xref: /llvm-project-15.0.7/openmp/libomptarget/plugins/cuda/src/rtl.cpp (revision a697a0a4)

//===----RTLs/cuda/src/rtl.cpp - Target RTLs Implementation ------- 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
//
//===----------------------------------------------------------------------===//
//
// RTL for CUDA machine
//
//===----------------------------------------------------------------------===//

#include <cassert>
#include <cstddef>
#include <cuda.h>
#include <list>
#include <memory>
#include <mutex>
#include <string>
#include <unordered_map>
#include <vector>

#include "Debug.h"
#include "DeviceEnvironment.h"
#include "omptargetplugin.h"

#define TARGET_NAME CUDA
#define DEBUG_PREFIX "Target " GETNAME(TARGET_NAME) " RTL"

#include "MemoryManager.h"

#include "llvm/Frontend/OpenMP/OMPConstants.h"

// Utility for retrieving and printing CUDA error string.
#ifdef OMPTARGET_DEBUG
#define CUDA_ERR_STRING(err)                                                   \
  do {                                                                         \
    if (getDebugLevel() > 0) {                                                 \
      const char *errStr = nullptr;                                            \
      CUresult errStr_status = cuGetErrorString(err, &errStr);                 \
      if (errStr_status == CUDA_ERROR_INVALID_VALUE)                           \
        REPORT("Unrecognized CUDA error code: %d\n", err);                     \
      else if (errStr_status == CUDA_SUCCESS)                                  \
        REPORT("CUDA error is: %s\n", errStr);                                 \
      else {                                                                   \
        REPORT("Unresolved CUDA error code: %d\n", err);                       \
        REPORT("Unsuccessful cuGetErrorString return status: %d\n",            \
               errStr_status);                                                 \
      }                                                                        \
    } else {                                                                   \
      const char *errStr = nullptr;                                            \
      CUresult errStr_status = cuGetErrorString(err, &errStr);                 \
      if (errStr_status == CUDA_SUCCESS)                                       \
        REPORT("%s \n", errStr);                                               \
    }                                                                          \
  } while (false)
#else // OMPTARGET_DEBUG
#define CUDA_ERR_STRING(err)                                                   \
  do {                                                                         \
    const char *errStr = nullptr;                                              \
    CUresult errStr_status = cuGetErrorString(err, &errStr);                   \
    if (errStr_status == CUDA_SUCCESS)                                         \
      REPORT("%s \n", errStr);                                                 \
  } while (false)
#endif // OMPTARGET_DEBUG

#define BOOL2TEXT(b) ((b) ? "Yes" : "No")

#include "elf_common.h"

/// Keep entries table per device.
struct FuncOrGblEntryTy {
  __tgt_target_table Table;
  std::vector<__tgt_offload_entry> Entries;
};

/// Use a single entity to encode a kernel and a set of flags.
struct KernelTy {
  CUfunction Func;

  // execution mode of kernel
  llvm::omp::OMPTgtExecModeFlags ExecutionMode;

  /// Maximal number of threads per block for this kernel.
  int MaxThreadsPerBlock = 0;

  KernelTy(CUfunction _Func, llvm::omp::OMPTgtExecModeFlags _ExecutionMode)
      : Func(_Func), ExecutionMode(_ExecutionMode) {}
};

namespace {
bool checkResult(CUresult Err, const char *ErrMsg) {
  if (Err == CUDA_SUCCESS)
    return true;

  REPORT("%s", ErrMsg);
  CUDA_ERR_STRING(Err);
  return false;
}

int memcpyDtoD(const void *SrcPtr, void *DstPtr, int64_t Size,
               CUstream Stream) {
  CUresult Err =
      cuMemcpyDtoDAsync((CUdeviceptr)DstPtr, (CUdeviceptr)SrcPtr, Size, Stream);

  if (Err != CUDA_SUCCESS) {
    DP("Error when copying data from device to device. Pointers: src "
       "= " DPxMOD ", dst = " DPxMOD ", size = %" PRId64 "\n",
       DPxPTR(SrcPtr), DPxPTR(DstPtr), Size);
    CUDA_ERR_STRING(Err);
    return OFFLOAD_FAIL;
  }

  return OFFLOAD_SUCCESS;
}

int createEvent(void **P) {
  CUevent Event = nullptr;

  CUresult Err = cuEventCreate(&Event, CU_EVENT_DEFAULT);
  if (Err != CUDA_SUCCESS) {
    DP("Error when creating event event = " DPxMOD "\n", DPxPTR(Event));
    CUDA_ERR_STRING(Err);
    return OFFLOAD_FAIL;
  }

  *P = Event;

  return OFFLOAD_SUCCESS;
}

int recordEvent(void *EventPtr, __tgt_async_info *AsyncInfo) {
  CUstream Stream = reinterpret_cast<CUstream>(AsyncInfo->Queue);
  CUevent Event = reinterpret_cast<CUevent>(EventPtr);

  CUresult Err = cuEventRecord(Event, Stream);
  if (Err != CUDA_SUCCESS) {
    DP("Error when recording event. stream = " DPxMOD ", event = " DPxMOD "\n",
       DPxPTR(Stream), DPxPTR(Event));
    CUDA_ERR_STRING(Err);
    return OFFLOAD_FAIL;
  }

  return OFFLOAD_SUCCESS;
}

int syncEvent(void *EventPtr) {
  CUevent Event = reinterpret_cast<CUevent>(EventPtr);

  CUresult Err = cuEventSynchronize(Event);
  if (Err != CUDA_SUCCESS) {
    DP("Error when syncing event = " DPxMOD "\n", DPxPTR(Event));
    CUDA_ERR_STRING(Err);
    return OFFLOAD_FAIL;
  }

  return OFFLOAD_SUCCESS;
}

int destroyEvent(void *EventPtr) {
  CUevent Event = reinterpret_cast<CUevent>(EventPtr);

  CUresult Err = cuEventDestroy(Event);
  if (Err != CUDA_SUCCESS) {
    DP("Error when destroying event = " DPxMOD "\n", DPxPTR(Event));
    CUDA_ERR_STRING(Err);
    return OFFLOAD_FAIL;
  }

  return OFFLOAD_SUCCESS;
}

// Structure contains per-device data
struct DeviceDataTy {
  /// List that contains all the kernels.
  std::list<KernelTy> KernelsList;

  std::list<FuncOrGblEntryTy> FuncGblEntries;

  CUcontext Context = nullptr;
  // Device properties
  int ThreadsPerBlock = 0;
  int BlocksPerGrid = 0;
  int WarpSize = 0;
  // OpenMP properties
  int NumTeams = 0;
  int NumThreads = 0;
};

/// Resource allocator where \p T is the resource type.
/// Functions \p create and \p destroy return OFFLOAD_SUCCESS and OFFLOAD_FAIL
/// accordingly. The implementation should not raise any exception.
template <typename T> class AllocatorTy {
public:
  /// Create a resource and assign to R.
  int create(T &R) noexcept;
  /// Destroy the resource.
  int destroy(T) noexcept;
};

/// Allocator for CUstream.
template <> class AllocatorTy<CUstream> {
  CUcontext Context;

public:
  AllocatorTy(CUcontext C) noexcept : Context(C) {}

  /// See AllocatorTy<T>::create.
  int create(CUstream &Stream) noexcept {
    if (!checkResult(cuCtxSetCurrent(Context),
                     "Error returned from cuCtxSetCurrent\n"))
      return OFFLOAD_FAIL;

    if (!checkResult(cuStreamCreate(&Stream, CU_STREAM_NON_BLOCKING),
                     "Error returned from cuStreamCreate\n"))
      return OFFLOAD_FAIL;

    return OFFLOAD_SUCCESS;
  }

  /// See AllocatorTy<T>::destroy.
  int destroy(CUstream Stream) noexcept {
    if (!checkResult(cuCtxSetCurrent(Context),
                     "Error returned from cuCtxSetCurrent\n"))
      return OFFLOAD_FAIL;
    if (!checkResult(cuStreamDestroy(Stream),
                     "Error returned from cuStreamDestroy\n"))
      return OFFLOAD_FAIL;

    return OFFLOAD_SUCCESS;
  }
};

/// A generic pool of resources where \p T is the resource type.
/// \p T should be copyable as the object is stored in \p std::vector .
template <typename T> class ResourcePoolTy {
  /// Index of the next available resource.
  size_t Next = 0;
  /// Mutex to guard the pool.
  std::mutex Mutex;
  /// Pool of resources.
  std::vector<T> Resources;
  /// A reference to the corresponding allocator.
  AllocatorTy<T> Allocator;

  /// If `Resources` is used up, we will fill in more resources. It assumes that
  /// the new size `Size` should be always larger than the current size.
  bool resize(size_t Size) {
    auto CurSize = Resources.size();
    assert(Size > CurSize && "Unexpected smaller size");
    Resources.reserve(Size);
    for (auto I = CurSize; I < Size; ++I) {
      T NewItem;
      int Ret = Allocator.create(NewItem);
      if (Ret != OFFLOAD_SUCCESS)
        return false;
      Resources.push_back(NewItem);
    }
    return true;
  }

public:
  ResourcePoolTy(AllocatorTy<T> &&A, size_t Size = 0) noexcept
      : Allocator(std::move(A)) {
    (void)resize(Size);
  }

  ~ResourcePoolTy() noexcept {
    for (auto &R : Resources)
      (void)Allocator.destroy(R);
  }

