mirror of
https://github.com/nlohmann/json.git
synced 2026-10-11 17:07:15 +00:00
Speed up json_view's parser with SIMD scanning and a hash table for large objects. Long runs of string bytes are scanned 16 bytes at a time with NEON (AArch64, GCC and Clang) and SSE2 (x86-64), both baseline instruction sets. Keys keep 16 table checks before the vector loop, because their lengths repeat from record to record; string values get 8, because their lengths vary more. Non-ASCII text is validated 16 bytes at a time with simdjson's "lookup4" check (Keiser and Lemire, 2021), with NEON on AArch64 and, on x86-64, with SSSE3. SSSE3 is not part of baseline x86-64, so the check is compiled for SSSE3 with a function attribute and used only where CPUID reports it, which all x86-64 CPUs since about 2011 do; the answer is cached in a statically initialized atomic, so there is no guard of a local static and no global constructor. The same input is accepted either way. JSON_VIEW_NO_SIMD selects the portable code. On x86-64, string runs are now checked vector-first: one SSE2 compare from the first byte finds the end of most keys and short values, instead of a branch per byte for the first 8-16 bytes. AArch64 keeps the byte-wise steps, where a NEON mask costs more and the branches predict well. Entering an object or array no longer stalls: open() stores the parent's frame field by field instead of building it on the stack and reading it back with wider loads, which waited for the narrower stores to retire. Objects with 128 members or more get an open-addressing hash table built when the object closes, so operator[], at(), find(), contains(), count(), value(), and JSON pointers take constant time on average in such objects; of duplicate keys, the first is kept, as for the linear search. The idea comes from Boost.JSON. simdjson is credited in simd.hpp's SPDX block, the README, and license.md. Signed-off-by: Niels Lohmann <mail@nlohmann.me>
167 lines
5.4 KiB
C++
167 lines
5.4 KiB
C++
// __ _____ _____ _____
|
|
// __| | __| | | | JSON for Modern C++
|
|
// | | |__ | | | | | | version 3.12.0
|
|
// |_____|_____|_____|_|___| https://github.com/nlohmann/json
|
|
//
|
|
// SPDX-FileCopyrightText: 2013-2026 Niels Lohmann <https://nlohmann.me>
|
|
// SPDX-License-Identifier: MIT
|
|
|
|
#pragma once
|
|
|
|
#include <cstddef> // size_t
|
|
#include <cstdint> // uint16_t, uint32_t, uint64_t
|
|
#include <cstring> // memcmp, memcpy
|
|
#include <limits> // numeric_limits
|
|
#include <type_traits> // integral_constant, is_integral, is_same
|
|
|
|
#include <nlohmann/json.hpp>
|
|
#include <nlohmann/detail/view/document_data.hpp>
|
|
#include <nlohmann/detail/view/macro_scope.hpp>
|
|
#include <nlohmann/detail/view/node.hpp>
|
|
#include <nlohmann/detail/view/object_index.hpp>
|
|
|
|
NLOHMANN_JSON_NAMESPACE_BEGIN
|
|
namespace detail
|
|
{
|
|
namespace view
|
|
{
|
|
|
|
/// equality test for strings of one length n <= 16: two overlapping loads per
|
|
/// string (the first and the last 8, 4, or 2 bytes) replace a memcmp, and no
|
|
/// byte outside [s, s + n) is read
|
|
class short_key
|
|
{
|
|
public:
|
|
short_key(const unsigned char* k, std::size_t n) noexcept
|
|
: m_n(n)
|
|
{
|
|
load(k, m_a, m_b);
|
|
}
|
|
|
|
NLOHMANN_VIEW_ALWAYS_INLINE bool matches(const unsigned char* s) const noexcept
|
|
{
|
|
std::uint64_t a = 0;
|
|
std::uint64_t b = 0;
|
|
load(s, a, b);
|
|
return a == m_a && b == m_b;
|
|
}
|
|
|
|
private:
|
|
template<typename T>
|
|
static NLOHMANN_VIEW_ALWAYS_INLINE std::uint64_t load_word(const unsigned char* s) noexcept
|
|
{
|
|
T w = 0;
|
|
std::memcpy(&w, s, sizeof(T));
|
|
return w;
|
|
}
|
|
|
|
NLOHMANN_VIEW_ALWAYS_INLINE void load(const unsigned char* s, std::uint64_t& a, std::uint64_t& b) const noexcept
|
|
