early-access version 4115

README.md

@@ -1,7 +1,7 @@
 yuzu emulator early access
 =============
 
-This is the source code for early-access 4114.
+This is the source code for early-access 4115.
 
 ## Legal Notice
 

src/common/CMakeLists.txt

@@ -107,6 +107,8 @@ add_library(common STATIC
     quaternion.h
     range_map.h
     range_mutex.h
+    range_sets.h
+    range_sets.inc
     reader_writer_queue.h
     ring_buffer.h
     ${CMAKE_CURRENT_BINARY_DIR}/scm_rev.cpp
@@ -121,6 +123,7 @@ add_library(common STATIC
     settings_input.cpp
     settings_input.h
     settings_setting.h
+    slot_vector.h
     socket_types.h
     spin_lock.cpp
     spin_lock.h

src/common/range_sets.h

@@ -0,0 +1,73 @@
+// SPDX-FileCopyrightText: 2024 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#include <memory>
+
+#include "common/common_types.h"
+
+namespace Common {
+
+template <typename AddressType>
+class RangeSet {
+public:
+    RangeSet();
+    ~RangeSet();
+
+    RangeSet(RangeSet const&) = delete;
+    RangeSet& operator=(RangeSet const&) = delete;
+
+    RangeSet(RangeSet&& other);
+    RangeSet& operator=(RangeSet&& other);
+
+    void Add(AddressType base_address, size_t size);
+    void Subtract(AddressType base_address, size_t size);
+    void Clear();
+    bool Empty() const;
+
+    template <typename Func>
+    void ForEach(Func&& func) const;
+
+    template <typename Func>
+    void ForEachInRange(AddressType device_addr, size_t size, Func&& func) const;
+
+private:
+    struct RangeSetImpl;
+    std::unique_ptr<RangeSetImpl> m_impl;
+};
+
+template <typename AddressType>
+class OverlapRangeSet {
+public:
+    OverlapRangeSet();
+    ~OverlapRangeSet();
+
+    OverlapRangeSet(OverlapRangeSet const&) = delete;
+    OverlapRangeSet& operator=(OverlapRangeSet const&) = delete;
+
+    OverlapRangeSet(OverlapRangeSet&& other);
+    OverlapRangeSet& operator=(OverlapRangeSet&& other);
+
+    void Add(AddressType base_address, size_t size);
+    void Subtract(AddressType base_address, size_t size);
+
+    template <typename Func>
+    void Subtract(AddressType base_address, size_t size, Func&& on_delete);
+
+    void DeleteAll(AddressType base_address, size_t size);
+    void Clear();
+    bool Empty() const;
+
+    template <typename Func>
+    void ForEach(Func&& func) const;
+
+    template <typename Func>
+    void ForEachInRange(AddressType device_addr, size_t size, Func&& func) const;
+
+private:
+    struct OverlapRangeSetImpl;
+    std::unique_ptr<OverlapRangeSetImpl> m_impl;
+};
+
+} // namespace Common