  /// Get a resource from pool. `Next` always points to the next available
  /// resource. That means, `[0, next-1]` have been assigned, and `[id,]` are
  /// still available. If there is no resource left, we will ask for more. Each
  /// time a resource is assigned, the id will increase one.
  /// xxxxxs+++++++++
  ///      ^
  ///      Next
  /// After assignment, the pool becomes the following and s is assigned.
  /// xxxxxs+++++++++
  ///       ^
  ///       Next
  int acquire(T &R) noexcept {
    std::lock_guard<std::mutex> LG(Mutex);
    if (Next == Resources.size() && !resize(Resources.size() * 2))
      return OFFLOAD_FAIL;

    R = Resources[Next++];

    return OFFLOAD_SUCCESS;
  }

  /// Return the resource back to the pool. When we return a resource, we need
  /// to first decrease `Next`, and then copy the resource back. It is worth
  /// noting that, the order of resources return might be different from that
  /// they're assigned, that saying, at some point, there might be two identical
  /// resources.
  /// xxax+a+++++
  ///     ^
  ///     Next
  /// However, it doesn't matter, because they're always on the two sides of
  /// `Next`. The left one will in the end be overwritten by another resource.
  /// Therefore, after several execution, the order of pool might be different
  /// from its initial state.
  void release(T R) noexcept {
    std::lock_guard<std::mutex> LG(Mutex);
    Resources[--Next] = R;
  }
};

class DeviceRTLTy {
  int NumberOfDevices;
  // OpenMP environment properties
  int EnvNumTeams;
  int EnvTeamLimit;
  int EnvTeamThreadLimit;
  // OpenMP requires flags
  int64_t RequiresFlags;
  // Amount of dynamic shared memory to use at launch.
  uint64_t DynamicMemorySize;
  // Number of initial streams for each device.
  int NumInitialStreams = 32;

  static constexpr const int HardTeamLimit = 1U << 16U; // 64k
  static constexpr const int HardThreadLimit = 1024;
  static constexpr const int DefaultNumTeams = 128;
  static constexpr const int DefaultNumThreads = 128;

  using StreamPoolTy = ResourcePoolTy<CUstream>;
  using StreamAllocatorTy = AllocatorTy<CUstream>;
  std::vector<std::unique_ptr<StreamPoolTy>> StreamPool;

  std::vector<DeviceDataTy> DeviceData;
  std::vector<CUmodule> Modules;

  /// A class responsible for interacting with device native runtime library to
  /// allocate and free memory.
  class CUDADeviceAllocatorTy : public DeviceAllocatorTy {
    const int DeviceId;
    const std::vector<DeviceDataTy> &DeviceData;
    std::unordered_map<void *, TargetAllocTy> HostPinnedAllocs;

  public:
    CUDADeviceAllocatorTy(int DeviceId, std::vector<DeviceDataTy> &DeviceData)
        : DeviceId(DeviceId), DeviceData(DeviceData) {}

    void *allocate(size_t Size, void *, TargetAllocTy Kind) override {
      if (Size == 0)
        return nullptr;

      CUresult Err = cuCtxSetCurrent(DeviceData[DeviceId].Context);
      if (!checkResult(Err, "Error returned from cuCtxSetCurrent\n"))
        return nullptr;

      void *MemAlloc = nullptr;
      switch (Kind) {
      case TARGET_ALLOC_DEFAULT:
      case TARGET_ALLOC_DEVICE:
        CUdeviceptr DevicePtr;
        Err = cuMemAlloc(&DevicePtr, Size);
        MemAlloc = (void *)DevicePtr;
        if (!checkResult(Err, "Error returned from cuMemAlloc\n"))
          return nullptr;
        break;
      case TARGET_ALLOC_HOST:
        void *HostPtr;
        Err = cuMemAllocHost(&HostPtr, Size);
        MemAlloc = HostPtr;
        if (!checkResult(Err, "Error returned from cuMemAllocHost\n"))
          return nullptr;
        HostPinnedAllocs[MemAlloc] = Kind;
        break;
      case TARGET_ALLOC_SHARED:
        CUdeviceptr SharedPtr;
        Err = cuMemAllocManaged(&SharedPtr, Size, CU_MEM_ATTACH_GLOBAL);
        MemAlloc = (void *)SharedPtr;
        if (!checkResult(Err, "Error returned from cuMemAllocManaged\n"))
          return nullptr;
        break;
      }

      return MemAlloc;
    }

    int free(void *TgtPtr) override {
      CUresult Err = cuCtxSetCurrent(DeviceData[DeviceId].Context);
      if (!checkResult(Err, "Error returned from cuCtxSetCurrent\n"))
        return OFFLOAD_FAIL;

      // Host pinned memory must be freed differently.
      TargetAllocTy Kind =
          (HostPinnedAllocs.find(TgtPtr) == HostPinnedAllocs.end())
              ? TARGET_ALLOC_DEFAULT
              : TARGET_ALLOC_HOST;
      switch (Kind) {
      case TARGET_ALLOC_DEFAULT:
      case TARGET_ALLOC_DEVICE:
      case TARGET_ALLOC_SHARED:
        Err = cuMemFree((CUdeviceptr)TgtPtr);
        if (!checkResult(Err, "Error returned from cuMemFree\n"))
          return OFFLOAD_FAIL;
        break;
      case TARGET_ALLOC_HOST:
        Err = cuMemFreeHost(TgtPtr);
        if (!checkResult(Err, "Error returned from cuMemFreeHost\n"))
          return OFFLOAD_FAIL;
        break;
      }

      return OFFLOAD_SUCCESS;
    }
  };

  /// A vector of device allocators
  std::vector<CUDADeviceAllocatorTy> DeviceAllocators;

  /// A vector of memory managers. Since the memory manager is non-copyable and
  // non-removable, we wrap them into std::unique_ptr.
  std::vector<std::unique_ptr<MemoryManagerTy>> MemoryManagers;

  /// Whether use memory manager
  bool UseMemoryManager = true;

  // Record entry point associated with device
  void addOffloadEntry(const int DeviceId, const __tgt_offload_entry entry) {
    FuncOrGblEntryTy &E = DeviceData[DeviceId].FuncGblEntries.back();
    E.Entries.push_back(entry);
  }

  // Return a pointer to the entry associated with the pointer
  const __tgt_offload_entry *getOffloadEntry(const int DeviceId,
                                             const void *Addr) const {
    for (const __tgt_offload_entry &Itr :
         DeviceData[DeviceId].FuncGblEntries.back().Entries)
      if (Itr.addr == Addr)
        return &Itr;

    return nullptr;
  }

  // Return the pointer to the target entries table
  __tgt_target_table *getOffloadEntriesTable(const int DeviceId) {
    FuncOrGblEntryTy &E = DeviceData[DeviceId].FuncGblEntries.back();

    if (E.Entries.empty())
      return nullptr;

    // Update table info according to the entries and return the pointer
    E.Table.EntriesBegin = E.Entries.data();
    E.Table.EntriesEnd = E.Entries.data() + E.Entries.size();

    return &E.Table;
  }

  // Clear entries table for a device
  void clearOffloadEntriesTable(const int DeviceId) {
    DeviceData[DeviceId].FuncGblEntries.emplace_back();
    FuncOrGblEntryTy &E = DeviceData[DeviceId].FuncGblEntries.back();
    E.Entries.clear();
    E.Table.EntriesBegin = E.Table.EntriesEnd = nullptr;
  }

  CUstream getStream(const int DeviceId, __tgt_async_info *AsyncInfo) const {
    assert(AsyncInfo && "AsyncInfo is nullptr");

    if (!AsyncInfo->Queue) {
      CUstream S;
      if (StreamPool[DeviceId]->acquire(S) != OFFLOAD_SUCCESS)
        return nullptr;

      AsyncInfo->Queue = S;
    }

    return reinterpret_cast<CUstream>(AsyncInfo->Queue);
  }

public:
  // This class should not be copied
  DeviceRTLTy(const DeviceRTLTy &) = delete;
  DeviceRTLTy(DeviceRTLTy &&) = delete;

  DeviceRTLTy()
      : NumberOfDevices(0), EnvNumTeams(-1), EnvTeamLimit(-1),
        EnvTeamThreadLimit(-1), RequiresFlags(OMP_REQ_UNDEFINED),
        DynamicMemorySize(0) {

    DP("Start initializing CUDA\n");

    CUresult Err = cuInit(0);
    if (Err == CUDA_ERROR_INVALID_HANDLE) {
      // Can't call cuGetErrorString if dlsym failed
      DP("Failed to load CUDA shared library\n");
      return;
    }
    if (!checkResult(Err, "Error returned from cuInit\n")) {
      return;
    }

    Err = cuDeviceGetCount(&NumberOfDevices);
    if (!checkResult(Err, "Error returned from cuDeviceGetCount\n"))
      return;

    if (NumberOfDevices == 0) {
      DP("There are no devices supporting CUDA.\n");
      return;
    }

    DeviceData.resize(NumberOfDevices);
    StreamPool.resize(NumberOfDevices);

