[OpenMP] Use function tracing RAII for runtime functions.

[jhuber6]

Summary

This patch adds support for using function tracing features to track the
executino of runtime functions in the device runtime library. This is
enabled by first compiling the new runtime with
-fopenmp-target-debug=3 and running with
LIBOMPTARGET_DEVICE_RTL_DEBUG=3. The output only tracks team 0 and
thread 0 so there isn't much output when using a generic region.

[jdoerfert]

The output only tracks team 0 and thread 0 so there isn't much output when using a generic region.

We could even expose this via the device_environment to the user as a runtime choice (env vars).
In case this is not eliminated at compile time it's costly either way.

openmp/libomptarget/DeviceRTL/src/Debug.cpp

48

make it omp_get_thread_num though, and omp_get_team_num

[jhuber6]

Changing the interface to use the OpenMP standard functions for getting thread
numbers and team numbers. This in theory will allow the implementation to more
generally follow the thread doing the actual execution. The implementation of
these on the device are not currently correct however, so this won't work as
intended until omp_get_thread_num() is fixed on the device.

[jhuber6]

Changing back to use the mapping functions, can be fixed later once there's a
better interface to getting the effective initial thread in generic and SPMD
regions from any arbitrary location.

[jdoerfert]

LG

openmp/libomptarget/DeviceRTL/include/Debug.h

 #define PRINT(str)
 #endif
 
 ///}
 
 /// Enter a debugging scope for performing function traces. Enabled with
 /// FunctionTracting set in the debug kind.
 #define FunctionTracingRAII()                                                  \
-  DebugEntryRAII Entry(__LINE__, __PRETTY_FUNCTION__);
+  DebugEntryRAII Entry(__FILE__, __LINE__, __PRETTY_FUNCTION__);
 
 /// An RAII class for handling entries to debug locations. The current location
 /// and function will be printed on entry. Nested levels increase the
 /// indentation shown in the debugging output.
 struct DebugEntryRAII {
-  DebugEntryRAII(const unsigned Line, const char *Function);
+  DebugEntryRAII(const char *File, const unsigned Line, const char *Function);
   ~DebugEntryRAII();
 };
 
 #endif

openmp/libomptarget/DeviceRTL/src/Debug.cpp

 //===--- Debug.cpp -------- Debug utilities ----------------------- C++ -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 // This file contains debug utilities
 //
 //===----------------------------------------------------------------------===//
 
 #include "Debug.h"
 #include "Configuration.h"
+#include "Interface.h"
 #include "Mapping.h"
 #include "Types.h"
 
 using namespace _OMP;
 
 #pragma omp declare target
 
 extern "C" {
void __assert_fail(const char *assertion, const char *file, unsigned line,
   __builtin_trap();
 }
 }
 
 /// Current indentation level for the function trace. Only accessed by thread 0.
 static uint32_t Level = 0;
 #pragma omp allocate(Level) allocator(omp_pteam_mem_alloc)
 
-DebugEntryRAII::DebugEntryRAII(const unsigned Line, const char *Function) {
+DebugEntryRAII::DebugEntryRAII(const char *File, const unsigned Line,
+                               const char *Function) {
   if (config::isDebugMode(config::DebugKind::FunctionTracing) &&
-      mapping::getThreadIdInBlock() == 0) {
+      mapping::getThreadIdInBlock() == 0 && mapping::getBlockId() == 0) {
 
     for (int I = 0; I < Level; ++I)
       PRINTF("%s", "  ");
 
-    PRINTF("Line %u: Thread %u Entering %s:%u\n", Line,
+    PRINTF("%s:%u: Thread %u Entering %s\n", File, Line,
            mapping::getThreadIdInBlock(), Function);
     Level++;
   }
 }
 
 DebugEntryRAII::~DebugEntryRAII() {
   if (config::isDebugMode(config::DebugKind::FunctionTracing) &&
-      mapping::getThreadIdInBlock() == 0)
+      mapping::getThreadIdInBlock() == 0 && mapping::getBlockId() == 0)
     Level--;
 }
 
 #pragma omp end declare target

openmp/libomptarget/DeviceRTL/src/Kernel.cpp

static void inititializeRuntime(bool IsSPMD) {
   // Order is important here.
   synchronize::init(IsSPMD);
   mapping::init(IsSPMD);
   state::init(IsSPMD);
 }
 
 /// Simple generic state machine for worker threads.
 static void genericStateMachine(IdentTy *Ident) {
+  FunctionTracingRAII();
 
   uint32_t TId = mapping::getThreadIdInBlock();
 
   do {
     ParallelRegionFnTy WorkFn = 0;
 
     // Wait for the signal that we have a new work function.
     synchronize::threads();

 extern "C" {
 
 /// Initialization
 ///
 /// \param Ident               Source location identification, can be NULL.
 ///
 int32_t __kmpc_target_init(IdentTy *Ident, int8_t Mode,
                            bool UseGenericStateMachine, bool) {
+  FunctionTracingRAII();
   const bool IsSPMD = Mode & OMP_TGT_EXEC_MODE_SPMD;
   if (IsSPMD) {
     inititializeRuntime(/* IsSPMD */ true);
     synchronize::threads();
   } else {
     inititializeRuntime(/* IsSPMD */ false);
     // No need to wait since only the main threads will execute user
     // code and workers will run into a barrier right away.