src/common/range_sets.inc

@@ -0,0 +1,304 @@
+// SPDX-FileCopyrightText: 2024 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#include <limits>
+#include <utility>
+
+#include <boost/icl/interval.hpp>
+#include <boost/icl/interval_base_set.hpp>
+#include <boost/icl/interval_map.hpp>
+#include <boost/icl/interval_set.hpp>
+#include <boost/icl/split_interval_map.hpp>
+#include <boost/pool/pool.hpp>
+#include <boost/pool/pool_alloc.hpp>
+#include <boost/pool/poolfwd.hpp>
+
+#include "common/range_sets.h"
+
+namespace Common {
+
+namespace {
+template <class T>
+using RangeSetsAllocator =
+    boost::fast_pool_allocator<T, boost::default_user_allocator_new_delete,
+                               boost::details::pool::default_mutex, 1024, 2048>;
+}
+
+template <typename AddressType>
+struct RangeSet<AddressType>::RangeSetImpl {
+    using IntervalSet = boost::icl::interval_set<
+        AddressType, std::less, ICL_INTERVAL_INSTANCE(ICL_INTERVAL_DEFAULT, AddressType, std::less),
+        RangeSetsAllocator>;
+    using IntervalType = typename IntervalSet::interval_type;
+
+    RangeSetImpl() = default;
+    ~RangeSetImpl() = default;
+
+    void Add(AddressType base_address, size_t size) {
+        AddressType end_address = base_address + static_cast<AddressType>(size);
+        IntervalType interval{base_address, end_address};
+        m_ranges_set.add(interval);
+    }
+
+    void Subtract(AddressType base_address, size_t size) {
+        AddressType end_address = base_address + static_cast<AddressType>(size);
+        IntervalType interval{base_address, end_address};
+        m_ranges_set.subtract(interval);
+    }
+
+    template <typename Func>
+    void ForEach(Func&& func) const {
+        if (m_ranges_set.empty()) {
+            return;
+        }
+        auto it = m_ranges_set.begin();
+        auto end_it = m_ranges_set.end();
+        for (; it != end_it; it++) {
+            const AddressType inter_addr_end = it->upper();
+            const AddressType inter_addr = it->lower();
+            func(inter_addr, inter_addr_end);
+        }
+    }
+
+    template <typename Func>
+    void ForEachInRange(AddressType base_addr, size_t size, Func&& func) const {
+        if (m_ranges_set.empty()) {
+            return;
+        }
+        const AddressType start_address = base_addr;
+        const AddressType end_address = start_address + size;
+        const RangeSetImpl::IntervalType search_interval{start_address, end_address};
+        auto it = m_ranges_set.lower_bound(search_interval);
+        if (it == m_ranges_set.end()) {
+            return;
+        }
+        auto end_it = m_ranges_set.upper_bound(search_interval);
+        for (; it != end_it; it++) {
+            AddressType inter_addr_end = it->upper();
+            AddressType inter_addr = it->lower();
+            if (inter_addr_end > end_address) {
+                inter_addr_end = end_address;
+            }
+            if (inter_addr < start_address) {
+                inter_addr = start_address;
+            }
+            func(inter_addr, inter_addr_end);
+        }
+    }
+
+    IntervalSet m_ranges_set;
+};
+
+template <typename AddressType>
+struct OverlapRangeSet<AddressType>::OverlapRangeSetImpl {
+    using IntervalSet = boost::icl::split_interval_map<
+        AddressType, s32, boost::icl::partial_enricher, std::less, boost::icl::inplace_plus,
+        boost::icl::inter_section,
+        ICL_INTERVAL_INSTANCE(ICL_INTERVAL_DEFAULT, AddressType, std::less), RangeSetsAllocator>;
+    using IntervalType = typename IntervalSet::interval_type;
+
+    OverlapRangeSetImpl() = default;
+    ~OverlapRangeSetImpl() = default;
+
+    void Add(AddressType base_address, size_t size) {
+        AddressType end_address = base_address + static_cast<AddressType>(size);
+        IntervalType interval{base_address, end_address};
+        m_split_ranges_set += std::make_pair(interval, 1);
+    }
+
+    template <bool has_on_delete, typename Func>
+    void Subtract(AddressType base_address, size_t size, s32 amount,
+                  [[maybe_unused]] Func&& on_delete) {
+        if (m_split_ranges_set.empty()) {
+            return;
+        }
+        AddressType end_address = base_address + static_cast<AddressType>(size);
+        IntervalType interval{base_address, end_address};
+        bool any_removals = false;
+        m_split_ranges_set += std::make_pair(interval, -amount);
+        do {
+            any_removals = false;
+            auto it = m_split_ranges_set.lower_bound(interval);
+            if (it == m_split_ranges_set.end()) {
+                return;
+            }
+            auto end_it = m_split_ranges_set.upper_bound(interval);
+            for (; it != end_it; it++) {
+                if (it->second <= 0) {
+                    if constexpr (has_on_delete) {
+                        if (it->second == 0) {
+                            on_delete(it->first.lower(), it->first.upper());
+                        }
+                    }
+                    any_removals = true;
+                    m_split_ranges_set.erase(it);
+                    break;
+                }
+            }
+        } while (any_removals);
+    }
+
+    template <typename Func>
+    void ForEach(Func&& func) const {
+        if (m_split_ranges_set.empty()) {
+            return;
+        }
+        auto it = m_split_ranges_set.begin();
+        auto end_it = m_split_ranges_set.end();
