Merge branch 'main' into users/mingmingl-llvm/samplefdo-profile-formatusers/mingmingl-llvm/samplefdo-profile-format

13 files changed, 0 insertions, 2362 deletions

diff --git a/offload/DeviceRTL/include/Allocator.h b/offload/DeviceRTL/include/Allocator.h
deleted file mode 100644
index dc4d029ed75f..000000000000
--- a/offload/DeviceRTL/include/Allocator.h
+++ /dev/null
@@ -1,45 +0,0 @@
-//===-------- Allocator.h - OpenMP memory allocator interface ---- 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
-//
-//===----------------------------------------------------------------------===//
-//
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef OMPTARGET_ALLOCATOR_H
-#define OMPTARGET_ALLOCATOR_H
-
-#include "DeviceTypes.h"
-
-// Forward declaration.
-struct KernelEnvironmentTy;
-
-namespace ompx {
-
-namespace allocator {
-
-static uint64_t constexpr ALIGNMENT = 16;
-
-/// Initialize the allocator according to \p KernelEnvironment
-void init(bool IsSPMD, KernelEnvironmentTy &KernelEnvironment);
-
-/// Allocate \p Size bytes.
-[[gnu::alloc_size(1), gnu::assume_aligned(ALIGNMENT), gnu::malloc]] void *
-alloc(uint64_t Size);
-
-/// Free the allocation pointed to by \p Ptr.
-void free(void *Ptr);
-
-} // namespace allocator
-
-} // namespace ompx
-
-extern "C" {
-void *malloc(size_t Size);
-void free(void *Ptr);
-}
-
-#endif
diff --git a/offload/DeviceRTL/include/Configuration.h b/offload/DeviceRTL/include/Configuration.h
deleted file mode 100644
index 95408933dd86..000000000000
--- a/offload/DeviceRTL/include/Configuration.h
+++ /dev/null
@@ -1,68 +0,0 @@
-//===--- Configuration.h - OpenMP device configuration interface -- 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
-//
-//===----------------------------------------------------------------------===//
-//
-// API to query the global (constant) device environment.
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef OMPTARGET_CONFIGURATION_H
-#define OMPTARGET_CONFIGURATION_H
-
-#include "Shared/Environment.h"
-
-#include "DeviceTypes.h"
-
-namespace ompx {
-namespace config {
-
-/// Return the number of devices in the system, same number as returned on the
-/// host by omp_get_num_devices.
-uint32_t getNumDevices();
-
-/// Return the device number in the system for omp_get_device_num.
-uint32_t getDeviceNum();
-
-/// Return the user chosen debug level.
-uint32_t getDebugKind();
-
-/// Return if teams oversubscription is assumed
-uint32_t getAssumeTeamsOversubscription();
-
-/// Return if threads oversubscription is assumed
-uint32_t getAssumeThreadsOversubscription();
-
-/// Return the amount of dynamic shared memory that was allocated at launch.
-uint64_t getDynamicMemorySize();
-
-/// Returns the cycles per second of the device's fixed frequency clock.
-uint64_t getClockFrequency();
-
-/// Returns the pointer to the beginning of the indirect call table.
-void *getIndirectCallTablePtr();
-
-/// Returns the size of the indirect call table.
-uint64_t getIndirectCallTableSize();
-
-/// Returns the size of the indirect call table.
-uint64_t getHardwareParallelism();
-
-/// Return if debugging is enabled for the given debug kind.
-bool isDebugMode(DeviceDebugKind Level);
-
-/// Indicates if this kernel may require thread-specific states, or if it was
-/// explicitly disabled by the user.
-bool mayUseThreadStates();
-
-/// Indicates if this kernel may require data environments for nested
-/// parallelism, or if it was explicitly disabled by the user.
-bool mayUseNestedParallelism();
-
-} // namespace config
-} // namespace ompx
-
-#endif
diff --git a/offload/DeviceRTL/include/Debug.h b/offload/DeviceRTL/include/Debug.h
deleted file mode 100644
index 98d0fa498d95..000000000000
--- a/offload/DeviceRTL/include/Debug.h
+++ /dev/null
@@ -1,44 +0,0 @@
-//===-------- Debug.h ---- 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
-//
-//===----------------------------------------------------------------------===//
-//
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef OMPTARGET_DEVICERTL_DEBUG_H
-#define OMPTARGET_DEVICERTL_DEBUG_H
-
-#include "Configuration.h"
-#include "LibC.h"
-
-/// Assertion
-///
-/// {
-extern "C" {
-void __assert_assume(bool condition);
-void __assert_fail(const char *expr, const char *file, unsigned line,
-                   const char *function);
-void __assert_fail_internal(const char *expr, const char *msg, const char *file,
-                            unsigned line, const char *function);
-}
-
-#define ASSERT(expr, msg)                                                      \
-  {                                                                            \
-    if (config::isDebugMode(DeviceDebugKind::Assertion) && !(expr))            \
-      __assert_fail_internal(#expr, msg, __FILE__, __LINE__,                   \
-                             __PRETTY_FUNCTION__);                             \
-    else                                                                       \
-      __assert_assume(expr);                                                   \
-  }
-#define UNREACHABLE(msg)                                                       \
-  printf(msg);                                                                 \
-  __builtin_trap();                                                            \
-  __builtin_unreachable();
-
-///}
-
-#endif
diff --git a/offload/DeviceRTL/include/DeviceTypes.h b/offload/DeviceRTL/include/DeviceTypes.h
deleted file mode 100644
index 2e5d92380f04..000000000000
--- a/offload/DeviceRTL/include/DeviceTypes.h
+++ /dev/null
@@ -1,166 +0,0 @@
-//===---------- DeviceTypes.h - OpenMP types ---------------------- 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
-//
-//===----------------------------------------------------------------------===//
-//
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef OMPTARGET_TYPES_H
-#define OMPTARGET_TYPES_H
-
-#include <gpuintrin.h>
-#include <stddef.h>
-#include <stdint.h>
-
-template <typename T> using Private = __gpu_private T;
-template <typename T> using Constant = __gpu_constant T;
-template <typename T> using Local = __gpu_local T;
-template <typename T> using Global = __gpu_local T;
-
-enum omp_proc_bind_t {
-  omp_proc_bind_false = 0,
-  omp_proc_bind_true = 1,
-  omp_proc_bind_master = 2,
-  omp_proc_bind_close = 3,
-  omp_proc_bind_spread = 4
-};
-
-enum omp_sched_t {
-  omp_sched_static = 1,  /* chunkSize >0 */
-  omp_sched_dynamic = 2, /* chunkSize >0 */
-  omp_sched_guided = 3,  /* chunkSize >0 */
-  omp_sched_auto = 4,    /* no chunkSize */
-};
-
-enum kmp_sched_t {
-  kmp_sched_static_chunk = 33,
-  kmp_sched_static_nochunk = 34,
-  kmp_sched_dynamic = 35,
-  kmp_sched_guided = 36,
-  kmp_sched_runtime = 37,
-  kmp_sched_auto = 38,
-
-  kmp_sched_static_balanced_chunk = 45,
-
-  kmp_sched_static_ordered = 65,
-  kmp_sched_static_nochunk_ordered = 66,
-  kmp_sched_dynamic_ordered = 67,
-  kmp_sched_guided_ordered = 68,
-  kmp_sched_runtime_ordered = 69,
-  kmp_sched_auto_ordered = 70,
-
-  kmp_sched_distr_static_chunk = 91,
-  kmp_sched_distr_static_nochunk = 92,
-  kmp_sched_distr_static_chunk_sched_static_chunkone = 93,
-
-  kmp_sched_default = kmp_sched_static_nochunk,
-  kmp_sched_unordered_first = kmp_sched_static_chunk,
-  kmp_sched_unordered_last = kmp_sched_auto,
-  kmp_sched_ordered_first = kmp_sched_static_ordered,
-  kmp_sched_ordered_last = kmp_sched_auto_ordered,
-  kmp_sched_distribute_first = kmp_sched_distr_static_chunk,
-  kmp_sched_distribute_last =
-      kmp_sched_distr_static_chunk_sched_static_chunkone,
-
-  /* Support for OpenMP 4.5 monotonic and nonmonotonic schedule modifiers.
-   * Since we need to distinguish the three possible cases (no modifier,
-   * monotonic modifier, nonmonotonic modifier), we need separate bits for
-   * each modifier. The absence of monotonic does not imply nonmonotonic,
-   * especially since 4.5 says that the behaviour of the "no modifier" case
-   * is implementation defined in 4.5, but will become "nonmonotonic" in 5.0.
-   *
-   * Since we're passing a full 32 bit value, we can use a couple of high
-   * bits for these flags; out of paranoia we avoid the sign bit.
-   *
-   * These modifiers can be or-ed into non-static schedules by the compiler
-   * to pass the additional information. They will be stripped early in the
-   * processing in __kmp_dispatch_init when setting up schedules, so
-   * most of the code won't ever see schedules with these bits set.
-   */
-  kmp_sched_modifier_monotonic = (1 << 29),
-  /**< Set if the monotonic schedule modifier was present */
-  kmp_sched_modifier_nonmonotonic = (1 << 30),
-/**< Set if the nonmonotonic schedule modifier was present */
-
-#define SCHEDULE_WITHOUT_MODIFIERS(s)                                          \
-  (enum kmp_sched_t)(                                                          \
-      (s) & ~(kmp_sched_modifier_nonmonotonic | kmp_sched_modifier_monotonic))
-#define SCHEDULE_HAS_MONOTONIC(s) (((s) & kmp_sched_modifier_monotonic) != 0)
-#define SCHEDULE_HAS_NONMONOTONIC(s)                                           \
-  (((s) & kmp_sched_modifier_nonmonotonic) != 0)
-#define SCHEDULE_HAS_NO_MODIFIERS(s)                                           \
-  (((s) & (kmp_sched_modifier_nonmonotonic | kmp_sched_modifier_monotonic)) == \
-   0)
-
-};
-
-struct TaskDescriptorTy;
-using TaskFnTy = int32_t (*)(int32_t global_tid, TaskDescriptorTy *taskDescr);
-struct TaskDescriptorTy {
-  void *Payload;
-  TaskFnTy TaskFn;
-};
-
-using LaneMaskTy = uint64_t;
-
-namespace lanes {
-enum : LaneMaskTy { All = ~(LaneMaskTy)0 };
-} // namespace lanes
-
-/// The ident structure that describes a source location. The struct is
-/// identical to the one in the kmp.h file. We maintain the same data structure
-/// for compatibility.
-struct IdentTy {
-  int32_t reserved_1;  /**<  might be used in Fortran; see above  */
-  int32_t flags;       /**<  also f.flags; KMP_IDENT_xxx flags; KMP_IDENT_KMPC
-                            identifies this union member  */
-  int32_t reserved_2;  /**<  not really used in Fortran any more; see above */
-  int32_t reserved_3;  /**<  source[4] in Fortran, do not use for C++  */
-  char const *psource; /**<  String describing the source location.
