[Openmp-commits] [PATCH] D95550: [OpenMP][deviceRTLs] Added `[[clang::loader_uninitialized]]` explicitly

[Shilei Tian via Phabricator via Openmp-commits]

tianshilei1992 created this revision.
tianshilei1992 added reviewers: jdoerfert, JonChesterfield.
Herald added subscribers: guansong, yaxunl.
tianshilei1992 requested review of this revision.
Herald added subscribers: openmp-commits, sstefan1.
Herald added a project: OpenMP.

`[[clang::loader_uninitialized]]` is in macro `SHARED` but it doesn't
work for array like `parallelLevel`, so the variable will be zero initialized.
There is also a similar issue for `omptarget_nvptx_device_State` which is in
global address space. Its c'tor is also generated, which was not in the past when
building the `deviceRTLs` with CUDA. In this patch, we added the attribute to
the two variables explicitly.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D95550

Files:
  openmp/libomptarget/deviceRTLs/common/src/omp_data.cu


Index: openmp/libomptarget/deviceRTLs/common/src/omp_data.cu
===================================================================
--- openmp/libomptarget/deviceRTLs/common/src/omp_data.cu
+++ openmp/libomptarget/deviceRTLs/common/src/omp_data.cu
@@ -25,7 +25,8 @@
 // global data holding OpenMP state information
 ////////////////////////////////////////////////////////////////////////////////
 
-DEVICE
+// OpenMP will try to call its ctor if we don't add the attribute explicitly
+[[clang::loader_uninitialized]] DEVICE
     omptarget_nvptx_Queue<omptarget_nvptx_ThreadPrivateContext, OMP_STATE_COUNT>
         omptarget_nvptx_device_State[MAX_SM];
 
@@ -33,7 +34,9 @@
 DEVICE uint32_t SHARED(usedMemIdx);
 DEVICE uint32_t SHARED(usedSlotIdx);
 
-DEVICE uint8_t parallelLevel[MAX_THREADS_PER_TEAM / WARPSIZE];
+// SHARED doesn't work with array so we add the attribute explicitly.
+[[clang::loader_uninitialized]] DEVICE uint8_t
+    parallelLevel[MAX_THREADS_PER_TEAM / WARPSIZE];
 #pragma omp allocate(parallelLevel) allocator(omp_pteam_mem_alloc)
 DEVICE uint16_t SHARED(threadLimit);
 DEVICE uint16_t SHARED(threadsInTeam);


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D95550.319630.patch
Type: text/x-patch
Size: 1130 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/openmp-commits/attachments/20210127/4cb14121/attachment.bin>