path: root/offload/plugins-nextgen/level_zero/src/L0Memory.cpp

//===--- Level Zero Target RTL Implementation -----------------------------===//
//
// 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
//
//===----------------------------------------------------------------------===//
//
// Memory related support for SPIR-V/Xe machine.
//
//===----------------------------------------------------------------------===//

#include "L0Memory.h"
#include "L0Device.h"
#include "L0Plugin.h"

namespace llvm::omp::target::plugin {

static const char *allocKindToStr(int32_t Kind) {
  switch (Kind) {
  case TARGET_ALLOC_DEVICE:
    return "DEVICE";
  case TARGET_ALLOC_HOST:
    return "HOST";
  case TARGET_ALLOC_SHARED:
    return "SHARED";
  default:
    return "DEFAULT";
  }
}

void *MemAllocatorTy::MemPoolTy::BlockTy::alloc() {
  if (isFull())
    return nullptr;

  if (FreeSlot != MaxSlots) {
    const uint32_t Slot = FreeSlot;
    FreeSlot = MaxSlots;
    UsedSlots[Slot] = true;
    NumUsedSlots++;
    return reinterpret_cast<void *>(Base + Slot * ChunkSize);
  }
  for (uint32_t I = 0; I < NumSlots; I++) {
    if (UsedSlots[I])
      continue;
    UsedSlots[I] = true;
    NumUsedSlots++;
    return reinterpret_cast<void *>(Base + I * ChunkSize);
  }
  // Should not reach here.
  assert(false && "Inconsistent memory pool state");
  return nullptr;
}

/// Deallocate the given memory.
void MemAllocatorTy::MemPoolTy::BlockTy::dealloc(void *Mem) {
  if (!contains(Mem))
    assert(0 && "Inconsistent memory pool state");
  const uint32_t Slot = (reinterpret_cast<uintptr_t>(Mem) - Base) / ChunkSize;
  UsedSlots[Slot] = false;
  NumUsedSlots--;
  FreeSlot = Slot;
}

Error MemAllocatorTy::MemPoolTy::init(int32_t Kind, MemAllocatorTy *AllocatorIn,
                                      const L0OptionsTy &Option) {
  AllocKind = Kind;
  Allocator = AllocatorIn;

  // Read user-defined options.
  const auto &UserOptions = Option.MemPoolConfig[AllocKind];
  const size_t UserAllocMax = UserOptions.AllocMax;
  const size_t UserCapacity = UserOptions.Capacity;
  const size_t UserPoolSize = UserOptions.PoolSize;

  BlockCapacity = UserCapacity;
  PoolSizeMax = UserPoolSize << 20; // Covert MB to B.
  PoolSize = 0;

  auto Context = Allocator->L0Context->getZeContext();
  const auto Device = Allocator->Device;

  // Check page size used for this allocation kind to decide minimum.
  // allocation size when allocating from L0.
  auto MemOrErr = Allocator->allocFromL0(8, 0, AllocKind);
  if (!MemOrErr)
    return MemOrErr.takeError();
  void *Mem = *MemOrErr;
  ze_memory_allocation_properties_t AP{
      ZE_STRUCTURE_TYPE_MEMORY_ALLOCATION_PROPERTIES, nullptr,
      ZE_MEMORY_TYPE_UNKNOWN, 0, 0};
  CALL_ZE_RET_ERROR(zeMemGetAllocProperties, Context, Mem, &AP, nullptr);
  AllocUnit = (std::max)(AP.pageSize, AllocUnit);
  if (auto Err = Allocator->deallocFromL0(Mem))
    return Err;

  bool IsDiscrete = false;
  if (Device) {
    ze_device_properties_t Properties{};
    Properties.deviceId = 0;
    Properties.stype = ZE_STRUCTURE_TYPE_DEVICE_PROPERTIES;
    Properties.pNext = nullptr;
    CALL_ZE_RET_ERROR(zeDeviceGetProperties, Device->getZeDevice(),
                      &Properties);
    IsDiscrete = Device->isDiscreteDevice();

    if (AllocKind == TARGET_ALLOC_SHARED && IsDiscrete) {
      // Use page size as minimum chunk size for USM shared on discrete.
      // device.
      // FIXME: pageSize is not returned correctly (=0) on some new devices,
      //        so use fallback value for now.
      AllocMin = (std::max)(AP.pageSize, AllocUnit);
      AllocUnit = AllocMin * BlockCapacity;
    }
  }

