From d708b77186c1685dcbd2298246ada7d204acec2f Mon Sep 17 00:00:00 2001 From: Maxime Coste Date: Mon, 13 Feb 2023 22:51:53 +1100 Subject: [PATCH] Refactor DualThreadStack as a RingBuffer Instead of two stacks growing from the two ends of a buffer, use a ring buffer growing from the same mid spot. This avoids the costly memory copy every step when we set next threads as the current ones. --- src/regex_impl.hh | 79 +++++++++++++++++++++++++---------------------- 1 file changed, 42 insertions(+), 37 deletions(-) diff --git a/src/regex_impl.hh b/src/regex_impl.hh index 3593b9b0..cb5acd05 100644 --- a/src/regex_impl.hh +++ b/src/regex_impl.hh @@ -197,6 +197,7 @@ template struct SentinelType> { using Type = typename It::Sentinel; }; template + requires (has_direction(mode)) class ThreadedRegexVM { public: @@ -450,7 +451,7 @@ private: kak_assert(m_threads.current_is_empty() and m_threads.next_is_empty()); release_saves(m_captures); m_captures = -1; - m_threads.grow_ifn(); + m_threads.ensure_initial_capacity(); const int16_t first_inst = forward ? 0 : m_program.first_backward_inst; m_threads.push_current({first_inst, -1}); @@ -478,9 +479,8 @@ private: (m_threads.next_is_empty() and (not search or m_found_match)) or (m_found_match and any_match)) { - for (auto& t : m_threads.next_threads()) - release_saves(t.saves); - m_threads.clear_next(); + while (not m_threads.next_is_empty()) + release_saves(m_threads.pop_next().saves); return m_found_match; } @@ -491,7 +491,6 @@ private: { if (start_desc and m_threads.next_is_empty()) to_next_start(pos, config, *start_desc); - m_threads.grow_ifn(); m_threads.push_next({first_inst, -1}); } m_threads.swap_next(); @@ -607,62 +606,68 @@ private: struct DualThreadStack { - DualThreadStack() = default; - DualThreadStack(const DualThreadStack&) = delete; - DualThreadStack(DualThreadStack&& other) - : m_data{other.m_data}, m_capacity{other.m_capacity}, m_current{other.m_current}, m_next{other.m_next} - { - other.m_data = nullptr; - } - ~DualThreadStack() { delete[] m_data; } + bool current_is_empty() const { return m_current == m_next_begin; } + bool next_is_empty() const { return m_next_end == m_next_begin; } - bool current_is_empty() const { return m_current == 0; } - bool next_is_empty() const { return m_next == m_capacity; } + void push_current(Thread thread) { m_data[decrement(m_current)] = thread; grow_ifn(); } + Thread pop_current() { auto res = m_data[m_current]; increment(m_current); return res; } - void push_current(Thread thread) { kak_assert(m_current < m_next); m_data[m_current++] = thread; grow_ifn(); } - Thread pop_current() { kak_assert(m_current > 0); return m_data[--m_current]; } - - void push_next(Thread thread) { kak_assert(m_current < m_next); m_data[--m_next] = thread; } - void clear_next() { m_next = m_capacity; } - ConstArrayView next_threads() const { return { m_data + m_next, m_data + m_capacity }; } + void push_next(Thread thread) { m_data[m_next_end] = thread; increment(m_next_end); } + Thread pop_next() { return m_data[decrement(m_next_end)]; } void swap_next() { - kak_assert(m_next < m_capacity); - const int32_t count = m_capacity - m_next; - std::copy_n(m_data + m_next, count, m_data); - m_current = count; - m_next = m_capacity; + m_current = m_next_begin; + m_next_begin = m_next_end; + } + + void ensure_initial_capacity() { + if (m_capacity == 0) + grow_ifn(); } void grow_ifn() { - if (m_current != m_next) + if (m_current != m_next_end) return; constexpr int32_t initial_capacity = 64 / sizeof(Thread); static_assert(initial_capacity >= 4); const auto new_capacity = m_capacity ? m_capacity * 2 : initial_capacity; - const auto next_count = m_capacity - m_next; - const auto new_next = new_capacity - next_count; Thread* new_data = new Thread[new_capacity]; - std::copy_n(m_data, m_current, new_data); - std::copy_n(m_data + m_next, next_count, new_data + new_next); - delete[] m_data; - m_data = new_data; + if (m_current < m_next_end) + m_next_end = std::copy(m_data.get() + m_current, m_data.get() + m_next_end, new_data) - new_data; + else + m_next_end = std::copy(m_data.get(), m_data.get() + m_next_end, std::copy(m_data.get() + m_current, m_data.get() + m_capacity, new_data)) - new_data; + + m_next_begin = m_next_begin >= m_current ? m_next_begin - m_current : m_capacity - (m_current - m_next_begin); + m_current = 0; + + m_data.reset(new_data); m_capacity = new_capacity; - m_next = new_next; } private: - Thread* m_data = nullptr; + int32_t decrement(int32_t& index) { + if (index == 0) + index = m_capacity; + return --index; + } + + int32_t increment(int32_t& index) { + if (++index == m_capacity) + index = 0; + return index; + } + + std::unique_ptr m_data; int32_t m_capacity = 0; // Maximum capacity should be 2*instruction count, so 65536 int32_t m_current = 0; - int32_t m_next = 0; + int32_t m_next_begin = 0; + int32_t m_next_end = 0; }; - static_assert(has_direction(mode)); static constexpr bool forward = mode & RegexMode::Forward; DualThreadStack m_threads;