-                       The string is composed of semi-colon separated fields
-                       which describe the source file, the function and a pair
-                       of line numbers that delimit the construct. */
-};
-
-using __kmpc_impl_lanemask_t = LaneMaskTy;
-
-using ParallelRegionFnTy = void *;
-
-using CriticalNameTy = int32_t[8];
-
-struct omp_lock_t {
-  void *Lock;
-};
-
-using InterWarpCopyFnTy = void (*)(void *src, int32_t warp_num);
-using ShuffleReductFnTy = void (*)(void *rhsData, int16_t lane_id,
-                                   int16_t lane_offset, int16_t shortCircuit);
-using ListGlobalFnTy = void (*)(void *buffer, int idx, void *reduce_data);
-
-/// Macros for allocating variables in different address spaces.
-///{
-
-// Follows the pattern in interface.h
-typedef enum omp_allocator_handle_t {
-  omp_null_allocator = 0,
-  omp_default_mem_alloc = 1,
-  omp_large_cap_mem_alloc = 2,
-  omp_const_mem_alloc = 3,
-  omp_high_bw_mem_alloc = 4,
-  omp_low_lat_mem_alloc = 5,
-  omp_cgroup_mem_alloc = 6,
-  omp_pteam_mem_alloc = 7,
-  omp_thread_mem_alloc = 8,
-  KMP_ALLOCATOR_MAX_HANDLE = ~(0LU)
-} omp_allocator_handle_t;
-
-#define __PRAGMA(STR) _Pragma(#STR)
-#define OMP_PRAGMA(STR) __PRAGMA(omp STR)
-
-///}
-
-#endif
diff --git a/offload/DeviceRTL/include/DeviceUtils.h b/offload/DeviceRTL/include/DeviceUtils.h
deleted file mode 100644
index b92514ee9838..000000000000
--- a/offload/DeviceRTL/include/DeviceUtils.h
+++ /dev/null
@@ -1,96 +0,0 @@
-//===--- DeviceUtils.h - 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
-//
-//===----------------------------------------------------------------------===//
-//
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef OMPTARGET_DEVICERTL_DEVICE_UTILS_H
-#define OMPTARGET_DEVICERTL_DEVICE_UTILS_H
-
-#include "DeviceTypes.h"
-#include "Shared/Utils.h"
-
-namespace utils {
-
-template <typename T> struct type_identity {
-  using type = T;
-};
-
-template <typename T, T v> struct integral_constant {
-  inline static constexpr T value = v;
-};
-
-/// Freestanding SFINAE helpers.
-template <class T> struct remove_cv : type_identity<T> {};
-template <class T> struct remove_cv<const T> : type_identity<T> {};
-template <class T> struct remove_cv<volatile T> : type_identity<T> {};
-template <class T> struct remove_cv<const volatile T> : type_identity<T> {};
-template <class T> using remove_cv_t = typename remove_cv<T>::type;
-
-using true_type = integral_constant<bool, true>;
-using false_type = integral_constant<bool, false>;
-
-template <typename T, typename U> struct is_same : false_type {};
-template <typename T> struct is_same<T, T> : true_type {};
-template <typename T, typename U>
-inline constexpr bool is_same_v = is_same<T, U>::value;
-
-template <typename T> struct is_floating_point {
-  inline static constexpr bool value =
-      is_same_v<remove_cv_t<T>, float> || is_same_v<remove_cv_t<T>, double>;
-};
-template <typename T>
-inline constexpr bool is_floating_point_v = is_floating_point<T>::value;
-
-template <bool B, typename T = void> struct enable_if;
-template <typename T> struct enable_if<true, T> : type_identity<T> {};
-template <bool B, typename T = void>
-using enable_if_t = typename enable_if<B, T>::type;
-
-template <class T> struct remove_addrspace : type_identity<T> {};
-template <class T, int N>
-struct remove_addrspace<T [[clang::address_space(N)]]> : type_identity<T> {};
-template <class T>
-using remove_addrspace_t = typename remove_addrspace<T>::type;
-
-template <typename To, typename From> inline To bitCast(From V) {
-  static_assert(sizeof(To) == sizeof(From), "Bad conversion");
-  return __builtin_bit_cast(To, V);
-}
-
-/// Return the value \p Var from thread Id \p SrcLane in the warp if the thread
-/// is identified by \p Mask.
-int32_t shuffle(uint64_t Mask, int32_t Var, int32_t SrcLane, int32_t Width);
-
-int32_t shuffleDown(uint64_t Mask, int32_t Var, uint32_t Delta, int32_t Width);
-
-int64_t shuffleDown(uint64_t Mask, int64_t Var, uint32_t Delta, int32_t Width);
-
-uint64_t ballotSync(uint64_t Mask, int32_t Pred);
-
-/// Return \p LowBits and \p HighBits packed into a single 64 bit value.
-uint64_t pack(uint32_t LowBits, uint32_t HighBits);
-
-/// Unpack \p Val into \p LowBits and \p HighBits.
-void unpack(uint64_t Val, uint32_t &LowBits, uint32_t &HighBits);
-
-/// Return true iff \p Ptr is pointing into shared (local) memory (AS(3)).
-bool isSharedMemPtr(void *Ptr);
-
-/// Return true iff \p Ptr is pointing into (thread) local memory (AS(5)).
-bool isThreadLocalMemPtr(void *Ptr);
-
-/// A  pointer variable that has by design an `undef` value. Use with care.
-[[clang::loader_uninitialized]] static void *const UndefPtr;
-
-#define OMP_LIKELY(EXPR) __builtin_expect((bool)(EXPR), true)
-#define OMP_UNLIKELY(EXPR) __builtin_expect((bool)(EXPR), false)
-
-} // namespace utils
-
-#endif
diff --git a/offload/DeviceRTL/include/Interface.h b/offload/DeviceRTL/include/Interface.h
deleted file mode 100644
index c4bfaaa2404b..000000000000
--- a/offload/DeviceRTL/include/Interface.h
+++ /dev/null
@@ -1,366 +0,0 @@
-//===-------- Interface.h - OpenMP interface ---------------------- 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
-//
-//===----------------------------------------------------------------------===//
-//
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef OMPTARGET_DEVICERTL_INTERFACE_H
-#define OMPTARGET_DEVICERTL_INTERFACE_H
-
-#include "Shared/Environment.h"
-
-#include "DeviceTypes.h"
-
-/// External API
-///
-///{
-
-extern "C" {
-
-/// ICV: dyn-var, constant 0
-///
-/// setter: ignored.
-/// getter: returns 0.
-///
-///{
-void omp_set_dynamic(int);
-int omp_get_dynamic(void);
-///}
-
-/// ICV: nthreads-var, integer
-///
-/// scope: data environment
-///
-/// setter: ignored.
-/// getter: returns false.
-///
-/// implementation notes:
-///
-///
-///{
-void omp_set_num_threads(int);
-int omp_get_max_threads(void);
-///}
-
-/// ICV: thread-limit-var, computed
-///
-/// getter: returns thread limited defined during launch.
-///
-///{
-int omp_get_thread_limit(void);
-///}
-
-/// ICV: max-active-level-var, constant 1
-///
-/// setter: ignored.
-/// getter: returns 1.
-///
-///{
-void omp_set_max_active_levels(int);
-int omp_get_max_active_levels(void);
-///}
-
-/// ICV: places-partition-var
-///
-///
-///{
-///}
-
-/// ICV: active-level-var, 0 or 1
-///
-/// getter: returns 0 or 1.
-///
-///{
-int omp_get_active_level(void);
-///}
-
-/// ICV: level-var
-///
-/// getter: returns parallel region nesting
-///
-///{
-int omp_get_level(void);
-///}
-
-/// ICV: run-sched-var
-///
-///
-///{
-void omp_set_schedule(omp_sched_t, int);
-void omp_get_schedule(omp_sched_t *, int *);
-///}
-
-/// TODO this is incomplete.
-int omp_get_num_threads(void);
-int omp_get_thread_num(void);
-void omp_set_nested(int);
-
-int omp_get_nested(void);
-
-void omp_set_max_active_levels(int Level);
-
-int omp_get_max_active_levels(void);
-
-omp_proc_bind_t omp_get_proc_bind(void);
-
-int omp_get_num_places(void);
-
-int omp_get_place_num_procs(int place_num);
-
-void omp_get_place_proc_ids(int place_num, int *ids);
-
-int omp_get_place_num(void);
-
-int omp_get_partition_num_places(void);
-
-void omp_get_partition_place_nums(int *place_nums);
-
-int omp_get_cancellation(void);
-
-void omp_set_default_device(int deviceId);
-
-int omp_get_default_device(void);
-
-int omp_get_num_devices(void);
-
-int omp_get_device_num(void);
-
-int omp_get_num_teams(void);
-
-int omp_get_team_num();
-
-int omp_get_initial_device(void);
-
-void *llvm_omp_target_dynamic_shared_alloc();
-
-/// Synchronization
-///
-///{
-void omp_init_lock(omp_lock_t *Lock);
-
-void omp_destroy_lock(omp_lock_t *Lock);
-
-void omp_set_lock(omp_lock_t *Lock);
-
-void omp_unset_lock(omp_lock_t *Lock);
-
-int omp_test_lock(omp_lock_t *Lock);
-///}
-
-/// Tasking
-///
-///{
-int omp_in_final(void);
-
-int omp_get_max_task_priority(void);
-///}
-
-/// Misc
-///
-///{
-double omp_get_wtick(void);
-
-double omp_get_wtime(void);
-///}
-}
-
-extern "C" {
-/// 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*.
-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 has to be the
-/// allocated by __kmpc_alloc_shared by the same thread.
-void __kmpc_free_shared(void *Ptr, uint64_t Bytes);
-
-/// Get a pointer to the memory buffer containing dynamically allocated shared
-/// memory configured at launch.
-void *__kmpc_get_dynamic_shared();
-
-/// Allocate sufficient space for \p NumArgs sequential `void*` and store the
-/// allocation address in \p GlobalArgs.
-///
-/// Called by the main thread prior to a parallel region.
-///
-/// We also remember it in GlobalArgsPtr to ensure the worker threads and
-/// deallocation function know the allocation address too.
-void __kmpc_begin_sharing_variables(void ***GlobalArgs, uint64_t NumArgs);
-
-/// Deallocate the memory allocated by __kmpc_begin_sharing_variables.
-///
-/// Called by the main thread after a parallel region.
-void __kmpc_end_sharing_variables();
-
-/// Store the allocation address obtained via __kmpc_begin_sharing_variables in
-/// \p GlobalArgs.
-///
-/// Called by the worker threads in the parallel region (function).
-void __kmpc_get_shared_variables(void ***GlobalArgs);
-
-/// External interface to get the thread ID.
-uint32_t __kmpc_get_hardware_thread_id_in_block();
-
-/// External interface to get the number of threads.
-uint32_t __kmpc_get_hardware_num_threads_in_block();
-
-/// External interface to get the warp size.