  // Convert MB to B and round up to power of 2.
  AllocMax = AllocMin << getBucketId(UserAllocMax * (1 << 20));
  if (AllocMin >= AllocMax) {
    AllocMax = 2 * AllocMin;
    ODBG(OLDT_Alloc) << "Warning: Adjusting pool's AllocMax to " << AllocMax
                     << " for " << allocKindToStr(AllocKind)
                     << " due to device requirements.";
  }
  assert(AllocMin < AllocMax &&
         "Invalid parameters while initializing memory pool");
  const auto MinSize = getBucketId(AllocMin);
  const auto MaxSize = getBucketId(AllocMax);
  Buckets.resize(MaxSize - MinSize + 1);
  BucketStats.resize(Buckets.size(), {0, 0});

  // Set bucket parameters
  for (size_t I = 0; I < Buckets.size(); I++) {
    const size_t ChunkSize = AllocMin << I;
    size_t BlockSize = ChunkSize * BlockCapacity;
    // On discrete device, the cost of native L0 invocation doubles when the
    // the requested size doubles after certain threshold, so allocating
    // larger block does not pay off at all. It is better to keep a single
    // chunk in a single block in such cases.
    if (BlockSize <= AllocUnit) {
      BlockSize = AllocUnit; // Allocation unit is already large enough.
    } else if (IsDiscrete) {
      // Do not preallocate if it does not pay off.
      if (ChunkSize >= L0UsmPreAllocThreshold ||
          (AllocKind == TARGET_ALLOC_HOST &&
           ChunkSize >= L0HostUsmPreAllocThreshold))
        BlockSize = ChunkSize;
    }
    BucketParams.emplace_back(ChunkSize, BlockSize);
  }

  ODBG(OLDT_Alloc) << "Initialized " << allocKindToStr(AllocKind)
                   << " pool for device " << Device
                   << ": AllocUnit = " << AllocUnit
                   << ", AllocMax = " << AllocMax
                   << ", Capacity = " << BlockCapacity
                   << ", PoolSizeMax = " << PoolSizeMax;
  return Plugin::success();
}

// Used for reduction pool.
Error MemAllocatorTy::MemPoolTy::init(MemAllocatorTy *AllocatorIn,
                                      const L0OptionsTy &Option) {
  AllocKind = TARGET_ALLOC_DEVICE;
  Allocator = AllocatorIn;
  AllocMin = AllocUnit = 1024 << 6; // 64KB.
  AllocMax = Option.ReductionPoolInfo[0] << 20;
  BlockCapacity = Option.ReductionPoolInfo[1];
  PoolSize = 0;
  PoolSizeMax = (size_t)Option.ReductionPoolInfo[2] << 20;

  const auto MinSize = getBucketId(AllocMin);
  const auto MaxSize = getBucketId(AllocMax);
  Buckets.resize(MaxSize - MinSize + 1);
  BucketStats.resize(Buckets.size(), {0, 0});
  for (size_t I = 0; I < Buckets.size(); I++) {
    const size_t ChunkSize = AllocMin << I;
    BucketParams.emplace_back(ChunkSize, ChunkSize * BlockCapacity);
  }

  ODBG(OLDT_Alloc) << "Initialized reduction scratch pool for device "
                   << Allocator->Device << ": AllocMin = " << AllocMin
                   << ", AllocMax = " << AllocMax
                   << ", PoolSizeMax = " << PoolSizeMax;
  return Plugin::success();
}

