR_IDLE SLEEP(2s) setenv SUNW_MP_THR_IDLE SLEEP(20ms) setenv SUNW_MP_THR_IDLE SLEEP(150mc) SUNW_MP_PROCBIND Oracle Solaris Linux SUNW_MP_PROCBIND OpenMP SUNW_MP_PROCBIND OMP_PROC_BIND SUNW_MP_MAX_POOL_THREADS OpenMP SUNW_MP_MAX_POOL_THREADS OpenMP OMP_THREAD_LIMIT OpenMP SUNW_MP_MAX_POOL_THREADS OMP_THREAD_LIMIT OMP_THREAD_LIMIT SUNW_MP_MAX_POOL_THREADS SUNW_MP_MAX_NESTED_LEVELS STACKSIZE SUNW_MP_MAX_NESTED_LEVELS SUNW_MP_MAX_NESTED_LEVELS OMP_MAX_ACTIVE_LEVELS 32 SPARC V8 x86 4MB 64 SPARC V9 x86 8MB 2 OpenMP 17

18 2.2 OpenMP STACKSIZE B, K, M G setenv STACKSIZE 8192 // sets the thread stack size to 8 Megabytes setenv STACKSIZE 16M // sets the thread stack size to 16 Megabytes STACKSIZE OMP_STACKSIZE SUNW_MP_GUIDED_WEIGHT GUIDED GUIDED SUNW_MP_WAIT_POLICY CPU csh setenv SUNW_MP_WAIT_POLICY IDLE=val :BARRIER=val:TASKWAIT= val IDLE=val BARRIER= val TASKWAIT= val val SPIN YIELD SLEEP SPIN(time) CPU time s ms mc SPIN YIELD(number) CPU CPU YIELD number SLEEP SPIN SLEEP YIELD "SPIN(0),YIELD(0)" SLEEP IDLE BARRIER TASKWAIT SUNW_MP_WAIT_POLICY OMP_WAIT_POLICY OMP_WAIT_POLICY 18 Oracle Solaris Studio 12.3 OpenMP API E

19 2.3 % setenv SUNW_MP_WAIT_POLICY BARRIER=SPIN % setenv SUNW_MP_WAIT_POLICY IDLE=SPIN(10ms),YIELD(5) 10 CPU 5 % setenv SUNW_MP_WAIT_POLICY \ IDLE=SPIN(10ms),YIELD(2):BARRIER=SLEEP:TASKWAIT=YIELD(10) 10 CPU 2 CPU n n SPARC T3 Oracle Solaris 64 Oracle Solaris psrinfo(1m) Linux /proc/cpuinfo SUNW_MP_PROCBIND OpenMP SUNW_MP_PROCBIND "TRUE" "FALSE" "COMPACT" "SCATTER" "true" "false" "compact" "scatter" % setenv SUNW_MP_PROCBIND "TRUE" 2 OpenMP 19

20 2.3 % setenv SUNW_MP_PROCBIND "2" % setenv SUNW_MP_PROCBIND "2 246" n1 n2 ("-") n1 n2 % setenv SUNW_MP_PROCBIND "0-6" (ID) ID ID ID ID Oracle Solaris psrinfo(1m) ID prtdiag(1m) psrinfo -pv ID 8 UltraSPARC IV 16 Sun Fire 4810 ID ID 0 n ID n-1 psrinfo(1m) SUNW_MP_PROCBIND SUNW_MP_PROCBIND SUNW_MP_PROCBIND SUNW_MP_PROCBIND FALSE SUNW_MP_PROCBIND TRUE SUNW_MP_PROCBIND COMPACT COMPACT SUNW_MP_PROCBIND SCATTER SCATTER COMPACT 20 Oracle Solaris Studio 12.3 OpenMP API E

21 2.4 SUNW_MP_PROCBIND ID ID ID n-1 ID 0 SUNW_MP_PROCBIND ID ID ID SUNW_MP_PROCBIND ("-") ID ID ID SUNW_MP_PROCBIND ID OpenMP OS Oracle Solaris psrset Linux taskset SUNW_MP_PROCBIND SUNW_MP_PROCBIND Linux SUNW_MP_PROCBIND Oracle Solaris 2.4 f95 -stackvar Fortran OpenMP -stackvar -stackvar Fortran limit C-shell ulimit Bourne Korn shell 8MB OpenMP PRIVATE 2 OpenMP 21

