Openmp Berkeley Chapman Slides 0

7/31/2019 Openmp Berkeley Chapman Slides 0

1/61

! !

!"#$%&'"()"*+',-.

"!#$%&'!()!*+',-.

./'0',1'&!234

5/6!27/87/7!9:7+;7,

9/'71'&!234

2/'11!


2/61

! !

!"#$%&'"()"*+',-.

",!'!B:7/'&!-';)/3!B301';!E!97C:'!=%'/7/C:3

FG FG-7%,!-';)/3

C+$!H

C+$!I C+$!J

C+$!G

FI

FI

FI

FI


3/61

! !

!"#$%&'"()"*+',-.

-7%,!-';)/3

cpu 1

cpu 2

cpu 3

cpu 0

cpu 0

*+',-.K0!!L!/)01"2"3)%,

M

1J

1G

1I

1H

1J

1G

1I

Do stuff 1H

Do stuff

Do stuff

Do stuff

N 1H1H

process 0

(:/'7&!H!%0!),!9.?!HO!71!1:'!A)/PQ!J!,'R!1:/'7&0!7/'!C/'71'&7,&!7/'!&%01/%8$1'&!1)!1:'!/';7%,%,S!J!9.?06


4/61

! !

!"#$%&'"()"*+',-.

T:71!%0!*+',-.U

! ",!%,&$01/3!017,&7/&!A)/!0:7/'&!;';)/3!+7/7>>'>!+/)S/7;;%,S

" *+',-.!"/C:%1'C1$/'!V'@%'R!2)7/&

" "-5Q!W,1'>Q!W2-Q!=.Q!-%C/)0)A1Q!B$,X*/7C>'Q!M$Y%10$Q!Z


5/61

! !

!"#$%&'"()"*+',-.

(:'!",71);3!)A!7,!*+',-.!./)S/7;

#include #include

int main (int argc, char *argv[]) {int tid, numt;numt = omp_get_num_threads();#pragma omp parallel private(tid) shared(numt){

tid = omp_get_thread_num();printf("hi, from %d\n", tid);

#pragma omp barrierif ( tid == 0 ) {printf("%d threads say hi!\n",numt);

}} return 0;}

parallel!]A)/P^&%/'C1%@'

/$,1%;'!A$,C1%),

01/$C1$/'&!parallel!!8>)CP!

&%/'C1%@'!]1:/'7&!barrier^

C>7$0'0


6/61

! !

!"#$%&'"()"*+',-.

*+',-.!%0!V'7>>3!7!B17,&7/&!B+'C%_C71%),

(:'!1%;'>%,'!)A!1:'!*+',-.!B17,&7/&!B+'C%_C71%),


7/61

! !

!"#$%&'"()"*+',-.

B);'!2','_10!)A!$0%,S!*+',-.

! W1K0!!"#$%&'(Q!0$++)/1'&!83!;)01!9X9\\!E!M)/1/7,!C);+%>'/0

! *A1',Q!;$C:!)A!0'`$',1%7>!C)&'!C7,!8'!>'A1!$,1)$C:'&!

! (:'!&'@'>)+;',1!C3C>'!%0!7!A/%',&>3!),'

"

97,!8'!%,1/)&$C'&!)$(#%$)*('+!%,1)!'D%01%,S!C)&'" 9)//'C1,'00!C7,!8'!@'/%_'&!7>),S!1:'!R73

" F%P'R%0'Q!+'/A)/;7,C'!8','_10!C7,!8'!S7$S'&

! *+1%;%a%,S!;';)/3!7CC'00!%,!1:'!0'/%7>!+/)S/7;!R%>>!8','_1!

1:'!1:/'7&'&!@'/0%),!]'6S6Q!A7>0'!0:7/%,SQ!'1C^! W1!C7,!8'!A$,!1)!$0'!]%;;'&%71'!S/71%_C71%),^


8/61

! !

!"#$%&'"()"*+',-.

T:71!5)'0!*+',-.!./)@%&'U

! ",!7801/7C1%),!78)@'!>)R!>'@'>!1:/'7&!>%8/7/%'0

! 5%/'C1%@'0Q!:%&&',!%,0%&'!)A!01/$C1$/'&!C);;',10

! "!0$,(%5'!>%8/7/3!1:71!;7,7S'0!'D'C$1%),!&3,7;%C7>>3

!

"&&%1%),7>!C),1/)>!@%7!',@%/),;',1!@7/%78>'0!E!7!0$,(%5'!".W

! %13O


9/61

! !

!"#$%&'"()"*+',-.

T:71!9);+%>'/0!B$++)/1!*+',-.U

!"#$%& '(#)*()"+ ,*--%&."$/,-"01210(.1%#

!"# $%$&&'()*(+,-./0/12'(3*(+ -45563*)67488./0

942%:/1;5< $%$&&,-./0/12'(=*(+ -45563*)67488./0

!20


10/61

! !

!"#$%&'"()"*+',-.

9);+%>%,S!7,&!


11/61

! !

!"#$%&'"()"*+',-.

*+',-.!5%/'C1%@'0!E!9),01/$C10

! 9),17%,'&!%,0%&'!9(0$4($0'&"4)55',(9

9X9\\4!!!

#pragma omp

M)/1/7,4!

