https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83812
Bug ID: 83812
Summary: nvptx-run: error getting kernel result: operation not
supported on global/shared address space
Product: gcc
Version: 8.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: vries at gcc dot gnu.org
Target Milestone: ---
When running make check for an nvptx build (with newlib sources linked in), we
get:
...
build-gcc/nvptx-none/libgomp/testsuite/libgomp.sum
build-gcc/nvptx-none/mgomp/newlib/testsuite/newlib.sum
build-gcc/nvptx-none/newlib/testsuite/newlib.sum
build-gcc/gcc/testsuite/g++/g++.sum
build-gcc/gcc/testsuite/gfortran/gfortran.sum
build-gcc/gcc/testsuite/gcc/gcc.sum
...
A bit surprising is the libgomp testsuite. [ Note: this is not testing a config
of a host with nvptx accelerator, this is testing nvptx as host using the
nvptx-run tool. ]
The test results are not good:
...
=== libgomp Summary ===
# of expected passes 272
# of unexpected failures 2081
# of unresolved testcases 2043
# of unsupported tests 270
...
Mostly test-cases fail to build due to missing functions.
However, when we look at tests that have been executed, we see a different
picture:
...
$ grep -c ^PASS:.*execution build-gcc/nvptx-none/libgomp/testsuite/libgomp.sum
106
$ grep -c ^FAIL:.*execution build-gcc/nvptx-none/libgomp/testsuite/libgomp.sum
31
...
The failures are (note: libgomp.c/pr64824.c contains omp atomic capture):
...
FAIL: libgomp.c/pr64824.c execution test
FAIL: libgomp.fortran/omp_atomic1.f90 -O0 execution test
FAIL: libgomp.fortran/omp_atomic1.f90 -O1 execution test
FAIL: libgomp.fortran/omp_atomic1.f90 -O2 execution test
FAIL: libgomp.fortran/omp_atomic1.f90 -O3 -fomit-frame-pointer -funroll-loops
-fpeel-loops -ftracer -finline-functions execution test
FAIL: libgomp.fortran/omp_atomic1.f90 -O3 -g execution test
FAIL: libgomp.fortran/omp_atomic1.f90 -Os execution test
FAIL: libgomp.fortran/omp_atomic2.f90 -O0 execution test
FAIL: libgomp.fortran/omp_atomic2.f90 -O1 execution test
FAIL: libgomp.fortran/omp_atomic2.f90 -O2 execution test
FAIL: libgomp.fortran/omp_atomic2.f90 -O3 -fomit-frame-pointer -funroll-loops
-fpeel-loops -ftracer -finline-functions execution test
FAIL: libgomp.fortran/omp_atomic2.f90 -O3 -g execution test
FAIL: libgomp.fortran/omp_atomic2.f90 -Os execution test
FAIL: libgomp.fortran/omp_atomic3.f90 -O0 execution test
FAIL: libgomp.fortran/omp_atomic3.f90 -O1 execution test
FAIL: libgomp.fortran/omp_atomic3.f90 -O2 execution test
FAIL: libgomp.fortran/omp_atomic3.f90 -O3 -fomit-frame-pointer -funroll-loops
-fpeel-loops -ftracer -finline-functions execution test
FAIL: libgomp.fortran/omp_atomic3.f90 -O3 -g execution test
FAIL: libgomp.fortran/omp_atomic3.f90 -Os execution test
FAIL: libgomp.fortran/omp_atomic4.f90 -O0 execution test
FAIL: libgomp.fortran/omp_atomic4.f90 -O1 execution test
FAIL: libgomp.fortran/omp_atomic4.f90 -O2 execution test
FAIL: libgomp.fortran/omp_atomic4.f90 -O3 -fomit-frame-pointer -funroll-loops
-fpeel-loops -ftracer -finline-functions execution test
FAIL: libgomp.fortran/omp_atomic4.f90 -O3 -g execution test
FAIL: libgomp.fortran/omp_atomic4.f90 -Os execution test
FAIL: libgomp.fortran/omp_atomic5.f90 -O0 execution test
FAIL: libgomp.fortran/omp_atomic5.f90 -O1 execution test
FAIL: libgomp.fortran/omp_atomic5.f90 -O2 execution test
FAIL: libgomp.fortran/omp_atomic5.f90 -O3 -fomit-frame-pointer -funroll-loops
-fpeel-loops -ftracer -finline-functions execution test
FAIL: libgomp.fortran/omp_atomic5.f90 -O3 -g execution test
FAIL: libgomp.fortran/omp_atomic5.f90 -Os execution test
...
Looking in more detail at libgomp.c/pr64824.c:
...
nvptx-run: error getting kernel result: operation not supported on
global/shared address space (CUDA_ERROR_INVALID_ADDRESS_SPACE, 717)
FAIL: libgomp.c/pr64824.c execution test
...
The rest of the test-cases fails with the same error.
I see the same error only once gcc/g++/gfortran testing:
...
PASS: gfortran.dg/coarray/atomic_2.f90 -fcoarray=single -O2 (test for excess
errors)
spawn nvptx-none-run ./atomic_2.exe^M
nvptx-run: error getting kernel result: operation not supported on
global/shared address space (CUDA_ERROR_INVALID_ADDRESS_SPACE, 717)^M
FAIL: gfortran.dg/coarray/atomic_2.f90 -fcoarray=single -O2 execution test
...