    // Get environment variables regarding teams
    if (const char *EnvStr = getenv("OMP_TEAM_LIMIT")) {
      // OMP_TEAM_LIMIT has been set
      EnvTeamLimit = std::stoi(EnvStr);
      DP("Parsed OMP_TEAM_LIMIT=%d\n", EnvTeamLimit);
    }
    if (const char *EnvStr = getenv("OMP_TEAMS_THREAD_LIMIT")) {
      // OMP_TEAMS_THREAD_LIMIT has been set
      EnvTeamThreadLimit = std::stoi(EnvStr);
      DP("Parsed OMP_TEAMS_THREAD_LIMIT=%d\n", EnvTeamThreadLimit);
    }
    if (const char *EnvStr = getenv("OMP_NUM_TEAMS")) {
      // OMP_NUM_TEAMS has been set
      EnvNumTeams = std::stoi(EnvStr);
      DP("Parsed OMP_NUM_TEAMS=%d\n", EnvNumTeams);
    }
    if (const char *EnvStr = getenv("LIBOMPTARGET_SHARED_MEMORY_SIZE")) {
      // LIBOMPTARGET_SHARED_MEMORY_SIZE has been set
      DynamicMemorySize = std::stoi(EnvStr);
      DP("Parsed LIBOMPTARGET_SHARED_MEMORY_SIZE = %" PRIu64 "\n",
         DynamicMemorySize);
    }
    if (const char *EnvStr = getenv("LIBOMPTARGET_NUM_INITIAL_STREAMS")) {
      // LIBOMPTARGET_NUM_INITIAL_STREAMS has been set
      NumInitialStreams = std::stoi(EnvStr);
      DP("Parsed LIBOMPTARGET_NUM_INITIAL_STREAMS=%d\n", NumInitialStreams);
    }

    for (int I = 0; I < NumberOfDevices; ++I)
      DeviceAllocators.emplace_back(I, DeviceData);

    // Get the size threshold from environment variable
    std::pair<size_t, bool> Res = MemoryManagerTy::getSizeThresholdFromEnv();
    UseMemoryManager = Res.second;
    size_t MemoryManagerThreshold = Res.first;

    if (UseMemoryManager)
      for (int I = 0; I < NumberOfDevices; ++I)
        MemoryManagers.emplace_back(std::make_unique<MemoryManagerTy>(
            DeviceAllocators[I], MemoryManagerThreshold));
  }

  ~DeviceRTLTy() {
    // We first destruct memory managers in case that its dependent data are
    // destroyed before it.
    for (auto &M : MemoryManagers)
      M.release();

    for (CUmodule &M : Modules)
      // Close module
      if (M)
        checkResult(cuModuleUnload(M), "Error returned from cuModuleUnload\n");

    for (auto &S : StreamPool)
      S = nullptr;

    for (DeviceDataTy &D : DeviceData) {
      // Destroy context
      if (D.Context) {
        checkResult(cuCtxSetCurrent(D.Context),
                    "Error returned from cuCtxSetCurrent\n");
        CUdevice Device;
        checkResult(cuCtxGetDevice(&Device),
                    "Error returned from cuCtxGetDevice\n");
        checkResult(cuDevicePrimaryCtxRelease(Device),
                    "Error returned from cuDevicePrimaryCtxRelease\n");
      }
    }
  }

  // Check whether a given DeviceId is valid
  bool isValidDeviceId(const int DeviceId) const {
    return DeviceId >= 0 && DeviceId < NumberOfDevices;
  }

  int getNumOfDevices() const { return NumberOfDevices; }

  void setRequiresFlag(const int64_t Flags) { this->RequiresFlags = Flags; }

  int initDevice(const int DeviceId) {
    CUdevice Device;

    DP("Getting device %d\n", DeviceId);
    CUresult Err = cuDeviceGet(&Device, DeviceId);
    if (!checkResult(Err, "Error returned from cuDeviceGet\n"))
      return OFFLOAD_FAIL;

    // Query the current flags of the primary context and set its flags if
    // it is inactive
    unsigned int FormerPrimaryCtxFlags = 0;
    int FormerPrimaryCtxIsActive = 0;
    Err = cuDevicePrimaryCtxGetState(Device, &FormerPrimaryCtxFlags,
                                     &FormerPrimaryCtxIsActive);
    if (!checkResult(Err, "Error returned from cuDevicePrimaryCtxGetState\n"))
      return OFFLOAD_FAIL;

    if (FormerPrimaryCtxIsActive) {
      DP("The primary context is active, no change to its flags\n");
      if ((FormerPrimaryCtxFlags & CU_CTX_SCHED_MASK) !=
          CU_CTX_SCHED_BLOCKING_SYNC)
        DP("Warning the current flags are not CU_CTX_SCHED_BLOCKING_SYNC\n");
    } else {
      DP("The primary context is inactive, set its flags to "
         "CU_CTX_SCHED_BLOCKING_SYNC\n");
      Err = cuDevicePrimaryCtxSetFlags(Device, CU_CTX_SCHED_BLOCKING_SYNC);
      if (!checkResult(Err, "Error returned from cuDevicePrimaryCtxSetFlags\n"))
        return OFFLOAD_FAIL;
    }

    // Retain the per device primary context and save it to use whenever this
    // device is selected.
    Err = cuDevicePrimaryCtxRetain(&DeviceData[DeviceId].Context, Device);
    if (!checkResult(Err, "Error returned from cuDevicePrimaryCtxRetain\n"))
      return OFFLOAD_FAIL;

    Err = cuCtxSetCurrent(DeviceData[DeviceId].Context);
    if (!checkResult(Err, "Error returned from cuCtxSetCurrent\n"))
      return OFFLOAD_FAIL;

    // Initialize stream pool
    if (!StreamPool[DeviceId])
      StreamPool[DeviceId] = std::make_unique<StreamPoolTy>(
          StreamAllocatorTy(DeviceData[DeviceId].Context), NumInitialStreams);

    // Query attributes to determine number of threads/block and blocks/grid.
    int MaxGridDimX;
    Err = cuDeviceGetAttribute(&MaxGridDimX, CU_DEVICE_ATTRIBUTE_MAX_GRID_DIM_X,
                               Device);
    if (Err != CUDA_SUCCESS) {
      DP("Error getting max grid dimension, use default value %d\n",
         DeviceRTLTy::DefaultNumTeams);
      DeviceData[DeviceId].BlocksPerGrid = DeviceRTLTy::DefaultNumTeams;
    } else if (MaxGridDimX <= DeviceRTLTy::HardTeamLimit) {
      DP("Using %d CUDA blocks per grid\n", MaxGridDimX);
      DeviceData[DeviceId].BlocksPerGrid = MaxGridDimX;
    } else {
      DP("Max CUDA blocks per grid %d exceeds the hard team limit %d, capping "
         "at the hard limit\n",
         MaxGridDimX, DeviceRTLTy::HardTeamLimit);
      DeviceData[DeviceId].BlocksPerGrid = DeviceRTLTy::HardTeamLimit;
    }

    // We are only exploiting threads along the x axis.
    int MaxBlockDimX;
    Err = cuDeviceGetAttribute(&MaxBlockDimX,
                               CU_DEVICE_ATTRIBUTE_MAX_BLOCK_DIM_X, Device);
    if (Err != CUDA_SUCCESS) {
      DP("Error getting max block dimension, use default value %d\n",
         DeviceRTLTy::DefaultNumThreads);
      DeviceData[DeviceId].ThreadsPerBlock = DeviceRTLTy::DefaultNumThreads;
    } else {
      DP("Using %d CUDA threads per block\n", MaxBlockDimX);
      DeviceData[DeviceId].ThreadsPerBlock = MaxBlockDimX;

      if (EnvTeamThreadLimit > 0 &&
          DeviceData[DeviceId].ThreadsPerBlock > EnvTeamThreadLimit) {
        DP("Max CUDA threads per block %d exceeds the thread limit %d set by "
           "OMP_TEAMS_THREAD_LIMIT, capping at the limit\n",
           DeviceData[DeviceId].ThreadsPerBlock, EnvTeamThreadLimit);
        DeviceData[DeviceId].ThreadsPerBlock = EnvTeamThreadLimit;
      }
      if (DeviceData[DeviceId].ThreadsPerBlock > DeviceRTLTy::HardThreadLimit) {
        DP("Max CUDA threads per block %d exceeds the hard thread limit %d, "
           "capping at the hard limit\n",
           DeviceData[DeviceId].ThreadsPerBlock, DeviceRTLTy::HardThreadLimit);
        DeviceData[DeviceId].ThreadsPerBlock = DeviceRTLTy::HardThreadLimit;
      }
    }

    // Get and set warp size
    int WarpSize;
    Err =
        cuDeviceGetAttribute(&WarpSize, CU_DEVICE_ATTRIBUTE_WARP_SIZE, Device);
    if (Err != CUDA_SUCCESS) {
      DP("Error getting warp size, assume default value 32\n");
      DeviceData[DeviceId].WarpSize = 32;
    } else {
      DP("Using warp size %d\n", WarpSize);
      DeviceData[DeviceId].WarpSize = WarpSize;
    }

