I noticed the other day that the USM bullet point only talked about
the 'unified_shared_memory' clause and not about the newer 'self_maps'
clause. That's now fixed by the attached patch.
-> https://gcc.gnu.org/onlinedocs/libgomp/Offload-Target-Specifics.html
Committed asr15-9506-g1ff4a22103767c Tobias
commit 1ff4a22103767cd0133f0c1db6e85220f28ab3fa
Author: Tobias Burnus <tbur...@baylibre.com>
Date: Tue Apr 15 23:19:50 2025 +0200
libgomp.texi (gcn, nvptx): Mention self_maps alongside USM
libgomp/ChangeLog:
* libgomp.texi (gcn, nvptx): Mention self_maps clause
besides unified_shared_memory in the requirements item.
---
libgomp/libgomp.texi | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/libgomp/libgomp.texi b/libgomp/libgomp.texi
index 3d3a56cc29a..dfd189b646e 100644
--- a/libgomp/libgomp.texi
+++ b/libgomp/libgomp.texi
@@ -6888,7 +6888,7 @@ The implementation remark:
@code{device(ancestor:1)}) are processed serially per @code{target} region
such that the next reverse offload region is only executed after the previous
one returned.
-@item OpenMP code that has a @code{requires} directive with
+@item OpenMP code that has a @code{requires} directive with @code{self_maps} or
@code{unified_shared_memory} is only supported if all AMD GPUs have the
@code{HSA_AMD_SYSTEM_INFO_SVM_ACCESSIBLE_BY_DEFAULT} property; for
discrete GPUs, this may require setting the @code{HSA_XNACK} environment
@@ -7045,7 +7045,7 @@ The implementation remark:
Per device, reverse offload regions are processed serially such that
the next reverse offload region is only executed after the previous
one returned.
-@item OpenMP code that has a @code{requires} directive with
+@item OpenMP code that has a @code{requires} directive with @code{self_maps} or
@code{unified_shared_memory} runs on nvptx devices if and only if
all of those support the @code{pageableMemoryAccess} property;@footnote{
@uref{https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#um-requirements}}
