[Bug libgomp/81886] Means to determine at runtime foffload targets specified at compile time

["vries at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=81886

--- Comment #4 from Tom de Vries <vries at gcc dot gnu.org> ---
(In reply to Tom de Vries from comment #2)
> FAIL: libgomp.oacc-c-c++-common/parallel-dims.c -foffload=nvptx-none  -O0 
> (test for excess errors)
> FAIL: libgomp.oacc-c-c++-common/parallel-dims.c -foffload=nvptx-none  -O2 
> (test for excess errors)

Failure in more detail:
...
FAIL: libgomp.oacc-c++/../libgomp.oacc-c-c++-common/parallel-dims.c
-DACC_DEVICE_TYPE_nvidia=1 -DACC_MEM_SHARED=0 -foffload=nvptx-none  -O0  (test
for excess errors)
Excess errors:
libgomp/testsuite/libgomp.oacc-c++/../libgomp.oacc-c-c++-common/parallel-dims.c:168:9:
warning: using vector_length (32), ignoring 1
...

At line 168, we find:
...
#pragma acc parallel copy (vectors_actual) /* { dg-warning "using vector_length
\\(32\\), ignoring 1" "" { target openacc_nvidia_accel_configured } } */ \
  vector_length (VECTORS) /* { dg-warning "'vector_length' value must be
positive" "" { target c++ } } */
...

AFAICT, openacc_nvidia_accel_configured returns false with the patch because
offload_targets is now nvptx-none instead of nvptx:
...
proc check_effective_target_openacc_nvidia_accel_configured { } {
    global offload_targets
    if { ![string match "*,nvptx,*" ",$offload_targets,"] } {
        return 0
    }
...
Note: the gomp-4_0-branch uses openacc_nvidia_accel_selected instead in the
test-case.