about summary refs log tree commit diff
path: root/immer/experimental
diff options
context:
space:
mode:
authorVincent Ambo <mail@tazj.in>2020-07-15T07·20+0100
committerVincent Ambo <mail@tazj.in>2020-07-15T07·20+0100
commit7f19d641647ac4ef313ed88d6b5c140983ce5436 (patch)
tree31b66c81465293da5c093c5dde3e419758c0d6cc /immer/experimental
Squashed 'third_party/immer/' content from commit ad3e3556d
git-subtree-dir: third_party/immer
git-subtree-split: ad3e3556d38bb75966dd24c61a774970a7c7957e
Diffstat (limited to 'immer/experimental')
-rw-r--r--immer/experimental/detail/dvektor_impl.hpp498
-rw-r--r--immer/experimental/dvektor.hpp69
2 files changed, 567 insertions, 0 deletions
diff --git a/immer/experimental/detail/dvektor_impl.hpp b/immer/experimental/detail/dvektor_impl.hpp
new file mode 100644
index 000000000000..81dbbc59f5f3
--- /dev/null
+++ b/immer/experimental/detail/dvektor_impl.hpp
@@ -0,0 +1,498 @@
+//
+// immer: immutable data structures for C++
+// Copyright (C) 2016, 2017, 2018 Juan Pedro Bolivar Puente
+//
+// This software is distributed under the Boost Software License, Version 1.0.
+// See accompanying file LICENSE or copy at http://boost.org/LICENSE_1_0.txt
+//
+
+#pragma once
+
+#include <immer/heap/heap_policy.hpp>
+#include <immer/refcount/enable_intrusive_ptr.hpp>
+#include <immer/refcount/refcount_policy.hpp>
+
+#include <boost/intrusive_ptr.hpp>
+#include <boost/iterator/iterator_facade.hpp>
+#include <boost/smart_ptr/intrusive_ref_counter.hpp>
+
+#include <cassert>
+#include <limits>
+
+namespace immer {
+namespace detail {
+namespace dvektor {
+
+constexpr auto fast_log2(std::size_t x)
+{
+    return x == 0 ? 0 : sizeof(std::size_t) * 8 - 1 - __builtin_clzl(x);
+}
+
+template <int B, typename T = std::size_t>
+constexpr T branches = T{1} << B;
+
+template <int B, typename T = std::size_t>
+constexpr T mask = branches<B, T> - 1;
+
+template <int B, typename T = std::size_t>
+constexpr auto
+    max_depth = fast_log2(std::numeric_limits<std::size_t>::max()) / B;
+
+template <typename T, int B, typename MP>
+struct node;
+
+template <typename T, int B, typename MP>
+using node_ptr = boost::intrusive_ptr<node<T, B, MP>>;
+
+template <typename T, int B>
+using leaf_node = std::array<T, 1 << B>;
+
+template <typename T, int B, typename MP>
+using inner_node = std::array<node_ptr<T, B, MP>, 1 << B>;
+
+template <typename T, int B, typename MP>
+struct node
+    : enable_intrusive_ptr<node<T, B, MP>, typename MP::refcount>
+    , enable_optimized_heap_policy<node<T, B, MP>, typename MP::heap>
+{
+    using leaf_node_t  = leaf_node<T, B>;
+    using inner_node_t = inner_node<T, B, MP>;
+
+    enum
+    {
+        leaf_kind,
+        inner_kind
+    } kind;
+
+    union data_t
+    {
+        leaf_node_t leaf;
+        inner_node_t inner;
+        data_t(leaf_node_t n)
+            : leaf(std::move(n))
+        {}
+        data_t(inner_node_t n)
+            : inner(std::move(n))
+        {}
+        ~data_t() {}
+    } data;
+
+    ~node()
+    {
+        switch (kind) {
+        case leaf_kind:
+            data.leaf.~leaf_node_t();
+            break;
+        case inner_kind:
+            data.inner.~inner_node_t();
+            break;
+        }
+    }
+
+    node(leaf_node<T, B> n)
+        : kind{leaf_kind}
+        , data{std::move(n)}
+    {}
+
+    node(inner_node<T, B, MP> n)
+        : kind{inner_kind}
+        , data{std::move(n)}
+    {}
+
+    inner_node_t& inner() &
+    {
+        assert(kind == inner_kind);
+        return data.inner;
+    }