 /// De-Initialization
 ///
 /// In non-SPMD, this function releases the workers trapped in a state machine
 /// and also any memory dynamically allocated by the runtime.
 ///
 /// \param Ident Source location identification, can be NULL.
 ///
 void __kmpc_target_deinit(IdentTy *Ident, int8_t Mode, bool) {
+  FunctionTracingRAII();
   const bool IsSPMD = Mode & OMP_TGT_EXEC_MODE_SPMD;
   state::assumeInitialState(IsSPMD);
   if (IsSPMD)
     return;
 
   // Signal the workers to exit the state machine and exit the kernel.
   state::ParallelRegionFn = nullptr;
 }
 
-int8_t __kmpc_is_spmd_exec_mode() { return mapping::isSPMDMode(); }
+int8_t __kmpc_is_spmd_exec_mode() {
+  FunctionTracingRAII();
+  return mapping::isSPMDMode();
+}
 }
 
 #pragma omp end declare target

openmp/libomptarget/DeviceRTL/src/Mapping.cpp

 
 bool mapping::isSPMDMode() { return IsSPMDMode; }
 
 bool mapping::isGenericMode() { return !isSPMDMode(); }
 ///}
 
 extern "C" {
 __attribute__((noinline)) uint32_t __kmpc_get_hardware_thread_id_in_block() {
+  FunctionTracingRAII();
   return mapping::getThreadIdInBlock();
 }
 
 __attribute__((noinline)) uint32_t __kmpc_get_hardware_num_threads_in_block() {
+  FunctionTracingRAII();
   return mapping::getNumberOfProcessorElements();
 }
 }
 #pragma omp end declare target

openmp/libomptarget/DeviceRTL/src/Misc.cpp

 //===--------- Misc.cpp - OpenMP device misc interfaces ----------- C++ -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 //
 //===----------------------------------------------------------------------===//
 
 #include "Types.h"
 
+#include "Debug.h"
+
 #pragma omp declare target
 
 namespace _OMP {
 namespace impl {
 
 /// AMDGCN Implementation
 ///
 ///{

 } // namespace impl
 } // namespace _OMP
 
 /// Interfaces
 ///
 ///{
 
 extern "C" {
-int32_t __kmpc_cancellationpoint(IdentTy *, int32_t, int32_t) { return 0; }
+int32_t __kmpc_cancellationpoint(IdentTy *, int32_t, int32_t) {
+  FunctionTracingRAII();
+  return 0;
+}
 
-int32_t __kmpc_cancel(IdentTy *, int32_t, int32_t) { return 0; }
+int32_t __kmpc_cancel(IdentTy *, int32_t, int32_t) {
+  FunctionTracingRAII();
+  return 0;
+}
 
 double omp_get_wtick(void) { return _OMP::impl::getWTick(); }
 
 double omp_get_wtime(void) { return _OMP::impl::getWTime(); }
 }
 
 ///}
 #pragma omp end declare target

openmp/libomptarget/DeviceRTL/src/Parallelism.cpp

else
     NumThreads = (NumThreads & ~((uint32_t)mapping::getWarpSize() - 1));
 
   return NumThreads;
 }
 
 // Invoke an outlined parallel function unwrapping arguments (up to 32).
 void invokeMicrotask(int32_t global_tid, int32_t bound_tid, void *fn,
                      void **args, int64_t nargs) {
+  DebugEntryRAII Entry(__FILE__, __LINE__, "<OpenMP Outlined Function>");
   switch (nargs) {
 #include "generated_microtask_cases.gen"
   default:
     PRINT("Too many arguments in kmp_invoke_microtask, aborting execution.\n");
     __builtin_trap();
   }
 }
 