-uint32_t __kmpc_get_warp_size();
-
-/// Kernel
-///
-///{
-// Forward declaration
-struct KernelEnvironmentTy;
-
-int8_t __kmpc_is_spmd_exec_mode();
-
-int32_t __kmpc_target_init(KernelEnvironmentTy &KernelEnvironment,
-                           KernelLaunchEnvironmentTy &KernelLaunchEnvironment);
-
-void __kmpc_target_deinit();
-
-///}
-
-/// Reduction
-///
-///{
-void *__kmpc_reduction_get_fixed_buffer();
-
-int32_t __kmpc_nvptx_parallel_reduce_nowait_v2(IdentTy *Loc,
-                                               uint64_t reduce_data_size,
-                                               void *reduce_data,
-                                               ShuffleReductFnTy shflFct,
-                                               InterWarpCopyFnTy cpyFct);
-
-int32_t __kmpc_nvptx_teams_reduce_nowait_v2(
-    IdentTy *Loc, void *GlobalBuffer, uint32_t num_of_records,
-    uint64_t reduce_data_size, void *reduce_data, ShuffleReductFnTy shflFct,
-    InterWarpCopyFnTy cpyFct, ListGlobalFnTy lgcpyFct, ListGlobalFnTy lgredFct,
-    ListGlobalFnTy glcpyFct, ListGlobalFnTy glredFct);
-///}
-
-/// Synchronization
-///
-///{
-void __kmpc_ordered(IdentTy *Loc, int32_t TId);
-
-void __kmpc_end_ordered(IdentTy *Loc, int32_t TId);
-
-int32_t __kmpc_cancel_barrier(IdentTy *Loc_ref, int32_t TId);
-
-void __kmpc_barrier(IdentTy *Loc_ref, int32_t TId);
-
-void __kmpc_barrier_simple_spmd(IdentTy *Loc_ref, int32_t TId);
-
-void __kmpc_barrier_simple_generic(IdentTy *Loc_ref, int32_t TId);
-
-int32_t __kmpc_master(IdentTy *Loc, int32_t TId);
-
-void __kmpc_end_master(IdentTy *Loc, int32_t TId);
-
-int32_t __kmpc_masked(IdentTy *Loc, int32_t TId, int32_t Filter);
-
-void __kmpc_end_masked(IdentTy *Loc, int32_t TId);
-
-int32_t __kmpc_single(IdentTy *Loc, int32_t TId);
-
-void __kmpc_end_single(IdentTy *Loc, int32_t TId);
-
-void __kmpc_flush(IdentTy *Loc);
-
-uint64_t __kmpc_warp_active_thread_mask(void);
-
-void __kmpc_syncwarp(uint64_t Mask);
-
-void __kmpc_critical(IdentTy *Loc, int32_t TId, CriticalNameTy *Name);
-
-void __kmpc_end_critical(IdentTy *Loc, int32_t TId, CriticalNameTy *Name);
-///}
-
-/// Parallelism
-///
-///{
-/// TODO
-void __kmpc_kernel_prepare_parallel(ParallelRegionFnTy WorkFn);
-
-/// TODO
-bool __kmpc_kernel_parallel(ParallelRegionFnTy *WorkFn);
-
-/// TODO
-void __kmpc_kernel_end_parallel();
-
-/// TODO
-void __kmpc_push_proc_bind(IdentTy *Loc, uint32_t TId, int ProcBind);
-
-/// TODO
-void __kmpc_push_num_teams(IdentTy *Loc, int32_t TId, int32_t NumTeams,
-                           int32_t ThreadLimit);
-
-/// TODO
-uint16_t __kmpc_parallel_level(IdentTy *Loc, uint32_t);
-
-///}
-
-/// Tasking
-///
-///{
-TaskDescriptorTy *__kmpc_omp_task_alloc(IdentTy *, int32_t, int32_t,
-                                        size_t TaskSizeInclPrivateValues,
-                                        size_t SharedValuesSize,
-                                        TaskFnTy TaskFn);
-
-int32_t __kmpc_omp_task(IdentTy *Loc, uint32_t TId,
-                        TaskDescriptorTy *TaskDescriptor);
-
-int32_t __kmpc_omp_task_with_deps(IdentTy *Loc, uint32_t TId,
-                                  TaskDescriptorTy *TaskDescriptor, int32_t,
-                                  void *, int32_t, void *);
-
-void __kmpc_omp_task_begin_if0(IdentTy *Loc, uint32_t TId,
-                               TaskDescriptorTy *TaskDescriptor);
-
-void __kmpc_omp_task_complete_if0(IdentTy *Loc, uint32_t TId,
-                                  TaskDescriptorTy *TaskDescriptor);
-
-void __kmpc_omp_wait_deps(IdentTy *Loc, uint32_t TId, int32_t, void *, int32_t,
-                          void *);
-
-void __kmpc_taskgroup(IdentTy *Loc, uint32_t TId);
-
-void __kmpc_end_taskgroup(IdentTy *Loc, uint32_t TId);
-
-int32_t __kmpc_omp_taskyield(IdentTy *Loc, uint32_t TId, int);
-
-int32_t __kmpc_omp_taskwait(IdentTy *Loc, uint32_t TId);
-
-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 *);
-///}
-
-/// Misc
-///
-///{
-int32_t __kmpc_cancellationpoint(IdentTy *Loc, int32_t TId, int32_t CancelVal);
-
-int32_t __kmpc_cancel(IdentTy *Loc, int32_t TId, int32_t CancelVal);
-///}
-
-/// Shuffle
-///
-///{
-int32_t __kmpc_shuffle_int32(int32_t val, int16_t delta, int16_t size);
-int64_t __kmpc_shuffle_int64(int64_t val, int16_t delta, int16_t size);
-
-///}
-}
-
-#endif
diff --git a/offload/DeviceRTL/include/LibC.h b/offload/DeviceRTL/include/LibC.h
deleted file mode 100644
index 94b5e6519606..000000000000
--- a/offload/DeviceRTL/include/LibC.h
+++ /dev/null
@@ -1,23 +0,0 @@
-//===--------- LibC.h - Simple implementation of libc 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
-//
-//===----------------------------------------------------------------------===//
-//
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef OMPTARGET_LIBC_H
-#define OMPTARGET_LIBC_H
-
-#include "DeviceTypes.h"
-
-namespace ompx {
-
-int printf(const char *Format, ...);
-
-} // namespace ompx
-
-#endif
diff --git a/offload/DeviceRTL/include/Mapping.h b/offload/DeviceRTL/include/Mapping.h
deleted file mode 100644
index 8ba018b5314a..000000000000
--- a/offload/DeviceRTL/include/Mapping.h
+++ /dev/null
@@ -1,108 +0,0 @@
-//===--------- Mapping.h - OpenMP device runtime mapping helpers -- 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
-//
-//===----------------------------------------------------------------------===//
-//
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef OMPTARGET_MAPPING_H
-#define OMPTARGET_MAPPING_H
-
-#include "DeviceTypes.h"
-
-namespace ompx {
-
-namespace mapping {
-
-enum {
-  DIM_X = __GPU_X_DIM,
-  DIM_Y = __GPU_Y_DIM,
-  DIM_Z = __GPU_Z_DIM,
-};
-
-inline constexpr uint32_t MaxThreadsPerTeam = 1024;
-
-/// Initialize the mapping machinery.
-void init(bool IsSPMD);
-
-/// Return true if the kernel is executed in SPMD mode.
-bool isSPMDMode();
-
-/// Return true if the kernel is executed in generic mode.
-bool isGenericMode();
-
-/// Return true if the executing thread is the main thread in generic mode.
-/// These functions will lookup state and it is required that that is OK for the
-/// thread and location. See also `isInitialThreadInLevel0` for a stateless
-/// alternative for certain situations, e.g. during initialization.
-bool isMainThreadInGenericMode();
-bool isMainThreadInGenericMode(bool IsSPMD);
-
-/// Return true if this thread is the initial thread in parallel level 0.
-///
-/// The thread for which this returns true should be used for single threaded
-/// initialization tasks. We pick a special thread to ensure there are no
-/// races between the initialization and the first read of initialized state.
-bool isInitialThreadInLevel0(bool IsSPMD);
-
-/// Return true if the executing thread has the lowest Id of the active threads
-/// in the warp.
-bool isLeaderInWarp();
-
-/// Return a mask describing all active threads in the warp.
-LaneMaskTy activemask();
-
-/// Return a mask describing all threads with a smaller Id in the warp.
-LaneMaskTy lanemaskLT();
-
-/// Return a mask describing all threads with a larger Id in the warp.
-LaneMaskTy lanemaskGT();
-
-/// Return the thread Id in the warp, in [0, getWarpSize()).
-uint32_t getThreadIdInWarp();
-
-/// Return the warp size, thus number of threads in the warp.
-uint32_t getWarpSize();
-
-/// Return the warp id in the block, in [0, getNumberOfWarpsInBlock()]
-uint32_t getWarpIdInBlock();
-
-/// Return the number of warps in the block.
-uint32_t getNumberOfWarpsInBlock();
-
-/// Return the thread Id in the block, in [0, getNumberOfThreadsInBlock(Dim)).
-uint32_t getThreadIdInBlock(int32_t Dim = DIM_X);
-
-/// Return the block size, thus number of threads in the block.
-uint32_t getNumberOfThreadsInBlock(int32_t Dim = DIM_X);
-
-/// Return the block Id in the kernel, in [0, getNumberOfBlocksInKernel(Dim)).
-uint32_t getBlockIdInKernel(int32_t Dim = DIM_X);
-
-/// Return the number of blocks in the kernel.
-uint32_t getNumberOfBlocksInKernel(int32_t Dim = DIM_X);
-
-/// Return the kernel size, thus number of threads in the kernel.
-uint32_t getNumberOfThreadsInKernel();
-
-/// Return the maximal number of threads in the block usable for a team (=
-/// parallel region).
-///
-/// Note: The version taking \p IsSPMD mode explicitly can be used during the
-/// initialization of the target region, that is before `mapping::isSPMDMode()`
-/// can be called by any thread other than the main one.
-uint32_t getMaxTeamThreads();
-uint32_t getMaxTeamThreads(bool IsSPMD);
-
-/// Return the number of processing elements on the device.
-uint32_t getNumberOfProcessorElements();
-
-} // namespace mapping
-
-} // namespace ompx
-
-#endif
diff --git a/offload/DeviceRTL/include/Profiling.h b/offload/DeviceRTL/include/Profiling.h
deleted file mode 100644
index d99475225412..000000000000
--- a/offload/DeviceRTL/include/Profiling.h
+++ /dev/null
@@ -1,21 +0,0 @@
-//===-------- Profiling.h - OpenMP interface ---------------------- 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
-//
-//===----------------------------------------------------------------------===//
-//
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef OMPTARGET_DEVICERTL_PROFILING_H
-#define OMPTARGET_DEVICERTL_PROFILING_H
-
-extern "C" {
-void __llvm_profile_register_function(void *Ptr);
-void __llvm_profile_register_names_function(void *Ptr, long int I);
-void __llvm_profile_instrument_memop(long int I, void *Ptr, int I2);
-}
-
-#endif
diff --git a/offload/DeviceRTL/include/State.h b/offload/DeviceRTL/include/State.h
deleted file mode 100644
index db396dae6e44..000000000000
--- a/offload/DeviceRTL/include/State.h
+++ /dev/null
@@ -1,377 +0,0 @@
-//===-------- State.h - OpenMP State & ICV interface ------------- 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
-//
-//===----------------------------------------------------------------------===//
-//
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef OMPTARGET_STATE_H
-#define OMPTARGET_STATE_H
-
-#include "Shared/Environment.h"
-
-#include "Debug.h"
-#include "DeviceTypes.h"
-#include "DeviceUtils.h"
-#include "Mapping.h"
-
-// Forward declaration.