+    const inner_node_t& inner() const&
+    {
+        assert(kind == inner_kind);
+        return data.inner;
+    }
+    inner_node_t&& inner() &&
+    {
+        assert(kind == inner_kind);
+        return std::move(data.inner);
+    }
+
+    leaf_node_t& leaf() &
+    {
+        assert(kind == leaf_kind);
+        return data.leaf;
+    }
+    const leaf_node_t& leaf() const&
+    {
+        assert(kind == leaf_kind);
+        return data.leaf;
+    }
+    leaf_node_t&& leaf() &&
+    {
+        assert(kind == leaf_kind);
+        return std::move(data.leaf);
+    }
+};
+
+template <typename T, int B, typename MP, typename... Ts>
+auto make_node(Ts&&... xs) -> boost::intrusive_ptr<node<T, B, MP>>
+{
+    return new node<T, B, MP>(std::forward<Ts>(xs)...);
+}
+
+template <typename T, int B, typename MP>
+struct ref
+{
+    using inner_t    = inner_node<T, B, MP>;
+    using leaf_t     = leaf_node<T, B>;
+    using node_t     = node<T, B, MP>;
+    using node_ptr_t = node_ptr<T, B, MP>;
+
+    unsigned depth;
+    std::array<node_ptr_t, max_depth<B>> display;
+
+    template <typename... Ts>
+    static auto make_node(Ts&&... xs)
+    {
+        return dvektor::make_node<T, B, MP>(std::forward<Ts>(xs)...);
+    }
+
+    const T& get_elem(std::size_t index, std::size_t xr) const
+    {
+        auto display_idx = fast_log2(xr) / B;
+        auto node        = display[display_idx].get();
+        auto shift       = display_idx * B;
+        while (display_idx--) {
+            node = node->inner()[(index >> shift) & mask<B>].get();
+            shift -= B;
+        }
+        return node->leaf()[index & mask<B>];
+    }
+
+    node_ptr_t null_slot_and_copy_inner(node_ptr_t& node, std::size_t idx)
+    {
+        auto& n = node->inner();
+        auto x  = node_ptr_t{};
+        x.swap(n[idx]);
+        return copy_of_inner(x);
+    }
+
+    node_ptr_t null_slot_and_copy_leaf(node_ptr_t& node, std::size_t idx)
+    {
+        auto& n = node->inner();
+        auto x  = node_ptr_t{};
+        x.swap(n[idx]);
+        return copy_of_leaf(x);
+    }
+
+    node_ptr_t copy_of_inner(const node_ptr_t& n)
+    {
+        return make_node(n->inner());
+    }
+
+    node_ptr_t copy_of_leaf(const node_ptr_t& n)
+    {
+        return make_node(n->leaf());
+    }
+
+    void stabilize(std::size_t index)
+    {
+        auto shift = B;
+        for (auto i = 1u; i < depth; ++i) {
+            display[i] = copy_of_inner(display[i]);
+            display[i]->inner()[(index >> shift) & mask<B>] = display[i - 1];
+            shift += B;
+        }
+    }
+
+    void goto_pos_writable_from_clean(std::size_t old_index,
+                                      std::size_t index,
+                                      std::size_t xr)
+    {
+        assert(depth);
+        auto d = depth - 1;
+        if (d == 0) {
+            display[0] = copy_of_leaf(display[0]);
+        } else {
+            IMMER_UNREACHABLE;
+            display[d] = copy_of_inner(display[d]);
+            auto shift = B * d;
+            while (--d) {
+                display[d] = null_slot_and_copy_inner(
+                    display[d + 1], (index >> shift) & mask<B>);
+                shift -= B;
+            }
+            display[0] =
+                null_slot_and_copy_leaf(display[1], (index >> B) & mask<B>);
+        }
+    }
+
+    void goto_pos_writable_from_dirty(std::size_t old_index,
+                                      std::size_t new_index,
+                                      std::size_t xr)
+    {
+        assert(depth);
