source: misc/tools/acceptance_tests/quadrature_parallel_interleaved.c @ 7659

Last change on this file since 7659 was 7276, checked in by ole, 15 years ago

Merged numpy branch back into the trunk.

In ~/sandpit/anuga/anuga_core/source
svn merge -r 6246:HEAD ../../branches/numpy .

In ~/sandpit/anuga/anuga_validation
svn merge -r 6417:HEAD ../branches/numpy_anuga_validation .

In ~/sandpit/anuga/misc
svn merge -r 6809:HEAD ../branches/numpy_misc .

For all merges, I used numpy version where conflicts existed

The suites test_all.py (in source/anuga) and validate_all.py passed using Python2.5 with numpy on my Ubuntu Linux box.

  • Property svn:executable set to *
File size: 2.8 KB
Line 
1// This is an alternative parallel version of quadrature_sequential.c
2//
3// It differs from quadratur_parallel.c by the way work is distributed.
4// This one uses a 'round robin' strategy.
5//
6// To compile: mpicc quadrature_parallel.c -lm
7//
8// This code also exhibits linear speedup. See listed output at the
9// bottom of this file.
10//
11// Ole Nielsen, SUT 2003
12
13       
14#include <math.h>
15#include <stdio.h>
16#include <time.h>
17#include <mpi.h>
18
19
20//Define the circle arc to be integrated.
21long double fct(long double x) {
22  return sqrt(1.0-x*x);
23}
24
25
26int main(argc,argv)
27     int argc;
28     char *argv[];
29{
30  int N, i, lo, hi;
31  long double result=0.0, my_result=0.0;
32  long double reference = 3.1415926535897932; //Reference value, 17 digits
33
34 
35  double t0;
36  int p;  // Myid
37  int P;  // Number of processors
38
39  int namelen, tag=0, j, k;
40  char processor_name[MPI_MAX_PROCESSOR_NAME]; 
41  int *C;
42
43  MPI_Status status;
44   
45  MPI_Init(&argc, &argv);
46  MPI_Comm_rank(MPI_COMM_WORLD, &p);
47  MPI_Comm_size(MPI_COMM_WORLD, &P);
48  MPI_Get_processor_name(processor_name, &namelen); 
49   
50  printf("This is process %d of %d running on %s\n", 
51         p, P, processor_name);
52
53  N = 100000000;   // Resolution 
54  //N = 29;   // Resolution 
55 
56  if (p == 0) {
57    printf("Resolution = %d\n", N);
58  }
59
60  // Begin computation
61  t0 = MPI_Wtime();
62
63 
64  for (k=0; k<15; k++) {
65    //Each process p handles own stride of points
66    my_result = 0.0;
67    for (i=p; i<N; i+=P) {
68      my_result += fct( (i+0.5)/N );
69    } 
70
71    my_result = 4*my_result/N;
72  }
73         
74  printf("P%d: Partial result = %.16Lf\n", p, my_result);   
75
76 
77  // Communication phase
78  if (p == 0) {
79 
80    //Receive results from others
81    result = my_result; 
82    for (j=1; j<P; j++) {         
83      printf("P%d: receiving from P%d\n", p, j);
84      MPI_Recv(&my_result, 1, MPI_LONG_DOUBLE, j, tag,
85               MPI_COMM_WORLD, &status);
86      result += my_result;
87    }
88     
89    printf("Error  = %.2e\n", fabs(result-reference)); 
90    printf("Time   = %.2f sec\n", MPI_Wtime() - t0);
91   
92  } else {
93    // send my_result to intended proc 0
94    MPI_Send(&my_result, 1, MPI_LONG_DOUBLE, 0, tag,
95             MPI_COMM_WORLD);     
96  } 
97
98  MPI_Finalize();
99  return 0;
100}
101
102//Observed output, Resolution = 100000000
103//P = 1
104//Error = 3.44e-13
105//Time = 13.52 sec
106//
107//P = 2
108//Error = 3.44e-13
109//Time = 6.76 sec
110//
111//P = 4
112//Error = 3.44e-13
113//Time = 3.39 sec
114//
115//P = 8
116//Error = 3.44e-13
117//Time = 1.69 sec
118//
119//
120//SPEEDUP & EFFICIENCY
121//t = array([13.52, 6.76, 3.39, 1.69])
122//
123//t[0]/t
124//
125//array([ 1.        ,  2.        ,  3.98820059,  8.        ])
126//
127//p = [1, 2, 4, 8]
128//
129//t[0]/t/p
130//array([ 1.        ,  1.        ,  0.99705015,  1.        ])
131
132
133
134
Note: See TracBrowser for help on using the repository browser.