    // Adjust teams to the env variables
    if (EnvTeamLimit > 0 && DeviceData[DeviceId].BlocksPerGrid > EnvTeamLimit) {
      DP("Capping max CUDA blocks per grid to OMP_TEAM_LIMIT=%d\n",
         EnvTeamLimit);
      DeviceData[DeviceId].BlocksPerGrid = EnvTeamLimit;
    }

    size_t StackLimit;
    size_t HeapLimit;
    if (const char *EnvStr = getenv("LIBOMPTARGET_STACK_SIZE")) {
      StackLimit = std::stol(EnvStr);
      if (cuCtxSetLimit(CU_LIMIT_STACK_SIZE, StackLimit) != CUDA_SUCCESS)
        return OFFLOAD_FAIL;
    } else {
      if (cuCtxGetLimit(&StackLimit, CU_LIMIT_STACK_SIZE) != CUDA_SUCCESS)
        return OFFLOAD_FAIL;
    }
    if (const char *EnvStr = getenv("LIBOMPTARGET_HEAP_SIZE")) {
      HeapLimit = std::stol(EnvStr);
      if (cuCtxSetLimit(CU_LIMIT_MALLOC_HEAP_SIZE, HeapLimit) != CUDA_SUCCESS)
        return OFFLOAD_FAIL;
    } else {
      if (cuCtxGetLimit(&HeapLimit, CU_LIMIT_MALLOC_HEAP_SIZE) != CUDA_SUCCESS)
        return OFFLOAD_FAIL;
    }

    INFO(OMP_INFOTYPE_PLUGIN_KERNEL, DeviceId,
         "Device supports up to %d CUDA blocks and %d threads with a "
         "warp size of %d\n",
         DeviceData[DeviceId].BlocksPerGrid,
         DeviceData[DeviceId].ThreadsPerBlock, DeviceData[DeviceId].WarpSize);
    INFO(OMP_INFOTYPE_PLUGIN_KERNEL, DeviceId,
         "Device heap size is %d Bytes, device stack size is %d Bytes per "
         "thread\n",
         (int)HeapLimit, (int)StackLimit);

    // Set default number of teams
    if (EnvNumTeams > 0) {
      DP("Default number of teams set according to environment %d\n",
         EnvNumTeams);
      DeviceData[DeviceId].NumTeams = EnvNumTeams;
    } else {
      DeviceData[DeviceId].NumTeams = DeviceRTLTy::DefaultNumTeams;
      DP("Default number of teams set according to library's default %d\n",
         DeviceRTLTy::DefaultNumTeams);
    }

    if (DeviceData[DeviceId].NumTeams > DeviceData[DeviceId].BlocksPerGrid) {
      DP("Default number of teams exceeds device limit, capping at %d\n",
         DeviceData[DeviceId].BlocksPerGrid);
      DeviceData[DeviceId].NumTeams = DeviceData[DeviceId].BlocksPerGrid;
    }

    // Set default number of threads
    DeviceData[DeviceId].NumThreads = DeviceRTLTy::DefaultNumThreads;
    DP("Default number of threads set according to library's default %d\n",
       DeviceRTLTy::DefaultNumThreads);
    if (DeviceData[DeviceId].NumThreads >
        DeviceData[DeviceId].ThreadsPerBlock) {
      DP("Default number of threads exceeds device limit, capping at %d\n",
         DeviceData[DeviceId].ThreadsPerBlock);
      DeviceData[DeviceId].NumThreads = DeviceData[DeviceId].ThreadsPerBlock;
    }

    return OFFLOAD_SUCCESS;
  }

  __tgt_target_table *loadBinary(const int DeviceId,
                                 const __tgt_device_image *Image) {
    // Set the context we are using
    CUresult Err = cuCtxSetCurrent(DeviceData[DeviceId].Context);
    if (!checkResult(Err, "Error returned from cuCtxSetCurrent\n"))
      return nullptr;

    // Clear the offload table as we are going to create a new one.
    clearOffloadEntriesTable(DeviceId);

    // Create the module and extract the function pointers.
    CUmodule Module;
    DP("Load data from image " DPxMOD "\n", DPxPTR(Image->ImageStart));
    Err = cuModuleLoadDataEx(&Module, Image->ImageStart, 0, nullptr, nullptr);
    if (!checkResult(Err, "Error returned from cuModuleLoadDataEx\n"))
      return nullptr;

    DP("CUDA module successfully loaded!\n");

    Modules.push_back(Module);

    // Find the symbols in the module by name.
    const __tgt_offload_entry *HostBegin = Image->EntriesBegin;
    const __tgt_offload_entry *HostEnd = Image->EntriesEnd;

    std::list<KernelTy> &KernelsList = DeviceData[DeviceId].KernelsList;
    for (const __tgt_offload_entry *E = HostBegin; E != HostEnd; ++E) {
      if (!E->addr) {
        // We return nullptr when something like this happens, the host should
        // have always something in the address to uniquely identify the target
        // region.
        DP("Invalid binary: host entry '<null>' (size = %zd)...\n", E->size);
        return nullptr;
      }

      if (E->size) {
        __tgt_offload_entry Entry = *E;
        CUdeviceptr CUPtr;
        size_t CUSize;
        Err = cuModuleGetGlobal(&CUPtr, &CUSize, Module, E->name);
        // We keep this style here because we need the name
        if (Err != CUDA_SUCCESS) {
          REPORT("Loading global '%s' Failed\n", E->name);
          CUDA_ERR_STRING(Err);
          return nullptr;
        }

        if (CUSize != E->size) {
          DP("Loading global '%s' - size mismatch (%zd != %zd)\n", E->name,
             CUSize, E->size);
          return nullptr;
        }

        DP("Entry point " DPxMOD " maps to global %s (" DPxMOD ")\n",
           DPxPTR(E - HostBegin), E->name, DPxPTR(CUPtr));

        Entry.addr = (void *)(CUPtr);

        // Note: In the current implementation declare target variables
        // can either be link or to. This means that once unified
        // memory is activated via the requires directive, the variable
        // can be used directly from the host in both cases.
        // TODO: when variables types other than to or link are added,
        // the below condition should be changed to explicitly
        // check for to and link variables types:
        // (RequiresFlags & OMP_REQ_UNIFIED_SHARED_MEMORY && (e->flags &
        // OMP_DECLARE_TARGET_LINK || e->flags == OMP_DECLARE_TARGET_TO))
        if (RequiresFlags & OMP_REQ_UNIFIED_SHARED_MEMORY) {
          // If unified memory is present any target link or to variables
          // can access host addresses directly. There is no longer a
          // need for device copies.
          cuMemcpyHtoD(CUPtr, E->addr, sizeof(void *));
          DP("Copy linked variable host address (" DPxMOD
             ") to device address (" DPxMOD ")\n",
             DPxPTR(*((void **)E->addr)), DPxPTR(CUPtr));
        }

        addOffloadEntry(DeviceId, Entry);

        continue;
      }

      CUfunction Func;
      Err = cuModuleGetFunction(&Func, Module, E->name);
      // We keep this style here because we need the name
      if (Err != CUDA_SUCCESS) {
        REPORT("Loading '%s' Failed\n", E->name);
        CUDA_ERR_STRING(Err);
        return nullptr;
      }

      DP("Entry point " DPxMOD " maps to %s (" DPxMOD ")\n",
         DPxPTR(E - HostBegin), E->name, DPxPTR(Func));

      // default value GENERIC (in case symbol is missing from cubin file)
      llvm::omp::OMPTgtExecModeFlags ExecModeVal;
      std::string ExecModeNameStr(E->name);
      ExecModeNameStr += "_exec_mode";
      const char *ExecModeName = ExecModeNameStr.c_str();

      CUdeviceptr ExecModePtr;
      size_t CUSize;
      Err = cuModuleGetGlobal(&ExecModePtr, &CUSize, Module, ExecModeName);
      if (Err == CUDA_SUCCESS) {
        if (CUSize != sizeof(llvm::omp::OMPTgtExecModeFlags)) {
          DP("Loading global exec_mode '%s' - size mismatch (%zd != %zd)\n",
             ExecModeName, CUSize, sizeof(llvm::omp::OMPTgtExecModeFlags));
          return nullptr;
        }

        Err = cuMemcpyDtoH(&ExecModeVal, ExecModePtr, CUSize);
        if (Err != CUDA_SUCCESS) {
          REPORT("Error when copying data from device to host. Pointers: "
                 "host = " DPxMOD ", device = " DPxMOD ", size = %zd\n",
                 DPxPTR(&ExecModeVal), DPxPTR(ExecModePtr), CUSize);
          CUDA_ERR_STRING(Err);
          return nullptr;
        }
      } else {
        DP("Loading global exec_mode '%s' - symbol missing, using default "
           "value GENERIC (1)\n",
           ExecModeName);
      }

      KernelsList.emplace_back(Func, ExecModeVal);

      __tgt_offload_entry Entry = *E;
      Entry.addr = &KernelsList.back();
      addOffloadEntry(DeviceId, Entry);
    }

    // send device environment data to the device
    {
      // TODO: The device ID used here is not the real device ID used by OpenMP.
      DeviceEnvironmentTy DeviceEnv{0, static_cast<uint32_t>(NumberOfDevices),
                                    static_cast<uint32_t>(DeviceId),
                                    static_cast<uint32_t>(DynamicMemorySize)};

      if (const char *EnvStr = getenv("LIBOMPTARGET_DEVICE_RTL_DEBUG"))
        DeviceEnv.DebugKind = std::stoi(EnvStr);

      const char *DeviceEnvName = "omptarget_device_environment";
      CUdeviceptr DeviceEnvPtr;
      size_t CUSize;

      Err = cuModuleGetGlobal(&DeviceEnvPtr, &CUSize, Module, DeviceEnvName);
      if (Err == CUDA_SUCCESS) {
        if (CUSize != sizeof(DeviceEnv)) {
          REPORT(
              "Global device_environment '%s' - size mismatch (%zu != %zu)\n",
              DeviceEnvName, CUSize, sizeof(int32_t));
          CUDA_ERR_STRING(Err);
          return nullptr;
        }