22 2.5 OpenMP 32 SPARC V8 x86 4MB 64 SPARC V9 x86 8MB OMP_STACKSIZE demo% setenv OMP_STACKSIZE <-Set thread stack size to 16 Mb (C shell) demo$ OMP_STACKSIZE=16384 demo$ export OMP_STACKSIZE <-Set thread stack size to 16 Mb (Bourne/Korn shell) -xcheck=stkovf Fortran C C OpenMP Oracle Solaris Studio OpenMP dbx dbx(1) tha(1) libtha(3) OpenMP collect(1) analyzer(1) 22 Oracle Solaris Studio 12.3 OpenMP API E

23 3 Oracle Solaris Studio OpenMP OpenMP 3.1 API E (untied) (tied) (untied) OpenMP (taskwait) (taskyield) (untied) load/store load/store 23

24 nthreads-var nthreads-var 1 dyn-var dyn-var TRUE run-sched-var run-sched-var static def-sched-var def-sched-var static bind-var bind-var FALSE stacksize-var OpenMP stacksize-var 4 MB 32 8MB 64 wait-policy-var wait-policy-var PASSIVE thread-limit-var OpenMP thread-limit-var 1024 max-active-levels-var max-active-levels-var OMP_DYNAMIC FALSE omp_set_dynamic() OpenMP SUNW_MP_WARN TRUE sunw_mp_register_warn() 3.5 (collapsed) long run-sched-var auto schedule(runtime) static 24 Oracle Solaris Studio 12.3 OpenMP API E

25 SECTIONS SINGLE ATOMIC 3.7 sections single critical atomic omp_set_num_threads() omp_set_num_threads() SUNW_MP_WARN TRUE sunw_mp_register_warn() omp_set_schedule() Oracle Solaris Studio sunw_mp_sched_reserved static omp_set_max_active_levels() omp_set_max_active_levels() SUNW_MP_WARN TRUE sunw_mp_register_warn() 3 25

26 3.8 omp_set_max_active_levels() SUNW_MP_WARN TRUE sunw_mp_register_warn() omp_get_max_active_levels() omp_get_max_active_levels() max-active-levels-var 3.8 OMP_SCHEDULE OMP_SCHEDULE static dynamic guided auto static SUNW_MP_WARN TRUE sunw_mp_register_warn() OMP_SCHEDULE static dynamic guided static dynamic guided 1 SUNW_MP_WARN TRUE sunw_mp_register_warn() OMP_NUM_THREADS SUNW_MP_WARN TRUE sunw_mp_register_warn() SUNW_MP_WARN TRUE sunw_mp_register_warn() OMP_PROC_BIND OMP_DYNAMIC OMP_NESTED OMP_PROC_BIND TRUE FALSE OMP_DYNAMIC TRUE FALSE TRUE SUNW_MP_WARN TRUE sunw_mp_register_warn() OMP_NESTED TRUE FALSE FALSE SUNW_MP_WARN TRUE sunw_mp_register_warn() 26 Oracle Solaris Studio 12.3 OpenMP API E

27 3.9 Fortran OMP_STACKSIZE OMP_WAIT_POLICY OMP_STACKSIZE 32 4MB 64 8MB SUNW_MP_WARN TRUE sunw_mp_register_warn() ACTIVE PASSIVE OMP_MAX_ACTIVE_LEVELS OMP_MAX_ACTIVE_LEVELS (4) SUNW_MP_WARN TRUE sunw_mp_register_warn() OMP_THREAD_LIMIT OMP_THREAD_LIMIT (1024) SUNW_MP_WARN TRUE sunw_mp_register_warn() 3.9 Fortran Fortran THREADPRIVATE (thread-private) SHARED OpenMP 3.1 shared 3 27

28 3.9 Fortran omp_lib.h omp_lib Oracle Solaris OpenMP Fortran KIND 28 Oracle Solaris Studio 12.3 OpenMP API E

29 4 4.1 OpenMP OpenMP fork-join - OpenMP OpenMP OMP_NESTED OMP_NESTED omp_set_nested() 29

