| 1 |
/* { dg-additional-options "--param=openacc-kernels=decompose" } */ |
| 2 |
|
| 3 |
/* { dg-additional-options "-fopt-info-all-omp" } |
| 4 |
{ dg-additional-options "-foffload=-fopt-info-all-omp" } */ |
| 5 |
|
| 6 |
/* { dg-additional-options "--param=openacc-privatization=noisy" } |
| 7 |
{ dg-additional-options "-foffload=--param=openacc-privatization=noisy" } |
| 8 |
Prune a few: uninteresting, and potentially varying depending on GCC configuration (data types): |
| 9 |
{ dg-prune-output {note: variable 'D\.[0-9]+' declared in block isn't candidate for adjusting OpenACC privatization level: not addressable} } */ |
| 10 |
|
| 11 |
/* It's only with Tcl 8.5 (released in 2007) that "the variable 'varName' |
| 12 |
passed to 'incr' may be unset, and in that case, it will be set to [...]", |
| 13 |
so to maintain compatibility with earlier Tcl releases, we manually |
| 14 |
initialize counter variables: |
| 15 |
{ dg-line l_dummy[variable c_compute 0 c_loop_i 0] } |
| 16 |
{ dg-message dummy {} { target iN-VAl-Id } l_dummy } to avoid |
| 17 |
"WARNING: dg-line var l_dummy defined, but not used". */ |
| 18 |
|
| 19 |
#include <openacc.h> |
| 20 |
|
| 21 |
int test_parallel () |
| 22 |
{ |
| 23 |
int ok = 1; |
| 24 |
int val = 2; |
| 25 |
int ary[32]; |
| 26 |
int ondev = 0; |
| 27 |
|
| 28 |
for (int i = 0; i < 32; i++) |
| 29 |
ary[i] = ~0; |
| 30 |
|
| 31 |
/* val defaults to firstprivate, ary defaults to copy. */ |
| 32 |
#pragma acc parallel num_gangs (32) copy (ok) copy(ondev) /* { dg-line l_compute[incr c_compute] } */ |
| 33 |
/* { dg-note {variable 'i' declared in block isn't candidate for adjusting OpenACC privatization level: not addressable} {} { target *-*-* } l_compute$c_compute } */ |
| 34 |
{ |
| 35 |
ondev = acc_on_device (acc_device_not_host); |
| 36 |
/* { dg-optimized {assigned OpenACC seq loop parallelism} {} { target { c++ && { ! __OPTIMIZE__ } } } .-1 } |
| 37 |
..., as without optimizations, we're not inlining the C++ 'acc_on_device' wrapper. */ |
| 38 |
#pragma acc loop gang(static:1) /* { dg-line l_loop_i[incr c_loop_i] } */ |
| 39 |
/* { dg-note {variable 'i' in 'private' clause isn't candidate for adjusting OpenACC privatization level: not addressable} {} { target *-*-* } l_loop_i$c_loop_i } */ |
| 40 |
/* { dg-optimized {assigned OpenACC gang loop parallelism} {} { target *-*-* } l_loop_i$c_loop_i } */ |
| 41 |
for (unsigned i = 0; i < 32; i++) |
| 42 |
{ |
| 43 |
if (val != 2) |
| 44 |
ok = 0; |
| 45 |
val += i; |
| 46 |
ary[i] = val; |
| 47 |
} |
| 48 |
} |
| 49 |
|
| 50 |
if (ondev) |
| 51 |
{ |
| 52 |
if (!ok) |
| 53 |
return 1; |
| 54 |
if (val != 2) |
| 55 |
return 1; |
| 56 |
|
| 57 |
for (int i = 0; i < 32; i++) |
| 58 |
if (ary[i] != 2 + i) |
| 59 |
return 1; |
| 60 |
} |
| 61 |
|
| 62 |
return 0; |
| 63 |
} |
| 64 |
|
| 65 |
int test_kernels () |
| 66 |
{ |
| 67 |
int val = 2; |
| 68 |
int ary[32]; |
| 69 |
int ondev = 0; |
| 70 |
|
| 71 |
for (int i = 0; i < 32; i++) |
| 72 |
ary[i] = ~0; |
| 73 |
|
| 74 |
/* val defaults to copy, ary defaults to copy. */ |
| 75 |
#pragma acc kernels copy(ondev) /* { dg-line l_compute[incr c_compute] } */ |
| 76 |
/* { dg-note {OpenACC 'kernels' decomposition: variable 'val' in 'copy' clause requested to be made addressable} {} { target *-*-* } l_compute$c_compute } |
| 77 |
{ dg-note {variable 'val' made addressable} {} { target *-*-* } l_compute$c_compute } */ |
| 78 |
/* { dg-note {variable 'ondev\.[0-9]+' declared in block isn't candidate for adjusting OpenACC privatization level: not addressable} {} { target *-*-* } l_compute$c_compute } */ |
| 79 |
{ |
| 80 |
/* { dg-note {beginning 'gang-single' part in OpenACC 'kernels' region} {} { target *-*-* } .+1 } */ |
| 81 |
ondev = acc_on_device (acc_device_not_host); |
| 82 |
/* { dg-optimized {assigned OpenACC seq loop parallelism} {} { target { c++ && { ! __OPTIMIZE__ } } } .-1 } |
| 83 |
..., as without optimizations, we're not inlining the C++ 'acc_on_device' wrapper. */ |
| 84 |
#pragma acc loop /* { dg-line l_loop_i[incr c_loop_i] } */ |
| 85 |
/* { dg-note {forwarded loop nest in OpenACC 'kernels' region to 'parloops' for analysis} {} { target *-*-* } l_loop_i$c_loop_i } */ |
| 86 |
/* { dg-note {variable 'i' declared in block isn't candidate for adjusting OpenACC privatization level: not addressable} {} { target *-*-* } l_loop_i$c_loop_i } */ |
| 87 |
/* { dg-note {variable 'i' in 'private' clause isn't candidate for adjusting OpenACC privatization level: not addressable} {} { target *-*-* } l_loop_i$c_loop_i } */ |
| 88 |
/* { dg-optimized {assigned OpenACC seq loop parallelism} {} { target *-*-* } l_loop_i$c_loop_i } */ |
| 89 |
for (unsigned i = 0; i < 32; i++) |
| 90 |
{ |
| 91 |
ary[i] = val; |
| 92 |
val++; |
| 93 |
} |
| 94 |
} |
| 95 |
|
| 96 |
if (ondev) |
| 97 |
{ |
| 98 |
if (val != 2 + 32) |
| 99 |
return 1; |
| 100 |
|
| 101 |
for (int i = 0; i < 32; i++) |
| 102 |
if (ary[i] != 2 + i) |
| 103 |
return 1; |
| 104 |
} |
| 105 |
|
| 106 |
return 0; |
| 107 |
} |
| 108 |
|
| 109 |
int main () |
| 110 |
{ |
| 111 |
if (test_parallel ()) |
| 112 |
return 1; |
| 113 |
|
| 114 |
if (test_kernels ()) |
| 115 |
return 1; |
| 116 |
|
| 117 |
return 0; |
| 118 |
} |