+        for (; it != end_it; it++) {
+            const AddressType inter_addr_end = it->first.upper();
+            const AddressType inter_addr = it->first.lower();
+            func(inter_addr, inter_addr_end, it->second);
+        }
+    }
+
+    template <typename Func>
+    void ForEachInRange(AddressType base_address, size_t size, Func&& func) const {
+        if (m_split_ranges_set.empty()) {
+            return;
+        }
+        const AddressType start_address = base_address;
+        const AddressType end_address = start_address + size;
+        const OverlapRangeSetImpl::IntervalType search_interval{start_address, end_address};
+        auto it = m_split_ranges_set.lower_bound(search_interval);
+        if (it == m_split_ranges_set.end()) {
+            return;
+        }
+        auto end_it = m_split_ranges_set.upper_bound(search_interval);
+        for (; it != end_it; it++) {
+            auto& inter = it->first;
+            AddressType inter_addr_end = inter.upper();
+            AddressType inter_addr = inter.lower();
+            if (inter_addr_end > end_address) {
+                inter_addr_end = end_address;
+            }
+            if (inter_addr < start_address) {
+                inter_addr = start_address;
+            }
+            func(inter_addr, inter_addr_end, it->second);
+        }
+    }
+
+    IntervalSet m_split_ranges_set;
+};
+
+template <typename AddressType>
+RangeSet<AddressType>::RangeSet() {
+    m_impl = std::make_unique<RangeSet<AddressType>::RangeSetImpl>();
+}
+
+template <typename AddressType>
+RangeSet<AddressType>::~RangeSet() = default;
+
+template <typename AddressType>
+RangeSet<AddressType>::RangeSet(RangeSet&& other) {
+    m_impl = std::make_unique<RangeSet<AddressType>::RangeSetImpl>();
+    m_impl->m_ranges_set = std::move(other.m_impl->m_ranges_set);
+}
+
+template <typename AddressType>
+RangeSet<AddressType>& RangeSet<AddressType>::operator=(RangeSet&& other) {
+    m_impl->m_ranges_set = std::move(other.m_impl->m_ranges_set);
+}
+
+template <typename AddressType>
+void RangeSet<AddressType>::Add(AddressType base_address, size_t size) {
+    m_impl->Add(base_address, size);
+}
+
+template <typename AddressType>
+void RangeSet<AddressType>::Subtract(AddressType base_address, size_t size) {
+    m_impl->Subtract(base_address, size);
+}
+
+template <typename AddressType>
+void RangeSet<AddressType>::Clear() {
+    m_impl->m_ranges_set.clear();
+}
+
+template <typename AddressType>
+bool RangeSet<AddressType>::Empty() const {
+    return m_impl->m_ranges_set.empty();
+}
+
+template <typename AddressType>
+template <typename Func>
+void RangeSet<AddressType>::ForEach(Func&& func) const {
+    m_impl->ForEach(std::move(func));
+}
+
+template <typename AddressType>
+template <typename Func>
+void RangeSet<AddressType>::ForEachInRange(AddressType base_address, size_t size,
+                                           Func&& func) const {
+    m_impl->ForEachInRange(base_address, size, std::move(func));
+}
+
+template <typename AddressType>
+OverlapRangeSet<AddressType>::OverlapRangeSet() {
+    m_impl = std::make_unique<OverlapRangeSet<AddressType>::OverlapRangeSetImpl>();
+}
+
+template <typename AddressType>
+OverlapRangeSet<AddressType>::~OverlapRangeSet() = default;
+
+template <typename AddressType>
+OverlapRangeSet<AddressType>::OverlapRangeSet(OverlapRangeSet&& other) {
+    m_impl = std::make_unique<OverlapRangeSet<AddressType>::OverlapRangeSetImpl>();
+    m_impl->m_split_ranges_set = std::move(other.m_impl->m_split_ranges_set);
+}
+
+template <typename AddressType>
+OverlapRangeSet<AddressType>& OverlapRangeSet<AddressType>::operator=(OverlapRangeSet&& other) {
+    m_impl->m_split_ranges_set = std::move(other.m_impl->m_split_ranges_set);
+}
+
+template <typename AddressType>
+void OverlapRangeSet<AddressType>::Add(AddressType base_address, size_t size) {
+    m_impl->Add(base_address, size);
+}
+
+template <typename AddressType>
+void OverlapRangeSet<AddressType>::Subtract(AddressType base_address, size_t size) {
+    m_impl->template Subtract<false>(base_address, size, 1, [](AddressType, AddressType) {});
+}
+
+template <typename AddressType>
+template <typename Func>
+void OverlapRangeSet<AddressType>::Subtract(AddressType base_address, size_t size,
+                                            Func&& on_delete) {
+    m_impl->template Subtract<true, Func>(base_address, size, 1, std::move(on_delete));
+}
+
+template <typename AddressType>
+void OverlapRangeSet<AddressType>::DeleteAll(AddressType base_address, size_t size) {
+    m_impl->template Subtract<false>(base_address, size, std::numeric_limits<s32>::max(),
+                                     [](AddressType, AddressType) {});
+}
+
+template <typename AddressType>
+void OverlapRangeSet<AddressType>::Clear() {
+    m_impl->m_split_ranges_set.clear();
+}
+
+template <typename AddressType>
+bool OverlapRangeSet<AddressType>::Empty() const {
+    return m_impl->m_split_ranges_set.empty();
+}
+
+template <typename AddressType>
+template <typename Func>
+void OverlapRangeSet<AddressType>::ForEach(Func&& func) const {
+    m_impl->ForEach(func);
+}
+
+template <typename AddressType>
+template <typename Func>
+void OverlapRangeSet<AddressType>::ForEachInRange(AddressType base_address, size_t size,
+                                                  Func&& func) const {
+    m_impl->ForEachInRange(base_address, size, std::move(func));
+}
+
+} // namespace Common