{
|
|
if (m_n >= 8)
|
|
{
|
|
a = load_word<std::uint64_t>(s);
|
|
b = load_word<std::uint64_t>(s + m_n - 8);
|
|
}
|
|
else if (m_n >= 4)
|
|
{
|
|
a = load_word<std::uint32_t>(s);
|
|
b = load_word<std::uint32_t>(s + m_n - 4);
|
|
}
|
|
else if (m_n >= 2)
|
|
{
|
|
a = load_word<std::uint16_t>(s);
|
|
b = load_word<std::uint16_t>(s + m_n - 2);
|
|
}
|
|
else
|
|
{
|
|
a = m_n == 1 ? s[0] : 0;
|
|
b = 0;
|
|
}
|
|
}
|
|
|
|
std::size_t m_n;
|
|
std::uint64_t m_a = 0;
|
|
std::uint64_t m_b = 0;
|
|
};
|
|
|
|
/// the key node of the first member of an object with the given key, or
|
|
/// nullptr (the search stops at the first match; materialize() and parse()
|
|
/// keep the last value of a duplicate key instead); most keys are rejected by
|
|
/// their length, from the index alone
|
|
inline const node* find_member(const document_data& d, const node* object, const char* key, std::size_t n) noexcept
|
|
{
|
|
if (NLOHMANN_VIEW_UNLIKELY(object->extra != 0))
|
|
{
|
|
return find_indexed(d, object, key, n); // a large object
|
|
}
|
|
const node* const end = document_data::child_end(object);
|
|
const auto* const k = reinterpret_cast<const unsigned char*>(key); // NOLINT(cppcoreguidelines-pro-type-reinterpret-cast)
|
|
if (NLOHMANN_VIEW_LIKELY(n <= 16))
|
|
{
|
|
const short_key probe(k, n);
|
|
for (const node* m = document_data::first_child(object); m != end; m = document_data::after(m + 1))
|
|
{
|
|
if (m->len == n && probe.matches(reinterpret_cast<const unsigned char*>(d.str(*m)))) // NOLINT(cppcoreguidelines-pro-type-reinterpret-cast)
|
|
{
|
|
return m;
|
|
}
|
|
}
|
|
return nullptr;
|
|
}
|
|
for (const node* m = document_data::first_child(object); m != end; m = document_data::after(m + 1))
|
|
{
|
|
if (m->len == n && std::memcmp(d.str(*m), key, n) == 0)
|
|
{
|
|
return m;
|
|
}
|
|
}
|
|
return nullptr;
|
|
}
|
|
|
|
/// whether an integer type is accepted as an array index by the view's
|
|
/// operator[] and at(): every integer type but bool and size_t, which has its
|
|
/// own overload
|
|
template<typename T>
|
|
struct is_index_type : std::integral_constant < bool,
|
|
std::is_integral<T>::value && !std::is_same<T, bool>::value && !std::is_same<T, std::size_t>::value >
|
|
{};
|
|
|
|
/// an integer as an index: negative values, and values that do not fit a
|
|
/// size_t, map to the largest size_t (out of range for every array)
|
|
template<typename SizeType, typename IntegerType>
|
|
SizeType to_index(IntegerType idx) noexcept
|
|
{
|
|
const IntegerType zero = 0;
|
|
const auto result = static_cast<SizeType>(idx); // NOLINT(bugprone-signed-char-misuse,cert-str34-c): negative values are mapped below
|
|
return (idx < zero || static_cast<IntegerType>(result) != idx) ? (std::numeric_limits<SizeType>::max)() : result;
|
|
}
|
|
|
|
/// the element of an array at an index below its size
|
|
inline const node* element_at(const node* array, std::size_t idx) noexcept
|
|
{
|
|
const node* e = document_data::first_child(array);
|
|
for (std::size_t i = 0; i < idx; ++i)
|
|
{
|
|
e = document_data::after(e);
|
|
}
|
|
return e;
|
|
}
|
|
|
|
/// the last element of a non-empty array, or the key of the last member of a
|
|
/// non-empty object
|
|
inline const node* last_child(const node* container) noexcept
|
|
{
|
|
const std::size_t value_offset = container->kind == static_cast<std::uint8_t>(value_t::object) ? 1 : 0;
|
|
const node* const end = document_data::child_end(container);
|
|
const node* last = document_data::first_child(container);
|
|
for (const node* c = document_data::after(last + value_offset); c != end; c = document_data::after(c + value_offset))
|
|
{
|
|
last = c;
|
|
}
|
|
return last;
|
|
}
|
|
|
|
} // namespace view
|
|
} // namespace detail
|
|
NLOHMANN_JSON_NAMESPACE_END
|