Sunstudio on Linux using OpenMP

Question

I am new to use SunStudio.
Rightnow, I want to use openmp to parallel my code.
To do that, I begin from simple example.
PROGRAM calculate
USE omp_lib
INTEGER :: i, max_number
DOUBLE PRECISION :: step, x, value, sum
DOUBLE PRECISION :: num_steps = 1000000000
REAL(8) :: start_time, end_time, run_time, timearray1(2), timearray2(2)
INTEGER :: j
DOUBLE PRECISION :: y, value2 = 0.0
INTEGER :: id, nthrds, nthreads
sum = 0.0
step = 1.0/num_steps;
start_time = dtime(timearray1)
DO  i = 1, int(num_steps), 1
    x = (i - 0.5)*step;
    sum = sum + 4.0/(1.0 + x*x)
END DO
value = step*sum
end_time = dtime(timearray1)
run_time = end_time - start_time
write(*,*) 'run_time is ', end_time, 'seconds'
write(*,*) 'calcuated value is ', value
WRite(*,*) '-------------------------------'
write(*,*) '**      Parallel Test      **'
write(*,*) '-------------------------------'
max_number = OMP_GET_MAX_THREADS()
write(*,*) 'Max Threads = ', max_number
CALL OMP_SET_NUM_THREADS(4)
sum = 0.0
start_time = dtime(timearray2)
!$OMP PARALLEL PRIVATE(id,sum)
id = OMP_get_thread_num()
     DO j = id + 1, int(num_steps), 4
         y = (j - 0.5)*step
         sum = sum + 4.0/(1.0 + y*y)
     END DO
!$OMP CRITICAL
     value2 = value2 + sum*step
     write(*,*) "My thread is", ID, "value2 is", value2
!$OMP END CRITICAL
!$OMP END PARALLEL
end_time = dtime(timearray2)
run_time = end_time - start_time
write(*,*) value2, end_time
END PROGRAMCompilation proceed with no errors or warnings. I used 'f95 test.f90 -xopenmp -O3 -fast'
But the result is strange, it is
run_time is 10.717370986938477 seconds
calcuated value is 3.1415926525857966
** Parallel Test **
Max Threads = 4
My thread is 1 value2 is 0.7853981633965759
My thread is 0 value2 is 1.57079632680205
My thread is 3 value2 is 2.3561944891987165
My thread is 2 value2 is 3.1415926525868203
3.1415926525868203 11.203296661376953
looks like not parallel at all, I test the same code by GCC, it did parallel.
Anybody can help me out? thank you

Answer 1

Thank you very much, Peter,
I try your code, the output is strange, the time is wired
'3.1415926525874704 -188950.34375 '
I am using RedHat Enterperise Linux 5.
The purpose here is I want to see how much performance can be improved by openmp, the similar problem coding up by C and compiled by gcc gave
'using 1000000000 steps to calculate Pi
num_threads = 1
pi is 3.141593 in 7.146756 seconds 1 threds
num_threads = 2
pi is 3.141593 in 4.289904 seconds 2 threds
num_threads = 3
pi is 3.141593 in 2.787567 seconds 3 threds
num_threads = 4
pi is 3.141593 in 1.698492 seconds 4 threds
Thanks again
Edited by: frankqi on Feb 11, 2009 12:21 PM