!$OMP

! *+',-.!C);+>%7,1!C);+%>'/0!_,&!7,&!+7/0'!&%/'C1%@'0

! Z),[C);+>%7,1!9:)$;&!07A'>3!%S,)/'!1:';!70!C);;',10

! "!4),9(0$4(!%0!7!&%/'C1%@'!1:71!7e'C10!1:'!',C>)0%,S!C)&'

! W;+'/71%@'!]017,&7>),'^!&%/'C1%@'0!'D%01

! 7$0'0!:70!,)!8'7/%,S!),!'e'C1


12/61

! !

!"#$%&'"()"*+',-.

B$;;7/3!)A!*+',-.K0!5%/'C1%@'0

! M)/P%,S!(:/'7&0

parallel

! 5%01/%8$1%,S!T)/P

for]9X9\\^!

DO!]M)/1/7,^

sections/section

WORKSHARE]M)/1/7,^

! B%,S>%,S!*$1!(:/'7&0

singleMaster

! -$1$7>!$0%),

" critical

"

atomic

! B3,C:/),%a71%),

barrier

flush

ordered

taskwait! "03,C:/),)$0!(70P%,S

task

! 5717!


13/61

! !

!"#$%&'"()"*+',-.

?0'A$>!*+',-.!


14/61

! !

!"#$%&'"()"*+',-.

J!(3+'0!)A!V$,1%;'!M$,C1%),0

'&"5$($8;"'?4;$9%),^O!'6S6Q

L omp_{init,set,test,unset,destroy}_lock

" omp_{...}_nest_lock

(%;%,S!/)$1%,'0O!'6S6QL omp_get_wtime

L omp_get_wtick

* 9 %> &U * & %


15/61

! !

!"#$%&'"()"*+',-.

=)R!W0!7,!*+',-.!./)S/7;!9);+%>'&U!='/'K0!=)R!*+',?=!&)'0!%16

@%8)A"'(B"8;B4!:11+4XXRRRG6C06$:6'&$XgC)++'/X)+',$:6+&A

.7/0'!%,1)!f'/3!=%S:

!T=WVF

.$1!!*+',-.!C),01/$C10!%,1)!017,&7/&!A)/;71!

=%S:!>'@'>Q!0'/%7>!+/)S/7;)+1%;%a71%),0

F))+!>'@'>!0'/%7>!+/)S/7;)+1%;%a71%),0

(/7,0A)/;!*+',-.!%,1)1:/'7&'&!C)&'

*+1%;%a'Q!,)RQ!+)1',1%7>>3Q1:/'7&'&!C)&'

*$1+$1!,71%@'!W"hi!%,01/6)/!%,1'/A7C'!R%1:!,71%@'C);+%>'/!R%1:!0)$/C'!1)0)$/C'

T: 1 5 1: ( A & 9 & F P >%P U
http://www2.cs.uh.edu/~copper/openuh.pdfhttp://www2.cs.uh.edu/~copper/openuh.pdf


16/61

! !

!"#$%&'"()"*+',-.

T:71!5)'0!1:'!(/7,0A)/;'&!9)&'!F))P!>%P'U

! W,1'/;'&%71'!C)&'QjTG9k![!T=WVF!1)!9

" uhcc -mp -gnu3 -CLIST:emit_nested_pu simple.c

" :11+4XXRRRG6C06$:6'&$Xg'01/78&X*+',-.X0%;+>'X!

static void __omprg_main_1(__ompv_gtid_a, __ompv_slink_a)_INT32 __ompv_gtid_a;_UINT64 __ompv_slink_a;

{register _INT32 _w2c___comma;_UINT64 _temp___slink_sym0;_INT32 __ompv_temp_gtid;_INT32 __mplocal_my_id;

/*Begin_of_nested_Program_Unit(s)*/

_temp___slink_sym0 = __ompv_slink_a;__ompv_temp_gtid = __ompv_gtid_a;_w2c___comma = omp_get_thread_num();__mplocal_my_id = _w2c___comma;

printf("hello from %d\n", __mplocal_my_id);return;} /* __omprg_main_1 */

#include int main(int argc, char *argv[]) {

int my_id;#pragma omp parallel default(none) private(my_id){

my_id = omp_get_thread_num();printf("hello from %d\n",my_id);

}return 0;

}

(:'!)/%S%,7>!main()

+7/7>>'>!/'S%),!%,!main!%0!)$1>%,'&!1)!__omprg_main_1()

" * -. jC (% D @%E FCG@H
http://www2.cs.uh.edu/~estrabd/OpenMP/simple/http://www2.cs.uh.edu/~estrabd/OpenMP/simple/


17/61

! !

!"#$%&'"()"*+',-.

",!*+',-.!jC$,(%5'D"@%E0806"FCG@H

! (:'!j0$,(%5'D"!;7,7S'0!1:'!;$>1%[1:/'7&'&!'D'C$1%),4

" W1K0!$0'&!83!1:'!/'0$>1%,S!'D'C$178>'!*+',-.!+/)S/7;

" W1K0!R:71!0+7R,0!1:/'7&0!]'6S6Q!C7>>0!+1:/'7&0^!

