| 1 | // dot product implementation with critical region.
|
|---|
| 2 |
|
|---|
| 3 | /*
|
|---|
| 4 | OpenMP example program which computes the dot product of two arrays a and b
|
|---|
| 5 | (that is sum(a[i]*b[i]) ) using explicit synchronization with a critical region.
|
|---|
| 6 | Compile with gcc -O3 -fopenmp omp_critical.c -o omp_critical
|
|---|
| 7 | */
|
|---|
| 8 | // Online source: http://users.abo.fi/mats/PP2012/examples/OpenMP/omp_critical.c
|
|---|
| 9 | /*
|
|---|
| 10 | * civl verify -inputNTHREADS=2 -inputN=8 dotProduct_critical.cvl
|
|---|
| 11 | * */
|
|---|
| 12 | #include <civlc.h>
|
|---|
| 13 | #include <stdio.h>
|
|---|
| 14 | #include <stdlib.h>
|
|---|
| 15 | #include "omp.cvh"
|
|---|
| 16 | $input int NTHREADS;
|
|---|
| 17 | $input int N;
|
|---|
| 18 |
|
|---|
| 19 | int main (int argc, char *argv[]) {
|
|---|
| 20 | /* implicitly initializing the number of threads */
|
|---|
| 21 | omp_set_num_threads(NTHREADS);
|
|---|
| 22 |
|
|---|
| 23 | double a[N], b[N];
|
|---|
| 24 | double localsum, sum = 0.0;
|
|---|
| 25 | int i, tid, nthreads;
|
|---|
| 26 |
|
|---|
| 27 | /* helper variables shared by all threads*/
|
|---|
| 28 | _Bool __critical_0 = $false; //default crticial region variable name
|
|---|
| 29 |
|
|---|
| 30 | /* parallel procedure */
|
|---|
| 31 | void __parallel_0(int __tid) {
|
|---|
| 32 | #include "omp_thread.cvh"
|
|---|
| 33 | int __i;
|
|---|
| 34 | double __localsum; //private(i, localsum)
|
|---|
| 35 |
|
|---|
| 36 | tid = omp_get_thread_num();
|
|---|
| 37 | if(tid == 0) { //#pragma omp master
|
|---|
| 38 | nthreads = omp_get_num_threads();
|
|---|
| 39 | printf("Number of threads = %d\n", nthreads);
|
|---|
| 40 | }
|
|---|
| 41 |
|
|---|
| 42 | /* The procedure of each thread for the first for loop */
|
|---|
| 43 | void __for_0(int __tid, int __start, int __end, int __extra){
|
|---|
| 44 | for(__i = __start; __i < __end; __i++) {
|
|---|
| 45 | a[__i] = b[__i] = (double)__i;
|
|---|
| 46 | }
|
|---|
| 47 | if(__extra > 0) {
|
|---|
| 48 | a[__extra] = b[__extra] = (double)__extra;
|
|---|
| 49 | }
|
|---|
| 50 | /* implicit barrier at the end of the for construct */
|
|---|
| 51 | __barrier(__tid);
|
|---|
| 52 | }
|
|---|
| 53 |
|
|---|
| 54 | { //#pragma omp for
|
|---|
| 55 | // for (i=0; i < N; i++)
|
|---|
| 56 | // localsum = localsum + (a[i] * b[i]);
|
|---|
| 57 | int __start = __for_start(__tid, N);
|
|---|
| 58 | int __end = __for_end(__tid, N);
|
|---|
| 59 | int __extra = __for_extra(__tid, N);
|
|---|
| 60 |
|
|---|
| 61 | __for_0(__tid, __start, __end, __extra);
|
|---|
| 62 | }
|
|---|
| 63 |
|
|---|
| 64 | __localsum = 0.0;
|
|---|
| 65 |
|
|---|
| 66 | /* The procedure of each thread for the second for loop */
|
|---|
| 67 | void __for_1(int __tid, int __start, int __end, int __extra){
|
|---|
| 68 | for(__i = __start; __i < __end; __i++) {
|
|---|
| 69 | __localsum = __localsum + (a[__i] * b[__i]);
|
|---|
| 70 | }
|
|---|
| 71 | if(__extra > 0) {
|
|---|
| 72 | __localsum = __localsum + (a[__extra] * b[__extra]);
|
|---|
| 73 | }
|
|---|
| 74 | /* implicit barrier at the end of the for construct */
|
|---|
| 75 | __barrier(__tid);
|
|---|
| 76 | }
|
|---|
| 77 |
|
|---|
| 78 | {//#pragma omp for
|
|---|
| 79 | // for (i=0; i < N; i++)
|
|---|
| 80 | // localsum = localsum + (a[i] * b[i]);
|
|---|
| 81 | int __start = __for_start(__tid, N);
|
|---|
| 82 | int __end = __for_end(__tid, N);
|
|---|
| 83 | int __extra = __for_extra(__tid, N);
|
|---|
| 84 |
|
|---|
| 85 | __for_1(__tid, __start, __end, __extra);
|
|---|
| 86 | }
|
|---|
| 87 |
|
|---|
| 88 | {//#pragma omp critical
|
|---|
| 89 | $when(!__critical_0) __critical_0 = $true;
|
|---|
| 90 | sum = sum+__localsum;
|
|---|
| 91 | __critical_0 = $false;
|
|---|
| 92 | }
|
|---|
| 93 | /* implicit barrier at the end of the parallel construct */
|
|---|
| 94 | __barrier(__tid);
|
|---|
| 95 | }
|
|---|
| 96 |
|
|---|
| 97 | $proc __parallel_0_procs[OMP_NUM_THREADS];
|
|---|
| 98 |
|
|---|
| 99 | __barrier_init();
|
|---|
| 100 | /* Create processes to conduct the parallel region */
|
|---|
| 101 | for(i = 0; i < OMP_NUM_THREADS; i++) {
|
|---|
| 102 | __parallel_0_procs[i] = $spawn __parallel_0(i);
|
|---|
| 103 | }
|
|---|
| 104 | for(i = 0; i < OMP_NUM_THREADS; i++) {
|
|---|
| 105 | $wait(__parallel_0_procs[i]);
|
|---|
| 106 | }
|
|---|
| 107 | printf(" Sum = %f\n",sum);
|
|---|
| 108 | }
|
|---|