30 #include <omp.h> #include <stdio.h> void report_num_threads(int level) #pragma omp single printf("level %d: number of threads in the team - %d\n", level, omp_get_num_threads()); int main() omp_set_dynamic(0); #pragma omp parallel num_threads(2) report_num_threads(1); #pragma omp parallel num_threads(2) report_num_threads(2); #pragma omp parallel num_threads(2) report_num_threads(3); return(0); % setenv OMP_NESTED TRUE % a.out Level 1: number of threads in the team - 2 Level 2: number of threads in the team - 2 Level 2: number of threads in the team - 2 Level 3: number of threads in the team - 2 Level 3: number of threads in the team - 2 Level 3: number of threads in the team - 2 Level 3: number of threads in the team - 2 % setenv OMP_NESTED FALSE % a.out Level 1: number of threads in the team - 2 Level 2: number of threads in the team - 1 Level 3: number of threads in the team - 1 Level 2: number of threads in the team - 1 Level 3: number of threads in the team Oracle Solaris Studio 12.3 OpenMP API E

31 OMP_THREAD_LIMIT OpenMP OMP_THREAD_LIMIT 1023 OMP_THREAD_LIMIT 1 SUNW_MP_MAX_POOL_THREADS OMP_THREAD_LIMIT 6 SUNW_MP_MAX_POOL_THREADS 5 5 % setenv OMP_NESTED TRUE % OMP_THREAD_LIMIT 6 % a.out Level 1: number of threads in the team - 2 Level 2: number of threads in the team - 2 Level 2: number of threads in the team - 2 Level 3: number of threads in the team - 2 Level 3: number of threads in the team - 2 Level 3: number of threads in the team - 1 Level 3: number of threads in the team OMP_MAX_ACTIVE_LEVELS OMP_MAX_ACTIVE_LEVELS 4 OMP_NESTED TRUE true omp_set_nested() 4 OMP_MAX_ACTIVE_LEVELS #include <omp.h> #include <stdio.h> #define DEPTH 5 void report_num_threads(int level) 4 31

32 4.2 #pragma omp single printf("level %d: number of threads in the team - %d\n", level, omp_get_num_threads()); void nested(int depth) if (depth == DEPTH) return; #pragma omp parallel num_threads(2) report_num_threads(depth); nested(depth+1); int main() omp_set_dynamic(0); omp_set_nested(1); nested(1); return(0); 4 OS % setenv OMP_MAX_ACTIVE_LEVELS 4 % a.out sort Level 1: number of threads in the team - 2 Level 2: number of threads in the team - 2 Level 2: number of threads in the team - 2 Level 3: number of threads in the team - 2 Level 3: number of threads in the team - 2 Level 3: number of threads in the team - 2 Level 3: number of threads in the team - 2 Level 4: number of threads in the team - 2 Level 4: number of threads in the team - 2 Level 4: number of threads in the team - 2 Level 4: number of threads in the team - 2 Level 4: number of threads in the team - 2 Level 4: number of threads in the team - 2 Level 4: number of threads in the team - 2 Level 4: number of threads in the team % setenv OMP_MAX_ACTIVE_LEVELS 2 % a.out sort Level 1: number of threads in the team - 2 Level 2: number of threads in the team - 2 Level 2: number of threads in the team - 2 Level 3: number of threads in the team - 1 Level 3: number of threads in the team - 1 Level 3: number of threads in the team - 1 Level 3: number of threads in the team Oracle Solaris Studio 12.3 OpenMP API E

33 4.3 OpenMP Level 4: number of threads in the team - 1 Level 4: number of threads in the team - 1 Level 4: number of threads in the team - 1 Level 4: number of threads in the team - 1 OS 4.3 OpenMP OpenMP - omp_set_num_threads() - omp_get_max_threads() - omp_set_dynamic() - omp_get_dynamic() - omp_set_nested() - omp_get_nested() set get 4 2 OpenMP #include <stdio.h> #include <omp.h> int main() omp_set_nested(1); omp_set_dynamic(0); #pragma omp parallel num_threads(2) if (omp_get_thread_num() == 0) omp_set_num_threads(4); /* line A */ else omp_set_num_threads(6); /* line B */ /* The following statement will print out * *0:24 *1:26 * * omp_get_num_threads() returns the number * of the threads in the team, so it is * the same for the two threads in the team. */ printf("%d: %d %d\n", omp_get_thread_num(), omp_get_num_threads(), omp_get_max_threads()); 4 33