" W1K0!R:71!;7,7S'0!0:7/'&!E!+/%@71'!;';)/3

" W1K0!R:71!&%01/%8$1'0!]0:7/'0^!R)/P!7;),S!1:/'7&0

" W1K0!R:71!03,C:/),%a'0!1:/'7&0!E!170P0

" W1K0!R:71!/'&$C'0!@7/%78>'0!7,&!P''+0!lastprivate

" W1K0!R:71!%0!%,l$',C'&!83!',@7/0!E!1:'!$0'/!>'@'>!".W

! 5)D3S',!&)C0!)A!*+',?=K0!*+',-.!V(FQ!libopenmp

" :11+4XXRRRG6C06$:6'&$Xg'01/78&X*+',?=X/mdJX:1;>[>%8)+',;+X

! (:'!5)D3S',!C7>>!S/7+:!A)/!cc);+cA)/P!%,!>%8)+',;+X1:/'7&06C

" __omp_fork(...) call graph

(:' ,'R i ()
http://www2.cs.uh.edu/~estrabd/OpenUH/r593/html-libopenmp/http://www2.cs.uh.edu/~estrabd/OpenUH/r593/html-libopenmp/dc/d87/a00032_acb1e4cd1e1a77010203f0ae270c3ace3_cgraph.pnghttp://www2.cs.uh.edu/~estrabd/OpenUH/r593/html-libopenmp/http://www2.cs.uh.edu/~estrabd/OpenUH/r593/html-libopenmp/dc/d87/a00032_acb1e4cd1e1a77010203f0ae270c3ace3_cgraph.png


18/61

! !

!"#$%&'"()"*+',-.

(:'!,'R!main()

extern _INT32 main() {register _INT32 _w2c___ompv_ok_to_fork;register _UINT64 _w2c_reg3;register _INT32 _w2c___comma;

_INT32 my_id;_INT32 __ompv_gtid_s1;/*Begin_of_nested_PU(s)*/

_w2c___ompv_ok_to_fork = 1;if(_w2c___ompv_ok_to_fork){

_w2c___ompv_ok_to_fork = __ompc_can_fork();}if(_w2c___ompv_ok_to_fork){

__ompc_fork(0, &__omprg_main_1, _w2c_reg3);}else{

__ompv_gtid_s1 = __ompc_get_local_thread_num();__ompc_serialized_parallel();_w2c___comma = omp_get_thread_num();

my_id = _w2c___comma;printf("hello from %d\n", my_id);__ompc_end_serialized_parallel();

}return 0;

} /* main */

C7>>0!V(F!A)/P!7,&!+700'0A$,C1%),!+)%,1'/!1)!)$1>%,'&!main()

0'/%7>!@'/0%),

__omprg_main_1K0!A/7;'!+)%,1'/

Z)8)&3!R7,10!1)!C)&'!>%P'!1:%0Q!0)!>'1!1:'!C);+%>'/!7,&!/$,1%;'!&)!;)01!7>>!1:%0!1'&%)$0!R)/Pn


19/61

! !

!"#$%&'"()"*+',-.

!

./)S/7;;%,S!R%1:!*+',-.!J6H

(:' parallel 9),01/$C1


20/61

! !

!"#$%&'"()"*+',-.

(:'!parallel!9),01/$C1

! T:'/'!1:'!jA)/Pk!)CC$/0!]'6S6Q__ompc_fork(...)^

! )0'0!7>>!)1:'/!*+',-.!C),01/$C10!E!&%/'C1%@'0

! (:%0!C),01/$C1!7CC'+10!1:'!A)>>)R%,S!C>7$0'04!if,num_threads, private, firstprivate, shared,default, copyin, reduction

! 97,!C7>>!A$,C1%),0!1:71!C),17%,!j)/+:7,k!C),01/$C10

" B171%C7>>3!)$10%&'!)A!+7/7>>'>Q!8$1!&3,7;%C7>>3!%,0%&'!&$/%,S!/$,1%;'

! 97,!8'!,'01'&

" B%;+>' *+',-. '


21/61

! !

!"#$%&'"()"*+',-.

"!B%;+>'!*+',-.!'

#include #include



#pragma omp barrier

if ( tid == 0 ) {printf("%d threads say hi!\n",numt);}

} return 0;}

9X9\\

*$1+$1!$0%,S!i!(:0'8&94

hi, from 3hi, from 0hi, from 2hi, from 14 threads say hi!

Z)1'Q!1:/'7&!)/&'/,)1!S$7/7,1''&n

S'1!,$;8'/!)A!1:/'7&0

A)/P

R7%1!A)/!7>>!1:/'7&0

S'1!1:/'7&!%&

Y)%,!]%;+>%C%1!87//%'/Q!7>>!R7%1^

(:' M)/1/7, f'/0%),


22/61

! !

!"#$%&'"()"*+',-.

(:'!M)/1/7,!f'/0%),

#include #include



#pragma omp barrier


} return 0;}

program hello90use omp_libinteger:: tid, numtnumt = omp_get_num_threads()

!$omp parallel private(id) shared(numt)tid = omp_get_thread_num()

write (*,*) 'hi, from', tid!$omp barrier if ( tid == 0 ) then

write (*,*) numt,'threads say hi!'end if

!$omp end parallelend program

9X9\\ MdH

*$1+$1!$0%,S!i!(:0'8&94


Z)1'Q!1:/'7&!)/&'/,)1!S$7/7,1''&n

Z)R N$01 1:' .7/7>>'>%a'& 9)&'


23/61

! !

