* Copyright (c) 2025 Huawei Technologies Co., Ltd.
* This program is free software, you can redistribute it and/or modify it under the terms and conditions of
* CANN Open Software License Agreement Version 2.0 (the "License").
* Please refer to the License for details. You may not use this file except in compliance with the License.
* THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
* INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
* See LICENSE in the root of the software repository for the full text of the License.
*/
* \file tile_tensor.h
* \brief
*/
#ifndef TILEOP_UTILS_TILE_TENSOR_H
#define TILEOP_UTILS_TILE_TENSOR_H
#include "common_type.h"
template <typename... Indexs>
using Offsets = Std::tuple<Indexs...>;
using TileOffset = Offsets<size_t, size_t, size_t>;
using TileOffset4Dim = Offsets<size_t, size_t, size_t, size_t>;
template <typename T, typename LA, Hardware FMT = Hardware::UB>
struct TileTensor {
using Type = T;
using LayoutType = LA;
using Shape = typename LA::Shape;
using Stride = typename LA::Stride;
using TileShape = typename LA::TileShape;
static constexpr Hardware FORMAT = FMT;
__aicore__ inline TileTensor() {}
__aicore__ inline TileTensor(uint64_t addr, LA layout) : addr_(addr), layout_(layout) {}
__aicore__ inline TileTensor(uint64_t addr, Shape shape) : addr_(addr), layout_(LA(shape)) {}
__aicore__ inline TileTensor(uint64_t addr) : addr_(addr) {}
__aicore__ inline uint64_t GetAddr() { return addr_; }
__aicore__ inline void SetAddr(uint64_t addr) { addr_ = addr; }
__aicore__ inline LA GetLayout() { return layout_; }
__aicore__ inline const LA GetLayout() const { return layout_; }
__aicore__ inline static constexpr bool IsStaticLayout() { return LayoutType::IsStaticLayout(); }
__aicore__ inline constexpr Hardware GetPhyType() { return FORMAT; }
__aicore__ inline Shape GetShape() { return layout_.GetShape(); }
__aicore__ inline Stride GetStride() { return layout_.GetStride(); }
__aicore__ inline uint64_t GetLinearAddr(const TileOffset& offsets = TileOffset(0, 0, 0))
{
size_t offset = Std::get<DIM_1ST>(offsets) * layout_.template GetStrideDim<DIM_1ST, MAX_DIMS>();
offset += Std::get<DIM_2ND>(offsets) * layout_.template GetStrideDim<DIM_2ND, MAX_DIMS>();
offset += Std::get<DIM_3RD>(offsets) * layout_.template GetStrideDim<DIM_3RD, MAX_DIMS>();
return addr_ + offset * sizeof(Type);
}
private:
uint64_t addr_;
LA layout_;
};
template <typename T, typename LA>
struct TileTensor<T, LA, Hardware::GM> {
using Type = T;
using LayoutType = LA;
using Shape = typename LA::Shape;
using Stride = typename LA::Stride;
using TileShape = typename LA::TileShape;
static constexpr Hardware FORMAT = Hardware::GM;
__aicore__ inline TileTensor() {}
__aicore__ inline TileTensor(T* addr, LA layout) : addr_(addr), layout_(layout) {}
__aicore__ inline TileTensor(T* addr, Shape shape) : addr_(addr), layout_(LA(shape)) {}
__aicore__ inline TileTensor(T* addr) : addr_(addr) {}
__aicore__ inline T* GetAddr() { return addr_; }
__aicore__ inline void SetAddr(T* addr) { addr_ = addr; }
__aicore__ inline LA GetLayout() { return layout_; }
__aicore__ inline const LA GetLayout() const { return layout_; }
__aicore__ inline static constexpr bool IsStaticLayout() { return LayoutType::IsStaticLayout(); }
__aicore__ inline constexpr Hardware GetPhyType() { return FORMAT; }
__aicore__ inline Shape GetShape() { return layout_.GetShape(); }
__aicore__ inline Stride GetStride() { return layout_.GetStride(); }
private:
T* addr_;
LA layout_;
};
template <typename T>
struct TensorTileInfo {
static constexpr auto tile0 = TileOp::GetTensorTileShapeDim<T, DIM_1ST, MAX_DIMS>();
static constexpr auto tile1 = TileOp::GetTensorTileShapeDim<T, DIM_2ND, MAX_DIMS>();
static constexpr auto tile2 = TileOp::GetTensorTileShapeDim<T, DIM_3RD, MAX_DIMS>();
static constexpr auto tileH = TileOp::GetTensorTileShapeDim<T, DIM_4TH, MAX_DIMS>();
static constexpr auto tileW = TileOp::GetTensorTileShapeDim<T, DIM_5TH, MAX_DIMS>();
};
template <typename T, typename LA, Hardware FMT = Hardware::UB>
TILEOP TileTensor<T, LA, FMT> MakeTensor(__ubuf__ T* addr, LA layout)
{
return TileTensor<T, LA, FMT>(addr, layout);
}
#endif