34 4.3 OpenMP 4 2 OpenMP /* Two inner parallel regions will be created * one with a team of 4 threads, and the other * with a team of 6 threads. */ #pragma omp parallel #pragma omp master /* The following statement will print out * * Inner: 4 * Inner: 6 */ printf("inner: %d\n", omp_get_num_threads()); omp_set_num_threads(7); /* line C */ /* Again two inner parallel regions will be created, * one with a team of 4 threads, and the other * with a team of 6 threads. * * The omp_set_num_threads(7) call at line C * has no effect here, since it affects only * parallel regions at the same or inner nesting * level as line C. */ #pragma omp parallel printf("count me.\n"); return(0); % a.out 0:24 Inner: 4 1:26 Inner: 6 count me. count me. count me. count me. count me. count me. count me. count me. count me. count me. 34 Oracle Solaris Studio 12.3 OpenMP API E

35 CPU CPU CPU OMP_THREAD_LIMIT OMP_MAX_ACTIVE_LEVELS CPU CPU 4 35

36 36

37 5 OpenMP 5.1 while OpenMP task task (tied) (untied) (tied) untied task (untied) (tied) (untied) OpenMP (tied) (taskwait) (taskyield) 37

38 5.1 Oracle Solaris Studio (untied) task OpenMP parallel (tied) parallel OpenMP 3.1 (undeferred) (undeferred) if false (undeferred) (undeferred) if (included) (final) (included) (included) (undeferred) (undeferred) (undeferred) (undeferred) (undeferred) (undeferred) (included) (included) (included) (included) (merged) mergeable task (undeferred) (included) (merged) (merged) task (final) (final) (included) final task final TRUE (final) (final) (final) (included) 38 Oracle Solaris Studio 12.3 OpenMP API E

39 task default (private firstprivate shared none) private (list) firstprivate (list) shared (list) shared task private firstprivate task firstprivate OpenMP OpenMP C/C++ OpenMP task taskwait parallel single fib(n) print fib(n) task fib(n-1) fib(n-2) fib(n) fib(n-1) fib(n-2) fib() 2 task final final (n <= THRESHOLD) true final (final) task (included) (final) n = 9, 8,..., 2 fib (included) (included) 5 39

40 5.3 taskwait fib() i j fib() single fib(n) Oracle Solaris Studio C #include <stdio.h> #include <omp.h> #define THRESHOLD 5 int fib(int n) int i, j; if (n<2) return n; #pragma omp task shared(i) firstprivate(n) final(n <= THRESHOLD) i=fib(n-1); #pragma omp task shared(j) firstprivate(n) final(n <= THRESHOLD) j=fib(n-2); #pragma omp taskwait return i+j; int main() int n = 30; omp_set_dynamic(0); omp_set_num_threads(4); #pragma omp parallel shared(n) #pragma omp single printf ("fib(%d) = %d\n", n, fib(n)); % CC -xopenmp -xo3 task_example.cc % a.out fib(30) = Oracle Solaris Studio 12.3 OpenMP API E

41 OpenMP THREADPRIVATE threadprivate (tied) threadprivate threadprivate (untied) threadprivate OpenMP critical OpenMP 3.1 A OpenMP 2.5 lck OpenMP 3.1 lck 5 2 OpenMP 3.0 #include <stdlib.h> #include <stdio.h> #include <omp.h> int main() int x; omp_lock_t lck; omp_init_lock (&lck); omp_set_lock (&lck); x=0; 5 41

42 OpenMP 3.0 #pragma omp parallel shared (x) #pragma omp master x=x+1; omp_unset_lock (&lck); omp_destroy_lock (&lck); task i shared work() i work() main() i work() task taskwait task i firstprivate shared 5 3 #include <stdio.h> #include <omp.h> void work() int i; i = 10; #pragma omp task shared(i) #pragma omp critical printf("in Task, i = %d\n",i); int main(int argc, char** argv) 42 Oracle Solaris Studio 12.3 OpenMP API E

43 omp_set_num_threads(8); omp_set_dynamic(0); #pragma omp parallel work(); 5 4 #include <stdio.h> #include <omp.h> void work() int i; i = 10; #pragma omp task shared(i) #pragma omp critical printf("in Task, i = %d\n",i); /* Use TASKWAIT for synchronization. */ #pragma omp taskwait int main(int argc, char** argv) omp_set_num_threads(8); omp_set_dynamic(0); #pragma omp parallel work(); task j sections j sections j firstprivate Oracle Solaris Studio sections sections sections j sections task taskwait task j firstprivate shared 5 43