// Used for small memory pool with fixed parameters.
Error MemAllocatorTy::MemPoolTy::init(MemAllocatorTy *AllocatorIn) {
  AllocKind = TARGET_ALLOC_DEVICE;
  Allocator = AllocatorIn;
  AllocMax = AllocMin;
  BlockCapacity = AllocUnit / AllocMax;
  PoolSize = 0;
  PoolSizeMax = (1 << 20); // This should be sufficiently large.
  Buckets.resize(1);
  BucketStats.resize(1, {0, 0});
  BucketParams.emplace_back(AllocMax, AllocUnit);
  ZeroInit = true;
  ODBG(OLDT_Alloc) << "Initialized zero-initialized reduction counter pool for "
                   << "device " << Allocator->Device
                   << ": AllocMin = " << AllocMin << ", AllocMax = " << AllocMax
                   << ", PoolSizeMax = " << PoolSizeMax;
  return Plugin::success();
}

void MemAllocatorTy::MemPoolTy::printUsage() {
  ODBG_OS([&](llvm::raw_ostream &Os) {
    auto PrintNum = [&](uint64_t Num) {
      if (Num > 1e9)
        Os << llvm::format("%.2e", float(Num));
      else
        Os << llvm::format("%11" PRIu64, Num);
    };

    bool HasPoolAlloc = false;
    for (auto &Stat : BucketStats) {
      if (Stat.first > 0 || Stat.second > 0) {
        HasPoolAlloc = true;
        break;
      }
    }

    Os << "MemPool usage for " << allocKindToStr(AllocKind) << ", device "
       << Allocator->Device << "\n";

    if (HasPoolAlloc) {
      Os << "-- AllocMax=" << (AllocMax >> 20)
         << "(MB), Capacity=" << BlockCapacity
         << ", PoolSizeMax=" << (PoolSizeMax >> 20) << "(MB)\n";
      Os << "-- "
         << llvm::format("%18s:%11s%11s%11s\n", "", "NewAlloc", "Reuse",
                         "Hit(%)");
      for (size_t I = 0; I < Buckets.size(); I++) {
        const auto &Stat = BucketStats[I];
        if (Stat.first > 0 || Stat.second > 0) {
          Os << "-- Bucket[" << llvm::format("%10zu", BucketParams[I].first)
             << "]:";
          PrintNum(Stat.first);
          PrintNum(Stat.second);
          Os << llvm::format("%11.2f\n", float(Stat.second) /
                                             float(Stat.first + Stat.second) *
                                             100);
        }
      }
    } else {
      Os << "-- Not used\n";
    }
  });
}

/// Release resources used in the pool.
Error MemAllocatorTy::MemPoolTy::deinit() {
  printUsage();
  for (auto &Bucket : Buckets) {
    for (auto *Block : Bucket) {
      ODBG_IF([&]() { Allocator->log(0, Block->Size, AllocKind); });
      auto Err =
          Allocator->deallocFromL0(reinterpret_cast<void *>(Block->Base));
      delete Block;
      if (Err)
        return Err;
    }
  }
  return Plugin::success();
}

/// Allocate the requested size of memory from this pool.
/// AllocSize is the chunk size internally used for the returned memory.
Expected<void *> MemAllocatorTy::MemPoolTy::alloc(size_t Size,
                                                  size_t &AllocSize) {
  if (Size == 0 || Size > AllocMax)
    return nullptr;

  const uint32_t BucketId = getBucketId(Size);
  auto &Blocks = Buckets[BucketId];
  void *Mem = nullptr;

  for (auto *Block : Blocks) {
    if (Block->isFull())
      continue;
    Mem = Block->alloc();
    assert(Mem && "Inconsistent state while allocating memory from pool");
    PtrToBlock.try_emplace(Mem, Block);
    break;
  }

