helpers.hpp

Go to the documentation of this file.

//==---------------- helpers.hpp - SYCL helpers ----------------------------==//
//
// 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
//
//===----------------------------------------------------------------------===//
 
#pragma once
 
#include <CL/__spirv/spirv_types.hpp> // for MemorySemanticsMask
#include <sycl/access/access.hpp>     // for fence_space
#include <sycl/detail/export.hpp>     // for __SYCL_EXPORT
#include <sycl/detail/pi.hpp>         // for PiEvent
#include <sycl/memory_enums.hpp>      // for memory_order
 
#ifdef __SYCL_DEVICE_ONLY__
#include <CL/__spirv/spirv_vars.hpp>
#endif
 
#include <cstddef>     // for size_t
#include <memory>      // for shared_ptr
#include <stdint.h>    // for uint32_t
#include <type_traits> // for enable_if_t, integral_constant
#include <utility>     // for forward, integer_sequence, mak...
#include <vector>      // for vector
 
namespace sycl {
inline namespace _V1 {
class context;
class event;
template <int Dims, bool WithOffset> class item;
template <int Dims> class group;
template <int Dims> class range;
template <int Dims> class id;
template <int Dims> class nd_item;
template <int Dims> class h_item;
template <typename Type, std::size_t NumElements> class marray;
enum class memory_order;
 
namespace detail {
 
class buffer_impl;
class context_impl;
// The function returns list of events that can be passed to OpenCL API as
// dependency list and waits for others.
__SYCL_EXPORT std::vector<sycl::detail::pi::PiEvent>
getOrWaitEvents(std::vector<sycl::event> DepEvents,
                std::shared_ptr<sycl::detail::context_impl> Context);
 
__SYCL_EXPORT void waitEvents(std::vector<sycl::event> DepEvents);
 
__SYCL_EXPORT void
markBufferAsInternal(const std::shared_ptr<buffer_impl> &BufImpl);
 
template <typename T> T *declptr() { return static_cast<T *>(nullptr); }
 
// Function to get of store id, item, nd_item, group for the host implementation
// Pass nullptr to get stored object. Pass valid address to store object
template <typename T> T get_or_store(const T *obj) {
  static thread_local auto stored = *obj;
  if (obj != nullptr) {
    stored = *obj;
  }
  return stored;
}
 
class Builder {
public:
  Builder() = delete;
 
  template <int Dims>
  static group<Dims>
  createGroup(const range<Dims> &Global, const range<Dims> &Local,
              const range<Dims> &Group, const id<Dims> &Index) {
    return group<Dims>(Global, Local, Group, Index);
  }
 
  template <int Dims>
  static group<Dims> createGroup(const range<Dims> &Global,
                                 const range<Dims> &Local,
                                 const id<Dims> &Index) {
    return group<Dims>(Global, Local, Global / Local, Index);
  }
 
  template <class ResType, typename BitsType>
  static ResType createSubGroupMask(BitsType Bits, size_t BitsNum) {
    return ResType(Bits, BitsNum);
  }
 
  template <int Dims, bool WithOffset>
  static std::enable_if_t<WithOffset, item<Dims, WithOffset>>
  createItem(const range<Dims> &Extent, const id<Dims> &Index,
             const id<Dims> &Offset) {
    return item<Dims, WithOffset>(Extent, Index, Offset);
  }
 
  template <int Dims, bool WithOffset>
  static std::enable_if_t<!WithOffset, item<Dims, WithOffset>>
  createItem(const range<Dims> &Extent, const id<Dims> &Index) {
    return item<Dims, WithOffset>(Extent, Index);
  }
 
  template <int Dims>
  static nd_item<Dims> createNDItem(const item<Dims, true> &Global,
                                    const item<Dims, false> &Local,
                                    const group<Dims> &Group) {
    return nd_item<Dims>(Global, Local, Group);
  }
 
  template <int Dims>
  static h_item<Dims> createHItem(const item<Dims, false> &Global,
                                  const item<Dims, false> &Local) {
    return h_item<Dims>(Global, Local);
  }
 
  template <int Dims>
  static h_item<Dims> createHItem(const item<Dims, false> &Global,
                                  const item<Dims, false> &Local,
                                  const range<Dims> &Flex) {
    return h_item<Dims>(Global, Local, Flex);
  }
 
  template <int Dims, bool WithOffset>
  static void updateItemIndex(sycl::item<Dims, WithOffset> &Item,
                              const id<Dims> &NextIndex) {
    Item.MImpl.MIndex = NextIndex;
  }
 
#ifdef __SYCL_DEVICE_ONLY__
 
  template <int N>
  using is_valid_dimensions = std::integral_constant<bool, (N > 0) && (N < 4)>;
 
  template <int Dims> static const id<Dims> getElement(id<Dims> *) {
    static_assert(is_valid_dimensions<Dims>::value, "invalid dimensions");
    return __spirv::initGlobalInvocationId<Dims, id<Dims>>();
  }
 
  template <int Dims> static const group<Dims> getElement(group<Dims> *) {
    static_assert(is_valid_dimensions<Dims>::value, "invalid dimensions");
    range<Dims> GlobalSize{__spirv::initGlobalSize<Dims, range<Dims>>()};
    range<Dims> LocalSize{__spirv::initWorkgroupSize<Dims, range<Dims>>()};
    range<Dims> GroupRange{__spirv::initNumWorkgroups<Dims, range<Dims>>()};
    id<Dims> GroupId{__spirv::initWorkgroupId<Dims, id<Dims>>()};
    return createGroup<Dims>(GlobalSize, LocalSize, GroupRange, GroupId);
  }
 