        Err = cuMemcpyHtoD(DeviceEnvPtr, &DeviceEnv, CUSize);
        if (Err != CUDA_SUCCESS) {
          REPORT("Error when copying data from host to device. Pointers: "
                 "host = " DPxMOD ", device = " DPxMOD ", size = %zu\n",
                 DPxPTR(&DeviceEnv), DPxPTR(DeviceEnvPtr), CUSize);
          CUDA_ERR_STRING(Err);
          return nullptr;
        }

        DP("Sending global device environment data %zu bytes\n", CUSize);
      } else {
        DP("Finding global device environment '%s' - symbol missing.\n",
           DeviceEnvName);
        DP("Continue, considering this is a device RTL which does not accept "
           "environment setting.\n");
      }
    }

    return getOffloadEntriesTable(DeviceId);
  }

  void *dataAlloc(const int DeviceId, const int64_t Size,
                  const TargetAllocTy Kind) {
    switch (Kind) {
    case TARGET_ALLOC_DEFAULT:
    case TARGET_ALLOC_DEVICE:
      if (UseMemoryManager)
        return MemoryManagers[DeviceId]->allocate(Size, nullptr);
      else
        return DeviceAllocators[DeviceId].allocate(Size, nullptr, Kind);
    case TARGET_ALLOC_HOST:
    case TARGET_ALLOC_SHARED:
      return DeviceAllocators[DeviceId].allocate(Size, nullptr, Kind);
    }

    REPORT("Invalid target data allocation kind or requested allocator not "
           "implemented yet\n");

    return nullptr;
  }

  int dataSubmit(const int DeviceId, const void *TgtPtr, const void *HstPtr,
                 const int64_t Size, __tgt_async_info *AsyncInfo) const {
    assert(AsyncInfo && "AsyncInfo is nullptr");

    CUresult Err = cuCtxSetCurrent(DeviceData[DeviceId].Context);
    if (!checkResult(Err, "Error returned from cuCtxSetCurrent\n"))
      return OFFLOAD_FAIL;

    CUstream Stream = getStream(DeviceId, AsyncInfo);

    Err = cuMemcpyHtoDAsync((CUdeviceptr)TgtPtr, HstPtr, Size, Stream);
    if (Err != CUDA_SUCCESS) {
      DP("Error when copying data from host to device. Pointers: host "
         "= " DPxMOD ", device = " DPxMOD ", size = %" PRId64 "\n",
         DPxPTR(HstPtr), DPxPTR(TgtPtr), Size);
      CUDA_ERR_STRING(Err);
      return OFFLOAD_FAIL;
    }

    return OFFLOAD_SUCCESS;
  }

  int dataRetrieve(const int DeviceId, void *HstPtr, const void *TgtPtr,
                   const int64_t Size, __tgt_async_info *AsyncInfo) const {
    assert(AsyncInfo && "AsyncInfo is nullptr");

    CUresult Err = cuCtxSetCurrent(DeviceData[DeviceId].Context);
    if (!checkResult(Err, "Error returned from cuCtxSetCurrent\n"))
      return OFFLOAD_FAIL;

    CUstream Stream = getStream(DeviceId, AsyncInfo);

    Err = cuMemcpyDtoHAsync(HstPtr, (CUdeviceptr)TgtPtr, Size, Stream);
    if (Err != CUDA_SUCCESS) {
      DP("Error when copying data from device to host. Pointers: host "
         "= " DPxMOD ", device = " DPxMOD ", size = %" PRId64 "\n",
         DPxPTR(HstPtr), DPxPTR(TgtPtr), Size);
      CUDA_ERR_STRING(Err);
      return OFFLOAD_FAIL;
    }

    return OFFLOAD_SUCCESS;
  }

  int dataExchange(int SrcDevId, const void *SrcPtr, int DstDevId, void *DstPtr,
                   int64_t Size, __tgt_async_info *AsyncInfo) const {
    assert(AsyncInfo && "AsyncInfo is nullptr");

    CUresult Err = cuCtxSetCurrent(DeviceData[SrcDevId].Context);
    if (!checkResult(Err, "Error returned from cuCtxSetCurrent\n"))
      return OFFLOAD_FAIL;

    CUstream Stream = getStream(SrcDevId, AsyncInfo);

    // If they are two devices, we try peer to peer copy first
    if (SrcDevId != DstDevId) {
      int CanAccessPeer = 0;
      Err = cuDeviceCanAccessPeer(&CanAccessPeer, SrcDevId, DstDevId);
      if (Err != CUDA_SUCCESS) {
        REPORT("Error returned from cuDeviceCanAccessPeer. src = %" PRId32
               ", dst = %" PRId32 "\n",
               SrcDevId, DstDevId);
        CUDA_ERR_STRING(Err);
        return memcpyDtoD(SrcPtr, DstPtr, Size, Stream);
      }

      if (!CanAccessPeer) {
        DP("P2P memcpy not supported so fall back to D2D memcpy");
        return memcpyDtoD(SrcPtr, DstPtr, Size, Stream);
      }

      Err = cuCtxEnablePeerAccess(DeviceData[DstDevId].Context, 0);
      if (Err != CUDA_SUCCESS) {
        REPORT("Error returned from cuCtxEnablePeerAccess. src = %" PRId32
               ", dst = %" PRId32 "\n",
               SrcDevId, DstDevId);
        CUDA_ERR_STRING(Err);
        return memcpyDtoD(SrcPtr, DstPtr, Size, Stream);
      }

      Err = cuMemcpyPeerAsync((CUdeviceptr)DstPtr, DeviceData[DstDevId].Context,
                              (CUdeviceptr)SrcPtr, DeviceData[SrcDevId].Context,
                              Size, Stream);
      if (Err == CUDA_SUCCESS)
        return OFFLOAD_SUCCESS;

      DP("Error returned from cuMemcpyPeerAsync. src_ptr = " DPxMOD
         ", src_id =%" PRId32 ", dst_ptr = " DPxMOD ", dst_id =%" PRId32 "\n",
         DPxPTR(SrcPtr), SrcDevId, DPxPTR(DstPtr), DstDevId);
      CUDA_ERR_STRING(Err);
    }

    return memcpyDtoD(SrcPtr, DstPtr, Size, Stream);
  }

  int dataDelete(const int DeviceId, void *TgtPtr) {
    if (UseMemoryManager)
      return MemoryManagers[DeviceId]->free(TgtPtr);

    return DeviceAllocators[DeviceId].free(TgtPtr);
  }

  int runTargetTeamRegion(const int DeviceId, void *TgtEntryPtr, void **TgtArgs,
                          ptrdiff_t *TgtOffsets, const int ArgNum,
                          const int TeamNum, const int ThreadLimit,
                          const unsigned int LoopTripCount,
                          __tgt_async_info *AsyncInfo) const {
    CUresult Err = cuCtxSetCurrent(DeviceData[DeviceId].Context);
    if (!checkResult(Err, "Error returned from cuCtxSetCurrent\n"))
      return OFFLOAD_FAIL;

    // All args are references.
    std::vector<void *> Args(ArgNum);
    std::vector<void *> Ptrs(ArgNum);

    for (int I = 0; I < ArgNum; ++I) {
      Ptrs[I] = (void *)((intptr_t)TgtArgs[I] + TgtOffsets[I]);
      Args[I] = &Ptrs[I];
    }

    KernelTy *KernelInfo = reinterpret_cast<KernelTy *>(TgtEntryPtr);

    const bool IsSPMDGenericMode =
        KernelInfo->ExecutionMode == llvm::omp::OMP_TGT_EXEC_MODE_GENERIC_SPMD;
    const bool IsSPMDMode =
        KernelInfo->ExecutionMode == llvm::omp::OMP_TGT_EXEC_MODE_SPMD;
    const bool IsGenericMode =
        KernelInfo->ExecutionMode == llvm::omp::OMP_TGT_EXEC_MODE_GENERIC;

    int CudaThreadsPerBlock;
    if (ThreadLimit > 0) {
      DP("Setting CUDA threads per block to requested %d\n", ThreadLimit);
      CudaThreadsPerBlock = ThreadLimit;
      // Add master warp if necessary
      if (IsGenericMode) {
        DP("Adding master warp: +%d threads\n", DeviceData[DeviceId].WarpSize);
        CudaThreadsPerBlock += DeviceData[DeviceId].WarpSize;
      }
    } else {
      DP("Setting CUDA threads per block to default %d\n",
         DeviceData[DeviceId].NumThreads);
      CudaThreadsPerBlock = DeviceData[DeviceId].NumThreads;
    }

    if (CudaThreadsPerBlock > DeviceData[DeviceId].ThreadsPerBlock) {
      DP("Threads per block capped at device limit %d\n",
         DeviceData[DeviceId].ThreadsPerBlock);
      CudaThreadsPerBlock = DeviceData[DeviceId].ThreadsPerBlock;
    }

    if (!KernelInfo->MaxThreadsPerBlock) {
      Err = cuFuncGetAttribute(&KernelInfo->MaxThreadsPerBlock,
                               CU_FUNC_ATTRIBUTE_MAX_THREADS_PER_BLOCK,
                               KernelInfo->Func);
      if (!checkResult(Err, "Error returned from cuFuncGetAttribute\n"))
        return OFFLOAD_FAIL;
    }