44 #include <stdio.h> #include <omp.h> int main(int argc, char** argv) omp_set_num_threads(2); omp_set_dynamic(0); int j=100; #pragma omp parallel shared(j) #pragma omp sections firstprivate(j) #pragma omp section #pragma omp task shared(j) #pragma omp critical printf("in Task, j = %d\n",j); printf("after parallel, j = %d\n",j); 5 6 #include <stdio.h> #include <omp.h> int main(int argc, char** argv) omp_set_num_threads(2); omp_set_dynamic(0); int j=100; #pragma omp parallel shared(j) #pragma omp sections firstprivate(j) #pragma omp section #pragma omp task shared(j) #pragma omp critical printf("in Task, j = %d\n",j); /* Use TASKWAIT for synchronization. */ #pragma omp taskwait printf("after parallel, j = %d\n",j); 44 Oracle Solaris Studio 12.3 OpenMP API E

46 46

47 6 OpenMP OpenMP 2 OpenMP OpenMP OpenMP API OpenMP 2 OpenMP OpenMP 3.0 OpenMP 3.0 task Oracle Solaris Studio parallel Oracle Solaris Studio task 6.1 auto default( auto) Oracle Solaris Studio OpenMP 47

48 auto auto(list-of-variables) Fortran AUTO(list-of-variables) auto auto auto PARALLEL PARALLEL DO/for PARALLEL SECTIONS Fortran 95 PARALLEL WORKSHARE TASK auto default( auto) default( auto) Fortran DEFAULT( AUTO) default( auto) default( auto) PARALLEL PARALLEL DO/for PARALLEL SECTIONS Fortran 95 PARALLEL WORKSHARE TASK 6.2 OpenMP DO FOR PS1-PS3 PS1 SHARED 48 Oracle Solaris Studio 12.3 OpenMP API E

49 6.3 PS2 PRIVATE PRIVATE PARALLEL DO PARALLEL SECTIONS LASTPRIVATE PS3 REDUCTION PA1 SHARED 6.3 task OpenMP PARALLEL DO/for TS1-TS5 TS1 task FIRSTPRIVATE TS2 SHARED TS3 FIRSTPRIVATE TS4 PRIVATE TS5 FIRSTPRIVATE

50 _auto(list-of-variables) default(_auto) parallel default(_auto) _auto(list-of-variables) parallel task SHARED parallel task IF (.FALSE.) if(0) 6.5 -xopenmp -xo3 -xopenmp=noopt C C++ C C++ #pragma omp task /* task1 */... #pragma omp task /* task 2 */ task2 task1 task2 SHARED task2 IF(.FALSE.) if(0) 50 Oracle Solaris Studio 12.3 OpenMP API E

51 6.6 OpenMP OpenMP omp_set_lock() omp_unset_lock() CRITICAL END CRITICAL OpenMP BARRIER MASTER 6.6 -g er_src er_src Oracle Solaris Studio er_src(1) Oracle Solaris Studio -xvpara xvpara 6 1 -vpara %cat source1.f INTEGER X(100), Y(100), I, T C$OMP PARALLEL DO DEFAULT( AUTO) DO I=1, 100 T = Y(I) X(I) = T*T END DO C$OMP END PARALLEL DO END %f95 -xopenmp -xo3 -vpara -c -g source1.f "source1.f", line 2: Autoscoping for OpenMP construct succeeded. Check er_src for details -xvpara 6 2 -vpara %cat source2.f INTEGER X(100), Y(100), I, T C$OMP PARALLEL DO DEFAULT( AUTO) DO I=1, 100 T = Y(I) CALL FOO(X) X(I) = T*T END DO C$OMP END PARALLEL DO END 6 51

