libgomp: Add no-target-region rev offload test + fix plugin-nvptx
Checks
Commit Message
The nvptx reverse-offload code mishandled the case that there was a reverse
offload function that isn't called inside a target region. In that case,
the linker did not include GOMP_target_ext and the global variable it uses.
But the plugin-nvptx.c code expected that the latter is present.
Found via sollve_vv's tests/5.0/requires/test_requires_reverse_offload.c which is
similar to the new testcase. (Albeit the 'if' and comments imply that the sollve_vv
author did not intend this.)
Solution: Handle it gracefully that the global variable does not exist - and
do this check first - and only when successful allocate dev->rev_data. If not,
deallocate rev_fn_table to disable reverse offload handling.
OK for mainline?
Tobias
PS: Admittedly, the nvptx code is not yet exercised as I still have to submit the
libgomp/target.c code handling the reverse offload (+ enabling requires reverse_offload
in plugin-nvptx.c). As it is obvious from this patch, the target.c patch is nearly but
not yet completely ready. - That patch passes the three sollve_vv testcases and also
the existing libgomp testcases, but some corner cases and more testcases are missing.
-----------------
Siemens Electronic Design Automation GmbH; Anschrift: Arnulfstraße 201, 80634 München; Gesellschaft mit beschränkter Haftung; Geschäftsführer: Thomas Heurung, Frank Thürauf; Sitz der Gesellschaft: München; Registergericht München, HRB 106955
Comments
On Thu, Nov 24, 2022 at 06:48:01PM +0100, Tobias Burnus wrote:
> libgomp: Add no-target-region rev offload test + fix plugin-nvptx
>
> OpenMP permits that a 'target device(ancestor:1)' is called without being
> enclosed in a target region - using the current device (i.e. the host) in
> that case. This commit adds a testcase for this.
>
> In case of nvptx, the missing on-device 'GOMP_target_ext' call causes that
> it and also the associated on-device GOMP_REV_OFFLOAD_VAR variable are not
> linked in from nvptx's libgomp.a. Thus, handle the failing cuModuleGetGlobal
> gracefully by disabling reverse offload and assuming that the failure is fine.
>
> libgomp/ChangeLog:
>
> * plugin/plugin-nvptx.c (GOMP_OFFLOAD_load_image): Use unsigned int
> for 'i' to match 'fn_entries'; regard absent GOMP_REV_OFFLOAD_VAR
> as valid and the code having no reverse-offload code.
> * testsuite/libgomp.c-c++-common/reverse-offload-2.c: New test.
Ok, thanks.
Jakub
libgomp: Add no-target-region rev offload test + fix plugin-nvptx
OpenMP permits that a 'target device(ancestor:1)' is called without being
enclosed in a target region - using the current device (i.e. the host) in
that case. This commit adds a testcase for this.
In case of nvptx, the missing on-device 'GOMP_target_ext' call causes that
it and also the associated on-device GOMP_REV_OFFLOAD_VAR variable are not
linked in from nvptx's libgomp.a. Thus, handle the failing cuModuleGetGlobal
gracefully by disabling reverse offload and assuming that the failure is fine.
libgomp/ChangeLog:
* plugin/plugin-nvptx.c (GOMP_OFFLOAD_load_image): Use unsigned int
for 'i' to match 'fn_entries'; regard absent GOMP_REV_OFFLOAD_VAR
as valid and the code having no reverse-offload code.
* testsuite/libgomp.c-c++-common/reverse-offload-2.c: New test.
