* Copyright (c) 2026 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.
*/
#if !defined(ASCENDC_TENSOR_API_INCLUDE_COMPILER_INTERNAL_HEADERS)
#warning \
"impl/tensor_api/utils/constant_impl.h is an internal header file and must not be used directly. Functions or variables defined in this file maybe removed in the future. Please use "#include "tensor_api/tensor.h"" and use public functions or variables defined in interface headers files."
#define ASCENDC_TENSOR_API_INCLUDE_COMPILER_INTERNAL_HEADERS
#define UNDEF_ASCENDC_TENSOR_API_INCLUDE_COMPILER_INTERNAL_HEADERS_ASCENDC
#endif
* \file constant_impl.h
* \brief
*/
#ifndef IMPL_TENSOR_API_UTILS_CONSTANT_IMPL_H
#define IMPL_TENSOR_API_UTILS_CONSTANT_IMPL_H
#include "tensor_api/utils/utils.h"
#include "impl/tensor_api/utils/extra_impl.h"
#include "impl/tensor_api/utils/map_impl.h"
namespace asc {
namespace te {
constexpr size_t two_dim_data = 2;
constexpr size_t three_dim_data = 3;
constexpr size_t four_dim_data = 4;
constexpr size_t five_dim_data = 5;
constexpr size_t fractal_fixed = 16;
constexpr size_t mx_scale_k0 = 2;
constexpr uint32_t block_cube = 16;
constexpr uint64_t hifloat8_mmad_ctrl_mask = 0x200000000000ULL;
struct arch_version {
static constexpr uint32_t v3510 = 3510;
static constexpr uint32_t v2201 = 2201;
static constexpr uint32_t V3510 = 3510;
static constexpr uint32_t V2201 = 2201;
};
struct get_arch_version {
__aicore__ inline constexpr uint32_t operator()() const
{
#ifdef __NPU_ARCH__
return __NPU_ARCH__;
#else
return 0;
#endif
}
};
constexpr uint32_t current_arch_version = get_arch_version{}();
template <typename Hardware>
struct is_hardware {
private:
template <typename Target, typename... Candidates>
__aicore__ inline static constexpr bool is_unqualified_any_of()
{
return (... || Std::is_same_v<Std::remove_cvref_t<Target>, Candidates>);
}
public:
static constexpr bool value = is_unqualified_any_of<
Hardware, location::invalid, location::gm, location::ub, location::l1, location::l0a, location::l0b,
location::l0scalea, location::l0scaleb, location::l0c, location::bias, location::fixbuf, location::ssbuf>();
};
template <typename Hardware>
constexpr bool is_hardware_v = is_hardware<Hardware>::value;
template <typename TupleType>
using tuple_sequence = Std::make_index_sequence<Std::tuple_size_v<Std::remove_cvref_t<TupleType>>>;
template <typename DataType>
struct location_attr {
using gm_attr = __gm__ DataType*;
using l1_attr = __cbuf__ DataType*;
using l0a_attr = __ca__ DataType*;
using l0b_attr = __cb__ DataType*;
using l0c_attr = __cc__ DataType*;
using ub_attr = __ubuf__ DataType*;
using fixbuf_attr = __fbuf__ DataType*;
using ssbuf_attr = __ssbuf__ DataType*;
using bias_attr = __biasbuf__ DataType*;
using none_attr = DataType*;
using type =
Std::tuple<gm_attr, l1_attr, l0a_attr, l0b_attr, l0c_attr, ub_attr, fixbuf_attr, ssbuf_attr, bias_attr>;
using location_map = tuple_map<
Std::tuple<location::gm, gm_attr>, Std::tuple<location::l1, l1_attr>, Std::tuple<location::l0a, l0a_attr>,
Std::tuple<location::l0b, l0b_attr>, Std::tuple<location::l0scalea, none_attr>,
Std::tuple<location::l0scaleb, none_attr>, Std::tuple<location::l0c, l0c_attr>,
Std::tuple<location::ub, ub_attr>, Std::tuple<location::bias, bias_attr>,
Std::tuple<location::fixbuf, fixbuf_attr>>;
};
template <typename Target, typename ProcessedTuple>
struct all_elements_same_as_a;
template <typename Target, typename First, typename... Rest>
struct all_elements_same_as_a<Target, Std::tuple<First, Rest...>> {
static constexpr bool value = Std::is_same_v<Target, typename iter_ele<First>::type> ||
all_elements_same_as_a<Target, Std::tuple<Rest...>>::value;
};
template <typename Target>
struct all_elements_same_as_a<Target, Std::tuple<>> {
static constexpr bool value = false;
};
template <typename Target, typename... Candidates>
struct check_all_same;
template <typename Target, typename Candidate, typename... RestCandidates>
struct check_all_same<Target, Candidate, RestCandidates...> {
static constexpr bool value = Std::is_same_v<Target, Candidate> ||
all_elements_same_as_a<Target, typename location_attr<Candidate>::type>::value ||
check_all_same<Target, RestCandidates...>::value;
};
template <typename Target>
struct check_all_same<Target> {
static constexpr bool value = false;
};
template <typename Target, typename... Candidates>
constexpr bool is_one_of_attr_v = check_all_same<Target, Candidates...>::value;
template <typename DataType>
inline constexpr bool is_data_type =
is_one_of_attr_v<
Std::remove_cvref_t<DataType>, hifloat8_t, bfloat16_t, fp4x2_e1m2_t, fp4x2_e2m1_t, fp8_e5m2_t, fp8_e4m3fn_t,
fp8_e8m0_t> ||
Std::is_integral_v<Std::remove_cvref_t<DataType>> || Std::is_floating_point_v<Std::remove_cvref_t<DataType>>;
template <typename DataType>
inline constexpr bool is_b4_type = is_one_of_attr_v<DataType, fp4x2_e1m2_t, fp4x2_e2m1_t>;
template <typename DataType = Std::ignore_t>
__aicore__ inline constexpr size_t get_c0_size()
{
constexpr size_t c0_size = 32;
if constexpr (is_b4_type<DataType>) {
return c0_size * 2;
} else {
return c0_size;
}
}
template <typename DataType = Std::ignore_t>
constexpr size_t c0_size = get_c0_size<DataType>();
template <typename DataType>
constexpr size_t c0_element = c0_size<DataType> / sizeof(DataType);
template <typename Constant>
struct is_integral_constant : Std::false_type {};
template <size_t Value>
struct is_integral_constant<Std::Int<Value>> : Std::true_type {};
template <typename Constant>
constexpr bool is_integral_constant_v = is_integral_constant<Constant>::value;
}
}
#endif
#if defined(UNDEF_ASCENDC_TENSOR_API_INCLUDE_COMPILER_INTERNAL_HEADERS_ASCENDC)
#undef ASCENDC_TENSOR_API_INCLUDE_COMPILER_INTERNAL_HEADERS
#undef UNDEF_ASCENDC_TENSOR_API_INCLUDE_COMPILER_INTERNAL_HEADERS_ASCENDC
#endif