52 vpara %f95 -xopenmp -xo3 -vpara -c -g source2.f "source2.f", line 2: Warning: Autoscoping for OpenMP construct failed. Check er-src for details. Parallel region will be executed by a single thread. er_src 6 3 er_src % er_src source2.o Source file: source2.f Object file: source2.o Load Object: source2.o 1. INTEGER X(100), Y(100), I, T Source OpenMP region below has tag R1 Variables autoscoped as SHARED in R1: y Variables autoscoped as PRIVATE in R1: t, i Variables treated as shared because they cannot be autoscoped in R1: x R1 will be executed by a single thread because autoscoping for some variable s was not successful Private variables in R1: i, t Shared variables in R1: y, x 2. C$OMP PARALLEL DO DEFAULT( AUTO) Source loop below has tag L1 L1 parallelized by explicit user directive L1 autoparallelized L1 parallel loop-body code placed in function _$d1a2.main_ along with 0 inne r loops L1 could not be pipelined because it contains calls 3. DO I=1, T = Y(I) 5. CALL FOO(X) 6. X(I) = T*T 7. END DO 8. C$OMP END PARALLEL DO 9. END REAL FUNCTION FOO (N, X, Y) 2. INTEGER N, I 3. REAL X(*), Y(*) 4. REAL W, MM, M 52 Oracle Solaris Studio 12.3 OpenMP API E

53 W = C$OMP PARALLEL DEFAULT( AUTO) C$OMP SINGLE 11. M = C$OMP END SINGLE MM = C$OMP DO 17. DOI=1,N 18. T = X(I) 19. Y(I) = T 20. IF (MM.GT. T) THEN 21. W=W+T 22. MM = T 23. END IF 24. END DO 25. C$OMP END DO C$OMP CRITICAL 28. IF ( MM.GT. M ) THEN 29. M = MM 30. END IF 31. C$OMP END CRITICAL C$OMP END PARALLEL FOO=W-M RETURN 38. END FOO() SINGLE DO CRITICAL OpenMP 1. X Y 2. X M 3. X W I N MM T W M X Y I DO OpenMP I PRIVATE N S1 SHARED 6 53

54 MM 0.0 S1 MM S2 MM PRIVATE T PRIVATE W 21 S1 S2 S3 W REDUCTION(+) M 11 SINGLE SINGLE CRITICAL M M S1 M SHARED X A1 SHARED Y Y A1 Y SHARED 6 5 QuickSort static void par_quick_sort (int p, int r, float *data) if (p < r) int q = partition (p, r, data); #pragma omp task default( auto) if ((r-p)>=low_limit) par_quick_sort (p, q-1, data); #pragma omp task default( auto) if ((r-p)>=low_limit) par_quick_sort (q+1, r, data); int main ()... #pragma omp parallel #pragma omp single nowait par_quick_sort (0, N-1, &Data[0]);... er_src result: Source OpenMP region below has tag R1 Variables autoscoped as FIRSTPRIVATE in R1: p, q, data Firstprivate variables in R1: data, p, q 47. #pragma omp task default( auto) if ((r-p)>=low_limit) 48. par_quick_sort (p, q-1, data); 54 Oracle Solaris Studio 12.3 OpenMP API E

55 QuickSort Source OpenMP region below has tag R2 Variables autoscoped as FIRSTPRIVATE in R2: q, r, data Firstprivate variables in R2: data, q, r 49. #pragma omp task default( auto) if ((r-p)>=low_limit) 50. par_quick_sort (q+1, r, data); p q TS1 FIRSTPRIVATE 6 6 Fibbonacci int fib (int n) int x, y; if (n < 2) return n; #pragma omp task default( auto) x = fib(n - 1); #pragma omp task default( auto) y = fib(n - 2); #pragma omp taskwait return x+y; er_src result: Source OpenMP region below has tag R1 Variables autoscoped as SHARED in R1: x Variables autoscoped as FIRSTPRIVATE in R1: n Shared variables in R1: x Firstprivate variables in R1: n 24. #pragma omp task default( auto) /* shared(x) firstprivate(n) */ 25. x = fib(n - 1); Source OpenMP region below has tag R2 Variables autoscoped as SHARED in R2: y Variables autoscoped as FIRSTPRIVATE in R2: n Shared variables in R2: y Firstprivate variables in R2: n 26. #pragma omp task default( auto) /* shared(y) firstprivate(n) */ 27. y = fib(n - 2); #pragma omp taskwait 30. return x+y; 31. n TS1 n FIRSTPRIVATE x y fib() x y taskwait fib() fib() x y 6 55