!"#$%&'"()"*+',-.

Z)RQ!N$01!1:'!.7/7>>'>%a'&!9)&'

#include #include



#pragma omp barrier


}return 0;

}}






9X9\\ MdH

*$1+$1!$0%,S!i!(:0'8&94


Z)1'Q!1:/'7&!)/&'/,)1!S$7/7,1''&n

(/7C' )A (:'


24/61

! !

!"#$%&'"()"*+',-.

(/7C'!)A!(:'!>!1:/'7&0!C7>>!printf

1:/'7&!87//%'/

only thread withtid == 0 does this

)1:'/!1:/'7&0!R7%1

Y)%,

9),1/)>>%,S!1:'!M)/P!71!V$,1%;'


25/61

! !

!"#$%&'"()"*+',-.

9) ) S ' ) 7 $ '

! (:'!jifk!C>7$0'!C),17%,0!7!C),&%1%),7>!'D+/'00%),6

! WA!(V?7$0'!%0!7,)1:'/!R73!1)!C),1/)>!1:'!,$;8'/!)A!1:/'7&0!7C1%@'!%,!7!parallel!C),1/$C1

int n = some_func();#pragma omp parallel if(n>5){

do stuff in parallel}

int n = some_func();#pragma omp parallel num_threads(n){ do stuff in parallel

}

(:'!5717!


26/61

! !

!"#$%&'"()"*+',-.

S

! default([shared]|none|private)! shared(list,)![!0$++)/1'&!83!parallel!C),01/$C1!),>3

! private(list,)

! firstprivate(list,)

! lastprivate(list,)![!0$++)/1'&!83!>))+!E!sections!C),01/$C10!),>3

! reduction(:list,)

! copyprivate(list,)![!0$++)/1'&!83!single!C),01/$C1!),>3

! threadprivate!L!7!017,&7>),'!&%/'C1%@'Q!,"$!7!C>7$0'

#pragma omp threadprivate(list,)

!$omp threadprivate(list,)

! copyin(list,)![!0$++)/1'&!83!parallel!C),01/$C1!),>3

private!E!shared!%,!1:71!B%;+>'!*+',-.!'


27/61

! !

!"#$%&'"()"*+',-.

p + + +

#include #include



#pragma omp barrier


} return 0;}






9X9\\ MdH

*$1+$1!$0%,S!i!(:0'8&94

hi, from 3

hi, from 0hi, from 2hi, from 14 threads say hi!

Z)1'Q!1:/'7&!)/&'/,)1!S$7/7,1''&n

-';)/3!9),0%01',C3!-)&'>


28/61

! !

!"#$%&'"()"*+',-.

! *+',-.!$0'0!7!j/'>7D'&!C),0%01',C3k!;)&'>

" (:/'7&0!;73!1';+)/7/%>3!j0''k!&%e'/',1!@7>$'0!A)/!1:'!07;'!0:7/'&!@7/%78>'!

" 9)/'0!;73!:7@'!)$1!)A!&71'!@7>$'0!%,!1:'%/!C7C:'

! -)01!C),01/$C10!%;+>3!7!jflushk!)A!'7C:!1:/'7&K0!C7C:'

! (/'71'&!70!7!;';)/3!jA',C'k!83!C);+%>'/0!R:',!%1!C);'0!1)!/')/&'/%,S!)+'/71%),0

! *+',-.!+/)@%&'0!7,!'D+>%C%1!l$0:!&%/'C1%@'

#pragma flush (list,)

!$OMP FLUSH(list,)

-$>1%+>'!(:/'7&0!-73!=7@'!9)+%'0!)A!B:7/'&!f7/%78>'0!%,!97C:'!


29/61

! !

!"#$%&'"()"*+',-.

process 3

-7%,!-';)/3

cpu 1

cpu 2

cpu 3

cpu 0 Thread 0

Thread 1

Thread 2

Thread 3

p0

",!'D+>%C%1!barrier!%,!1:71!B%;+>'!*+',-.!'


30/61

! !

!"#$%&'"()"*+',-.

#include #include



#pragma omp barrier


} return 0;}






9X9\\ MdH

*$1+$1!$0%,S!i!(:0'8&94

hi, from 3

hi, from 0hi, from 2hi, from 14 threads say hi!

(/7C'!)A!(:'!


31/61

! !

!"#$%&'"()"*+',-.

0 B 0hi, from 0 0 == 04 threads say hi!

1 B 1hi, from 1 1 != 0

2 B 2hi, from 2 2 !=0

3 B 3hi, from 3 3 != 0

M N

B!o!R7%1!A)/!7>>!1:/'7&0!p!barrier!8'A)/'!+/)S/'00%,S!A$/1:'/6#pragma omp barrier

"!reduction!'!


32/61

! !

!"#$%&'"()"*+',-.

