diff --git a/openmp/libomptarget/deviceRTLs/common/src/data_sharing.cu b/openmp/libomptarget/deviceRTLs/common/src/data_sharing.cu
--- a/openmp/libomptarget/deviceRTLs/common/src/data_sharing.cu
+++ b/openmp/libomptarget/deviceRTLs/common/src/data_sharing.cu
@@ -36,6 +36,14 @@
   }
 }
 
+EXTERN void *__kmpc_alloc_shared(size_t DataSize) {
+  return (void *)SafeMalloc(DataSize, "Alloc Shared");
+}
+
+EXTERN void __kmpc_free_shared(void *FrameStart) {
+  SafeFree(FrameStart, "Free Shared");
+}
+
 // Initialize data sharing data structure. This function needs to be called
 // once at the beginning of a data sharing context (coincides with the kernel
 // initialization). This function is called only by the MASTER thread of each
@@ -207,24 +215,20 @@
   }
 }
 
-// Begin a data sharing context. Maintain a list of references to shared
-// variables. This list of references to shared variables will be passed
-// to one or more threads.
-// In L0 data sharing this is called by master thread.
-// In L1 data sharing this is called by active warp master thread.
-EXTERN void __kmpc_begin_sharing_variables(void ***GlobalArgs, size_t nArgs) {
-  omptarget_nvptx_globalArgs.EnsureSize(nArgs);
-  *GlobalArgs = omptarget_nvptx_globalArgs.GetArgs();
+void** SHARED(GlobalArgs);
+void* globalArgsStorage[64];
+#pragma omp allocate(globalArgsStorage) allocator(omp_pteam_mem_alloc)
+EXTERN void __kmpc_begin_sharing_variables(void ***GlobalArgsPtr, size_t nArgs) {
+  if (nArgs > 64) {
+    *GlobalArgsPtr = GlobalArgs = reinterpret_cast<void**>(__kmpc_alloc_shared(nArgs * sizeof(void*)));
+  } else {
+    *GlobalArgsPtr = GlobalArgs = &globalArgsStorage[0];
+  }
 }
 
-// End a data sharing context. There is no need to have a list of refs
-// to shared variables because the context in which those variables were
-// shared has now ended. This should clean-up the list of references only
-// without affecting the actual global storage of the variables.
-// In L0 data sharing this is called by master thread.
-// In L1 data sharing this is called by active warp master thread.
 EXTERN void __kmpc_end_sharing_variables() {
-  omptarget_nvptx_globalArgs.DeInit();
+  if (GlobalArgs != &globalArgsStorage[0])
+    __kmpc_free_shared(GlobalArgs);
 }
 
 // This function will return a list of references to global variables. This
@@ -232,8 +236,9 @@
 // members of this list will be passed to the outlined parallel function
 // preserving the order.
 // Called by all workers.
-EXTERN void __kmpc_get_shared_variables(void ***GlobalArgs) {
-  *GlobalArgs = omptarget_nvptx_globalArgs.GetArgs();
+
+EXTERN void __kmpc_get_shared_variables(void ***GlobalArgsPtr) {
+  *GlobalArgsPtr = GlobalArgs;
 }
 
 // This function is used to init static memory manager. This manager is used to
diff --git a/openmp/libomptarget/deviceRTLs/interface.h b/openmp/libomptarget/deviceRTLs/interface.h
--- a/openmp/libomptarget/deviceRTLs/interface.h
+++ b/openmp/libomptarget/deviceRTLs/interface.h
@@ -462,4 +462,14 @@
 EXTERN void __kmpc_restore_team_static_memory(int16_t isSPMDExecutionMode,
                                               int16_t is_shared);
 
+/// Allocate \p Bytes in "shareable" memory and return the address. Needs to be
+/// called balanced with __kmpc_free_shared like a stack (push/pop). Can be
+/// called by any thread, allocation happens per-thread.
+EXTERN void *__kmpc_alloc_shared(uint64_t Bytes);
+
+/// Deallocate \p Ptr. Needs to be called balanced with __kmpc_alloc_shared like
+/// a stack (push/pop). Can be called by any thread. \p Ptr must be allocated by
+/// __kmpc_alloc_shared by the same thread.
+EXTERN void __kmpc_free_shared(void *Ptr);
+
 #endif