-struct KernelEnvironmentTy;
-
-namespace ompx {
-
-namespace memory {
-
-/// Alloca \p Size bytes in shared memory, if possible, for \p Reason.
-///
-/// Note: See the restrictions on __kmpc_alloc_shared for proper usage.
-void *allocShared(uint64_t Size, const char *Reason);
-
-/// Free \p Ptr, allocated via allocShared, for \p Reason.
-///
-/// Note: See the restrictions on __kmpc_free_shared for proper usage.
-void freeShared(void *Ptr, uint64_t Bytes, const char *Reason);
-
-/// Alloca \p Size bytes in global memory, if possible, for \p Reason.
-void *allocGlobal(uint64_t Size, const char *Reason);
-
-/// Return a pointer to the dynamic shared memory buffer.
-void *getDynamicBuffer();
-
-/// Free \p Ptr, allocated via allocGlobal, for \p Reason.
-void freeGlobal(void *Ptr, const char *Reason);
-
-} // namespace memory
-
-namespace state {
-
-inline constexpr uint32_t SharedScratchpadSize = SHARED_SCRATCHPAD_SIZE;
-
-struct ICVStateTy {
-  uint32_t NThreadsVar;
-  uint32_t LevelVar;
-  uint32_t ActiveLevelVar;
-  uint32_t Padding0Val;
-  uint32_t MaxActiveLevelsVar;
-  uint32_t RunSchedVar;
-  uint32_t RunSchedChunkVar;
-
-  bool operator==(const ICVStateTy &Other) const;
-
-  void assertEqual(const ICVStateTy &Other) const;
-};
-
-struct TeamStateTy {
-  void init(bool IsSPMD);
-
-  bool operator==(const TeamStateTy &) const;
-
-  void assertEqual(TeamStateTy &Other) const;
-
-  /// ICVs
-  ///
-  /// Preallocated storage for ICV values that are used if the threads have not
-  /// set a custom default. The latter is supported but unlikely and slow(er).
-  ///
-  ///{
-  ICVStateTy ICVState;
-  ///}
-
-  uint32_t ParallelTeamSize;
-  uint32_t HasThreadState;
-  ParallelRegionFnTy ParallelRegionFnVar;
-};
-
-extern Local<TeamStateTy> TeamState;
-
-struct ThreadStateTy {
-
-  /// ICVs have preallocated storage in the TeamStateTy which is used if a
-  /// thread has not set a custom value. The latter is supported but unlikely.
-  /// When it happens we will allocate dynamic memory to hold the values of all
-  /// ICVs. Thus, the first time an ICV is set by a thread we will allocate an
-  /// ICV struct to hold them all. This is slower than alternatives but allows
-  /// users to pay only for what they use.
-  ///
-  state::ICVStateTy ICVState;
-
-  ThreadStateTy *PreviousThreadState;
-
-  void init() {
-    ICVState = TeamState.ICVState;
-    PreviousThreadState = nullptr;
-  }
-
-  void init(ThreadStateTy *PreviousTS) {
-    ICVState = PreviousTS ? PreviousTS->ICVState : TeamState.ICVState;
-    PreviousThreadState = PreviousTS;
-  }
-};
-
-extern Local<ThreadStateTy **> ThreadStates;
-
-/// Initialize the state machinery. Must be called by all threads.
-void init(bool IsSPMD, KernelEnvironmentTy &KernelEnvironment,
-          KernelLaunchEnvironmentTy &KernelLaunchEnvironment);
-
-/// Return the kernel and kernel launch environment associated with the current
-/// kernel. The former is static and contains compile time information that
-/// holds for all instances of the kernel. The latter is dynamic and provides
-/// per-launch information.
-KernelEnvironmentTy &getKernelEnvironment();
-KernelLaunchEnvironmentTy &getKernelLaunchEnvironment();
-
-/// TODO
-enum ValueKind {
-  VK_NThreads,
-  VK_Level,
-  VK_ActiveLevel,
-  VK_MaxActiveLevels,
-  VK_RunSched,
-  // ---
-  VK_RunSchedChunk,
-  VK_ParallelRegionFn,
-  VK_ParallelTeamSize,
-  VK_HasThreadState,
-};
-
-/// TODO
-void enterDataEnvironment(IdentTy *Ident);
-
-/// TODO
-void exitDataEnvironment();
-
-/// TODO
-struct DateEnvironmentRAII {
-  DateEnvironmentRAII(IdentTy *Ident) { enterDataEnvironment(Ident); }
-  ~DateEnvironmentRAII() { exitDataEnvironment(); }
-};
-
-/// TODO
-void resetStateForThread(uint32_t TId);
-
-// FIXME: https://github.com/llvm/llvm-project/issues/123241.
-#define lookupForModify32Impl(Member, Ident, ForceTeamState)                   \
-  {                                                                            \
-    if (OMP_LIKELY(ForceTeamState || !config::mayUseThreadStates() ||          \
-                   !TeamState.HasThreadState))                                 \
-      return TeamState.ICVState.Member;                                        \
-    uint32_t TId = mapping::getThreadIdInBlock();                              \
-    if (OMP_UNLIKELY(!ThreadStates[TId])) {                                    \
-      ThreadStates[TId] = reinterpret_cast<ThreadStateTy *>(                   \
-          memory::allocGlobal(sizeof(ThreadStateTy),                           \
-                              "ICV modification outside data environment"));   \
-      ASSERT(ThreadStates[TId] != nullptr, "Nullptr returned by malloc!");     \
-      TeamState.HasThreadState = true;                                         \
-      ThreadStates[TId]->init();                                               \
-    }                                                                          \
-    return ThreadStates[TId]->ICVState.Member;                                 \
-  }
-
-// FIXME: https://github.com/llvm/llvm-project/issues/123241.
-#define lookupImpl(Member, ForceTeamState)                                     \
-  {                                                                            \
-    auto TId = mapping::getThreadIdInBlock();                                  \
-    if (OMP_UNLIKELY(!ForceTeamState && config::mayUseThreadStates() &&        \
-                     TeamState.HasThreadState && ThreadStates[TId]))           \
-      return ThreadStates[TId]->ICVState.Member;                               \
-    return TeamState.ICVState.Member;                                          \
-  }
-
-[[gnu::always_inline, gnu::flatten]] inline uint32_t &
-lookup32(ValueKind Kind, bool IsReadonly, IdentTy *Ident, bool ForceTeamState) {
-  switch (Kind) {
-  case state::VK_NThreads:
-    if (IsReadonly)
-      lookupImpl(NThreadsVar, ForceTeamState);
-    lookupForModify32Impl(NThreadsVar, Ident, ForceTeamState);
-  case state::VK_Level:
-    if (IsReadonly)
-      lookupImpl(LevelVar, ForceTeamState);
-    lookupForModify32Impl(LevelVar, Ident, ForceTeamState);
-  case state::VK_ActiveLevel:
-    if (IsReadonly)
-      lookupImpl(ActiveLevelVar, ForceTeamState);
-    lookupForModify32Impl(ActiveLevelVar, Ident, ForceTeamState);
-  case state::VK_MaxActiveLevels:
-    if (IsReadonly)
-      lookupImpl(MaxActiveLevelsVar, ForceTeamState);
-    lookupForModify32Impl(MaxActiveLevelsVar, Ident, ForceTeamState);
-  case state::VK_RunSched:
-    if (IsReadonly)
-      lookupImpl(RunSchedVar, ForceTeamState);
-    lookupForModify32Impl(RunSchedVar, Ident, ForceTeamState);
-  case state::VK_RunSchedChunk:
-    if (IsReadonly)
-      lookupImpl(RunSchedChunkVar, ForceTeamState);
-    lookupForModify32Impl(RunSchedChunkVar, Ident, ForceTeamState);
-  case state::VK_ParallelTeamSize:
-    return TeamState.ParallelTeamSize;
-  case state::VK_HasThreadState:
-    return TeamState.HasThreadState;
-  default:
-    break;
-  }
-  __builtin_unreachable();
-}
-
-[[gnu::always_inline, gnu::flatten]] inline void *&
-lookupPtr(ValueKind Kind, bool IsReadonly, bool ForceTeamState) {
-  switch (Kind) {
-  case state::VK_ParallelRegionFn:
-    return TeamState.ParallelRegionFnVar;
-  default:
-    break;
-  }
-  __builtin_unreachable();
-}
-
-/// A class without actual state used to provide a nice interface to lookup and
-/// update ICV values we can declare in global scope.
-template <typename Ty, ValueKind Kind> struct Value {
-  [[gnu::flatten, gnu::always_inline]] operator Ty() {
-    return lookup(/*IsReadonly=*/true, /*IdentTy=*/nullptr,
-                  /*ForceTeamState=*/false);
-  }
-
-  [[gnu::flatten, gnu::always_inline]] Value &operator=(const Ty &Other) {
-    set(Other, /*IdentTy=*/nullptr);
-    return *this;
-  }
-
-  [[gnu::flatten, gnu::always_inline]] Value &operator++() {
-    inc(1, /*IdentTy=*/nullptr);
-    return *this;
-  }
-
-  [[gnu::flatten, gnu::always_inline]] Value &operator--() {
-    inc(-1, /*IdentTy=*/nullptr);
-    return *this;
-  }
-
-  [[gnu::flatten, gnu::always_inline]] void
-  assert_eq(const Ty &V, IdentTy *Ident = nullptr,
-            bool ForceTeamState = false) {
-    ASSERT(lookup(/*IsReadonly=*/true, Ident, ForceTeamState) == V, nullptr);
-  }
-
-private:
-  [[gnu::flatten, gnu::always_inline]] Ty &
-  lookup(bool IsReadonly, IdentTy *Ident, bool ForceTeamState) {
-    Ty &t = lookup32(Kind, IsReadonly, Ident, ForceTeamState);
-    return t;
-  }
-
-  [[gnu::flatten, gnu::always_inline]] Ty &inc(int UpdateVal, IdentTy *Ident) {
-    return (lookup(/*IsReadonly=*/false, Ident, /*ForceTeamState=*/false) +=
-            UpdateVal);
-  }
-
-  [[gnu::flatten, gnu::always_inline]] Ty &set(Ty UpdateVal, IdentTy *Ident) {
-    return (lookup(/*IsReadonly=*/false, Ident, /*ForceTeamState=*/false) =
-                UpdateVal);
-  }
-
-  template <typename VTy, typename Ty2> friend struct ValueRAII;
-};
-
-/// A mookup class without actual state used to provide
-/// a nice interface to lookup and update ICV values
-/// we can declare in global scope.