+        if (xr < (1 << B)) {
+            display[0] = copy_of_leaf(display[0]);
+        } else {
+            auto display_idx = fast_log2(xr) / B;
+            auto shift       = B;
+            for (auto i = 1u; i <= display_idx; ++i) {
+                display[i] = copy_of_inner(display[i]);
+                display[i]->inner()[(old_index >> shift) & mask<B>] =
+                    display[i - 1];
+                shift += B;
+            }
+            for (auto i = display_idx - 1; i > 0; --i) {
+                shift -= B;
+                display[i] = null_slot_and_copy_inner(
+                    display[i + 1], (new_index >> shift) & mask<B>);
+            }
+            display[0] =
+                null_slot_and_copy_leaf(display[1], (new_index >> B) & mask<B>);
+        }
+    }
+
+    void goto_fresh_pos_writable_from_clean(std::size_t old_index,
+                                            std::size_t new_index,
+                                            std::size_t xr)
+    {
+        auto display_idx = fast_log2(xr) / B;
+        if (display_idx > 0) {
+            auto shift = display_idx * B;
+            if (display_idx == depth) {
+                display[display_idx] = make_node(inner_t{});
+                display[display_idx]->inner()[(old_index >> shift) & mask<B>] =
+                    display[display_idx - 1];
+                ++depth;
+            }
+            while (--display_idx) {
+                auto node = display[display_idx + 1]
+                                ->inner()[(new_index >> shift) & mask<B>];
+                display[display_idx] =
+                    node ? std::move(node) : make_node(inner_t{});
+            }
+            display[0] = make_node(leaf_t{});
+        }
+    }
+
+    void goto_fresh_pos_writable_from_dirty(std::size_t old_index,
+                                            std::size_t new_index,
+                                            std::size_t xr)
+    {
+        stabilize(old_index);
+        goto_fresh_pos_writable_from_clean(old_index, new_index, xr);
+    }
+
+    void goto_next_block_start(std::size_t index, std::size_t xr)
+    {
+        auto display_idx = fast_log2(xr) / B;
+        auto shift       = display_idx * B;
+        if (display_idx > 0) {
+            display[display_idx - 1] =
+                display[display_idx]->inner()[(index >> shift) & mask<B>];
+            while (--display_idx)
+                display[display_idx - 1] = display[display_idx]->inner()[0];
+        }
+    }
+
+    void goto_pos(std::size_t index, std::size_t xr)
+    {
+        auto display_idx = fast_log2(xr) / B;
+        auto shift       = display_idx * B;
+        if (display_idx) {
+            do {
+                display[display_idx - 1] =
+                    display[display_idx]->inner()[(index >> shift) & mask<B>];
+                shift -= B;
+            } while (--display_idx);
+        }
+    }
+};
+
+template <typename T, int B, typename MP>
+struct impl
+{
+    using inner_t    = inner_node<T, B, MP>;
+    using leaf_t     = leaf_node<T, B>;
+    using node_t     = node<T, B, MP>;
+    using node_ptr_t = node_ptr<T, B, MP>;
+    using ref_t      = ref<T, B, MP>;
+
+    std::size_t size;
+    std::size_t focus;
+    bool dirty;
+    ref_t p;
+
+    template <typename... Ts>
+    static auto make_node(Ts&&... xs)
+    {
+        return dvektor::make_node<T, B, MP>(std::forward<Ts>(xs)...);
+    }
+
+    void goto_pos_writable(std::size_t old_index,
+                           std::size_t new_index,
