展开查看详情
1 .Parallel Programming in Ope
nMP
2009.10.09
JEONGSIK CHOI
(chjs@skku.edu)
ARCS
2 .Agenda
• Instruction
• Directives
• Clauses
ARCS 2
4 .What is OpenMP?
It is an API
that supports multi-platform
shared memory multiprocessing program
ming
in C, C++ and Fortran
on many architectures
ARCS 4
5 .Shared Memory Architecture
I/O
Memory
Bus or Crossbar switch
Cache Cache Cache Cache
Processor Processor Processor Processor
ARCS 5
6 .The Components of OpenMP
ARCS 6
7 .The Components of OpenMP
• Directives
!$OMP PARALLEL DO
• Runtime library routines
CALL omp_set_num_threads(128)
• Environment variables
Export OMP_NUM_THREADS=8
ARCS 7
8 .OpenMP Programming Model
• Thread-Based
• Fork-Join Model
ARCS 8
9 .OpenMP Fork-and-Join model
printf(“program begin\n”);
N = 1000; Serial
#pragma omp parallel for
for (i=0; i<N; i++) Parallel
A[i] = B[i] + C[i];
M = 500; Serial
#pragma omp parallel for
for (j=0; j<M; j++) Parallel
p[j] = q[j] – r[j];
printf(“program done\n”);
Serial
ARCS 9
10 .Parallelizing a Simple Loop
Serial version Parallel OpenMP version
PROGRAM exam PROGRAM exam
… …
ialpha = 2 ialpha = 2
DO i = 1, 100 !$OMP PARALLEL DO
a(i) = a(i) + ialpha * b(i) DO i = 1, 100
ENDDO a(i) = a(i) + ialpha * b(i)
PRINT *,a ENDDO
END !$OMP END PARALLEL DO
PRINT *,a
END
ARCS 10
11 .Parallelizing a Simple Loop
• Fork-Join ※ export OMP_NUM_THREADS = 4
ialpha = 2 (Master Thread)
Fork
DO i=1, 25 DO i=26, 50 DO i=51, 75 DO i=76, 100
... ... ... ...
(Master) (Slave) (Slave) (Slave)
Join
PRINT *, a (Master Thread)
ARCS 11
12 .Syntax
Fortran: f77 Fortran: f90 C
prefix !$OMP <directive> !$OMP <directive> #pragma omp <directive>
C$OMP <directive>
*$OMP <directive>
newline !$OMP <directive> !$OMP <directive> & #pragma omp … \
!$OMP&… … …
selective !$ … !$ … #ifdef _OPENMP
compile C$ … …
*$ … #endif
start first column any any
location
ARCS 12
13 .Syntax
• Commonly used directives
Fortran C
!$OMP PARALLEL #pragma omp parallel
!$OMP DO #pragma omp for
!$OMP PARALLEL DO #pragma omp parallel for
!$OMP CRITICAL #pragma omp critical
PRIVATE/SHARED private/shared
DEFAULT default
RECUCTION reduction
ARCS 13
14 .Shared Memory Model
private
private • Data can be shared or pr
thread2
ivate
thread1
• Shared data is accessibl
Shared
Shared
e by all threads
Memory
Memory • Private data can be acce
thread3
ssed only by the threads
thread5 that owns it
thread4 private
private
private
ARCS 14
15 .Data Scoping
/* hello_wrong */
I am = 3, tid = 3
I am = 0, tid = 1
#include <omp.h> I am = 1, tid = 1
main() { I am = 2, tid = 1
int i, a, tid;
#pragma omp parallel
{
tid = omp_get_thread_num();
for(i=0; i<10000; i++) a = i;
printf(“I am %d, tid = %d\n”,
omp_get_thread_num(), tid);
}
}
ARCS 15
16 .Data Scoping
/* hello_right */
I am = 3, tid = 3
I am = 0, tid = 0
#include <omp.h> I am = 1, tid = 1
main() { I am = 2, tid = 2
int i, a, tid;
#pragma omp parallel private(tid)
{
tid = omp_get_thread_num();
for(i=0; i<10000; i++) a = i;
printf(“I am %d, tid = %d\n”,
omp_get_thread_num(), tid);
}
}
ARCS 16
18 .Compile & Excute
• Compile
$ ifort –o hello hello.f -openmp
$ icc –o hello hello.c -openmp
• Excute
$ ./hello
ARCS 18
19 .Directives and Clauses
ARCS 19
20 .Major Directives
• Parallel Region Constructs
− parallel
• Work-Sharing Constructs
− do/for
− sections
− single
• Combined Parallel Work-Sharing Constructs
− parallel do/for
− parallel sections
ARCS 20
21 .Parallel Region Construct
• C format
#pragma omp parallel [clause ...] newline
if (scalar_expression)
private (list)
shared (list)
default (shared | none)
firstprivate (list)
reduction (operator: list)
copyin (list)
num_threads (integer-expression)
{
structured_block
}
ARCS 21
22 .Parallel Region Construct
• How Many Threads?
− Use of the omp_set_num_threads() library function
− Setting of the OMP_NUM_THREADS environment variable
• SPMD Style
− The code is duplicated and all threads will execute that cod
e.
• Fork-Join Model
− There is an implied barrier at the end of a parallel section.
Only the master thread continues execution past this point.
ARCS 22
23 .Parallel Region Construct
• Thread ID
− 0(Master Thread) ~[# of threads -1]
− omp_get_thread_num()
#pragma omp parallel
{
myid = omp_get_thread_num();
if( myid == 0 )
do_something();
else
do_something( myid );
}
ARCS 23
24 .Parallel Region Construct
• Dynamic Threads:
− If supported, the two methods available for enabling d
ynamic threads are:
• The omp_set_dynamic() library routine
• Setting of the OMP_DYNAMIC environment variable to TRUE
• Nested Parallel Regions:
− Use the omp_get_nested() library function to determin
e if nested parallel regions are enabled.
ARCS 24
25 .Work-Sharing Constructs
• do / for
• sections
• single
− A work-sharing construct divides the execution of the e
nclosed code region among the members of the team
that encounter it.
− Work-sharing constructs do not launch new threads
− There is no implied barrier upon entry to a work-sharin
g construct, however there is an implied barrier at the
end of a work sharing construct.
ARCS 25
26 .Work-Sharing Constructs
do / for sections single
ARCS 26
27 .Work-Sharing Constructs : DO / for
• The DO / for directive specifies that the iteration
s of the loop immediately following it must be ex
ecuted in parallel by the team.
#pragma omp for [clause ...] newline
schedule (type [,chunk])
ordered
private (list)
firstprivate (list)
lastprivate (list)
shared (list)
reduction (operator: list)
collapse (n)
nowait
for_loop
ARCS 27
28 .Work-Sharing Constructs : sections
• The SECTIONS directive is a non-iterative work-sharing co
nstruct. It specifies that the enclosed section(s) of code
are to be divided among the threads in the team.
#pragma omp sections [clause ...] newline
private (list)
firstprivate (list)
lastprivate (list)
reduction (operator: list)
nowait
{
#pragma omp section newline
structured_block
#pragma omp section newline
structured_block
}
ARCS 28
29 .Work-Sharing Constructs : single
• The SINGLE directive specifies that the enclosed code is
to be executed by only one thread in the team.
• May be useful when dealing with sections of code that ar
e not thread safe (such as I/O)
#pragma omp single [clause ...] newline
private (list)
firstprivate (list)
nowait
structured_block
ARCS 29
热门城市