-template <typename Ty, ValueKind Kind> struct PtrValue {
-  [[gnu::flatten, gnu::always_inline]] operator Ty() {
-    return lookup(/*IsReadonly=*/true, /*IdentTy=*/nullptr,
-                  /*ForceTeamState=*/false);
-  }
-
-  [[gnu::flatten, gnu::always_inline]] PtrValue &operator=(const Ty Other) {
-    set(Other);
-    return *this;
-  }
-
-private:
-  Ty &lookup(bool IsReadonly, IdentTy *, bool ForceTeamState) {
-    return lookupPtr(Kind, IsReadonly, ForceTeamState);
-  }
-
-  Ty &set(Ty UpdateVal) {
-    return (lookup(/*IsReadonly=*/false, /*IdentTy=*/nullptr,
-                   /*ForceTeamState=*/false) = UpdateVal);
-  }
-
-  template <typename VTy, typename Ty2> friend struct ValueRAII;
-};
-
-template <typename VTy, typename Ty> struct ValueRAII {
-  ValueRAII(VTy &V, Ty NewValue, Ty OldValue, bool Active, IdentTy *Ident,
-            bool ForceTeamState = false)
-      : Ptr(Active ? &V.lookup(/*IsReadonly=*/false, Ident, ForceTeamState)
-                   : (Ty *)utils::UndefPtr),
-        Val(OldValue), Active(Active) {
-    if (!Active)
-      return;
-    ASSERT(*Ptr == OldValue, "ValueRAII initialization with wrong old value!");
-    *Ptr = NewValue;
-  }
-  ~ValueRAII() {
-    if (Active)
-      *Ptr = Val;
-  }
-
-private:
-  Ty *Ptr;
-  Ty Val;
-  bool Active;
-};
-
-/// TODO
-inline state::Value<uint32_t, state::VK_RunSchedChunk> RunSchedChunk;
-
-/// TODO
-inline state::Value<uint32_t, state::VK_ParallelTeamSize> ParallelTeamSize;
-
-/// TODO
-inline state::Value<uint32_t, state::VK_HasThreadState> HasThreadState;
-
-/// TODO
-inline state::PtrValue<ParallelRegionFnTy, state::VK_ParallelRegionFn>
-    ParallelRegionFn;
-
-void runAndCheckState(void(Func(void)));
-
-void assumeInitialState(bool IsSPMD);
-
-/// Return the value of the ParallelTeamSize ICV.
-int getEffectivePTeamSize();
-
-} // namespace state
-
-namespace icv {
-
-/// TODO
-inline state::Value<uint32_t, state::VK_NThreads> NThreads;
-
-/// TODO
-inline state::Value<uint32_t, state::VK_Level> Level;
-
-/// The `active-level` describes which of the parallel level counted with the
-/// `level-var` is active. There can only be one.
-///
-/// active-level-var is 1, if ActiveLevelVar is not 0, otherwise it is 0.
-inline state::Value<uint32_t, state::VK_ActiveLevel> ActiveLevel;
-
-/// TODO
-inline state::Value<uint32_t, state::VK_MaxActiveLevels> MaxActiveLevels;
-
-/// TODO
-inline state::Value<uint32_t, state::VK_RunSched> RunSched;
-
-} // namespace icv
-
-} // namespace ompx
-
-#endif
diff --git a/offload/DeviceRTL/include/Synchronization.h b/offload/DeviceRTL/include/Synchronization.h
deleted file mode 100644
index 7e7c8eacb917..000000000000
--- a/offload/DeviceRTL/include/Synchronization.h
+++ /dev/null
@@ -1,225 +0,0 @@
-//===- Synchronization.h - OpenMP synchronization 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
-//
-//===----------------------------------------------------------------------===//
-//
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef OMPTARGET_DEVICERTL_SYNCHRONIZATION_H
-#define OMPTARGET_DEVICERTL_SYNCHRONIZATION_H
-
-#include "DeviceTypes.h"
-#include "DeviceUtils.h"
-
-namespace ompx {
-namespace atomic {
-
-enum OrderingTy {
-  relaxed = __ATOMIC_RELAXED,
-  acquire = __ATOMIC_ACQUIRE,
-  release = __ATOMIC_RELEASE,
-  acq_rel = __ATOMIC_ACQ_REL,
-  seq_cst = __ATOMIC_SEQ_CST,
-};
-
-enum MemScopeTy {
-  system = __MEMORY_SCOPE_SYSTEM,
-  device = __MEMORY_SCOPE_DEVICE,
-  workgroup = __MEMORY_SCOPE_WRKGRP,
-  wavefront = __MEMORY_SCOPE_WVFRNT,
-  single = __MEMORY_SCOPE_SINGLE,
-};
-
-/// Atomically increment \p *Addr and wrap at \p V with \p Ordering semantics.
-uint32_t inc(uint32_t *Addr, uint32_t V, OrderingTy Ordering,
-             MemScopeTy MemScope = MemScopeTy::device);
-
-/// Atomically perform <op> on \p V and \p *Addr with \p Ordering semantics. The
-/// result is stored in \p *Addr;
-/// {
-
-template <typename Ty, typename V = utils::remove_addrspace_t<Ty>>
-bool cas(Ty *Address, V ExpectedV, V DesiredV, atomic::OrderingTy OrderingSucc,
-         atomic::OrderingTy OrderingFail,
-         MemScopeTy MemScope = MemScopeTy::device) {
-  return __scoped_atomic_compare_exchange(Address, &ExpectedV, &DesiredV, false,
-                                          OrderingSucc, OrderingFail, MemScope);
-}
-
-template <typename Ty, typename V = utils::remove_addrspace_t<Ty>>
-V add(Ty *Address, V Val, atomic::OrderingTy Ordering,
-      MemScopeTy MemScope = MemScopeTy::device) {
-  return __scoped_atomic_fetch_add(Address, Val, Ordering, MemScope);
-}
-
-template <typename Ty, typename V = utils::remove_addrspace_t<Ty>>
-V load(Ty *Address, atomic::OrderingTy Ordering,
-       MemScopeTy MemScope = MemScopeTy::device) {
-#ifdef __NVPTX__
-  return __scoped_atomic_fetch_add(Address, V(0), Ordering, MemScope);
-#else
-  return __scoped_atomic_load_n(Address, Ordering, MemScope);
-#endif
-}
-
-template <typename Ty, typename V = utils::remove_addrspace_t<Ty>>
-void store(Ty *Address, V Val, atomic::OrderingTy Ordering,
-           MemScopeTy MemScope = MemScopeTy::device) {
-  __scoped_atomic_store_n(Address, Val, Ordering, MemScope);
-}
-
-template <typename Ty, typename V = utils::remove_addrspace_t<Ty>>
-V mul(Ty *Address, V Val, atomic::OrderingTy Ordering,
-      MemScopeTy MemScope = MemScopeTy::device) {
-  Ty TypedCurrentVal, TypedResultVal, TypedNewVal;
-  bool Success;
-  do {
-    TypedCurrentVal = atomic::load(Address, Ordering);
-    TypedNewVal = TypedCurrentVal * Val;
-    Success = atomic::cas(Address, TypedCurrentVal, TypedNewVal, Ordering,
-                          atomic::relaxed, MemScope);
-  } while (!Success);
-  return TypedResultVal;
-}
-
-template <typename Ty, typename V = utils::remove_addrspace_t<Ty>>
-utils::enable_if_t<!utils::is_floating_point_v<V>, V>
-max(Ty *Address, V Val, atomic::OrderingTy Ordering,
-    MemScopeTy MemScope = MemScopeTy::device) {
-  return __scoped_atomic_fetch_max(Address, Val, Ordering, MemScope);
-}
-
-template <typename Ty, typename V = utils::remove_addrspace_t<Ty>>
-utils::enable_if_t<utils::is_same_v<V, float>, V>
-max(Ty *Address, V Val, atomic::OrderingTy Ordering,
-    MemScopeTy MemScope = MemScopeTy::device) {
-  if (Val >= 0)
-    return utils::bitCast<float>(max(
-        (int32_t *)Address, utils::bitCast<int32_t>(Val), Ordering, MemScope));
-  return utils::bitCast<float>(min(
-      (uint32_t *)Address, utils::bitCast<uint32_t>(Val), Ordering, MemScope));
-}
-
-template <typename Ty, typename V = utils::remove_addrspace_t<Ty>>
-utils::enable_if_t<utils::is_same_v<V, double>, V>
-max(Ty *Address, V Val, atomic::OrderingTy Ordering,
-    MemScopeTy MemScope = MemScopeTy::device) {
-  if (Val >= 0)
-    return utils::bitCast<double>(max(
-        (int64_t *)Address, utils::bitCast<int64_t>(Val), Ordering, MemScope));
-  return utils::bitCast<double>(min(
-      (uint64_t *)Address, utils::bitCast<uint64_t>(Val), Ordering, MemScope));
-}
-
-template <typename Ty, typename V = utils::remove_addrspace_t<Ty>>
-utils::enable_if_t<!utils::is_floating_point_v<V>, V>
-min(Ty *Address, V Val, atomic::OrderingTy Ordering,
-    MemScopeTy MemScope = MemScopeTy::device) {
-  return __scoped_atomic_fetch_min(Address, Val, Ordering, MemScope);
-}
-
-// TODO: Implement this with __atomic_fetch_max and remove the duplication.
-template <typename Ty, typename V = utils::remove_addrspace_t<Ty>>
-utils::enable_if_t<utils::is_same_v<V, float>, V>
-min(Ty *Address, V Val, atomic::OrderingTy Ordering,
-    MemScopeTy MemScope = MemScopeTy::device) {
-  if (Val >= 0)
-    return utils::bitCast<float>(min(
-        (int32_t *)Address, utils::bitCast<int32_t>(Val), Ordering, MemScope));
-  return utils::bitCast<float>(max(
-      (uint32_t *)Address, utils::bitCast<uint32_t>(Val), Ordering, MemScope));
-}
-
-// TODO: Implement this with __atomic_fetch_max and remove the duplication.