56 Fibbonacci TS2 x y SHARED 6 7 int main(void) int yy = 0; #pragma omp parallel default( auto) shared(yy) int xx = 0; #pragma omp single #pragma omp task default( auto) // task1 xx = 20; #pragma omp task default( auto) // task2 yy = xx; return 0; er_src result: Source OpenMP region below has tag R1 Variables autoscoped as PRIVATE in R1: xx Private variables in R1: xx Shared variables in R1: yy 7. #pragma omp parallel default( auto) shared(yy) int xx = 0; 10. Source OpenMP region below has tag R2 11. #pragma omp single 12. Source OpenMP region below has tag R3 Variables autoscoped as SHARED in R3: xx Shared variables in R3: xx 13. #pragma omp task default( auto) // task xx = 20; Source OpenMP region below has tag R4 Variables autoscoped as PRIVATE in R4: yy 56 Oracle Solaris Studio 12.3 OpenMP API E

57 Variables autoscoped as FIRSTPRIVATE in R4: xx Private variables in R4: yy Firstprivate variables in R4: xx 19. #pragma omp task default( auto) // task yy = xx; xx xx task1 xx task2 task1 xx task1 xx task1 TS2 task1 xx SHARED task2 xx xx xx PRIVATE xx task2 TS3 task2 xx FIRSTPRIVATE task2 yy task2 yy TS4 task2 yy PRIVATE 6 8 int foo(void) intxx=1,yy=0; #pragma omp parallel shared(xx,yy) #pragma omp task default( auto) xx += 1; #pragma omp atomic yy += xx; #pragma omp taskwait return 0; er_src result: Source OpenMP region below has tag R1 Shared variables in R1: yy, xx 5. #pragma omp parallel shared(xx,yy)

58 Source OpenMP region below has tag R2 Variables autoscoped as SHARED in R2: yy Variables autoscoped as FIRSTPRIVATE in R2: xx Shared variables in R2: yy Firstprivate variables in R2: xx 7. #pragma omp task default( auto) xx += 1; #pragma omp atomic 12. yy += xx; #pragma omp taskwait 16. task xx x x xx SHARED x TS5 xx FIRSTPRIVATE task yy taskwait yy TS2 yy SHARED 58 Oracle Solaris Studio 12.3 OpenMP API E

59 7 SHARED Oracle Solaris Studio C C++ Fortran OpenMP 7.1 -xvpara -xopenmp -xo3 OpenMP -xopenmp=noopt -xo3 OpenMP 7 1 % cat t.c #include <omp.h> #include <string.h> int main() 59

60 int g[100], b, i; memset(g, 0, sizeof(int)*100); #pragma omp parallel for shared(b) for (i = 0; i < 100; i++) b += g[i]; return 0; % cc -xopenmp -xo3 -xvpara source1.c "source1.c", line 10: Warning: inappropriate scoping variable b may be scoped inappropriately as shared. write at line 13 and write at line 13 may cause data race "source1.c", line 10: Warning: inappropriate scoping variable b may be scoped inappropriately as shared. write at line 13 and read at line 13 may cause data race -xo3 % cc -xopenmp=noopt -xvpara source1.c "source1.c", line 10: Warning: Scope checking under vpara compiler option is supported with optimization level -xo3 or higher. Compile with a higher optimization level to enable this feature 7 2 % cat source2.c #include <omp.h> int main() int g[100]; int r=0, a=1, b, i; #pragma omp parallel for private(a) lastprivate(i) reduction(+:r) for (i = 0; i < 100; i++) g[i] = a; b=b+g[i]; r=r*g[i]; a=b; return 0; 60 Oracle Solaris Studio 12.3 OpenMP API E

61 % cc -xopenmp -xo3 -xvpara source2.c "source2.c", line 8: Warning: inappropriate scoping variable r may be scoped inappropriately as reduction. reference at line 13 may not be a reduction of the specified type "source2.c", line 8: Warning: inappropriate scoping variable a may be scoped inappropriately as private. read at line 11 may be undefined. consider firstprivate "source2.c", line 8: Warning: inappropriate scoping variable i may be scoped inappropriately as lastprivate. value defined inside the parallel construct is not used outside. consider private "source2.c", line 8: Warning: inappropriate scoping variable b may be scoped inappropriately as shared. write at line 12 and write at line 12 may cause data race "source2.c", line 8: Warning: inappropriate scoping variable b may be scoped inappropriately as shared. write at line 12 and read at line 12 may cause data race 1. r + * 2. a PRIVATE PRIVATE 11 a a FIRSTPRIVATE 3. i LASTPRIVATE i i PRIVATE PRIVATE LASTPRIVATE 4. b OpenMP b SHARED b SHARED b REDUCTION 7.2 -xo3 -xopenmp=noopt OpenMP BARRIER MASTER 7 61