  if (Mem == nullptr) {
    const bool IsSmallAllocatable =
        (Size <= SmallAllocMax && SmallPoolSize <= SmallPoolSizeMax);
    const bool IsFull = (PoolSize > PoolSizeMax);
    if (IsFull && !IsSmallAllocatable)
      return nullptr;
    // Bucket is empty or all blocks in the bucket are full.
    const auto ChunkSize = BucketParams[BucketId].first;
    const auto BlockSize = BucketParams[BucketId].second;
    auto BaseOrErr = Allocator->allocFromL0AndLog(BlockSize, 0, AllocKind);
    if (!BaseOrErr)
      return BaseOrErr.takeError();

    void *Base = *BaseOrErr;
    if (ZeroInit) {
      auto Err = Allocator->enqueueMemSet(Base, 0, BlockSize);
      if (Err)
        return Err;
    }

    BlockTy *Block = new BlockTy(Base, BlockSize, ChunkSize);
    Blocks.push_back(Block);
    Mem = Block->alloc();
    PtrToBlock.try_emplace(Mem, Block);
    if (IsFull)
      SmallPoolSize += BlockSize;
    else
      PoolSize += BlockSize;
    ODBG(OLDT_Alloc) << "New block allocation for " << allocKindToStr(AllocKind)
                     << " pool: base = " << Base << ", size = " << BlockSize
                     << ", pool size = " << PoolSize;
    BucketStats[BucketId].first++;
  } else {
    BucketStats[BucketId].second++;
  }

  AllocSize = (AllocMin << BucketId);

  return Mem;
}

/// Deallocate the specified memory and returns block size deallocated.
size_t MemAllocatorTy::MemPoolTy::dealloc(void *Ptr) {
  if (PtrToBlock.count(Ptr) == 0)
    return 0;
  PtrToBlock[Ptr]->dealloc(Ptr);
  const size_t Deallocated = PtrToBlock[Ptr]->ChunkSize;
  PtrToBlock.erase(Ptr);
  return Deallocated;
}

void MemAllocatorTy::MemAllocInfoMapTy::add(void *Ptr, void *Base,
                                            size_t ReqSize, size_t AllocSize,
                                            int32_t Kind, bool InPool,
                                            bool ImplicitArg) {
  const auto Inserted = Map.emplace(
      Ptr, MemAllocInfoTy{Base, ReqSize, AllocSize, Kind, InPool, ImplicitArg});
  // Check if we keep valid disjoint memory ranges.
  [[maybe_unused]] bool Valid = Inserted.second;
  if (Valid) {
    if (Inserted.first != Map.begin()) {
      const auto I = std::prev(Inserted.first, 1);
      Valid =
          Valid && (uintptr_t)I->first + I->second.ReqSize <= (uintptr_t)Ptr;
    }
    if (Valid) {
      const auto I = std::next(Inserted.first, 1);
      if (I != Map.end())
        Valid = Valid && (uintptr_t)Ptr + ReqSize <= (uintptr_t)I->first;
    }
  }
  assert(Valid && "Invalid overlapping memory allocation");
  assert(Kind >= 0 && Kind < MaxMemKind && "Invalid target allocation kind");
  if (ImplicitArg)
    NumImplicitArgs[Kind]++;
}

/// Remove allocation information for the given memory location.
bool MemAllocatorTy::MemAllocInfoMapTy::remove(void *Ptr,
                                               MemAllocInfoTy *Removed) {
  const auto AllocInfo = Map.find(Ptr);
  if (AllocInfo == Map.end())
    return false;
  if (AllocInfo->second.ImplicitArg)
    NumImplicitArgs[AllocInfo->second.Kind]--;
  if (Removed)
    *Removed = AllocInfo->second;
  Map.erase(AllocInfo);
  return true;
}