 } // namespace
 
 extern "C" {
 
 void __kmpc_parallel_51(IdentTy *ident, int32_t, int32_t if_expr,
                         int32_t num_threads, int proc_bind, void *fn,
                         void *wrapper_fn, void **args, int64_t nargs) {
+  FunctionTracingRAII();
 
   uint32_t TId = mapping::getThreadIdInBlock();
   // Handle the serialized case first, same for SPMD/non-SPMD.
   if (OMP_UNLIKELY(!if_expr || icv::Level)) {
     __kmpc_serialized_parallel(ident, TId);
     invokeMicrotask(TId, 0, fn, args, nargs);
     __kmpc_end_serialized_parallel(ident, TId);
     return;
#pragma unroll
   }
 
   if (nargs)
     __kmpc_end_sharing_variables();
 }
 
 __attribute__((noinline)) bool
 __kmpc_kernel_parallel(ParallelRegionFnTy *WorkFn) {
+  FunctionTracingRAII();
   // Work function and arguments for L1 parallel region.
   *WorkFn = state::ParallelRegionFn;
 
   // If this is the termination signal from the master, quit early.
   if (!*WorkFn)
     return false;
 
   // Set to true for workers participating in the parallel region.
   uint32_t TId = mapping::getThreadIdInBlock();
   bool ThreadIsActive = TId < state::ParallelTeamSize;
   return ThreadIsActive;
 }
 
 __attribute__((noinline)) void __kmpc_kernel_end_parallel() {
+  FunctionTracingRAII();
   // In case we have modified an ICV for this thread before a ThreadState was
   // created. We drop it now to not contaminate the next parallel region.
   ASSERT(!mapping::isSPMDMode());
   uint32_t TId = mapping::getThreadIdInBlock();
   state::resetStateForThread(TId);
   ASSERT(!mapping::isSPMDMode());
 }
 
 void __kmpc_serialized_parallel(IdentTy *, uint32_t TId) {
+  FunctionTracingRAII();
   state::enterDataEnvironment();
   ++icv::Level;
 }
 
 void __kmpc_end_serialized_parallel(IdentTy *, uint32_t TId) {
+  FunctionTracingRAII();
   state::exitDataEnvironment();
   --icv::Level;
 }
 
-uint16_t __kmpc_parallel_level(IdentTy *, uint32_t) { return omp_get_level(); }
+uint16_t __kmpc_parallel_level(IdentTy *, uint32_t) {
+  FunctionTracingRAII();
+  return omp_get_level();
+}
 
-int32_t __kmpc_global_thread_num(IdentTy *) { return omp_get_thread_num(); }
+int32_t __kmpc_global_thread_num(IdentTy *) {
+  FunctionTracingRAII();
+  return omp_get_thread_num();
+}
 
 void __kmpc_push_num_threads(IdentTy *, int32_t, int32_t NumThreads) {
+  FunctionTracingRAII();
   icv::NThreads = NumThreads;
 }
 
 void __kmpc_push_num_teams(IdentTy *loc, int32_t tid, int32_t num_teams,
-                           int32_t thread_limit) {}
+                           int32_t thread_limit) {
+  FunctionTracingRAII();
+}
 
-void __kmpc_push_proc_bind(IdentTy *loc, uint32_t tid, int proc_bind) {}
+void __kmpc_push_proc_bind(IdentTy *loc, uint32_t tid, int proc_bind) {
+  FunctionTracingRAII();
+}
 }
 
 #pragma omp end declare target

openmp/libomptarget/DeviceRTL/src/Reduction.cpp

 static volatile uint32_t Cnt = 0;
 
 } // namespace
 
 extern "C" {
 int32_t __kmpc_nvptx_parallel_reduce_nowait_v2(
     IdentTy *Loc, int32_t TId, int32_t num_vars, uint64_t reduce_size,
     void *reduce_data, ShuffleReductFnTy shflFct, InterWarpCopyFnTy cpyFct) {
+  FunctionTracingRAII();
   return nvptx_parallel_reduce_nowait(TId, num_vars, reduce_size, reduce_data,
                                       shflFct, cpyFct, mapping::isSPMDMode(),
                                       false);
 }
 
 int32_t __kmpc_nvptx_teams_reduce_nowait_v2(
     IdentTy *Loc, int32_t TId, void *GlobalBuffer, uint32_t num_of_records,
     void *reduce_data, ShuffleReductFnTy shflFct, InterWarpCopyFnTy cpyFct,
     ListGlobalFnTy lgcpyFct, ListGlobalFnTy lgredFct, ListGlobalFnTy glcpyFct,
     ListGlobalFnTy glredFct) {
+  FunctionTracingRAII();
 