    if (KernelInfo->MaxThreadsPerBlock < CudaThreadsPerBlock) {
      DP("Threads per block capped at kernel limit %d\n",
         KernelInfo->MaxThreadsPerBlock);
      CudaThreadsPerBlock = KernelInfo->MaxThreadsPerBlock;
    }

    unsigned int CudaBlocksPerGrid;
    if (TeamNum <= 0) {
      if (LoopTripCount > 0 && EnvNumTeams < 0) {
        if (IsSPMDGenericMode) {
          // If we reach this point, then we are executing a kernel that was
          // transformed from Generic-mode to SPMD-mode. This kernel has
          // SPMD-mode execution, but needs its blocks to be scheduled
          // differently because the current loop trip count only applies to the
          // `teams distribute` region and will create var too few blocks using
          // the regular SPMD-mode method.
          CudaBlocksPerGrid = LoopTripCount;
        } else if (IsSPMDMode) {
          // We have a combined construct, i.e. `target teams distribute
          // parallel for [simd]`. We launch so many teams so that each thread
          // will execute one iteration of the loop. round up to the nearest
          // integer
          CudaBlocksPerGrid = ((LoopTripCount - 1) / CudaThreadsPerBlock) + 1;
        } else if (IsGenericMode) {
          // If we reach this point, then we have a non-combined construct, i.e.
          // `teams distribute` with a nested `parallel for` and each team is
          // assigned one iteration of the `distribute` loop. E.g.:
          //
          // #pragma omp target teams distribute
          // for(...loop_tripcount...) {
          //   #pragma omp parallel for
          //   for(...) {}
          // }
          //
          // Threads within a team will execute the iterations of the `parallel`
          // loop.
          CudaBlocksPerGrid = LoopTripCount;
        } else {
          REPORT("Unknown execution mode: %d\n",
                 static_cast<int8_t>(KernelInfo->ExecutionMode));
          return OFFLOAD_FAIL;
        }
        DP("Using %d teams due to loop trip count %" PRIu32
           " and number of threads per block %d\n",
           CudaBlocksPerGrid, LoopTripCount, CudaThreadsPerBlock);
      } else {
        DP("Using default number of teams %d\n", DeviceData[DeviceId].NumTeams);
        CudaBlocksPerGrid = DeviceData[DeviceId].NumTeams;
      }
    } else if (TeamNum > DeviceData[DeviceId].BlocksPerGrid) {
      DP("Capping number of teams to team limit %d\n",
         DeviceData[DeviceId].BlocksPerGrid);
      CudaBlocksPerGrid = DeviceData[DeviceId].BlocksPerGrid;
    } else {
      DP("Using requested number of teams %d\n", TeamNum);
      CudaBlocksPerGrid = TeamNum;
    }

    INFO(OMP_INFOTYPE_PLUGIN_KERNEL, DeviceId,
         "Launching kernel %s with %d blocks and %d threads in %s mode\n",
         (getOffloadEntry(DeviceId, TgtEntryPtr))
             ? getOffloadEntry(DeviceId, TgtEntryPtr)->name
             : "(null)",
         CudaBlocksPerGrid, CudaThreadsPerBlock,
         (!IsSPMDMode ? (IsGenericMode ? "Generic" : "SPMD-Generic") : "SPMD"));

    CUstream Stream = getStream(DeviceId, AsyncInfo);
    Err = cuLaunchKernel(KernelInfo->Func, CudaBlocksPerGrid, /* gridDimY */ 1,
                         /* gridDimZ */ 1, CudaThreadsPerBlock,
                         /* blockDimY */ 1, /* blockDimZ */ 1,
                         DynamicMemorySize, Stream, &Args[0], nullptr);
    if (!checkResult(Err, "Error returned from cuLaunchKernel\n"))
      return OFFLOAD_FAIL;

    DP("Launch of entry point at " DPxMOD " successful!\n",
       DPxPTR(TgtEntryPtr));

    return OFFLOAD_SUCCESS;
  }

  int synchronize(const int DeviceId, __tgt_async_info *AsyncInfo) const {
    CUstream Stream = reinterpret_cast<CUstream>(AsyncInfo->Queue);
    CUresult Err = cuStreamSynchronize(Stream);

    // Once the stream is synchronized, return it to stream pool and reset
    // AsyncInfo. This is to make sure the synchronization only works for its
    // own tasks.
    StreamPool[DeviceId]->release(reinterpret_cast<CUstream>(AsyncInfo->Queue));
    AsyncInfo->Queue = nullptr;

    if (Err != CUDA_SUCCESS) {
      DP("Error when synchronizing stream. stream = " DPxMOD
         ", async info ptr = " DPxMOD "\n",
         DPxPTR(Stream), DPxPTR(AsyncInfo));
      CUDA_ERR_STRING(Err);
    }
    return (Err == CUDA_SUCCESS) ? OFFLOAD_SUCCESS : OFFLOAD_FAIL;
  }

  void printDeviceInfo(int32_t device_id) {
    char TmpChar[1000];
    std::string TmpStr;
    size_t TmpSt;
    int TmpInt, TmpInt2, TmpInt3;

    CUdevice Device;
    checkResult(cuDeviceGet(&Device, device_id),
                "Error returned from cuCtxGetDevice\n");