62 62

63 8 OpenMP OpenMP Oracle Solaris Studio OpenMP solarisstudio 8.1 OpenMP BARRIER CRITICAL ORDERED NOWAIT NOWAIT DO CRITICAL FLUSH SPIN SUNW_MP_THR_IDLE DO/FOR 63

64 8.1!$OMP PARALLEL...!$OMP DO...!$OMP END DO...!$OMP END PARALLEL!$OMP PARALLEL...!$OMP DO...!$OMP END DO...!$OMP END PARALLEL!$OMP PARALLEL...!$OMP DO...!$OMP END DO...!$OMP DO...!$OMP END DO!$OMP END PARALLEL PARALLEL DO/FOR DO/FOR PARALLEL DO/FOR!$OMP PARALLEL!$OMP DO...!$OMP END DO!$OMP END PARALLEL!$OMP PARALLEL DO...!$OMP END PARALLEL Oracle Solaris SUNW_MP_PROCBIND MASTER SINGLE MASTER BARRIER IF IF(omp_get_thread_num() == 0) Oracle Solaris Studio 12.3 OpenMP API E

65 8.1 SINGLE SINGLE NOWAIT BARRIER STATIC STATIC DYNAMIC GUIDED LASTPRIVATE DO/FOR malloc() free() / libc malloc() free() libmtmalloc -lmtmalloc libmtmalloc OpenMP PARALLEL if!$omp parallel do do i =......statements_1... end do!$omp parallel do do i =......statements_2... end do!$omp parallel do do i =......statements_ statements_2... end do OpenMP

66 OpenMP UltraSPARC CPU OpenMP chunksize 66 Oracle Solaris Studio 12.3 OpenMP API E

67 8.3 Oracle Solaris OS 8.3 Oracle Solaris OS Oracle Solaris OpenMP (Memory Placement Optimization, MPO) (Multiple Page Size Support, MPSS) MPO OS SunFire E20K SunFire E25K UniBoard UniBoard MPO UniBoard MPSS Oracle Solaris UltraSPARC 8KB AMD64 Opteron 4KB TLB Oracle Solaris Studio TLB Oracle Solaris OS /usr/bin/pagesize -a pagesize(1) Oracle Solaris OS ppgsz(1) -xpagesize -xpagesize_heap -xpagesize_stack MPSS mpss.so.1(1) 8 67

68 68

69 A A 1 Pragma /Pragma parallel do/for sections single parallel do/for parallel sections parallel workshare task if private shared firstprivate lastprivate default reduction copyin copyprivate + (1) ordered + + schedule + + nowait + (2) + (2) + (2) num_threads untied + final + mergeable + auto

70 1. Fortran COPYPRIVATE END SINGLE 2. Fortran NOWAIT END DO END SECTIONS END SINGLE END WORKSHARE 3. Fortran WORKSHARE PARALLEL WORKSHARE 70 Oracle Solaris Studio 12.3 OpenMP API E

71 A auto, 47 48, 48 D default(_auto), G guided, 18 O OMP_DYNAMIC, 15 OMP_MAX_ACTIVE_LEVELS, 15, 31 OMP_NESTED, 15, 29 OMP_NUM_THREADS, 15 OMP_PROC_BIND, 15 OMP_SCHEDULE, 15 OMP_STACKSIZE, 15 OMP_THREAD_LIMIT, 15 OMP_WAIT_POLICY, 15 OpenMP API, 11 Oracle Solaris OS, 67 S SLEEP, 16 SPIN, 16 STACKSIZE, 17 -stackvar, 21 SUNW_MP_MAX_POOL_THREADS, 31 SUNW_MP_THR_IDLE, 16 SUNW_MP_WAIT_POLICY, 18 SUNW_MP_WARN, 16 X -xopenmp, 13 OpenMP, 13 22, 51 52, 48 49, P PARALLEL, 16 pragma, 12,, 29 71

72 , OMP_SCHEDULE, 15 (memory placement optimization, MPO), 67, 15, 15, 29, 21, 17, 21,, 49, 15 19, 23, 66, 66, 18,, 7, 7, 16, 66, 16, 17, OMP_NUM_THREADS, 15, 63, Oracle Solaris Studio 12.3 OpenMP API E

73 ,

74 74