   // Terminate all threads in non-SPMD mode except for the master thread.
   uint32_t ThreadId = mapping::getThreadIdInBlock();
   if (mapping::isGenericMode()) {
     if (!mapping::isMainThreadInGenericMode())
       return 0;
     ThreadId = 0;
   }
if (IsMaster && ChunkTeamCount == num_of_records - 1) {
     // intermediate results to the global buffer.
     atomic::add((uint32_t *)&IterCnt, uint32_t(num_of_records),
                 __ATOMIC_SEQ_CST);
   }
 
   return 0;
 }
 
-void __kmpc_nvptx_end_reduce(int32_t TId) {}
+void __kmpc_nvptx_end_reduce(int32_t TId) { FunctionTracingRAII(); }
 
-void __kmpc_nvptx_end_reduce_nowait(int32_t TId) {}
+void __kmpc_nvptx_end_reduce_nowait(int32_t TId) { FunctionTracingRAII(); }
 }
 
 #pragma omp end declare target

openmp/libomptarget/DeviceRTL/src/State.cpp

 
 int omp_get_team_num() { return mapping::getBlockId(); }
 
 int omp_get_initial_device(void) { return -1; }
 }
 
 extern "C" {
 __attribute__((noinline)) void *__kmpc_alloc_shared(uint64_t Bytes) {
+  FunctionTracingRAII();
   return memory::allocShared(Bytes, "Frontend alloc shared");
 }
 
 __attribute__((noinline)) void __kmpc_free_shared(void *Ptr, uint64_t Bytes) {
+  FunctionTracingRAII();
   memory::freeShared(Ptr, Bytes, "Frontend free shared");
 }
 
 void *__kmpc_get_dynamic_shared() { return memory::getDynamicBuffer(); }
 
 void *llvm_omp_get_dynamic_shared() { return __kmpc_get_dynamic_shared(); }
 
 /// Allocate storage in shared memory to communicate arguments from the main
 /// thread to the workers in generic mode. If we exceed
 /// NUM_SHARED_VARIABLES_IN_SHARED_MEM we will malloc space for communication.
 constexpr uint64_t NUM_SHARED_VARIABLES_IN_SHARED_MEM = 64;
 
 [[clang::loader_uninitialized]] static void
     *SharedMemVariableSharingSpace[NUM_SHARED_VARIABLES_IN_SHARED_MEM];
 #pragma omp allocate(SharedMemVariableSharingSpace)                            \
     allocator(omp_pteam_mem_alloc)
 [[clang::loader_uninitialized]] static void **SharedMemVariableSharingSpacePtr;
 #pragma omp allocate(SharedMemVariableSharingSpacePtr)                         \
     allocator(omp_pteam_mem_alloc)
 
 void __kmpc_begin_sharing_variables(void ***GlobalArgs, uint64_t nArgs) {
+  FunctionTracingRAII();
   if (nArgs <= NUM_SHARED_VARIABLES_IN_SHARED_MEM) {
     SharedMemVariableSharingSpacePtr = &SharedMemVariableSharingSpace[0];
   } else {
     SharedMemVariableSharingSpacePtr = (void **)memory::allocGlobal(
         nArgs * sizeof(void *), "new extended args");
   }
   *GlobalArgs = SharedMemVariableSharingSpacePtr;
 }
 
 void __kmpc_end_sharing_variables() {
+  FunctionTracingRAII();
   if (SharedMemVariableSharingSpacePtr != &SharedMemVariableSharingSpace[0])
     memory::freeGlobal(SharedMemVariableSharingSpacePtr, "new extended args");
 }
 
 void __kmpc_get_shared_variables(void ***GlobalArgs) {
+  FunctionTracingRAII();
   *GlobalArgs = SharedMemVariableSharingSpacePtr;
 }
 }
 #pragma omp end declare target

openmp/libomptarget/DeviceRTL/src/Synchronization.cpp

uint32_t atomic::add(uint32_t *Addr, uint32_t V, int Ordering) {
   return impl::atomicAdd(Addr, V, Ordering);
 }
 
 uint64_t atomic::add(uint64_t *Addr, uint64_t V, int Ordering) {
   return impl::atomicAdd(Addr, V, Ordering);
 }
 
 extern "C" {
-void __kmpc_ordered(IdentTy *Loc, int32_t TId) {}
+void __kmpc_ordered(IdentTy *Loc, int32_t TId) { FunctionTracingRAII(); }
 
