Andrew Stubbs wrote:
Date: Mon, 8 Dec 2025 16:18:02 +0000
Subject: [PATCH v4] libgomp, amdgcn, nvptx: Improve omp_target_is_accessible
[PR121813]
This patch extends omp_target_is_accessible to check the actual device status
for the memory region, on amdgcn and nvptx devices (rather than just checking
if shared memory is enabled).
In both cases, we check the status of each 4k region within the given memory
range (assuming 4k pages should be safe for all the currently supported hosts)
and returns true if all of the pages report accessible.
I think 4k works well and is also the default on x86-64 Linux ('4 kB' for
KernelPageSize and MMUPageSize), albeit 'grep -ir pagesize /proc/self/smaps'
returns '64 kB' for ppc64le GNU/Linux. But that's just extra work and not
wrong.
* * *
The testcases have been modified to check that allocations marked accessible
actually are accessible (inaccessibility can't be checked without invoking
memory faults), and to understand that some parts of an array can be accessible
but other parts not (I have observed this intermittently for the stack memory
on amdgcn using the Fortran testcase, which can have the allocation span pages).
There's also new testcases for the various other memory modes, and for managed
memory.
include/ChangeLog:
* cuda/cuda.h (CUpointer_attribute): New enum.
(cuPointerGetAttribute): New prototype.
libgomp/ChangeLog:
PR libgomp/121813
* libgomp-plugin.h (GOMP_OFFLOAD_is_accessible_ptr): New prototype.
* libgomp.h
(struct gomp_device_descr): Add GOMP_OFFLOAD_is_accessible_ptr.
* libgomp.texi: Update omp_target_is_accessible docs.
* plugin/cuda-lib.def (cuPointerGetAttribute): New entry.
* plugin/plugin-gcn.c (struct hsa_runtime_fn_info): Add
hsa_amd_svm_attributes_get_fn and hsa_amd_pointer_info_fn.
(init_hsa_runtime_functions): Add hsa_amd_svm_attributes_get and
hsa_amd_pointer_info.
(enum accessible): New enum type.
(host_memory_is_accessible): New function.
(device_memory_is_accessible): New function.
(GOMP_OFFLOAD_is_accessible_ptr): New function.
* plugin/plugin-nvptx.c (GOMP_OFFLOAD_is_accessible_ptr): Likewise.
* target.c (omp_target_is_accessible): Call is_accessible_ptr_func.
(gomp_load_plugin_for_device): Add is_accessible_ptr.
* testsuite/libgomp.c-c++-common/target-is-accessible-1.c: Rework
to match more details of the GPU implementation.
* testsuite/libgomp.fortran/target-is-accessible-1.f90: Likewise.
* testsuite/libgomp.c-c++-common/target-is-accessible-2.c: New test.
* testsuite/libgomp.c-c++-common/target-is-accessible-3.c: New test.
* testsuite/libgomp.c-c++-common/target-is-accessible-4.c: New test.
* testsuite/libgomp.c-c++-common/target-is-accessible-5.c: New test.
LGTM.
BTW:the patch also fixes:
PR libgomp/113213
[OpenMP] Update omp_target_is_present / omp_target_is_accessible handling for
NULL
which you might want to list alongside PR121813.
IMHO, both PRs can be marked as fixed once the patch has landed.
[ There is some follow-up work required, e.g., OpenMP 6.x now required
that omp_target_is_accessible is available on non-host devices (can
(and should?!?) be implemented as always returning 0) and some more.
However, my intent is to track those in the (existing) PR113216, which
hopefully lists all that remains to be done. ]
Thanks for the patch,
Tobias