+                           std::size_t xr)
+    {
+        if (dirty) {
+            p.goto_pos_writable_from_dirty(old_index, new_index, xr);
+        } else {
+            p.goto_pos_writable_from_clean(old_index, new_index, xr);
+            dirty = true;
+        }
+    }
+
+    void goto_fresh_pos_writable(std::size_t old_index,
+                                 std::size_t new_index,
+                                 std::size_t xr)
+    {
+        if (dirty) {
+            p.goto_fresh_pos_writable_from_dirty(old_index, new_index, xr);
+        } else {
+            p.goto_fresh_pos_writable_from_clean(old_index, new_index, xr);
+            dirty = true;
+        }
+    }
+
+    impl push_back(T value) const
+    {
+        if (size) {
+            auto block_index = size & ~mask<B>;
+            auto lo          = size & mask<B>;
+            if (size != block_index) {
+                auto s = impl{size + 1, block_index, dirty, p};
+                s.goto_pos_writable(focus, block_index, focus ^ block_index);
+                s.p.display[0]->leaf()[lo] = std::move(value);
+                return s;
+            } else {
+                auto s = impl{size + 1, block_index, dirty, p};
+                s.goto_fresh_pos_writable(
+                    focus, block_index, focus ^ block_index);
+                s.p.display[0]->leaf()[lo] = std::move(value);
+                return s;
+            }
+        } else {
+            return impl{
+                1, 0, false, {1, {{make_node(leaf_t{{std::move(value)}})}}}};
+        }
+    }
+
+    const T& get(std::size_t index) const
+    {
+        return p.get_elem(index, index ^ focus);
+    }
+
+    template <typename FnT>
+    impl update(std::size_t idx, FnT&& fn) const
+    {
+        auto s = impl{size, idx, dirty, p};
+        s.goto_pos_writable(focus, idx, focus ^ idx);
+        auto& v = s.p.display[0]->leaf()[idx & mask<B>];
+        v       = fn(std::move(v));
+        return s;
+    }
+
+    impl assoc(std::size_t idx, T value) const
+    {
+        return update(idx, [&](auto&&) { return std::move(value); });
+    }
+};
+
+template <typename T, int B, typename MP>
+const impl<T, B, MP> empty = {0, 0, false, ref<T, B, MP>{1, {}}};
+
+template <typename T, int B, typename MP>
+struct iterator
+    : boost::iterator_facade<iterator<T, B, MP>,
+                             T,
+                             boost::random_access_traversal_tag,
+                             const T&>
+{
+    struct end_t
+    {};
+
+    iterator() = default;
+
+    iterator(const impl<T, B, MP>& v)
+        : p_{v.p}
+        , i_{0}
+        , base_{0}
+    {
+        if (v.dirty)
+            p_.stabilize(v.focus);
+        p_.goto_pos(0, 0 ^ v.focus);
+        curr_ = p_.display[0]->leaf().begin();
+    }
+
+    iterator(const impl<T, B, MP>& v, end_t)
+        : p_{v.p}
+        , i_{v.size}
+        , base_{(v.size - 1) & ~mask<B>}
+    {
+        if (v.dirty)
+            p_.stabilize(v.focus);
+        p_.goto_pos(base_, base_ ^ v.focus);
+        curr_ = p_.display[0]->leaf().begin() + (i_ - base_);
+    }
+
+private:
+    friend class boost::iterator_core_access;
+    using leaf_iterator = typename leaf_node<T, B>::const_iterator;
+
+    ref<T, B, MP> p_;
+    std::size_t i_;
+    std::size_t base_;
+    leaf_iterator curr_;
+
+    void increment()
+    {
+        ++i_;
+        if (i_ - base_ < branches<B>) {
+            ++curr_;
+        } else {
+            auto new_base = base_ + branches<B>;