-template <typename Ty, typename V = utils::remove_addrspace_t<Ty>>
-utils::enable_if_t<utils::is_same_v<V, double>, V>
-min(Ty *Address, utils::remove_addrspace_t<Ty> Val, atomic::OrderingTy Ordering,
-    MemScopeTy MemScope = MemScopeTy::device) {
-  if (Val >= 0)
-    return utils::bitCast<double>(min(
-        (int64_t *)Address, utils::bitCast<int64_t>(Val), Ordering, MemScope));
-  return utils::bitCast<double>(max(
-      (uint64_t *)Address, utils::bitCast<uint64_t>(Val), Ordering, MemScope));
-}
-
-template <typename Ty, typename V = utils::remove_addrspace_t<Ty>>
-V bit_or(Ty *Address, V Val, atomic::OrderingTy Ordering,
-         MemScopeTy MemScope = MemScopeTy::device) {
-  return __scoped_atomic_fetch_or(Address, Val, Ordering, MemScope);
-}
-
-template <typename Ty, typename V = utils::remove_addrspace_t<Ty>>
-V bit_and(Ty *Address, V Val, atomic::OrderingTy Ordering,
-          MemScopeTy MemScope = MemScopeTy::device) {
-  return __scoped_atomic_fetch_and(Address, Val, Ordering, MemScope);
-}
-
-template <typename Ty, typename V = utils::remove_addrspace_t<Ty>>
-V bit_xor(Ty *Address, V Val, atomic::OrderingTy Ordering,
-          MemScopeTy MemScope = MemScopeTy::device) {
-  return __scoped_atomic_fetch_xor(Address, Val, Ordering, MemScope);
-}
-
-static inline uint32_t
-atomicExchange(uint32_t *Address, uint32_t Val, atomic::OrderingTy Ordering,
-               MemScopeTy MemScope = MemScopeTy::device) {
-  uint32_t R;
-  __scoped_atomic_exchange(Address, &Val, &R, Ordering, MemScope);
-  return R;
-}
-
-///}
-
-} // namespace atomic
-
-namespace synchronize {
-
-/// Initialize the synchronization machinery. Must be called by all threads.
-void init(bool IsSPMD);
-
-/// Synchronize all threads in a warp identified by \p Mask.
-void warp(LaneMaskTy Mask);
-
-/// Synchronize all threads in a block and perform a fence before and after the
-/// barrier according to \p Ordering. Note that the fence might be part of the
-/// barrier.
-void threads(atomic::OrderingTy Ordering);
-
-/// Synchronizing threads is allowed even if they all hit different instances of
-/// `synchronize::threads()`. However, `synchronize::threadsAligned()` is more
-/// restrictive in that it requires all threads to hit the same instance. The
-/// noinline is removed by the openmp-opt pass and helps to preserve the
-/// information till then.
-///{
-
-/// Synchronize all threads in a block, they are reaching the same instruction
-/// (hence all threads in the block are "aligned"). Also perform a fence before
-/// and after the barrier according to \p Ordering. Note that the
-/// fence might be part of the barrier if the target offers this.
-[[gnu::noinline, omp::assume("ompx_aligned_barrier")]] void
-threadsAligned(atomic::OrderingTy Ordering);
-
-///}
-
-} // namespace synchronize
-
-namespace fence {
-
-/// Memory fence with \p Ordering semantics for the team.
-void team(atomic::OrderingTy Ordering);
-
-/// Memory fence with \p Ordering semantics for the contention group.
-void kernel(atomic::OrderingTy Ordering);
-
-/// Memory fence with \p Ordering semantics for the system.
-void system(atomic::OrderingTy Ordering);
-
-} // namespace fence
-
-} // namespace ompx
-
-#endif
diff --git a/offload/DeviceRTL/include/Workshare.h b/offload/DeviceRTL/include/Workshare.h
deleted file mode 100644
index 554c3271c334..000000000000
--- a/offload/DeviceRTL/include/Workshare.h
+++ /dev/null
@@ -1,26 +0,0 @@
-//===-------- Workshare.h - OpenMP Workshare interface ------------ 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
-//
-//===----------------------------------------------------------------------===//
-//
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef OMPTARGET_WORKSHARE_H
-#define OMPTARGET_WORKSHARE_H
-
-namespace ompx {
-
-namespace workshare {
-
-/// Initialize the worksharing machinery.
-void init(bool IsSPMD);
-
-} // namespace workshare
-
-} // namespace ompx
-
-#endif
diff --git a/offload/DeviceRTL/include/generated_microtask_cases.gen b/offload/DeviceRTL/include/generated_microtask_cases.gen
deleted file mode 100644
index a05f6da2f84f..000000000000
--- a/offload/DeviceRTL/include/generated_microtask_cases.gen
+++ /dev/null
@@ -1,797 +0,0 @@
-case 0:
-((void (*)(int32_t *, int32_t *))fn)(&global_tid, &bound_tid);
-break;
-case 1:
-((void (*)(int32_t *, int32_t *, void *))fn)(&global_tid, &bound_tid, args[0]);
-break;
-case 2:
-((void (*)(int32_t *, int32_t *, void *, void *))fn)(&global_tid, &bound_tid,
-                                                     args[0], args[1]);
-break;
-case 3:
-((void (*)(int32_t *, int32_t *, void *, void *,
-           void *))fn)(&global_tid, &bound_tid, args[0], args[1], args[2]);
-break;
-case 4:
-((void (*)(int32_t *, int32_t *, void *, void *, void *,
-           void *))fn)(&global_tid, &bound_tid, args[0], args[1], args[2],
-                       args[3]);
-break;
-case 5:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *,
-           void *))fn)(&global_tid, &bound_tid, args[0], args[1], args[2],
-                       args[3], args[4]);
-break;
-case 6:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *,
-           void *))fn)(&global_tid, &bound_tid, args[0], args[1], args[2],
-                       args[3], args[4], args[5]);
-break;
-case 7:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *))fn)(&global_tid, &bound_tid, args[0], args[1], args[2],
-                       args[3], args[4], args[5], args[6]);
-break;
-case 8:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *))fn)(&global_tid, &bound_tid, args[0], args[1],
-                               args[2], args[3], args[4], args[5], args[6],
-                               args[7]);
-break;
-case 9:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *))fn)(&global_tid, &bound_tid, args[0],
-                                       args[1], args[2], args[3], args[4],
-                                       args[5], args[6], args[7], args[8]);
-break;
-case 10:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *))fn)(&global_tid, &bound_tid, args[0],
-                                               args[1], args[2], args[3],
-                                               args[4], args[5], args[6],
-                                               args[7], args[8], args[9]);
-break;
-case 11:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *,
-           void *))fn)(&global_tid, &bound_tid, args[0], args[1], args[2],
-                       args[3], args[4], args[5], args[6], args[7], args[8],
-                       args[9], args[10]);
-break;
-case 12:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *,
-           void *))fn)(&global_tid, &bound_tid, args[0], args[1], args[2],
-                       args[3], args[4], args[5], args[6], args[7], args[8],
-                       args[9], args[10], args[11]);
-break;
-case 13:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *,
-           void *))fn)(&global_tid, &bound_tid, args[0], args[1], args[2],
-                       args[3], args[4], args[5], args[6], args[7], args[8],
-                       args[9], args[10], args[11], args[12]);
-break;
-case 14:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *,
-           void *))fn)(&global_tid, &bound_tid, args[0], args[1], args[2],
-                       args[3], args[4], args[5], args[6], args[7], args[8],
-                       args[9], args[10], args[11], args[12], args[13]);
-break;
-case 15:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *))fn)(&global_tid, &bound_tid, args[0], args[1], args[2],
-                       args[3], args[4], args[5], args[6], args[7], args[8],
-                       args[9], args[10], args[11], args[12], args[13],
-                       args[14]);
-break;
-case 16:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *))fn)(&global_tid, &bound_tid, args[0], args[1],
-                               args[2], args[3], args[4], args[5], args[6],
-                               args[7], args[8], args[9], args[10], args[11],
-                               args[12], args[13], args[14], args[15]);
-break;
-case 17:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *))fn)(&global_tid, &bound_tid, args[0],
-                                       args[1], args[2], args[3], args[4],
-                                       args[5], args[6], args[7], args[8],
-                                       args[9], args[10], args[11], args[12],
-                                       args[13], args[14], args[15], args[16]);
-break;
-case 18:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *,
-           void *))fn)(&global_tid, &bound_tid, args[0], args[1], args[2],
-                       args[3], args[4], args[5], args[6], args[7], args[8],
-                       args[9], args[10], args[11], args[12], args[13],
-                       args[14], args[15], args[16], args[17]);
-break;
-case 19:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *,
-           void *))fn)(&global_tid, &bound_tid, args[0], args[1], args[2],
-                       args[3], args[4], args[5], args[6], args[7], args[8],
-                       args[9], args[10], args[11], args[12], args[13],
-                       args[14], args[15], args[16], args[17], args[18]);
-break;
-case 20:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *))fn)(
-    &global_tid, &bound_tid, args[0], args[1], args[2], args[3], args[4],
-    args[5], args[6], args[7], args[8], args[9], args[10], args[11], args[12],
-    args[13], args[14], args[15], args[16], args[17], args[18], args[19]);
-break;
-case 21:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *,
-           void *))fn)(&global_tid, &bound_tid, args[0], args[1], args[2],
-                       args[3], args[4], args[5], args[6], args[7], args[8],
-                       args[9], args[10], args[11], args[12], args[13],
-                       args[14], args[15], args[16], args[17], args[18],
-                       args[19], args[20]);
-break;
-case 22:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *,
-           void *))fn)(&global_tid, &bound_tid, args[0], args[1], args[2],
-                       args[3], args[4], args[5], args[6], args[7], args[8],
-                       args[9], args[10], args[11], args[12], args[13],
-                       args[14], args[15], args[16], args[17], args[18],
-                       args[19], args[20], args[21]);
-break;
-case 23:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *))fn)(&global_tid, &bound_tid, args[0], args[1], args[2],
-                       args[3], args[4], args[5], args[6], args[7], args[8],
-                       args[9], args[10], args[11], args[12], args[13],
-                       args[14], args[15], args[16], args[17], args[18],
-                       args[19], args[20], args[21], args[22]);
-break;
-case 24:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *))fn)(&global_tid, &bound_tid, args[0], args[1],
-                               args[2], args[3], args[4], args[5], args[6],
-                               args[7], args[8], args[9], args[10], args[11],
-                               args[12], args[13], args[14], args[15], args[16],
-                               args[17], args[18], args[19], args[20], args[21],
-                               args[22], args[23]);
-break;
-case 25:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *))fn)(&global_tid, &bound_tid, args[0],
-                                       args[1], args[2], args[3], args[4],
-                                       args[5], args[6], args[7], args[8],
-                                       args[9], args[10], args[11], args[12],
-                                       args[13], args[14], args[15], args[16],
-                                       args[17], args[18], args[19], args[20],
-                                       args[21], args[22], args[23], args[24]);
-break;
-case 26:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *))fn)(
-    &global_tid, &bound_tid, args[0], args[1], args[2], args[3], args[4],