#include #include

int main (int argc, char *argv[]) {int t, i;i = 0;#pragma omp parallel private(t) reduction(+,i){

t = omp_get_thread_num();i = t + 1;printf("hi, from %d\n", t);

#pragma omp barrierif ( t == 0 ) {int numt = omp_get_num_threads();printf("%d threads say hi!\n",numt);

}}printf(i is reduced to %d\n,i);

return 0;}

program hello90use omp_libinteger:: t, i, numti = 0

!$omp parallel private(t) reduction(+:i)t = omp_get_thread_num()i = t + 1;

write (*,*) 'hi, from', t!$omp barrier if ( t == 0 ) then

numt = omp_get_num_threads()

write (*,*) numt,'threads say hi!'end if!$omp end parallel

write (*,*) 'i is reduced to ', iend program

9X9\\ MdH

*$1+$1!$0%,S!i!(:0'8&94

hi, from 3hi, from 0hi, from 2hi, from 14 threads say hi!i is reduced to 10

(/7C'!)A!7!@7/%78>'!/'&$C1%),


33/61

! !

!"#$%&'"()"*+',-.

0 i0 = 0 + 1

1 i1 = 1 + 1

2 i2 = 2 + 1

3 i3 = 3 + 1

M

i = 0

N

i = i+i0+i

1+i

2+i

3...

i = 1+2+3+4...

i = 10i = 10

7!0:7/'&!@7/%78>'!%0!%,%1%7>%a'&

'7C:!1:/'7&!R/%1'0!1)!%10

+/%@71'!@7/%78>'

/'&$C1%),!)CC$/071!',&!)A!1:'!C),01/$C1

_,7>!@7>$'!)A!i!%07@7%>78>'!7A1'/!',&)A!C),01/$C1

f7>%&!*+'/71%),0!A)/!1:'!reduction!9>7$0'


34/61

! !

!"#$%&'"()"*+',-.

! V'&$C1%),!)+'/71%),0!%,!9X9\\4

" "/%1:;'1%C4!"!#!$

" 2%1R%0'4!!!!!!!%!&!'

" F)S%C7>4!!!!!!!%%!!''

! ()*+,-./0!/1)23-./04!.0!5/2-230

" 67+.839)0-!32.-:;)-.,.4)

" min

! @4)2!*)A.0)*!2)*+,-./04!B@C(D!.4!30!32)3!/A!,+22)0-!2)4)32,:

! E/-)F!.0.-.39.G)*!839+)!;3--)24H!!

Z'01'&!parallel!9),01/$C10


35/61

! !

!"#$%&'"()"*+',-.

!

97,!8'!,'01'&Q!8$1!0+'C%_C71%),!;7P'0!%1!)+1%),7>" OMP_NESTED={true,false}

" OMP_MAX_ACTIVE_LEVELS={1,2,..}

" omp_{get,set}_nested()

" omp_get_level()

" omp_get_ancestor_thread_num(level)

!


36/61

! !

!"#$%&'"()"*+',-.

H

G I H

G I H G I H G I H

Z'01!F'@'>!H

Z'01!F'@'>!I

Z'01!F'@'>!G

__ompc_fork(..)

__ompc_fork(..) __ompc_fork(..) __ompc_fork(..)

2'C7$0'!)A!1:'!1/''!01/$C1$/'Q!'7C:!1:/'7&!C7,!8'!$,%`$'>3!%&',1%_'&!83!%10!A$>>!+71:!A/);!1:'!/))1!)A!%10!0$8[1/''O

(:%0!+71:!1$+>'!C7,!8'!C7>C$>71'&!%,!-F;'I';H!$0%,S!omp_get_level!7,&omp_get_ancestor_thread_num!!%,!C);8%,71%),6

qHQHQGrqHQGQGr

(:/'7&!,$;8'/0!7/'!,)1!$,%`$'O!+71:0!1)!'7C:!1:/'7&!7/'6

T)/P!B:7/%,S!9),01/$C10


37/61

! !

!"#$%&'"()"*+',-.

!

(:/'7&0!;$01!0:7/'!%,!1:'!R)/P!)A!+/)S/7;!! *+',-.!+/)@%&'0!jR)/P!0:7/%,Sk!C),01/$C10

! B+'C%_'0!1:'!R)/P!1)!8'!&%01/%8$1'&!7,&!:)R

!

(:'0'!C),01/$C10!%,C>$&'4" >))+0!(for, DO)

" sections

" WORKSHARE!]M)/1/7,!),>3^

" single, master

*+',-.!.7/7>>'>%a'0!F))+0!83!5%01/%8$1%,S!W1'/71%),0!1)!


38/61

! !

!"#$%&'"()"*+',-.

thread 1i = 34 thru 67



!"#$!%&'()*+)$,+'$-,($$-,($.!/0%!$1/$22%$!334$5$$$$66$7,$8#9--$$$:

-,($.!/0%!$1/$;;%$!334$5$$66$7,$8#9--:

-,($.!/;%$!334$5$$66$7,$8#9--:

-,($.!/=?%!$1/$22%$!334$5$$66$7,$8#9--:

.7/7>>'>%a%,S!F))+0![!9X9\\


39/61

! !

!"#$%&'"()"*+',-.

:11+4XX&'@'>)+'/060$,6C);X0)>7/%0X7/1%C>'0X01$&%)c)+',;+6:1;>

#include #include #define N 100

int main(void){float a[N], b[N], c[N];int i;omp_set_dynamic(0); // ensures use of all available threadsomp_set_num_threads(20); // sets number of all available threads to 20

/* Initialize arrays a and b. */for (i = 0; i < N; i++){

a[i] = i * 1.0;b[i] = i * 2.0;}

/* Compute values of array c in parallel. */

#pragma omp parallel shared(a, b, c) private(i){

#pragma omp for[nowait]for (i = 0; i < N; i++)

c[i] = a[i] + b[i];}

printf ("%f\n", c[10]);}

.7/7>>'>%a%,S!F))+0![!9X9\\


40/61

! !