libgomp/plugin/plugin-nvptx.c | 36 ++++++++++------
.../libgomp.c-c++-common/reverse-offload-2.c | 49 ++++++++++++++++++++++
2 files changed, 73 insertions(+), 12 deletions(-)
@@ -1390,7 +1390,8 @@ GOMP_OFFLOAD_load_image (int ord, unsigned version, const void *target_data,
else if (rev_fn_table)
{
CUdeviceptr var;
- size_t bytes, i;
+ size_t bytes;
+ unsigned int i;
r = CUDA_CALL_NOCHECK (cuModuleGetGlobal, &var, &bytes, module,
"$offload_func_table");
if (r != CUDA_SUCCESS)
@@ -1413,12 +1414,11 @@ GOMP_OFFLOAD_load_image (int ord, unsigned version, const void *target_data,
if (rev_fn_table && *rev_fn_table && dev->rev_data == NULL)
{
- /* cuMemHostAlloc memory is accessible on the device, if unified-shared
- address is supported; this is assumed - see comment in
- nvptx_open_device for CU_DEVICE_ATTRIBUTE_UNIFIED_ADDRESSING. */
- CUDA_CALL_ASSERT (cuMemHostAlloc, (void **) &dev->rev_data,
- sizeof (*dev->rev_data), CU_MEMHOSTALLOC_DEVICEMAP);
- CUdeviceptr dp = (CUdeviceptr) dev->rev_data;
+ /* Get the on-device GOMP_REV_OFFLOAD_VAR variable. It should be
+ available but it might be not. One reason could be: if the user code
+ has 'omp target device(ancestor:1)' in pure hostcode, GOMP_target_ext
+ is not called on the device and, hence, it and GOMP_REV_OFFLOAD_VAR
+ are not linked in. */
CUdeviceptr device_rev_offload_var;
size_t device_rev_offload_size;
CUresult r = CUDA_CALL_NOCHECK (cuModuleGetGlobal,
@@ -1426,11 +1426,23 @@ GOMP_OFFLOAD_load_image (int ord, unsigned version, const void *target_data,
&device_rev_offload_size, module,
XSTRING (GOMP_REV_OFFLOAD_VAR));
if (r != CUDA_SUCCESS)
- GOMP_PLUGIN_fatal ("cuModuleGetGlobal error - GOMP_REV_OFFLOAD_VAR: %s", cuda_error (r));
- r = CUDA_CALL_NOCHECK (cuMemcpyHtoD, device_rev_offload_var, &dp,
- sizeof (dp));
- if (r != CUDA_SUCCESS)
- GOMP_PLUGIN_fatal ("cuMemcpyHtoD error: %s", cuda_error (r));
+ {
+ free (*rev_fn_table);
+ *rev_fn_table = NULL;
+ }
+ else
+ {
+ /* cuMemHostAlloc memory is accessible on the device, if
+ unified-shared address is supported; this is assumed - see comment
+ in nvptx_open_device for CU_DEVICE_ATTRIBUTE_UNIFIED_ADDRESSING. */
+ CUDA_CALL_ASSERT (cuMemHostAlloc, (void **) &dev->rev_data,
+ sizeof (*dev->rev_data), CU_MEMHOSTALLOC_DEVICEMAP);
+ CUdeviceptr dp = (CUdeviceptr) dev->rev_data;
+ r = CUDA_CALL_NOCHECK (cuMemcpyHtoD, device_rev_offload_var, &dp,
+ sizeof (dp));
+ if (r != CUDA_SUCCESS)
+ GOMP_PLUGIN_fatal ("cuMemcpyHtoD error: %s", cuda_error (r));
+ }
}
nvptx_set_clocktick (module, dev);
new file mode 100644
@@ -0,0 +1,49 @@
+/* { dg-do run } */
+/* { dg-additional-options "-foffload-options=nvptx-none=-misa=sm_35" { target { offload_target_nvptx } } } */
+
+#pragma omp requires reverse_offload
+
+int
+main ()
+{
+ int A[10];
+ int y;
+
+ for (int i = 0; i < 10; i++)
+ A[i] = 2*i;
+
+ y = 42;
+
+ /* Pointlessly copy to the default device. */
+ #pragma omp target data map(to: A)
+ {
+ /* Not enclosed in a target region (= i.e. running on the host); the
+ following is valid - it runs on the current device (= host). */
+ #pragma omp target device ( ancestor:1 ) firstprivate(y) map(to: A)
+ {
+ if (y != 42)
+ __builtin_abort ();
+ for (int i = 0; i < 10; i++)
+ if (A[i] != 2*i)
+ __builtin_abort ();
+ for (int i = 0; i < 10; i++)
+ if (A[i] != 2*i)
+ A[i] = 4*i;
+ y = 31;
+ }
+
+ if (y != 42)
+ __builtin_abort ();
+ for (int i = 0; i < 10; i++)
+ if (A[i] != 2*i)
+ __builtin_abort ();
+ }
+
+ if (y != 42)
+ __builtin_abort ();
+ for (int i = 0; i < 10; i++)
+ if (A[i] != 2*i)
+ __builtin_abort ();
+
+ return 0;
+}