  template <int Dims, bool WithOffset>
  static std::enable_if_t<WithOffset, const item<Dims, WithOffset>> getItem() {
    static_assert(is_valid_dimensions<Dims>::value, "invalid dimensions");
    id<Dims> GlobalId{__spirv::initGlobalInvocationId<Dims, id<Dims>>()};
    range<Dims> GlobalSize{__spirv::initGlobalSize<Dims, range<Dims>>()};
    id<Dims> GlobalOffset{__spirv::initGlobalOffset<Dims, id<Dims>>()};
    return createItem<Dims, true>(GlobalSize, GlobalId, GlobalOffset);
  }
 
  template <int Dims, bool WithOffset>
  static std::enable_if_t<!WithOffset, const item<Dims, WithOffset>> getItem() {
    static_assert(is_valid_dimensions<Dims>::value, "invalid dimensions");
    id<Dims> GlobalId{__spirv::initGlobalInvocationId<Dims, id<Dims>>()};
    range<Dims> GlobalSize{__spirv::initGlobalSize<Dims, range<Dims>>()};
    return createItem<Dims, false>(GlobalSize, GlobalId);
  }
 
  template <int Dims> static const nd_item<Dims> getElement(nd_item<Dims> *) {
    static_assert(is_valid_dimensions<Dims>::value, "invalid dimensions");
    range<Dims> GlobalSize{__spirv::initGlobalSize<Dims, range<Dims>>()};
    range<Dims> LocalSize{__spirv::initWorkgroupSize<Dims, range<Dims>>()};
    range<Dims> GroupRange{__spirv::initNumWorkgroups<Dims, range<Dims>>()};
    id<Dims> GroupId{__spirv::initWorkgroupId<Dims, id<Dims>>()};
    id<Dims> GlobalId{__spirv::initGlobalInvocationId<Dims, id<Dims>>()};
    id<Dims> LocalId{__spirv::initLocalInvocationId<Dims, id<Dims>>()};
    id<Dims> GlobalOffset{__spirv::initGlobalOffset<Dims, id<Dims>>()};
    group<Dims> Group =
        createGroup<Dims>(GlobalSize, LocalSize, GroupRange, GroupId);
    item<Dims, true> GlobalItem =
        createItem<Dims, true>(GlobalSize, GlobalId, GlobalOffset);
    item<Dims, false> LocalItem = createItem<Dims, false>(LocalSize, LocalId);
    return createNDItem<Dims>(GlobalItem, LocalItem, Group);
  }
 
  template <int Dims, bool WithOffset>
  static auto getElement(item<Dims, WithOffset> *)
      -> decltype(getItem<Dims, WithOffset>()) {
    return getItem<Dims, WithOffset>();
  }
 
  template <int Dims>
  static auto getNDItem() -> decltype(getElement(declptr<nd_item<Dims>>())) {
    return getElement(declptr<nd_item<Dims>>());
  }
 
#endif // __SYCL_DEVICE_ONLY__
};
 
inline constexpr __spv::MemorySemanticsMask::Flag
getSPIRVMemorySemanticsMask(memory_order) {
  return __spv::MemorySemanticsMask::None;
}
 
inline constexpr uint32_t
getSPIRVMemorySemanticsMask(const access::fence_space AccessSpace,
                            const __spv::MemorySemanticsMask LocalScopeMask =
                                __spv::MemorySemanticsMask::WorkgroupMemory) {
  // Huge ternary operator below is a workaround for constexpr function
  // requirement that such function can only contain return statement and
  // nothing more
  //
  // It is equivalent to the following code:
  //
  // uint32_t Flags =
  //     static_cast<uint32_t>(__spv::MemorySemanticsMask::SequentiallyConsistent);
  // switch (AccessSpace) {
  // case access::fence_space::global_space:
  //   Flags |=
  //       static_cast<uint32_t>(__spv::MemorySemanticsMask::CrossWorkgroupMemory);
  //   break;
  // case access::fence_space::local_space:
  //   Flags |= static_cast<uint32_t>(LocalScopeMask);
  //   break;
  // case access::fence_space::global_and_local:
  // default:
  //   Flags |= static_cast<uint32_t>(
  //                __spv::MemorySemanticsMask::CrossWorkgroupMemory) |
  //            static_cast<uint32_t>(LocalScopeMask);
  //   break;
  // }
  // return Flags;
 
  return (AccessSpace == access::fence_space::global_space)
             ? static_cast<uint32_t>(
                   __spv::MemorySemanticsMask::SequentiallyConsistent |
                   __spv::MemorySemanticsMask::CrossWorkgroupMemory)
         : (AccessSpace == access::fence_space::local_space)
             ? static_cast<uint32_t>(
                   __spv::MemorySemanticsMask::SequentiallyConsistent |
                   LocalScopeMask)
             : /* default: (AccessSpace ==
                  access::fence_space::global_and_local) */
             static_cast<uint32_t>(
                 __spv::MemorySemanticsMask::SequentiallyConsistent |
                 __spv::MemorySemanticsMask::CrossWorkgroupMemory |
                 LocalScopeMask);
}
 
// To ensure loop unrolling is done when processing dimensions.
template <size_t... Inds, class F>
void loop_impl(std::integer_sequence<size_t, Inds...>, F &&f) {
  (f(std::integral_constant<size_t, Inds>{}), ...);
}
 
template <size_t count, class F> void loop(F &&f) {
  loop_impl(std::make_index_sequence<count>{}, std::forward<F>(f));
}
} // namespace detail
 
} // namespace _V1
} // namespace sycl