-void __kmpc_end_ordered(IdentTy *Loc, int32_t TId) {}
+void __kmpc_end_ordered(IdentTy *Loc, int32_t TId) { FunctionTracingRAII(); }
 
 int32_t __kmpc_cancel_barrier(IdentTy *Loc, int32_t TId) {
+  FunctionTracingRAII();
   __kmpc_barrier(Loc, TId);
   return 0;
 }
 
 void __kmpc_barrier(IdentTy *Loc, int32_t TId) {
+  FunctionTracingRAII();
   if (mapping::isMainThreadInGenericMode())
     return __kmpc_flush(Loc);
 
   if (mapping::isSPMDMode())
     return __kmpc_barrier_simple_spmd(Loc, TId);
 
   impl::namedBarrier();
 }
 
 __attribute__((noinline)) void __kmpc_barrier_simple_spmd(IdentTy *Loc,
                                                           int32_t TId) {
+  FunctionTracingRAII();
   synchronize::threads();
 }
 
 int32_t __kmpc_master(IdentTy *Loc, int32_t TId) {
+  FunctionTracingRAII();
   return omp_get_team_num() == 0;
 }
 
-void __kmpc_end_master(IdentTy *Loc, int32_t TId) {}
+void __kmpc_end_master(IdentTy *Loc, int32_t TId) { FunctionTracingRAII(); }
 
 int32_t __kmpc_single(IdentTy *Loc, int32_t TId) {
+  FunctionTracingRAII();
   return __kmpc_master(Loc, TId);
 }
 
 void __kmpc_end_single(IdentTy *Loc, int32_t TId) {
+  FunctionTracingRAII();
   // The barrier is explicitly called.
 }
 
-void __kmpc_flush(IdentTy *Loc) { fence::kernel(__ATOMIC_SEQ_CST); }
+void __kmpc_flush(IdentTy *Loc) {
+  FunctionTracingRAII();
+  fence::kernel(__ATOMIC_SEQ_CST);
+}
 
-uint64_t __kmpc_warp_active_thread_mask(void) { return mapping::activemask(); }
+uint64_t __kmpc_warp_active_thread_mask(void) {
+  FunctionTracingRAII();
+  return mapping::activemask();
+}
 
-void __kmpc_syncwarp(uint64_t Mask) { synchronize::warp(Mask); }
+void __kmpc_syncwarp(uint64_t Mask) {
+  FunctionTracingRAII();
+  synchronize::warp(Mask);
+}
 
 void __kmpc_critical(IdentTy *Loc, int32_t TId, CriticalNameTy *Name) {
+  FunctionTracingRAII();
   omp_set_lock(reinterpret_cast<omp_lock_t *>(Name));
 }
 
 void __kmpc_end_critical(IdentTy *Loc, int32_t TId, CriticalNameTy *Name) {
+  FunctionTracingRAII();
   omp_unset_lock(reinterpret_cast<omp_lock_t *>(Name));
 }
 
 void omp_init_lock(omp_lock_t *Lock) { impl::initLock(Lock); }
 
 void omp_destroy_lock(omp_lock_t *Lock) { impl::destroyLock(Lock); }
 
 void omp_set_lock(omp_lock_t *Lock) { impl::setLock(Lock); }
 
 void omp_unset_lock(omp_lock_t *Lock) { impl::unsetLock(Lock); }
 
 int omp_test_lock(omp_lock_t *Lock) { return impl::testLock(Lock); }
 } // extern "C"
 
 #pragma omp end declare target

openmp/libomptarget/DeviceRTL/src/Tasking.cpp

 using namespace _OMP;
 
 #pragma omp declare target
 
 TaskDescriptorTy *__kmpc_omp_task_alloc(IdentTy *, uint32_t, int32_t,
                                         uint64_t TaskSizeInclPrivateValues,
                                         uint64_t SharedValuesSize,
                                         TaskFnTy TaskFn) {
+  FunctionTracingRAII();
   auto TaskSizeInclPrivateValuesPadded =
       utils::roundUp(TaskSizeInclPrivateValues, uint64_t(sizeof(void *)));
   auto TaskSizeTotal = TaskSizeInclPrivateValuesPadded + SharedValuesSize;
   TaskDescriptorTy *TaskDescriptor = (TaskDescriptorTy *)memory::allocGlobal(
       TaskSizeTotal, "explicit task descriptor");
   TaskDescriptor->Payload =
       utils::advance(TaskDescriptor, TaskSizeInclPrivateValuesPadded);
   TaskDescriptor->TaskFn = TaskFn;
 
   return TaskDescriptor;
 }
 
 int32_t __kmpc_omp_task(IdentTy *Loc, uint32_t TId,
                         TaskDescriptorTy *TaskDescriptor) {
+  FunctionTracingRAII();
   return __kmpc_omp_task_with_deps(Loc, TId, TaskDescriptor, 0, 0, 0, 0);
 }
 
 int32_t __kmpc_omp_task_with_deps(IdentTy *Loc, uint32_t TId,
                                   TaskDescriptorTy *TaskDescriptor, int32_t,
                                   void *, int32_t, void *) {
+  FunctionTracingRAII();
   state::DateEnvironmentRAII DERAII;
 