-    args[5], args[6], args[7], args[8], args[9], args[10], args[11], args[12],
-    args[13], args[14], args[15], args[16], args[17], args[18], args[19],
-    args[20], args[21], args[22], args[23], args[24], args[25]);
-break;
-case 27:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *))fn)(
-    &global_tid, &bound_tid, args[0], args[1], args[2], args[3], args[4],
-    args[5], args[6], args[7], args[8], args[9], args[10], args[11], args[12],
-    args[13], args[14], args[15], args[16], args[17], args[18], args[19],
-    args[20], args[21], args[22], args[23], args[24], args[25], args[26]);
-break;
-case 28:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *,
-           void *))fn)(&global_tid, &bound_tid, args[0], args[1], args[2],
-                       args[3], args[4], args[5], args[6], args[7], args[8],
-                       args[9], args[10], args[11], args[12], args[13],
-                       args[14], args[15], args[16], args[17], args[18],
-                       args[19], args[20], args[21], args[22], args[23],
-                       args[24], args[25], args[26], args[27]);
-break;
-case 29:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *,
-           void *))fn)(&global_tid, &bound_tid, args[0], args[1], args[2],
-                       args[3], args[4], args[5], args[6], args[7], args[8],
-                       args[9], args[10], args[11], args[12], args[13],
-                       args[14], args[15], args[16], args[17], args[18],
-                       args[19], args[20], args[21], args[22], args[23],
-                       args[24], args[25], args[26], args[27], args[28]);
-break;
-case 30:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *))fn)(
-    &global_tid, &bound_tid, args[0], args[1], args[2], args[3], args[4],
-    args[5], args[6], args[7], args[8], args[9], args[10], args[11], args[12],
-    args[13], args[14], args[15], args[16], args[17], args[18], args[19],
-    args[20], args[21], args[22], args[23], args[24], args[25], args[26],
-    args[27], args[28], args[29]);
-break;
-case 31:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *))fn)(&global_tid, &bound_tid, args[0], args[1], args[2],
-                       args[3], args[4], args[5], args[6], args[7], args[8],
-                       args[9], args[10], args[11], args[12], args[13],
-                       args[14], args[15], args[16], args[17], args[18],
-                       args[19], args[20], args[21], args[22], args[23],
-                       args[24], args[25], args[26], args[27], args[28],
-                       args[29], args[30]);
-break;
-case 32:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *))fn)(&global_tid, &bound_tid, args[0], args[1],
-                               args[2], args[3], args[4], args[5], args[6],
-                               args[7], args[8], args[9], args[10], args[11],
-                               args[12], args[13], args[14], args[15], args[16],
-                               args[17], args[18], args[19], args[20], args[21],
-                               args[22], args[23], args[24], args[25], args[26],
-                               args[27], args[28], args[29], args[30],
-                               args[31]);
-break;
-case 33:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *))fn)(
-    &global_tid, &bound_tid, args[0], args[1], args[2], args[3], args[4],
-    args[5], args[6], args[7], args[8], args[9], args[10], args[11], args[12],
-    args[13], args[14], args[15], args[16], args[17], args[18], args[19],
-    args[20], args[21], args[22], args[23], args[24], args[25], args[26],
-    args[27], args[28], args[29], args[30], args[31], args[32]);
-break;
-case 34:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *))fn)(
-    &global_tid, &bound_tid, args[0], args[1], args[2], args[3], args[4],
-    args[5], args[6], args[7], args[8], args[9], args[10], args[11], args[12],
-    args[13], args[14], args[15], args[16], args[17], args[18], args[19],
-    args[20], args[21], args[22], args[23], args[24], args[25], args[26],
-    args[27], args[28], args[29], args[30], args[31], args[32], args[33]);
-break;
-case 35:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *))fn)(
-    &global_tid, &bound_tid, args[0], args[1], args[2], args[3], args[4],
-    args[5], args[6], args[7], args[8], args[9], args[10], args[11], args[12],
-    args[13], args[14], args[15], args[16], args[17], args[18], args[19],
-    args[20], args[21], args[22], args[23], args[24], args[25], args[26],
-    args[27], args[28], args[29], args[30], args[31], args[32], args[33],
-    args[34]);
-break;
-case 36:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *))fn)(
-    &global_tid, &bound_tid, args[0], args[1], args[2], args[3], args[4],
-    args[5], args[6], args[7], args[8], args[9], args[10], args[11], args[12],
-    args[13], args[14], args[15], args[16], args[17], args[18], args[19],
-    args[20], args[21], args[22], args[23], args[24], args[25], args[26],
-    args[27], args[28], args[29], args[30], args[31], args[32], args[33],
-    args[34], args[35]);
-break;
-case 37:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *))fn)(
-    &global_tid, &bound_tid, args[0], args[1], args[2], args[3], args[4],
-    args[5], args[6], args[7], args[8], args[9], args[10], args[11], args[12],
-    args[13], args[14], args[15], args[16], args[17], args[18], args[19],
-    args[20], args[21], args[22], args[23], args[24], args[25], args[26],
-    args[27], args[28], args[29], args[30], args[31], args[32], args[33],
-    args[34], args[35], args[36]);
-break;
-case 38:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *))fn)(
-    &global_tid, &bound_tid, args[0], args[1], args[2], args[3], args[4],
-    args[5], args[6], args[7], args[8], args[9], args[10], args[11], args[12],
-    args[13], args[14], args[15], args[16], args[17], args[18], args[19],
-    args[20], args[21], args[22], args[23], args[24], args[25], args[26],
-    args[27], args[28], args[29], args[30], args[31], args[32], args[33],
-    args[34], args[35], args[36], args[37]);
-break;
-case 39:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *))fn)(&global_tid, &bound_tid, args[0], args[1], args[2],
-                       args[3], args[4], args[5], args[6], args[7], args[8],
-                       args[9], args[10], args[11], args[12], args[13],
-                       args[14], args[15], args[16], args[17], args[18],
-                       args[19], args[20], args[21], args[22], args[23],
-                       args[24], args[25], args[26], args[27], args[28],
-                       args[29], args[30], args[31], args[32], args[33],
-                       args[34], args[35], args[36], args[37], args[38]);
-break;
-case 40:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *))fn)(&global_tid, &bound_tid, args[0], args[1],
-                               args[2], args[3], args[4], args[5], args[6],
-                               args[7], args[8], args[9], args[10], args[11],
-                               args[12], args[13], args[14], args[15], args[16],
-                               args[17], args[18], args[19], args[20], args[21],
-                               args[22], args[23], args[24], args[25], args[26],
-                               args[27], args[28], args[29], args[30], args[31],
-                               args[32], args[33], args[34], args[35], args[36],
-                               args[37], args[38], args[39]);
-break;
-case 41:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *))fn)(
-    &global_tid, &bound_tid, args[0], args[1], args[2], args[3], args[4],
-    args[5], args[6], args[7], args[8], args[9], args[10], args[11], args[12],
-    args[13], args[14], args[15], args[16], args[17], args[18], args[19],
-    args[20], args[21], args[22], args[23], args[24], args[25], args[26],
-    args[27], args[28], args[29], args[30], args[31], args[32], args[33],
-    args[34], args[35], args[36], args[37], args[38], args[39], args[40]);
-break;
-case 42:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *))fn)(
-    &global_tid, &bound_tid, args[0], args[1], args[2], args[3], args[4],
-    args[5], args[6], args[7], args[8], args[9], args[10], args[11], args[12],
-    args[13], args[14], args[15], args[16], args[17], args[18], args[19],
-    args[20], args[21], args[22], args[23], args[24], args[25], args[26],
-    args[27], args[28], args[29], args[30], args[31], args[32], args[33],
-    args[34], args[35], args[36], args[37], args[38], args[39], args[40],
-    args[41]);
-break;
-case 43:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *))fn)(
-    &global_tid, &bound_tid, args[0], args[1], args[2], args[3], args[4],
-    args[5], args[6], args[7], args[8], args[9], args[10], args[11], args[12],
-    args[13], args[14], args[15], args[16], args[17], args[18], args[19],
-    args[20], args[21], args[22], args[23], args[24], args[25], args[26],
-    args[27], args[28], args[29], args[30], args[31], args[32], args[33],
-    args[34], args[35], args[36], args[37], args[38], args[39], args[40],
-    args[41], args[42]);
-break;
-case 44:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *))fn)(
-    &global_tid, &bound_tid, args[0], args[1], args[2], args[3], args[4],
-    args[5], args[6], args[7], args[8], args[9], args[10], args[11], args[12],
-    args[13], args[14], args[15], args[16], args[17], args[18], args[19],
-    args[20], args[21], args[22], args[23], args[24], args[25], args[26],
-    args[27], args[28], args[29], args[30], args[31], args[32], args[33],
-    args[34], args[35], args[36], args[37], args[38], args[39], args[40],
-    args[41], args[42], args[43]);
-break;
-case 45:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *))fn)(
-    &global_tid, &bound_tid, args[0], args[1], args[2], args[3], args[4],
-    args[5], args[6], args[7], args[8], args[9], args[10], args[11], args[12],
-    args[13], args[14], args[15], args[16], args[17], args[18], args[19],
-    args[20], args[21], args[22], args[23], args[24], args[25], args[26],
-    args[27], args[28], args[29], args[30], args[31], args[32], args[33],
-    args[34], args[35], args[36], args[37], args[38], args[39], args[40],
-    args[41], args[42], args[43], args[44]);
-break;
-case 46:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *))fn)(
-    &global_tid, &bound_tid, args[0], args[1], args[2], args[3], args[4],
-    args[5], args[6], args[7], args[8], args[9], args[10], args[11], args[12],
-    args[13], args[14], args[15], args[16], args[17], args[18], args[19],
-    args[20], args[21], args[22], args[23], args[24], args[25], args[26],
-    args[27], args[28], args[29], args[30], args[31], args[32], args[33],
-    args[34], args[35], args[36], args[37], args[38], args[39], args[40],
-    args[41], args[42], args[43], args[44], args[45]);
-break;
-///  DONE TO HERE
-case 47:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *))fn)(&global_tid, &bound_tid, args[0], args[1], args[2],
-                       args[3], args[4], args[5], args[6], args[7], args[8],
-                       args[9], args[10], args[11], args[12], args[13],
-                       args[14], args[15], args[16], args[17], args[18],
-                       args[19], args[20], args[21], args[22], args[23],
-                       args[24], args[25], args[26], args[27], args[28],
-                       args[29], args[30], args[31], args[32], args[33],
-                       args[34], args[35], args[36], args[37], args[38],
-                       args[39], args[40], args[41], args[42], args[43],
-                       args[44], args[45], args[46]);
-break;
-case 48:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *))fn)(
-    &global_tid, &bound_tid, args[0], args[1], args[2], args[3], args[4],
-    args[5], args[6], args[7], args[8], args[9], args[10], args[11], args[12],
-    args[13], args[14], args[15], args[16], args[17], args[18], args[19],
-    args[20], args[21], args[22], args[23], args[24], args[25], args[26],