Error MemAllocatorTy::initDevicePools(L0DeviceTy &L0Device,
                                      const L0OptionsTy &Options) {
  SupportsLargeMem = L0Device.supportsLargeMem();
  IsHostMem = false;
  Device = &L0Device;
  L0Context = &L0Device.getL0Context();
  for (auto Kind : {TARGET_ALLOC_DEVICE, TARGET_ALLOC_SHARED}) {
    if (Options.MemPoolConfig[Kind].Use) {
      std::lock_guard<std::mutex> Lock(Mtx);
      Pools[Kind] = std::make_unique<MemPoolTy>();
      if (auto Err = Pools[Kind]->init(Kind, this, Options))
        return Err;
    }
  }
  ReductionPool = std::make_unique<MemPoolTy>();
  if (auto Err = ReductionPool->init(this, Options))
    return Err;
  CounterPool = std::make_unique<MemPoolTy>();
  if (auto Err = CounterPool->init(this))
    return Err;
  updateMaxAllocSize(L0Device);
  return Plugin::success();
}

Error MemAllocatorTy::initHostPool(L0ContextTy &Driver,
                                   const L0OptionsTy &Option) {
  SupportsLargeMem = Driver.supportsLargeMem();
  IsHostMem = true;
  L0Context = &Driver;
  if (Option.MemPoolConfig[TARGET_ALLOC_HOST].Use) {
    std::lock_guard<std::mutex> Lock(Mtx);
    Pools[TARGET_ALLOC_HOST] = std::make_unique<MemPoolTy>();
    if (auto Err =
            Pools[TARGET_ALLOC_HOST]->init(TARGET_ALLOC_HOST, this, Option))
      return Err;
  }
  return Plugin::success();
}

void MemAllocatorTy::updateMaxAllocSize(L0DeviceTy &L0Device) {
  // Update the maximum allocation size for this Allocator.
  auto maxMemAllocSize = L0Device.getMaxMemAllocSize();

  if (IsHostMem) {
    // MaxAllocSize should be the minimum of all devices from the driver.
    if (MaxAllocSize > maxMemAllocSize) {
      MaxAllocSize = maxMemAllocSize;
      ODBG(OLDT_Alloc) << "Updated MaxAllocSize for driver " << L0Context
                       << " to " << MaxAllocSize;
    }
    return;
  }

  MaxAllocSize = maxMemAllocSize;
  ODBG(OLDT_Alloc) << "Updated MaxAllocSize for device " << Device << " to "
                   << MaxAllocSize;
}

/// Release resources and report statistics if requested.
Error MemAllocatorTy::deinit() {
  if (!L0Context)
    return Plugin::success();

  std::lock_guard<std::mutex> Lock(Mtx);
  if (!L0Context)
    return Plugin::success();
  // Release RTL-owned memory.
  for (auto *M : MemOwned) {
    auto Err = deallocLocked(M);
    if (Err)
      return Err;
  }
  for (auto &Pool : Pools) {
    if (Pool) {
      if (auto Err = Pool->deinit())
        return Err;
      Pool.reset(nullptr);
    }
  }
  if (ReductionPool) {
    if (auto Err = ReductionPool->deinit())
      return Err;
    ReductionPool.reset(nullptr);
  }
  if (CounterPool) {
    if (auto Err = CounterPool->deinit())
      return Err;
    CounterPool.reset(nullptr);
  }
  // Report memory usage if requested.
  ODBG_OS([&](llvm::raw_ostream &Os) {
    for (size_t Kind = 0; Kind < MaxMemKind; Kind++) {
      auto &Stat = Stats[Kind];
      Os << "Memory usage for " << allocKindToStr(Kind) << ", device " << Device
         << "\n";
      if (Stat.NumAllocs[0] == 0 && Stat.NumAllocs[1] == 0) {
        Os << "-- Not used\n";
        continue;
      }
      Os << "-- Allocator: " << llvm::format("%12s", "Native") << ", "
         << llvm::format("%12s", "Pool") << "\n";
      Os << "-- Requested: " << llvm::format("%12zu", Stat.Requested[0]) << ", "
         << llvm::format("%12zu", Stat.Requested[1]) << "\n";
      Os << "-- Allocated: " << llvm::format("%12zu", Stat.Allocated[0]) << ", "
         << llvm::format("%12zu", Stat.Allocated[1]) << "\n";
      Os << "-- Freed    : " << llvm::format("%12zu", Stat.Freed[0]) << ", "
         << llvm::format("%12zu", Stat.Freed[1]) << "\n";
      Os << "-- InUse    : " << llvm::format("%12zu", Stat.InUse[0]) << ", "
         << llvm::format("%12zu", Stat.InUse[1]) << "\n";
      Os << "-- PeakUse  : " << llvm::format("%12zu", Stat.PeakUse[0]) << ", "
         << llvm::format("%12zu", Stat.PeakUse[1]) << "\n";
      Os << "-- NumAllocs: " << llvm::format("%12zu", Stat.NumAllocs[0]) << ", "
         << llvm::format("%12zu", Stat.NumAllocs[1]) << "\n";
    }
  });