   TaskDescriptor->TaskFn(0, TaskDescriptor);
 
   memory::freeGlobal(TaskDescriptor, "explicit task descriptor");
   return 0;
 }
 
 void __kmpc_omp_task_begin_if0(IdentTy *Loc, uint32_t TId,
                                TaskDescriptorTy *TaskDescriptor) {
+  FunctionTracingRAII();
   state::enterDataEnvironment();
 }
 
 void __kmpc_omp_task_complete_if0(IdentTy *Loc, uint32_t TId,
                                   TaskDescriptorTy *TaskDescriptor) {
+  FunctionTracingRAII();
   state::exitDataEnvironment();
 
   memory::freeGlobal(TaskDescriptor, "explicit task descriptor");
 }
 
 void __kmpc_omp_wait_deps(IdentTy *Loc, uint32_t TId, int32_t, void *, int32_t,
-                          void *) {}
+                          void *) {
+  FunctionTracingRAII();
+}
 
-void __kmpc_taskgroup(IdentTy *Loc, uint32_t TId) {}
+void __kmpc_taskgroup(IdentTy *Loc, uint32_t TId) { FunctionTracingRAII(); }
 
-void __kmpc_end_taskgroup(IdentTy *Loc, uint32_t TId) {}
+void __kmpc_end_taskgroup(IdentTy *Loc, uint32_t TId) { FunctionTracingRAII(); }
 
-int32_t __kmpc_omp_taskyield(IdentTy *Loc, uint32_t TId, int) { return 0; }
+int32_t __kmpc_omp_taskyield(IdentTy *Loc, uint32_t TId, int) {
+  FunctionTracingRAII();
+  return 0;
+}
 
-int32_t __kmpc_omp_taskwait(IdentTy *Loc, uint32_t TId) { return 0; }
+int32_t __kmpc_omp_taskwait(IdentTy *Loc, uint32_t TId) {
+  FunctionTracingRAII();
+  return 0;
+}
 
 void __kmpc_taskloop(IdentTy *Loc, uint32_t TId,
                      TaskDescriptorTy *TaskDescriptor, int,
                      uint64_t *LowerBound, uint64_t *UpperBound, int64_t, int,
                      int32_t, uint64_t, void *) {
+  FunctionTracingRAII();
   // Skip task entirely if empty iteration space.
   if (*LowerBound > *UpperBound)
     return;
 
   // The compiler has already stored lb and ub in the TaskDescriptorTy structure
   // as we are using a single task to execute the entire loop, we can leave
   // the initial task_t untouched
   __kmpc_omp_task_with_deps(Loc, TId, TaskDescriptor, 0, 0, 0, 0);

openmp/libomptarget/DeviceRTL/src/Utils.cpp

 //===------- Utils.cpp - OpenMP device runtime utility functions -- C++ -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 //
 //===----------------------------------------------------------------------===//
 
 #include "Utils.h"
 
+#include "Debug.h"
 #include "Interface.h"
 #include "Mapping.h"
 
 #pragma omp declare target
 
 using namespace _OMP;
 
 namespace _OMP {

 
 int32_t utils::shuffleDown(uint64_t Mask, int32_t Var, uint32_t Delta,
                            int32_t Width) {
   return impl::shuffleDown(Mask, Var, Delta, Width);
 }
 
 extern "C" {
 int32_t __kmpc_shuffle_int32(int32_t Val, int16_t Delta, int16_t SrcLane) {
+  FunctionTracingRAII();
   return impl::shuffleDown(lanes::All, Val, Delta, SrcLane);
 }
 
 int64_t __kmpc_shuffle_int64(int64_t Val, int16_t Delta, int16_t Width) {
+  FunctionTracingRAII();
   uint32_t lo, hi;
   utils::unpack(Val, lo, hi);
   hi = impl::shuffleDown(lanes::All, hi, Delta, Width);
   lo = impl::shuffleDown(lanes::All, lo, Delta, Width);
   return utils::pack(lo, hi);
 }
 }
 