    cuDriverGetVersion(&TmpInt);
    printf("    CUDA Driver Version: \t\t%d \n", TmpInt);
    printf("    CUDA Device Number: \t\t%d \n", device_id);
    checkResult(cuDeviceGetName(TmpChar, 1000, Device),
                "Error returned from cuDeviceGetName\n");
    printf("    Device Name: \t\t\t%s \n", TmpChar);
    checkResult(cuDeviceTotalMem(&TmpSt, Device),
                "Error returned from cuDeviceTotalMem\n");
    printf("    Global Memory Size: \t\t%zu bytes \n", TmpSt);
    checkResult(cuDeviceGetAttribute(
                    &TmpInt, CU_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT, Device),
                "Error returned from cuDeviceGetAttribute\n");
    printf("    Number of Multiprocessors: \t\t%d \n", TmpInt);
    checkResult(
        cuDeviceGetAttribute(&TmpInt, CU_DEVICE_ATTRIBUTE_GPU_OVERLAP, Device),
        "Error returned from cuDeviceGetAttribute\n");
    printf("    Concurrent Copy and Execution: \t%s \n", BOOL2TEXT(TmpInt));
    checkResult(cuDeviceGetAttribute(
                    &TmpInt, CU_DEVICE_ATTRIBUTE_TOTAL_CONSTANT_MEMORY, Device),
                "Error returned from cuDeviceGetAttribute\n");
    printf("    Total Constant Memory: \t\t%d bytes\n", TmpInt);
    checkResult(
        cuDeviceGetAttribute(
            &TmpInt, CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_BLOCK, Device),
        "Error returned from cuDeviceGetAttribute\n");
    printf("    Max Shared Memory per Block: \t%d bytes \n", TmpInt);
    checkResult(
        cuDeviceGetAttribute(
            &TmpInt, CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_BLOCK, Device),
        "Error returned from cuDeviceGetAttribute\n");
    printf("    Registers per Block: \t\t%d \n", TmpInt);
    checkResult(
        cuDeviceGetAttribute(&TmpInt, CU_DEVICE_ATTRIBUTE_WARP_SIZE, Device),
        "Error returned from cuDeviceGetAttribute\n");
    printf("    Warp Size: \t\t\t\t%d Threads \n", TmpInt);
    checkResult(cuDeviceGetAttribute(
                    &TmpInt, CU_DEVICE_ATTRIBUTE_MAX_THREADS_PER_BLOCK, Device),
                "Error returned from cuDeviceGetAttribute\n");
    printf("    Maximum Threads per Block: \t\t%d \n", TmpInt);
    checkResult(cuDeviceGetAttribute(
                    &TmpInt, CU_DEVICE_ATTRIBUTE_MAX_BLOCK_DIM_X, Device),
                "Error returned from cuDeviceGetAttribute\n");
    checkResult(cuDeviceGetAttribute(
                    &TmpInt2, CU_DEVICE_ATTRIBUTE_MAX_BLOCK_DIM_Y, Device),
                "Error returned from cuDeviceGetAttribute\n");
    checkResult(cuDeviceGetAttribute(
                    &TmpInt3, CU_DEVICE_ATTRIBUTE_MAX_BLOCK_DIM_Z, Device),
                "Error returned from cuDeviceGetAttribute\n");
    printf("    Maximum Block Dimensions: \t\t%d, %d, %d \n", TmpInt, TmpInt2,
           TmpInt3);
    checkResult(cuDeviceGetAttribute(
                    &TmpInt, CU_DEVICE_ATTRIBUTE_MAX_GRID_DIM_X, Device),
                "Error returned from cuDeviceGetAttribute\n");
    checkResult(cuDeviceGetAttribute(
                    &TmpInt2, CU_DEVICE_ATTRIBUTE_MAX_GRID_DIM_Y, Device),
                "Error returned from cuDeviceGetAttribute\n");
    checkResult(cuDeviceGetAttribute(
                    &TmpInt3, CU_DEVICE_ATTRIBUTE_MAX_GRID_DIM_Z, Device),
                "Error returned from cuDeviceGetAttribute\n");
    printf("    Maximum Grid Dimensions: \t\t%d x %d x %d \n", TmpInt, TmpInt2,
           TmpInt3);
    checkResult(
        cuDeviceGetAttribute(&TmpInt, CU_DEVICE_ATTRIBUTE_MAX_PITCH, Device),
        "Error returned from cuDeviceGetAttribute\n");
    printf("    Maximum Memory Pitch: \t\t%d bytes \n", TmpInt);
    checkResult(cuDeviceGetAttribute(
                    &TmpInt, CU_DEVICE_ATTRIBUTE_TEXTURE_ALIGNMENT, Device),
                "Error returned from cuDeviceGetAttribute\n");
    printf("    Texture Alignment: \t\t\t%d bytes \n", TmpInt);
    checkResult(
        cuDeviceGetAttribute(&TmpInt, CU_DEVICE_ATTRIBUTE_CLOCK_RATE, Device),
        "Error returned from cuDeviceGetAttribute\n");
    printf("    Clock Rate: \t\t\t%d kHz\n", TmpInt);
    checkResult(cuDeviceGetAttribute(
                    &TmpInt, CU_DEVICE_ATTRIBUTE_KERNEL_EXEC_TIMEOUT, Device),
                "Error returned from cuDeviceGetAttribute\n");
    printf("    Execution Timeout: \t\t\t%s \n", BOOL2TEXT(TmpInt));
    checkResult(
        cuDeviceGetAttribute(&TmpInt, CU_DEVICE_ATTRIBUTE_INTEGRATED, Device),
        "Error returned from cuDeviceGetAttribute\n");
    printf("    Integrated Device: \t\t\t%s \n", BOOL2TEXT(TmpInt));
    checkResult(cuDeviceGetAttribute(
                    &TmpInt, CU_DEVICE_ATTRIBUTE_CAN_MAP_HOST_MEMORY, Device),
                "Error returned from cuDeviceGetAttribute\n");
    printf("    Can Map Host Memory: \t\t%s \n", BOOL2TEXT(TmpInt));
    checkResult(
        cuDeviceGetAttribute(&TmpInt, CU_DEVICE_ATTRIBUTE_COMPUTE_MODE, Device),
        "Error returned from cuDeviceGetAttribute\n");
    if (TmpInt == CU_COMPUTEMODE_DEFAULT)
      TmpStr = "DEFAULT";
    else if (TmpInt == CU_COMPUTEMODE_PROHIBITED)
      TmpStr = "PROHIBITED";
    else if (TmpInt == CU_COMPUTEMODE_EXCLUSIVE_PROCESS)
      TmpStr = "EXCLUSIVE PROCESS";
    else
      TmpStr = "unknown";
    printf("    Compute Mode: \t\t\t%s \n", TmpStr.c_str());
    checkResult(cuDeviceGetAttribute(
                    &TmpInt, CU_DEVICE_ATTRIBUTE_CONCURRENT_KERNELS, Device),
                "Error returned from cuDeviceGetAttribute\n");
    printf("    Concurrent Kernels: \t\t%s \n", BOOL2TEXT(TmpInt));
    checkResult(
        cuDeviceGetAttribute(&TmpInt, CU_DEVICE_ATTRIBUTE_ECC_ENABLED, Device),
        "Error returned from cuDeviceGetAttribute\n");
    printf("    ECC Enabled: \t\t\t%s \n", BOOL2TEXT(TmpInt));
    checkResult(cuDeviceGetAttribute(
                    &TmpInt, CU_DEVICE_ATTRIBUTE_MEMORY_CLOCK_RATE, Device),
                "Error returned from cuDeviceGetAttribute\n");
    printf("    Memory Clock Rate: \t\t\t%d kHz\n", TmpInt);
    checkResult(
        cuDeviceGetAttribute(
            &TmpInt, CU_DEVICE_ATTRIBUTE_GLOBAL_MEMORY_BUS_WIDTH, Device),
        "Error returned from cuDeviceGetAttribute\n");
    printf("    Memory Bus Width: \t\t\t%d bits\n", TmpInt);
    checkResult(cuDeviceGetAttribute(&TmpInt, CU_DEVICE_ATTRIBUTE_L2_CACHE_SIZE,
                                     Device),
                "Error returned from cuDeviceGetAttribute\n");
    printf("    L2 Cache Size: \t\t\t%d bytes \n", TmpInt);
    checkResult(cuDeviceGetAttribute(
                    &TmpInt, CU_DEVICE_ATTRIBUTE_MAX_THREADS_PER_MULTIPROCESSOR,
                    Device),
                "Error returned from cuDeviceGetAttribute\n");
    printf("    Max Threads Per SMP: \t\t%d \n", TmpInt);
    checkResult(cuDeviceGetAttribute(
                    &TmpInt, CU_DEVICE_ATTRIBUTE_ASYNC_ENGINE_COUNT, Device),
                "Error returned from cuDeviceGetAttribute\n");
    printf("    Async Engines: \t\t\t%s (%d) \n", BOOL2TEXT(TmpInt), TmpInt);
    checkResult(cuDeviceGetAttribute(
                    &TmpInt, CU_DEVICE_ATTRIBUTE_UNIFIED_ADDRESSING, Device),
                "Error returned from cuDeviceGetAttribute\n");
    printf("    Unified Addressing: \t\t%s \n", BOOL2TEXT(TmpInt));
    checkResult(cuDeviceGetAttribute(
                    &TmpInt, CU_DEVICE_ATTRIBUTE_MANAGED_MEMORY, Device),
                "Error returned from cuDeviceGetAttribute\n");
    printf("    Managed Memory: \t\t\t%s \n", BOOL2TEXT(TmpInt));
    checkResult(
        cuDeviceGetAttribute(
            &TmpInt, CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS, Device),
        "Error returned from cuDeviceGetAttribute\n");
    printf("    Concurrent Managed Memory: \t\t%s \n", BOOL2TEXT(TmpInt));
    checkResult(
        cuDeviceGetAttribute(
            &TmpInt, CU_DEVICE_ATTRIBUTE_COMPUTE_PREEMPTION_SUPPORTED, Device),
        "Error returned from cuDeviceGetAttribute\n");
    printf("    Preemption Supported: \t\t%s \n", BOOL2TEXT(TmpInt));
    checkResult(cuDeviceGetAttribute(
                    &TmpInt, CU_DEVICE_ATTRIBUTE_COOPERATIVE_LAUNCH, Device),
                "Error returned from cuDeviceGetAttribute\n");
    printf("    Cooperative Launch: \t\t%s \n", BOOL2TEXT(TmpInt));
    checkResult(cuDeviceGetAttribute(
                    &TmpInt, CU_DEVICE_ATTRIBUTE_MULTI_GPU_BOARD, Device),
                "Error returned from cuDeviceGetAttribute\n");
    printf("    Multi-Device Boars: \t\t%s \n", BOOL2TEXT(TmpInt));
    checkResult(
        cuDeviceGetAttribute(
            &TmpInt, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR, Device),
        "Error returned from cuDeviceGetAttribute\n");
    checkResult(
        cuDeviceGetAttribute(
            &TmpInt2, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR, Device),
        "Error returned from cuDeviceGetAttribute\n");
    printf("    Compute Capabilities: \t\t%d%d \n", TmpInt, TmpInt2);
  }

  int waitEvent(const int DeviceId, __tgt_async_info *AsyncInfo,
                void *EventPtr) const {
    CUstream Stream = getStream(DeviceId, AsyncInfo);
    CUevent Event = reinterpret_cast<CUevent>(EventPtr);

    // We don't use CU_EVENT_WAIT_DEFAULT here as it is only available from
    // specific CUDA version, and defined as 0x0. In previous version, per CUDA
    // API document, that argument has to be 0x0.
    CUresult Err = cuStreamWaitEvent(Stream, Event, 0);
    if (Err != CUDA_SUCCESS) {
      DP("Error when waiting event. stream = " DPxMOD ", event = " DPxMOD "\n",
         DPxPTR(Stream), DPxPTR(Event));
      CUDA_ERR_STRING(Err);
      return OFFLOAD_FAIL;
    }

    return OFFLOAD_SUCCESS;
  }
};

DeviceRTLTy DeviceRTL;
} // namespace

// Exposed library API function
#ifdef __cplusplus
extern "C" {
#endif

int32_t __tgt_rtl_is_valid_binary(__tgt_device_image *image) {
  return elf_check_machine(image, /* EM_CUDA */ 190);
}

int32_t __tgt_rtl_number_of_devices() { return DeviceRTL.getNumOfDevices(); }

int64_t __tgt_rtl_init_requires(int64_t RequiresFlags) {
  DP("Init requires flags to %" PRId64 "\n", RequiresFlags);
  DeviceRTL.setRequiresFlag(RequiresFlags);
  return RequiresFlags;
}

int32_t __tgt_rtl_is_data_exchangable(int32_t src_dev_id, int dst_dev_id) {
  if (DeviceRTL.isValidDeviceId(src_dev_id) &&
      DeviceRTL.isValidDeviceId(dst_dev_id))
    return 1;

  return 0;
}

int32_t __tgt_rtl_init_device(int32_t device_id) {
  assert(DeviceRTL.isValidDeviceId(device_id) && "device_id is invalid");

  return DeviceRTL.initDevice(device_id);
}

__tgt_target_table *__tgt_rtl_load_binary(int32_t device_id,
                                          __tgt_device_image *image) {
  assert(DeviceRTL.isValidDeviceId(device_id) && "device_id is invalid");

  return DeviceRTL.loadBinary(device_id, image);
}

void *__tgt_rtl_data_alloc(int32_t device_id, int64_t size, void *,
                           int32_t kind) {
  assert(DeviceRTL.isValidDeviceId(device_id) && "device_id is invalid");

  return DeviceRTL.dataAlloc(device_id, size, (TargetAllocTy)kind);
}

int32_t __tgt_rtl_data_submit(int32_t device_id, void *tgt_ptr, void *hst_ptr,
                              int64_t size) {
  assert(DeviceRTL.isValidDeviceId(device_id) && "device_id is invalid");