!"#$%&'"()"*+',-.

:11+4XX&'@'>)+'/060$,6C);X0)>7/%0X7/1%C>'0X01$&%)c)+',;+6:1;>

{a[i] = i * 1.0;b[i] = i * 2.0;

}

/* Compute values o#include #include #define N 100

int main(void){float a[N], b[N], c[N];int i;omp_set_dynamic(0); // ensures use of all available threads

omp_set_num_threads(20); // sets number of all available threads to 20/* Initialize arrays a and b. */for (i = 0; i < N; i++)f array c in parallel. */


#pragma omp for[nowait]for (i = 0; i < N; i++)

c[i] = a[i] + b[i];}printf ("%f\n", c[10]);

}

jnowaitk!%0!)+1%),7>

! # %& * -.

.7/7>>'>%a%,S!F))+0![!M)/1/7,


41/61

! !

!"#$%&'"()"*+',-.

:11+4XX&'@'>)+'/060$,6C);X0)>7/%0X7/1%C>'0X01$&%)c)+',;+6:1;>

PROGRAM VECTOR_ADDUSE OMP_LIBPARAMETER (N=100)INTEGER N, I

REAL A(N), B(N), C(N)CALL MP_SET_DYNAMIC (.FALSE.) !ensures use of all available threadsCALL OMP_SET_NUM_THREADS (20) !sets number of available threads to 20

! Initialize arrays A and B.DO I = 1, NA(I) = I * 1.0B(I) = I * 2.0

ENDDO! Compute values of array C in parallel.

!$OMP PARALLEL SHARED(A, B, C), PRIVATE(I)!$OMP DODO I = 1, NC(I) = A(I) + B(I)

ENDDO!$OMP END DO[nowait]

! ... some more instructions!$OMP END PARALLEL

PRINT *, C(10)

END

! # %&' () *+',-.

.7/7>>'>%a%,S!F))+0![!M)/1/7,


42/61

! !

!"#$%&'"()"*+',-.

:11+4XX&'@'>)+'/060$,6C);X0)>7/%0X7/1%C>'0X01$&%)c)+',;+6:1;>

PROGRAM VECTOR_ADDUSE OMP_LIBPARAMETER (N=100)INTEGER N, I

REAL A(N), B(N), C(N)CALL MP_SET_DYNAMIC (.FALSE.) !ensures use of all available threadsCALL OMP_SET_NUM_THREADS (20) !sets number of available threads to 20

! Initialize arrays A and B.DO I = 1, NA(I) = I * 1.0B(I) = I * 2.0

ENDDO! Compute values of array C in parallel.

!$OMP PARALLEL SHARED(A, B, C), PRIVATE(I)!$OMP DODO I = 1, NC(I) = A(I) + B(I)

ENDDO!$OMP END DO[nowait]

! ... some more instructions!$OMP END PARALLEL

PRINT *, C(10)

END

jnowaitk!%0!)+1%),7>!E!7&&'&1)!1:'!C>)0%,S!&%/'C1%@'

! #$%&' () *+',-.

.7/7>>'>!F))+!BC:'&$>%,S


43/61

! !

!"#$%&'"()"*+',-.

! BC:'&$>%,S!/'A'/0!1)!:)R!%1'/71%),0!7/'!700%S,'&!1)!7!+7/1%C$>7/!1:/'7&O

! (:'/'!7/'!m!13+'04

" 9(8(%4!L!'7C:!1:/'7&!%0!78>'!1)!C7>C$>71'!%10!C:$,P!

" &6,85%4!L!_/01!C);'Q!_/01!0'/@'!;7,7S'&!83!/$,1%;'" =$%&'&!L!&'C/'70%,S!C:$,P!0%a'0Q!%,C/'70%,S!R)/P

" 8$()!L!&'1'/;%,'&!7$1);71%C7>>3!83!C);+%>'/!)/!/$,1%;'

" 0$,(%5'!L!&'_,'&!83!OMP_SCHEDULE!)/!omp_set_schedule

! F%;%171%),0" ),>3!),'!0C:'&$>'!13+'!;73!8'!$0'&!A)/!7!S%@',!>))+

" 1:'!C:$,P!0%a'!7++>%'0!1)!8;;!1:/'7&0

! #$%&' () *+',-.

.7/7>>'>!F))+!BC:'&$>%,S![!'


44/61

! !

!"#$%&'"()"*+',-.

!$OMP PARALLEL SHARED(A, B, C) PRIVATE(I)!$OMP DO SCHEDULE (DYNAMIC,4)

DO I = 1, NC(I) = A(I) + B(I)

ENDDO!$OMP END DO [nowait]

!$OMP END PARALLEL


#pragma omp for schedule (guided,4) [nowait]

for (i = 0; i < N; i++)c[i] = a[i] + b[i];}

M)/1/7,

9X9\\

0C:'&$>' C:$,P!0%a'

! #$%&' () *+',-.

"!section!9),01/$C1!'


45/61

! !

!"#$%&'"()"*+',-.