+            p_.goto_next_block_start(new_base, base_ ^ new_base);
+            base_ = new_base;
+            curr_ = p_.display[0]->leaf().begin();
+        }
+    }
+
+    void decrement()
+    {
+        assert(i_ > 0);
+        --i_;
+        if (i_ >= base_) {
+            --curr_;
+        } else {
+            auto new_base = base_ - branches<B>;
+            p_.goto_pos(new_base, base_ ^ new_base);
+            base_ = new_base;
+            curr_ = std::prev(p_.display[0]->leaf().end());
+        }
+    }
+
+    void advance(std::ptrdiff_t n)
+    {
+        i_ += n;
+        if (i_ <= base_ && i_ - base_ < branches<B>) {
+            curr_ += n;
+        } else {
+            auto new_base = i_ & ~mask<B>;
+            p_.goto_pos(new_base, base_ ^ new_base);
+            base_ = new_base;
+            curr_ = p_.display[0]->leaf().begin() + (i_ - base_);
+        }
+    }
+
+    bool equal(const iterator& other) const { return i_ == other.i_; }
+
+    std::ptrdiff_t distance_to(const iterator& other) const
+    {
+        return other.i_ > i_ ? static_cast<std::ptrdiff_t>(other.i_ - i_)
+                             : -static_cast<std::ptrdiff_t>(i_ - other.i_);
+    }
+
+    const T& dereference() const { return *curr_; }
+};
+
+} /* namespace dvektor */
+} /* namespace detail */
+} /* namespace immer */
diff --git a/immer/experimental/dvektor.hpp b/immer/experimental/dvektor.hpp
new file mode 100644
index 000000000000..e809e749d5df
--- /dev/null
+++ b/immer/experimental/dvektor.hpp
@@ -0,0 +1,69 @@
+//
+// immer: immutable data structures for C++
+// Copyright (C) 2016, 2017, 2018 Juan Pedro Bolivar Puente
+//
+// This software is distributed under the Boost Software License, Version 1.0.
+// See accompanying file LICENSE or copy at http://boost.org/LICENSE_1_0.txt
+//
+
+#pragma once
+
+#include <immer/experimental/detail/dvektor_impl.hpp>
+
+#include <immer/memory_policy.hpp>
+
+namespace immer {
+
+template <typename T, int B = 5, typename MemoryPolicy = default_memory_policy>
+class dvektor
+{
+    using impl_t = detail::dvektor::impl<T, B, MemoryPolicy>;
+
+public:
+    using value_type      = T;
+    using reference       = const T&;
+    using size_type       = std::size_t;
+    using difference_type = std::ptrdiff_t;
+    using const_reference = const T&;
+
+    using iterator         = detail::dvektor::iterator<T, B, MemoryPolicy>;
+    using const_iterator   = iterator;
+    using reverse_iterator = std::reverse_iterator<iterator>;
+
+    dvektor() = default;
+
+    iterator begin() const { return {impl_}; }
+    iterator end() const { return {impl_, typename iterator::end_t{}}; }
+
+    reverse_iterator rbegin() const { return reverse_iterator{end()}; }
+    reverse_iterator rend() const { return reverse_iterator{begin()}; }
+
+    std::size_t size() const { return impl_.size; }
+    bool empty() const { return impl_.size == 0; }
+
+    reference operator[](size_type index) const { return impl_.get(index); }
+
+    dvektor push_back(value_type value) const
+    {
+        return {impl_.push_back(std::move(value))};
+    }
+
+    dvektor assoc(std::size_t idx, value_type value) const
+    {
+        return {impl_.assoc(idx, std::move(value))};
+    }
+
+    template <typename FnT>
+    dvektor update(std::size_t idx, FnT&& fn) const
+    {
+        return {impl_.update(idx, std::forward<FnT>(fn))};
+    }
+
+private:
+    dvektor(impl_t impl)
+        : impl_(std::move(impl))
+    {}
+    impl_t impl_ = detail::dvektor::empty<T, B, MemoryPolicy>;
+};
+
+} // namespace immer