 #pragma omp end declare target

openmp/libomptarget/DeviceRTL/src/Workshare.cpp

static void popDST() {
   ThreadDSTPtr = OldDST;
 }
 
 extern "C" {
 
 // init
 void __kmpc_dispatch_init_4(IdentTy *loc, int32_t tid, int32_t schedule,
                             int32_t lb, int32_t ub, int32_t st, int32_t chunk) {
+  FunctionTracingRAII();
   DynamicScheduleTracker *DST = pushDST();
   omptarget_nvptx_LoopSupport<int32_t, int32_t>::dispatch_init(
       loc, tid, (kmp_sched_t)schedule, lb, ub, st, chunk, DST);
 }
 
 void __kmpc_dispatch_init_4u(IdentTy *loc, int32_t tid, int32_t schedule,
                              uint32_t lb, uint32_t ub, int32_t st,
                              int32_t chunk) {
+  FunctionTracingRAII();
   DynamicScheduleTracker *DST = pushDST();
   omptarget_nvptx_LoopSupport<uint32_t, int32_t>::dispatch_init(
       loc, tid, (kmp_sched_t)schedule, lb, ub, st, chunk, DST);
 }
 
 void __kmpc_dispatch_init_8(IdentTy *loc, int32_t tid, int32_t schedule,
                             int64_t lb, int64_t ub, int64_t st, int64_t chunk) {
+  FunctionTracingRAII();
   DynamicScheduleTracker *DST = pushDST();
   omptarget_nvptx_LoopSupport<int64_t, int64_t>::dispatch_init(
       loc, tid, (kmp_sched_t)schedule, lb, ub, st, chunk, DST);
 }
 
 void __kmpc_dispatch_init_8u(IdentTy *loc, int32_t tid, int32_t schedule,
                              uint64_t lb, uint64_t ub, int64_t st,
                              int64_t chunk) {
+  FunctionTracingRAII();
   DynamicScheduleTracker *DST = pushDST();
   omptarget_nvptx_LoopSupport<uint64_t, int64_t>::dispatch_init(
       loc, tid, (kmp_sched_t)schedule, lb, ub, st, chunk, DST);
 }
 
 // next
 int __kmpc_dispatch_next_4(IdentTy *loc, int32_t tid, int32_t *p_last,
                            int32_t *p_lb, int32_t *p_ub, int32_t *p_st) {
+  FunctionTracingRAII();
   DynamicScheduleTracker *DST = peekDST();
   return omptarget_nvptx_LoopSupport<int32_t, int32_t>::dispatch_next(
       loc, tid, p_last, p_lb, p_ub, p_st, DST);
 }
 
 int __kmpc_dispatch_next_4u(IdentTy *loc, int32_t tid, int32_t *p_last,
                             uint32_t *p_lb, uint32_t *p_ub, int32_t *p_st) {
+  FunctionTracingRAII();
   DynamicScheduleTracker *DST = peekDST();
   return omptarget_nvptx_LoopSupport<uint32_t, int32_t>::dispatch_next(
       loc, tid, p_last, p_lb, p_ub, p_st, DST);
 }
 
 int __kmpc_dispatch_next_8(IdentTy *loc, int32_t tid, int32_t *p_last,
                            int64_t *p_lb, int64_t *p_ub, int64_t *p_st) {
+  FunctionTracingRAII();
   DynamicScheduleTracker *DST = peekDST();
   return omptarget_nvptx_LoopSupport<int64_t, int64_t>::dispatch_next(
       loc, tid, p_last, p_lb, p_ub, p_st, DST);
 }
 
 int __kmpc_dispatch_next_8u(IdentTy *loc, int32_t tid, int32_t *p_last,
                             uint64_t *p_lb, uint64_t *p_ub, int64_t *p_st) {
+  FunctionTracingRAII();
   DynamicScheduleTracker *DST = peekDST();
   return omptarget_nvptx_LoopSupport<uint64_t, int64_t>::dispatch_next(
       loc, tid, p_last, p_lb, p_ub, p_st, DST);
 }
 
 // fini
 void __kmpc_dispatch_fini_4(IdentTy *loc, int32_t tid) {
+  FunctionTracingRAII();
   omptarget_nvptx_LoopSupport<int32_t, int32_t>::dispatch_fini();
   popDST();
 }
 
 void __kmpc_dispatch_fini_4u(IdentTy *loc, int32_t tid) {
+  FunctionTracingRAII();
   omptarget_nvptx_LoopSupport<uint32_t, int32_t>::dispatch_fini();
   popDST();
 }
 
 void __kmpc_dispatch_fini_8(IdentTy *loc, int32_t tid) {
+  FunctionTracingRAII();
   omptarget_nvptx_LoopSupport<int64_t, int64_t>::dispatch_fini();
   popDST();
 }
 
 void __kmpc_dispatch_fini_8u(IdentTy *loc, int32_t tid) {
+  FunctionTracingRAII();
   omptarget_nvptx_LoopSupport<uint64_t, int64_t>::dispatch_fini();
   popDST();
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 // KMP interface implementation (static loops)
 ////////////////////////////////////////////////////////////////////////////////
 