#include #include

int square(int n){return n*n;

}int main(void){int x, y, z, xs, ys, zs;omp_set_dynamic(0);omp_set_num_threads(3);x = 2; y = 3; z = 5;

#pragma omp parallel shared(xs,ys,zs) firstprivate (x, y, z){#pragma omp sections

{#pragma omp section{ xs = square(x);printf ("id = %d, xs = %d\n", omp_get_thread_num(), xs);

}#pragma omp section

{ ys = square(y);printf ("id = %d, ys = %d\n", omp_get_thread_num(), ys);


{ zs = square(z);printf ("id = %d, zs = %d\n", omp_get_thread_num(), zs);}

}}return 0;

}

:11+4XX&'@'>)+'/060$,6C);X0)>7/%0X7/1%C>'0X01$&%)c)+',;+6:1;>

!"#$%&'"()"*+',-.

"!section!9),01/$C1!'


46/61

! !

+

#pragma omp sections{

#pragma omp section{ xs = square(x);printf ("id = %d, xs = %d\n", omp_get_thread_num(), xs);

}#pragma omp section{ ys = square(y);printf ("id = %d, ys = %d\n", omp_get_thread_num(), ys);


{ zs = square(z);printf ("id = %d, zs = %d\n", omp_get_thread_num(), zs);

}}

1:/'7&!H

1:/'7&!I

1:/'7&!G

(%;'

1oH 1oI

!"#$%&'"()"*+',-.

(:'!task!9),01/$C1


47/61

! !

+

!

(70P0!R'/'!7&&'&!%,!J6H!1)!:7,&>'!&3,7;%C!7,&!$,01/$C1$/'&!7++>%C71%),0

" V'C$/0%),

" (/''!E!S/7+:!1/7@'/07>0

! *+',-.K0!'D'C$1%),!;)&'>!870'&!),!1:/'7&0!R70!/'&'_,'&

! "!1:/'7&!%0!C),0%&'/'&!1)!8'!7,!%5+;%4%(!170P

! (:'!task!C),01/$C1!&'_,'0!0%,S$>7/!170P0!'D+>%C%1>3

! F'00!)@'/:'7&!1:7,!,'01'&!parallel!/'S%),0

!"#$%&'"()"*+',-.


48/61

! !

process 3

-7%,!-';)/3

cpu 1

cpu 2

cpu 3

cpu 0 Implicit task 0

Implicit task 1

Implicit task 2

Implicit task 3

p0tied (private)untied (public)

tied (private)

untied (public)

tied (private)

untied (public)

tied (private)

untied (public)

!"#$%&'"()"*+',-.

(:'!task!9),01/$C1


49/61

! !

!

9>7$0'0!0$++)/1'&!7/'4!if, default, private,firstprivate shared, tied/untied

! 23!&'A7$>1Q!7>>!@7/%78>'0!7/'!firstprivate

! (70P0!C7,!8'!,'01'&!03,17C1%C7>>3Q!8$1!7/'!01%>>!

703,C:/),)$0! (:'!170PR7%1!&%/'C1%@'!C7$0'0!7!170P!1)!R7%1!$,1%>!7>>!%10!

C:%>&/',!:7@'!C);+>'1'&

!"#$%&'"()"*+',-.

"!task!9),01/$C1!9X9\\!'


50/61

! !

struct node {struct node *left;struct node *right;

};

extern void process(struct node *);

void traverse( struct node *p ) {if (p->left)

#pragma omp task // p is firstprivate by default traverse(p->left);if (p->right)

#pragma omp task // p is firstprivate by default traverse(p->right);process(p);

}

!"#$%&'"()"*+',-.

-$1$7>!$0%),4!criticalQ!atomicQ!7,&!omp_lock_t


51/61

! !

!

B);'!C)&'!;$01!8'!'D'C$1'&!83!),'!1:/'7&!71!7!1%;'! 3!0'/%7>%a'0!1:'!1:/'7&0

! ">0)!C7>>'&!C/%1%C7>!0'C1%),0

! *+',-.!+/)@%&'0!J!R730!1)!7C:%'@'!;$1$7>!'DC>$0%),

" (:'!critical!C),01/$C1!',C>)0'0!7!C/%1%C7>!0'C1%),

" (:'!atomic!C),01/$C1!',C>)0'!$+&71'0!1)!0:7/'&!@7/%78>'0

" "!>)R!>'@'>Q!S','/7>!+$/+)0'!>)CP%,S!;'C:7,%0;

!"#$%&'"()"*+',-.

"!critical!9),01/$C1!'


52/61

! !

#pragma omp parallel shared(c) private(a, b, i){

#pragma omp critical{ for (i = 0; i < N; i++)

C[ 0] += a[i] + b[i];printf("%f\n",c[0]);

}}

1:/'7&!I 1:/'7&!H 1:/'7&!G

1!o!H 1!o!I 1!o!G

(%;'Z)1'4


53/61

! !

#pragma omp critical(a){// some code

}#pragma omp critical(b)

{// some code

}#pragma omp critical(c)

{// some code

}

thread 1

1!o!H

(%;'

Z)1'4C$>71%,S!s

t ti l t 100000


54/61

! !

static long num_steps = 100000;double step;

void main (){ int i; double x, pi, sum = 0.0;

step = 1.0/(double) num_steps;for (i=0;i


55/61

! !