-    args[27], args[28], args[29], args[30], args[31], args[32], args[33],
-    args[34], args[35], args[36], args[37], args[38], args[39], args[40],
-    args[41], args[42], args[43], args[44], args[45], args[46], args[47]);
-break;
-case 49:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *))fn)(
-    &global_tid, &bound_tid, args[0], args[1], args[2], args[3], args[4],
-    args[5], args[6], args[7], args[8], args[9], args[10], args[11], args[12],
-    args[13], args[14], args[15], args[16], args[17], args[18], args[19],
-    args[20], args[21], args[22], args[23], args[24], args[25], args[26],
-    args[27], args[28], args[29], args[30], args[31], args[32], args[33],
-    args[34], args[35], args[36], args[37], args[38], args[39], args[40],
-    args[41], args[42], args[43], args[44], args[45], args[46], args[47],
-    args[48]);
-break;
-case 50:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *))fn)(
-    &global_tid, &bound_tid, args[0], args[1], args[2], args[3], args[4],
-    args[5], args[6], args[7], args[8], args[9], args[10], args[11], args[12],
-    args[13], args[14], args[15], args[16], args[17], args[18], args[19],
-    args[20], args[21], args[22], args[23], args[24], args[25], args[26],
-    args[27], args[28], args[29], args[30], args[31], args[32], args[33],
-    args[34], args[35], args[36], args[37], args[38], args[39], args[40],
-    args[41], args[42], args[43], args[44], args[45], args[46], args[47],
-    args[48], args[49]);
-break;
-case 51:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *))fn)(
-    &global_tid, &bound_tid, args[0], args[1], args[2], args[3], args[4],
-    args[5], args[6], args[7], args[8], args[9], args[10], args[11], args[12],
-    args[13], args[14], args[15], args[16], args[17], args[18], args[19],
-    args[20], args[21], args[22], args[23], args[24], args[25], args[26],
-    args[27], args[28], args[29], args[30], args[31], args[32], args[33],
-    args[34], args[35], args[36], args[37], args[38], args[39], args[40],
-    args[41], args[42], args[43], args[44], args[45], args[46], args[47],
-    args[48], args[49], args[50]);
-break;
-case 52:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *))fn)(
-    &global_tid, &bound_tid, args[0], args[1], args[2], args[3], args[4],
-    args[5], args[6], args[7], args[8], args[9], args[10], args[11], args[12],
-    args[13], args[14], args[15], args[16], args[17], args[18], args[19],
-    args[20], args[21], args[22], args[23], args[24], args[25], args[26],
-    args[27], args[28], args[29], args[30], args[31], args[32], args[33],
-    args[34], args[35], args[36], args[37], args[38], args[39], args[40],
-    args[41], args[42], args[43], args[44], args[45], args[46], args[47],
-    args[48], args[49], args[50], args[51]);
-break;
-case 53:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *))fn)(
-    &global_tid, &bound_tid, args[0], args[1], args[2], args[3], args[4],
-    args[5], args[6], args[7], args[8], args[9], args[10], args[11], args[12],
-    args[13], args[14], args[15], args[16], args[17], args[18], args[19],
-    args[20], args[21], args[22], args[23], args[24], args[25], args[26],
-    args[27], args[28], args[29], args[30], args[31], args[32], args[33],
-    args[34], args[35], args[36], args[37], args[38], args[39], args[40],
-    args[41], args[42], args[43], args[44], args[45], args[46], args[47],
-    args[48], args[49], args[50], args[51], args[52]);
-break;
-case 54:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *))fn)(
-    &global_tid, &bound_tid, args[0], args[1], args[2], args[3], args[4],
-    args[5], args[6], args[7], args[8], args[9], args[10], args[11], args[12],
-    args[13], args[14], args[15], args[16], args[17], args[18], args[19],
-    args[20], args[21], args[22], args[23], args[24], args[25], args[26],
-    args[27], args[28], args[29], args[30], args[31], args[32], args[33],
-    args[34], args[35], args[36], args[37], args[38], args[39], args[40],
-    args[41], args[42], args[43], args[44], args[45], args[46], args[47],
-    args[48], args[49], args[50], args[51], args[52], args[53]);
-break;
-case 55:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *))fn)(
-    &global_tid, &bound_tid, args[0], args[1], args[2], args[3], args[4],
-    args[5], args[6], args[7], args[8], args[9], args[10], args[11], args[12],
-    args[13], args[14], args[15], args[16], args[17], args[18], args[19],
-    args[20], args[21], args[22], args[23], args[24], args[25], args[26],
-    args[27], args[28], args[29], args[30], args[31], args[32], args[33],
-    args[34], args[35], args[36], args[37], args[38], args[39], args[40],
-    args[41], args[42], args[43], args[44], args[45], args[46], args[47],
-    args[48], args[49], args[50], args[51], args[52], args[53], args[54]);
-break;
-case 56:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *))fn)(&global_tid, &bound_tid, args[0], args[1],
-                               args[2], args[3], args[4], args[5], args[6],
-                               args[7], args[8], args[9], args[10], args[11],
-                               args[12], args[13], args[14], args[15], args[16],
-                               args[17], args[18], args[19], args[20], args[21],
-                               args[22], args[23], args[24], args[25], args[26],
-                               args[27], args[28], args[29], args[30], args[31],
-                               args[32], args[33], args[34], args[35], args[36],
-                               args[37], args[38], args[39], args[40], args[41],
-                               args[42], args[43], args[44], args[45], args[46],
-                               args[47], args[48], args[49], args[50], args[51],
-                               args[52], args[53], args[54], args[55]);
-break;
-case 57:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *))fn)(
-    &global_tid, &bound_tid, args[0], args[1], args[2], args[3], args[4],
-    args[5], args[6], args[7], args[8], args[9], args[10], args[11], args[12],
-    args[13], args[14], args[15], args[16], args[17], args[18], args[19],
-    args[20], args[21], args[22], args[23], args[24], args[25], args[26],
-    args[27], args[28], args[29], args[30], args[31], args[32], args[33],
-    args[34], args[35], args[36], args[37], args[38], args[39], args[40],
-    args[41], args[42], args[43], args[44], args[45], args[46], args[47],
-    args[48], args[49], args[50], args[51], args[52], args[53], args[54],
-    args[55], args[56]);
-break;
-case 58:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *))fn)(
-    &global_tid, &bound_tid, args[0], args[1], args[2], args[3], args[4],
-    args[5], args[6], args[7], args[8], args[9], args[10], args[11], args[12],
-    args[13], args[14], args[15], args[16], args[17], args[18], args[19],
-    args[20], args[21], args[22], args[23], args[24], args[25], args[26],
-    args[27], args[28], args[29], args[30], args[31], args[32], args[33],
-    args[34], args[35], args[36], args[37], args[38], args[39], args[40],
-    args[41], args[42], args[43], args[44], args[45], args[46], args[47],
-    args[48], args[49], args[50], args[51], args[52], args[53], args[54],
-    args[55], args[56], args[57]);
-break;
-case 59:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *))fn)(
-    &global_tid, &bound_tid, args[0], args[1], args[2], args[3], args[4],
-    args[5], args[6], args[7], args[8], args[9], args[10], args[11], args[12],
-    args[13], args[14], args[15], args[16], args[17], args[18], args[19],
-    args[20], args[21], args[22], args[23], args[24], args[25], args[26],
-    args[27], args[28], args[29], args[30], args[31], args[32], args[33],
-    args[34], args[35], args[36], args[37], args[38], args[39], args[40],
-    args[41], args[42], args[43], args[44], args[45], args[46], args[47],
-    args[48], args[49], args[50], args[51], args[52], args[53], args[54],
-    args[55], args[56], args[57], args[58]);
-break;
-case 60:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *))fn)(
-    &global_tid, &bound_tid, args[0], args[1], args[2], args[3], args[4],
-    args[5], args[6], args[7], args[8], args[9], args[10], args[11], args[12],
-    args[13], args[14], args[15], args[16], args[17], args[18], args[19],
-    args[20], args[21], args[22], args[23], args[24], args[25], args[26],
-    args[27], args[28], args[29], args[30], args[31], args[32], args[33],
-    args[34], args[35], args[36], args[37], args[38], args[39], args[40],
-    args[41], args[42], args[43], args[44], args[45], args[46], args[47],
-    args[48], args[49], args[50], args[51], args[52], args[53], args[54],
-    args[55], args[56], args[57], args[58], args[59]);
-break;
-case 61:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *))fn)(
-    &global_tid, &bound_tid, args[0], args[1], args[2], args[3], args[4],
-    args[5], args[6], args[7], args[8], args[9], args[10], args[11], args[12],
-    args[13], args[14], args[15], args[16], args[17], args[18], args[19],
-    args[20], args[21], args[22], args[23], args[24], args[25], args[26],
-    args[27], args[28], args[29], args[30], args[31], args[32], args[33],
-    args[34], args[35], args[36], args[37], args[38], args[39], args[40],
-    args[41], args[42], args[43], args[44], args[45], args[46], args[47],
-    args[48], args[49], args[50], args[51], args[52], args[53], args[54],
-    args[55], args[56], args[57], args[58], args[59], args[60]);
-break;
-case 62:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *))fn)(
-    &global_tid, &bound_tid, args[0], args[1], args[2], args[3], args[4],
-    args[5], args[6], args[7], args[8], args[9], args[10], args[11], args[12],
-    args[13], args[14], args[15], args[16], args[17], args[18], args[19],
-    args[20], args[21], args[22], args[23], args[24], args[25], args[26],
-    args[27], args[28], args[29], args[30], args[31], args[32], args[33],
-    args[34], args[35], args[36], args[37], args[38], args[39], args[40],
-    args[41], args[42], args[43], args[44], args[45], args[46], args[47],
-    args[48], args[49], args[50], args[51], args[52], args[53], args[54],
-    args[55], args[56], args[57], args[58], args[59], args[60], args[61]);
-break;
-case 63:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *))fn)(&global_tid, &bound_tid, args[0], args[1], args[2],
-                       args[3], args[4], args[5], args[6], args[7], args[8],
-                       args[9], args[10], args[11], args[12], args[13],
-                       args[14], args[15], args[16], args[17], args[18],
-                       args[19], args[20], args[21], args[22], args[23],
-                       args[24], args[25], args[26], args[27], args[28],
-                       args[29], args[30], args[31], args[32], args[33],
-                       args[34], args[35], args[36], args[37], args[38],
-                       args[39], args[40], args[41], args[42], args[43],
-                       args[44], args[45], args[46], args[47], args[48],
-                       args[49], args[50], args[51], args[52], args[53],
-                       args[54], args[55], args[56], args[57], args[58],
-                       args[59], args[60], args[61], args[62]);
-break;
-case 64:
-((void (*)(int32_t *, int32_t *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *, void *, void *, void *, void *, void *, void *,
-           void *, void *))fn)(
-    &global_tid, &bound_tid, args[0], args[1], args[2], args[3], args[4],
-    args[5], args[6], args[7], args[8], args[9], args[10], args[11], args[12],
-    args[13], args[14], args[15], args[16], args[17], args[18], args[19],
-    args[20], args[21], args[22], args[23], args[24], args[25], args[26],
-    args[27], args[28], args[29], args[30], args[31], args[32], args[33],
-    args[34], args[35], args[36], args[37], args[38], args[39], args[40],
-    args[41], args[42], args[43], args[44], args[45], args[46], args[47],
-    args[48], args[49], args[50], args[51], args[52], args[53], args[54],
-    args[55], args[56], args[57], args[58], args[59], args[60], args[61],
-    args[62], args[63]);
-break;