  // Mark as deinitialized.
  L0Context = nullptr;
  return Plugin::success();
}

/// Allocate memory with the specified information.
Expected<void *> MemAllocatorTy::allocFromPool(size_t Size, size_t Align,
                                               int32_t Kind, intptr_t Offset,
                                               bool UserAlloc, bool DevMalloc,
                                               uint32_t MemAdvice,
                                               AllocOptionTy AllocOpt) {
  assert((Kind == TARGET_ALLOC_DEVICE || Kind == TARGET_ALLOC_HOST ||
          Kind == TARGET_ALLOC_SHARED) &&
         "Unknown memory kind while allocating target memory");

  std::lock_guard<std::mutex> Lock(Mtx);

  // We do not expect meaningful Align parameter when Offset > 0, so the
  // following code does not handle such case.

  size_t AllocSize = Size + Offset;
  void *Mem = nullptr;
  void *AllocBase = nullptr;
  const bool UseScratchPool =
      (AllocOpt == AllocOptionTy::ALLOC_OPT_REDUCTION_SCRATCH);
  const bool UseZeroInitPool =
      (AllocOpt == AllocOptionTy::ALLOC_OPT_REDUCTION_COUNTER);
  const bool UseDedicatedPool = UseScratchPool || UseZeroInitPool;

  if ((Pools[Kind] &&
       MemAdvice == std::numeric_limits<decltype(MemAdvice)>::max()) ||
      UseDedicatedPool) {
    // Pool is enabled for the allocation kind, and we do not use any memory
    // advice. We should avoid using pool if there is any meaningful memory
    // advice not to affect sibling allocation in the same block.
    if (Align > 0)
      AllocSize += (Align - 1);
    size_t PoolAllocSize = 0;
    MemPoolTy *Pool = nullptr;

    if (UseScratchPool)
      Pool = ReductionPool.get();
    else if (UseZeroInitPool)
      Pool = CounterPool.get();
    else
      Pool = Pools[Kind].get();

    auto PtrOrErr = Pool->alloc(AllocSize, PoolAllocSize);
    if (!PtrOrErr)
      return PtrOrErr.takeError();
    AllocBase = *PtrOrErr;
    if (AllocBase) {
      uintptr_t Base = (uintptr_t)AllocBase;
      if (Align > 0)
        Base = (Base + Align) & ~(Align - 1);
      Mem = (void *)(Base + Offset);
      AllocInfo.add(Mem, AllocBase, Size, PoolAllocSize, Kind, true, UserAlloc);
      log(Size, PoolAllocSize, Kind, true /* Pool */);
      if (DevMalloc)
        MemOwned.push_back(AllocBase);
      if (UseDedicatedPool) {
        ODBG(OLDT_Alloc) << "Allocated " << Size << " bytes from "
                         << (UseScratchPool ? "scratch" : "zero-initialized")
                         << " pool";
      }
      return Mem;
    }
  }

