| 1 |
/* Verify that we diagnose "gang reduction on an orphan loop" for automatically |
| 2 |
assigned gang level of parallelism. */ |
| 3 |
|
| 4 |
/* { dg-do compile } */ |
| 5 |
/* { dg-additional-options "-fopt-info-optimized-omp" } */ |
| 6 |
/* { dg-additional-options "-Wopenacc-parallelism" } */ |
| 7 |
|
| 8 |
#pragma acc routine gang |
| 9 |
/* { dg-bogus "warning: region is worker partitioned but does not contain worker partitioned code" "TODO default 'gang' 'vector'" { xfail *-*-* } .+3 } |
| 10 |
TODO It's the compiler's own decision to not use 'worker' parallelism here, so it doesn't make sense to bother the user about it. */ |
| 11 |
int |
| 12 |
f1 () |
| 13 |
{ |
| 14 |
int sum = 0, i; |
| 15 |
|
| 16 |
/* { dg-error "gang reduction on an orphan loop" "" { target *-*-* } .+1 } */ |
| 17 |
#pragma acc loop reduction (+:sum) /* { dg-optimized "assigned OpenACC gang vector loop parallelism" } */ |
| 18 |
for (i = 0; i < 100; i++) |
| 19 |
sum++; |
| 20 |
|
| 21 |
return sum; |
| 22 |
} |
| 23 |
|
| 24 |
#pragma acc routine gang |
| 25 |
int |
| 26 |
f2 () |
| 27 |
{ |
| 28 |
int sum = 0, i, j; |
| 29 |
|
| 30 |
/* { dg-error "gang reduction on an orphan loop" "" { target *-*-* } .+1 } */ |
| 31 |
#pragma acc loop reduction (+:sum) /* { dg-optimized "assigned OpenACC gang worker loop parallelism" } */ |
| 32 |
for (i = 0; i < 100; i++) |
| 33 |
#pragma acc loop reduction (+:sum) /* { dg-optimized "assigned OpenACC vector loop parallelism" } */ |
| 34 |
for (j = 0; j < 100; j++) |
| 35 |
sum++; |
| 36 |
|
| 37 |
return sum; |
| 38 |
} |
| 39 |
|
| 40 |
#pragma acc routine gang |
| 41 |
int |
| 42 |
f3 () |
| 43 |
{ |
| 44 |
int sum = 0, i, j, k; |
| 45 |
|
| 46 |
/* { dg-error "gang reduction on an orphan loop" "" { target *-*-* } .+1 } */ |
| 47 |
#pragma acc loop reduction (+:sum) /* { dg-optimized "assigned OpenACC gang loop parallelism" } */ |
| 48 |
for (i = 0; i < 100; i++) |
| 49 |
#pragma acc loop reduction (+:sum) /* { dg-optimized "assigned OpenACC worker loop parallelism" } */ |
| 50 |
for (j = 0; j < 100; j++) |
| 51 |
#pragma acc loop reduction (+:sum) /* { dg-optimized "assigned OpenACC vector loop parallelism" } */ |
| 52 |
for (k = 0; k < 100; k++) |
| 53 |
sum++; |
| 54 |
|
| 55 |
return sum; |
| 56 |
} |
| 57 |
|
| 58 |
int |
| 59 |
main () |
| 60 |
{ |
| 61 |
int sum = 0, i, j, k; |
| 62 |
|
| 63 |
#pragma acc parallel copy (sum) |
| 64 |
{ |
| 65 |
#pragma acc loop reduction (+:sum) /* { dg-optimized "assigned OpenACC gang vector loop parallelism" } */ |
| 66 |
for (i = 0; i < 100; i++) |
| 67 |
sum++; |
| 68 |
} |
| 69 |
|
| 70 |
#pragma acc parallel copy (sum) |
| 71 |
{ |
| 72 |
#pragma acc loop reduction (+:sum) /* { dg-optimized "assigned OpenACC gang worker loop parallelism" } */ |
| 73 |
for (i = 0; i < 100; i++) |
| 74 |
#pragma acc loop reduction (+:sum) /* { dg-optimized "assigned OpenACC vector loop parallelism" } */ |
| 75 |
for (j = 0; j < 100; j++) |
| 76 |
sum++; |
| 77 |
} |
| 78 |
|
| 79 |
#pragma acc parallel copy (sum) |
| 80 |
{ |
| 81 |
#pragma acc loop reduction (+:sum) /* { dg-optimized "assigned OpenACC gang loop parallelism" } */ |
| 82 |
for (i = 0; i < 100; i++) |
| 83 |
#pragma acc loop reduction (+:sum) /* { dg-optimized "assigned OpenACC worker loop parallelism" } */ |
| 84 |
for (j = 0; j < 100; j++) |
| 85 |
#pragma acc loop reduction (+:sum) /* { dg-optimized "assigned OpenACC vector loop parallelism" } */ |
| 86 |
for (k = 0; k < 100; k++) |
| 87 |
sum++; |
| 88 |
} |
| 89 |
|
| 90 |
return sum; |
| 91 |
} |