  __tgt_async_info AsyncInfo;
  const int32_t rc = __tgt_rtl_data_submit_async(device_id, tgt_ptr, hst_ptr,
                                                 size, &AsyncInfo);
  if (rc != OFFLOAD_SUCCESS)
    return OFFLOAD_FAIL;

  return __tgt_rtl_synchronize(device_id, &AsyncInfo);
}

int32_t __tgt_rtl_data_submit_async(int32_t device_id, void *tgt_ptr,
                                    void *hst_ptr, int64_t size,
                                    __tgt_async_info *async_info_ptr) {
  assert(DeviceRTL.isValidDeviceId(device_id) && "device_id is invalid");
  assert(async_info_ptr && "async_info_ptr is nullptr");

  return DeviceRTL.dataSubmit(device_id, tgt_ptr, hst_ptr, size,
                              async_info_ptr);
}

int32_t __tgt_rtl_data_retrieve(int32_t device_id, void *hst_ptr, void *tgt_ptr,
                                int64_t size) {
  assert(DeviceRTL.isValidDeviceId(device_id) && "device_id is invalid");

  __tgt_async_info AsyncInfo;
  const int32_t rc = __tgt_rtl_data_retrieve_async(device_id, hst_ptr, tgt_ptr,
                                                   size, &AsyncInfo);
  if (rc != OFFLOAD_SUCCESS)
    return OFFLOAD_FAIL;

  return __tgt_rtl_synchronize(device_id, &AsyncInfo);
}

int32_t __tgt_rtl_data_retrieve_async(int32_t device_id, void *hst_ptr,
                                      void *tgt_ptr, int64_t size,
                                      __tgt_async_info *async_info_ptr) {
  assert(DeviceRTL.isValidDeviceId(device_id) && "device_id is invalid");
  assert(async_info_ptr && "async_info_ptr is nullptr");

  return DeviceRTL.dataRetrieve(device_id, hst_ptr, tgt_ptr, size,
                                async_info_ptr);
}

int32_t __tgt_rtl_data_exchange_async(int32_t src_dev_id, void *src_ptr,
                                      int dst_dev_id, void *dst_ptr,
                                      int64_t size,
                                      __tgt_async_info *AsyncInfo) {
  assert(DeviceRTL.isValidDeviceId(src_dev_id) && "src_dev_id is invalid");
  assert(DeviceRTL.isValidDeviceId(dst_dev_id) && "dst_dev_id is invalid");
  assert(AsyncInfo && "AsyncInfo is nullptr");

  return DeviceRTL.dataExchange(src_dev_id, src_ptr, dst_dev_id, dst_ptr, size,
                                AsyncInfo);
}

int32_t __tgt_rtl_data_exchange(int32_t src_dev_id, void *src_ptr,
                                int32_t dst_dev_id, void *dst_ptr,
                                int64_t size) {
  assert(DeviceRTL.isValidDeviceId(src_dev_id) && "src_dev_id is invalid");
  assert(DeviceRTL.isValidDeviceId(dst_dev_id) && "dst_dev_id is invalid");

  __tgt_async_info AsyncInfo;
  const int32_t rc = __tgt_rtl_data_exchange_async(
      src_dev_id, src_ptr, dst_dev_id, dst_ptr, size, &AsyncInfo);
  if (rc != OFFLOAD_SUCCESS)
    return OFFLOAD_FAIL;

  return __tgt_rtl_synchronize(src_dev_id, &AsyncInfo);
}

int32_t __tgt_rtl_data_delete(int32_t device_id, void *tgt_ptr) {
  assert(DeviceRTL.isValidDeviceId(device_id) && "device_id is invalid");

  return DeviceRTL.dataDelete(device_id, tgt_ptr);
}

int32_t __tgt_rtl_run_target_team_region(int32_t device_id, void *tgt_entry_ptr,
                                         void **tgt_args,
                                         ptrdiff_t *tgt_offsets,
                                         int32_t arg_num, int32_t team_num,
                                         int32_t thread_limit,
                                         uint64_t loop_tripcount) {
  assert(DeviceRTL.isValidDeviceId(device_id) && "device_id is invalid");

  __tgt_async_info AsyncInfo;
  const int32_t rc = __tgt_rtl_run_target_team_region_async(
      device_id, tgt_entry_ptr, tgt_args, tgt_offsets, arg_num, team_num,
      thread_limit, loop_tripcount, &AsyncInfo);
  if (rc != OFFLOAD_SUCCESS)
    return OFFLOAD_FAIL;

  return __tgt_rtl_synchronize(device_id, &AsyncInfo);
}

int32_t __tgt_rtl_run_target_team_region_async(
    int32_t device_id, void *tgt_entry_ptr, void **tgt_args,
    ptrdiff_t *tgt_offsets, int32_t arg_num, int32_t team_num,
    int32_t thread_limit, uint64_t loop_tripcount,
    __tgt_async_info *async_info_ptr) {
  assert(DeviceRTL.isValidDeviceId(device_id) && "device_id is invalid");

  return DeviceRTL.runTargetTeamRegion(
      device_id, tgt_entry_ptr, tgt_args, tgt_offsets, arg_num, team_num,
      thread_limit, loop_tripcount, async_info_ptr);
}

int32_t __tgt_rtl_run_target_region(int32_t device_id, void *tgt_entry_ptr,
                                    void **tgt_args, ptrdiff_t *tgt_offsets,
                                    int32_t arg_num) {
  assert(DeviceRTL.isValidDeviceId(device_id) && "device_id is invalid");

  __tgt_async_info AsyncInfo;
  const int32_t rc = __tgt_rtl_run_target_region_async(
      device_id, tgt_entry_ptr, tgt_args, tgt_offsets, arg_num, &AsyncInfo);
  if (rc != OFFLOAD_SUCCESS)
    return OFFLOAD_FAIL;

  return __tgt_rtl_synchronize(device_id, &AsyncInfo);
}

int32_t __tgt_rtl_run_target_region_async(int32_t device_id,
                                          void *tgt_entry_ptr, void **tgt_args,
                                          ptrdiff_t *tgt_offsets,
                                          int32_t arg_num,
                                          __tgt_async_info *async_info_ptr) {
  assert(DeviceRTL.isValidDeviceId(device_id) && "device_id is invalid");

  return __tgt_rtl_run_target_team_region_async(
      device_id, tgt_entry_ptr, tgt_args, tgt_offsets, arg_num,
      /* team num*/ 1, /* thread_limit */ 1, /* loop_tripcount */ 0,
      async_info_ptr);
}

int32_t __tgt_rtl_synchronize(int32_t device_id,
                              __tgt_async_info *async_info_ptr) {
  assert(DeviceRTL.isValidDeviceId(device_id) && "device_id is invalid");
  assert(async_info_ptr && "async_info_ptr is nullptr");
  assert(async_info_ptr->Queue && "async_info_ptr->Queue is nullptr");

  return DeviceRTL.synchronize(device_id, async_info_ptr);
}

void __tgt_rtl_set_info_flag(uint32_t NewInfoLevel) {
  std::atomic<uint32_t> &InfoLevel = getInfoLevelInternal();
  InfoLevel.store(NewInfoLevel);
}

void __tgt_rtl_print_device_info(int32_t device_id) {
  assert(DeviceRTL.isValidDeviceId(device_id) && "device_id is invalid");
  DeviceRTL.printDeviceInfo(device_id);
}

int32_t __tgt_rtl_create_event(int32_t device_id, void **event) {
  assert(event && "event is nullptr");
  return createEvent(event);
}

int32_t __tgt_rtl_record_event(int32_t device_id, void *event_ptr,
                               __tgt_async_info *async_info_ptr) {
  assert(async_info_ptr && "async_info_ptr is nullptr");
  assert(async_info_ptr->Queue && "async_info_ptr->Queue is nullptr");
  assert(event_ptr && "event_ptr is nullptr");

  return recordEvent(event_ptr, async_info_ptr);
}

int32_t __tgt_rtl_wait_event(int32_t device_id, void *event_ptr,
                             __tgt_async_info *async_info_ptr) {
  assert(DeviceRTL.isValidDeviceId(device_id) && "device_id is invalid");
  assert(async_info_ptr && "async_info_ptr is nullptr");
  assert(event_ptr && "event is nullptr");

  return DeviceRTL.waitEvent(device_id, async_info_ptr, event_ptr);
}

int32_t __tgt_rtl_sync_event(int32_t device_id, void *event_ptr) {
  assert(event_ptr && "event is nullptr");

  return syncEvent(event_ptr);
}

int32_t __tgt_rtl_destroy_event(int32_t device_id, void *event_ptr) {
  assert(event_ptr && "event is nullptr");

  return destroyEvent(event_ptr);
}

#ifdef __cplusplus
}
#endif