  auto AllocBaseOrErr =
      allocFromL0AndLog(AllocSize, Align, Kind, /*ActiveSize=*/Size);
  if (!AllocBaseOrErr)
    return AllocBaseOrErr.takeError();
  AllocBase = *AllocBaseOrErr;
  if (AllocBase) {
    Mem = (void *)((uintptr_t)AllocBase + Offset);
    AllocInfo.add(Mem, AllocBase, Size, AllocSize, Kind, false, UserAlloc);
    if (DevMalloc)
      MemOwned.push_back(AllocBase);
    if (UseDedicatedPool) {
      // We do not want this happen in general.
      ODBG(OLDT_Alloc) << "Allocated " << Size << " bytes from L0 for "
                       << (UseScratchPool ? "scratch" : "zero-initialized")
                       << " pool";
    }
  }
  return Mem;
}

/// Deallocate memory.
Error MemAllocatorTy::deallocLocked(void *Ptr) {
  MemAllocInfoTy Info;
  if (!AllocInfo.remove(Ptr, &Info)) {
    return Plugin::error(ErrorCode::BACKEND_FAILURE,
                         "Cannot find memory allocation information for " DPxMOD
                         "\n",
                         DPxPTR(Ptr));
  }
  if (Info.InPool) {
    size_t DeallocSize = 0;
    if (Pools[Info.Kind])
      DeallocSize = Pools[Info.Kind]->dealloc(Info.Base);
    if (DeallocSize == 0) {
      // Try reduction scratch pool.
      DeallocSize = ReductionPool->dealloc(Info.Base);
      // Try reduction counter pool.
      if (DeallocSize == 0)
        DeallocSize = CounterPool->dealloc(Info.Base);
      if (DeallocSize == 0) {
        return Plugin::error(ErrorCode::BACKEND_FAILURE,
                             "Cannot return memory " DPxMOD " to pool\n",
                             DPxPTR(Ptr));
      }
    }
    log(0, DeallocSize, Info.Kind, true /* Pool */);
    return Plugin::success();
  }
  if (!Info.Base) {
    ODBG(OLDT_Alloc) << "Error: Cannot find base address of " << Ptr;
    return Plugin::error(ErrorCode::INVALID_ARGUMENT,
                         "Cannot find base address of " DPxMOD "\n",
                         DPxPTR(Ptr));
  }
  log(/*NoReqSize*/ 0, Info.AllocSize, Info.Kind);

  if (auto Err = deallocFromL0(Info.Base))
    return Err;
  ODBG(OLDT_Alloc) << "Deleted device memory " << Ptr << " (Base: " << Info.Base
                   << ", Size: " << Info.AllocSize << ")";

  return Plugin::success();
}

Error MemAllocatorTy::enqueueMemSet(void *Dst, int8_t Value, size_t Size) {
  return Device->enqueueMemFill(Dst, &Value, sizeof(int8_t), Size);
}

Error MemAllocatorTy::enqueueMemCopy(void *Dst, const void *Src, size_t Size) {
  return Device->enqueueMemCopy(Dst, Src, Size);
}

Expected<void *> MemAllocatorTy::allocFromL0(size_t Size, size_t Align,
                                             int32_t Kind) {
  void *Mem = nullptr;
  ze_device_mem_alloc_desc_t DeviceDesc{ZE_STRUCTURE_TYPE_DEVICE_MEM_ALLOC_DESC,
                                        nullptr, 0, 0};
  ze_host_mem_alloc_desc_t HostDesc{ZE_STRUCTURE_TYPE_HOST_MEM_ALLOC_DESC,
                                    nullptr, 0};