#include static long num_steps = 100000; double step;#define NUM_THREADS 2void main (){

int i, id, nthreads; double x, pi, sum[NUM_THREADS];step = 1.0/(double) num_steps;omp_set_num_threads(NUM_THREADS);

#pragma omp parallel private (i, id, x){

id = omp_get_thread_num();#pragma omp single

nthreads = omp_get_num_threads();for (i=id, sum[id]=0.0;i< num_steps; i=i+nthreads){

x = (i+0.5)*step;sum[id] += 4.0/(1.0+x*x);}

}for(i=0, pi=0.0;i0"50$6(014/%##%+0?)''0%@(2$0

!(#5"#/%,2(

!"#$%&'"()"*+',-.

"!B>%S:1>3!-)/'!!./)S/7;![!97>C$>71%,S! s

#include


56/61

! !


int i, id, nthreads; double x, pi, sum;step = 1.0/(double) num_steps;omp_set_num_threads(NUM_THREADS);

#pragma omp parallel private (i, id, x, sum) {

id = omp_get_thread_num();#pragma omp single

nthreads = omp_get_num_threads();for (i=id, sum=0.0;i< num_steps; i=i+nthreads){

x = (i+0.5)*step;sum += 4.0/(1.0+x*x);}

#pragma omp critical pi += sum * step; }}

A"0%##%+B01"0,"05%'1(016%#),>

A"$(C0$6)10/($6"30"502"/&),),>0!%#$)%'014/103"(1,8$012%'(0*(#+0?('''!E!!./)S/7;![!97>C$>71%,S!s

#include


57/61

! !


int i; double x, pi, sum = 0.0;step = 1.0/(double) num_steps;omp_set_num_threads(NUM_THREADS);

#pragma omp parallel for private(x) reduction(+:sum) for (i=0;i>'>!+/)S/7;!%0!C/'71'&!R%1:)$1!C:7,S%,S!7,3!'D%01%,S!C)&'!7,&!7&&%,S!i!0%;+>'!>%,'06

!"#$%&'"()"*+',-.

#include

(:'!M%,7>!.7/7>>'>!./)S/7;![!97>C$>71%,S!s


58/61

! !


int i; double x, pi, sum = 0.0;step = 1.0/(double) num_steps;omp_set_num_threads(NUM_THREADS);

#pragma omp parallel for private(x) reduction(+:sum) for (i=0;i>3!0'1!$0%,S!1:'!',@%/),;',1!@7/%78>'Q!OMP_NUM_THREADS6

)0!#)*%$(0&+03(5%4'$

="#0>""30-!(,D.0)/!'(/(,$%$)",1B0reduction0)10/"#(012%'%&'(0$6%,0%0

critical02",1$#42$

!"#$%&'"()"*+',-.

#','/7>!./)S/7;;%,S!(%+0


59/61

! !

!

-%,%;%a'!parallel!C),01/$C10! ?0'!4)5E%,'&!C),01/$C10Q!%A!%1!&)'0,K1!@%)>71'!1:'!78)@'

! -%,%;%a'!0:7/'&!@7/%78>'0Q!;7D%;%a'!+/%@71'

! -%,%;%a'!87//%'/0Q!8$1!&),K1!07C/%_C'!07A'13

! T:',!%,0'/1%,S!*+',-.!%,1)!'D%01%,S!C)&'

" ?0'!7!&%0C%+>%,'&Q!%1'/71%@'!&'@'>)+;',1!C3C>'!L!1'01!7S7%,01!0'/%7>!@'/0%),

" ?0'!87//%'/0!>%8'/7>>3

" *+1%;%a'!*+',-.!E!703,C:/),%a'!'%1$

! T:',!017/1%,S!A/);!0C/71C:

" B17/1!R%1:!7,!)+1%;%a'&!0'/%7>!@'/0%),

!"#$%&'"()"*+',-.

(:'!M$1$/'!)A!*+',-.


60/61

! !

!

f',&)/!8$3[%,!7,&!VE5!0$++)/1!%0!70!01/),S!70!'@'/! -$01!/';7%,!/'>'@7,1

! "C1%@'!7/'70!)A!/'0'7/C:4

" V'_,';',1!1)!170P%,S!;)&'>!]0C:'&$>%,SQ!'1C^

" ?0'/!&'_,'&!/'&$C1%),0!]?5V0^" "CC'>'/71)/0!E!:'1'/)S',')$0!',@%/),;',10

" %,S

" =38/%&!;)&'>0

!

BC7>%,S!%00$'0!8'%,S!7&&/'00'&4" (:)$07,&0!)A!1:/'7&0

" 5717!>)C7>%13

" -)/'!l'D%8>'!E!'vC%',1!03,C:/),%a71%),

!"#$%&'"()"*+',-.

"&&%1%),7>!V'0)$/C'0


61/61

! !

!

:11+4XXRRR6C06$:6'&$Xg:+C1))>0! :11+4XXRRR6C);+$,%136)/S

! :11+4XXRRR6)+',;+6)/S

" B+'C%_C71%),!J6H

" -)/'!/'0)$/C'0

! j?0%,S!*+',-.kQ!9:7+;7,Q!'16!7>6

Openmp Berkeley Chapman Slides 0

Documents

Transcript of Openmp Berkeley Chapman Slides 0