 void __kmpc_for_static_init_4(IdentTy *loc, int32_t global_tid,
                               int32_t schedtype, int32_t *plastiter,
                               int32_t *plower, int32_t *pupper,
                               int32_t *pstride, int32_t incr, int32_t chunk) {
+  FunctionTracingRAII();
   omptarget_nvptx_LoopSupport<int32_t, int32_t>::for_static_init(
       global_tid, schedtype, plastiter, plower, pupper, pstride, chunk,
       mapping::isSPMDMode());
 }
 
 void __kmpc_for_static_init_4u(IdentTy *loc, int32_t global_tid,
                                int32_t schedtype, int32_t *plastiter,
                                uint32_t *plower, uint32_t *pupper,
                                int32_t *pstride, int32_t incr, int32_t chunk) {
+  FunctionTracingRAII();
   omptarget_nvptx_LoopSupport<uint32_t, int32_t>::for_static_init(
       global_tid, schedtype, plastiter, plower, pupper, pstride, chunk,
       mapping::isSPMDMode());
 }
 
 void __kmpc_for_static_init_8(IdentTy *loc, int32_t global_tid,
                               int32_t schedtype, int32_t *plastiter,
                               int64_t *plower, int64_t *pupper,
                               int64_t *pstride, int64_t incr, int64_t chunk) {
+  FunctionTracingRAII();
   omptarget_nvptx_LoopSupport<int64_t, int64_t>::for_static_init(
       global_tid, schedtype, plastiter, plower, pupper, pstride, chunk,
       mapping::isSPMDMode());
 }
 
 void __kmpc_for_static_init_8u(IdentTy *loc, int32_t global_tid,
                                int32_t schedtype, int32_t *plastiter,
                                uint64_t *plower, uint64_t *pupper,
                                int64_t *pstride, int64_t incr, int64_t chunk) {
+  FunctionTracingRAII();
   omptarget_nvptx_LoopSupport<uint64_t, int64_t>::for_static_init(
       global_tid, schedtype, plastiter, plower, pupper, pstride, chunk,
       mapping::isSPMDMode());
 }
 
 void __kmpc_distribute_static_init_4(IdentTy *loc, int32_t global_tid,
                                      int32_t schedtype, int32_t *plastiter,
                                      int32_t *plower, int32_t *pupper,
                                      int32_t *pstride, int32_t incr,
                                      int32_t chunk) {
+  FunctionTracingRAII();
   omptarget_nvptx_LoopSupport<int32_t, int32_t>::for_static_init(
       global_tid, schedtype, plastiter, plower, pupper, pstride, chunk,
       mapping::isSPMDMode());
 }
 
 void __kmpc_distribute_static_init_4u(IdentTy *loc, int32_t global_tid,
                                       int32_t schedtype, int32_t *plastiter,
                                       uint32_t *plower, uint32_t *pupper,
                                       int32_t *pstride, int32_t incr,
                                       int32_t chunk) {
+  FunctionTracingRAII();
   omptarget_nvptx_LoopSupport<uint32_t, int32_t>::for_static_init(
       global_tid, schedtype, plastiter, plower, pupper, pstride, chunk,
       mapping::isSPMDMode());
 }
 
 void __kmpc_distribute_static_init_8(IdentTy *loc, int32_t global_tid,
                                      int32_t schedtype, int32_t *plastiter,
                                      int64_t *plower, int64_t *pupper,
                                      int64_t *pstride, int64_t incr,
                                      int64_t chunk) {
+  FunctionTracingRAII();
   omptarget_nvptx_LoopSupport<int64_t, int64_t>::for_static_init(
       global_tid, schedtype, plastiter, plower, pupper, pstride, chunk,
       mapping::isSPMDMode());
 }
 
 void __kmpc_distribute_static_init_8u(IdentTy *loc, int32_t global_tid,
                                       int32_t schedtype, int32_t *plastiter,
                                       uint64_t *plower, uint64_t *pupper,
                                       int64_t *pstride, int64_t incr,
                                       int64_t chunk) {
+  FunctionTracingRAII();
   omptarget_nvptx_LoopSupport<uint64_t, int64_t>::for_static_init(
       global_tid, schedtype, plastiter, plower, pupper, pstride, chunk,
       mapping::isSPMDMode());
 }
 
-void __kmpc_for_static_fini(IdentTy *loc, int32_t global_tid) {}
+void __kmpc_for_static_fini(IdentTy *loc, int32_t global_tid) {
+  FunctionTracingRAII();
+}
 
-void __kmpc_distribute_static_fini(IdentTy *loc, int32_t global_tid) {}
+void __kmpc_distribute_static_fini(IdentTy *loc, int32_t global_tid) {
+  FunctionTracingRAII();
+}
 }
 
 #pragma omp end declare target