  // Use relaxed allocation limit if driver supports.
  ze_relaxed_allocation_limits_exp_desc_t RelaxedDesc{
      ZE_STRUCTURE_TYPE_RELAXED_ALLOCATION_LIMITS_EXP_DESC, nullptr,
      ZE_RELAXED_ALLOCATION_LIMITS_EXP_FLAG_MAX_SIZE};
  if (Size > MaxAllocSize && SupportsLargeMem) {
    DeviceDesc.pNext = &RelaxedDesc;
    HostDesc.pNext = &RelaxedDesc;
  }

  auto ZeDevice = Device ? Device->getZeDevice() : nullptr;
  auto ZeContext = L0Context->getZeContext();
  bool MakeResident = false;
  switch (Kind) {
  case TARGET_ALLOC_DEVICE:
    MakeResident = true;
    CALL_ZE_RET_ERROR(zeMemAllocDevice, ZeContext, &DeviceDesc, Size, Align,
                      ZeDevice, &Mem);
    ODBG(OLDT_Alloc) << "Allocated " << Size << " bytes of device memory "
                     << Mem;
    break;
  case TARGET_ALLOC_HOST:
    CALL_ZE_RET_ERROR(zeMemAllocHost, ZeContext, &HostDesc, Size, Align, &Mem);
    ODBG(OLDT_Alloc) << "Allocated " << Size << " bytes of host memory " << Mem;
    break;
  case TARGET_ALLOC_SHARED:
    CALL_ZE_RET_ERROR(zeMemAllocShared, ZeContext, &DeviceDesc, &HostDesc, Size,
                      Align, ZeDevice, &Mem);
    ODBG(OLDT_Alloc) << "Allocated " << Size << " bytes of shared memory "
                     << Mem;
    break;
  default:
    assert(0 && "Invalid target data allocation kind");
  }

  if (MakeResident) {
    assert(Device &&
           "Device is not set for memory allocation. Is this a Device Pool?");
    if (auto Err = Device->makeMemoryResident(Mem, Size)) {
      Mem = nullptr;
      return std::move(Err);
    }
  }
  return Mem;
}

Error MemAllocatorTy::deallocFromL0(void *Ptr) {
  CALL_ZE_RET_ERROR(zeMemFree, L0Context->getZeContext(), Ptr);
  ODBG(OLDT_Alloc) << "Freed device pointer " << Ptr;
  return Plugin::success();
}

Expected<ze_event_handle_t> EventPoolTy::getEvent() {
  std::lock_guard<std::mutex> Lock(*Mtx);

  if (Events.empty()) {
    // Need to create a new L0 pool.
    ze_event_pool_desc_t Desc{ZE_STRUCTURE_TYPE_EVENT_POOL_DESC, nullptr, 0, 0};
    Desc.flags = ZE_EVENT_POOL_FLAG_HOST_VISIBLE | Flags;
    Desc.count = PoolSize;
    ze_event_pool_handle_t Pool;
    CALL_ZE_RET_ERROR(zeEventPoolCreate, Context, &Desc, 0, nullptr, &Pool);
    Pools.push_back(Pool);

    // Create events.
    ze_event_desc_t EventDesc{ZE_STRUCTURE_TYPE_EVENT_DESC, nullptr, 0, 0, 0};
    EventDesc.wait = 0;
    EventDesc.signal = ZE_EVENT_SCOPE_FLAG_HOST;
    for (uint32_t I = 0; I < PoolSize; I++) {
      EventDesc.index = I;
      ze_event_handle_t Event;
      CALL_ZE_RET_ERROR(zeEventCreate, Pool, &EventDesc, &Event);
      Events.push_back(Event);
    }
  }

  auto Ret = Events.back();
  Events.pop_back();

  return Ret;
}

/// Return an event to the pool.
Error EventPoolTy::releaseEvent(ze_event_handle_t Event, L0DeviceTy &Device) {
  std::lock_guard<std::mutex> Lock(*Mtx);
  CALL_ZE_RET_ERROR(zeEventHostReset, Event);
  Events.push_back(Event);
  return Plugin::success();
}

} // namespace llvm::omp::target::plugin