src/common/slot_vector.h

@@ -0,0 +1,227 @@
+// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#include <algorithm>
+#include <bit>
+#include <numeric>
+#include <type_traits>
+#include <utility>
+#include <vector>
+
+#include "common/assert.h"
+#include "common/common_types.h"
+#include "common/polyfill_ranges.h"
+
+namespace Common {
+
+struct SlotId {
+    static constexpr u32 INVALID_INDEX = std::numeric_limits<u32>::max();
+
+    constexpr auto operator<=>(const SlotId&) const noexcept = default;
+
+    constexpr explicit operator bool() const noexcept {
+        return index != INVALID_INDEX;
+    }
+
+    u32 index = INVALID_INDEX;
+};
+
+template <class T>
+    requires std::is_nothrow_move_assignable_v<T> && std::is_nothrow_move_constructible_v<T>
+class SlotVector {
+public:
+    class Iterator {
+        friend SlotVector<T>;
+
+    public:
+        constexpr Iterator() = default;
+
+        Iterator& operator++() noexcept {
+            const u64* const bitset = slot_vector->stored_bitset.data();
+            const u32 size = static_cast<u32>(slot_vector->stored_bitset.size()) * 64;
+            if (id.index < size) {
+                do {
+                    ++id.index;
+                } while (id.index < size && !IsValid(bitset));
+                if (id.index == size) {
+                    id.index = SlotId::INVALID_INDEX;
+                }
+            }
+            return *this;
+        }
+
+        Iterator operator++(int) noexcept {
+            const Iterator copy{*this};
+            ++*this;
+            return copy;
+        }
+
+        bool operator==(const Iterator& other) const noexcept {
+            return id.index == other.id.index;
+        }
+
+        bool operator!=(const Iterator& other) const noexcept {
+            return id.index != other.id.index;
+        }
+
+        std::pair<SlotId, T*> operator*() const noexcept {
+            return {id, std::addressof((*slot_vector)[id])};
+        }
+
+        T* operator->() const noexcept {
+            return std::addressof((*slot_vector)[id]);
+        }
+
+    private:
+        Iterator(SlotVector<T>* slot_vector_, SlotId id_) noexcept
+            : slot_vector{slot_vector_}, id{id_} {}
+
+        bool IsValid(const u64* bitset) const noexcept {
+            return ((bitset[id.index / 64] >> (id.index % 64)) & 1) != 0;
+        }
+
+        SlotVector<T>* slot_vector;
+        SlotId id;
+    };
+
+    ~SlotVector() noexcept {
+        size_t index = 0;
+        for (u64 bits : stored_bitset) {
+            for (size_t bit = 0; bits; ++bit, bits >>= 1) {
+                if ((bits & 1) != 0) {
+                    values[index + bit].object.~T();
+                }
+            }
+            index += 64;
+        }
+        delete[] values;
+    }
+
+    [[nodiscard]] T& operator[](SlotId id) noexcept {
+        ValidateIndex(id);
+        return values[id.index].object;
+    }
+
+    [[nodiscard]] const T& operator[](SlotId id) const noexcept {
+        ValidateIndex(id);
+        return values[id.index].object;
+    }
+
+    template <typename... Args>
+    [[nodiscard]] SlotId insert(Args&&... args) noexcept {
+        const u32 index = FreeValueIndex();
+        new (&values[index].object) T(std::forward<Args>(args)...);
+        SetStorageBit(index);
+
+        return SlotId{index};
+    }
+
+    void erase(SlotId id) noexcept {
+        values[id.index].object.~T();
+        free_list.push_back(id.index);
+        ResetStorageBit(id.index);
+    }
+
+    [[nodiscard]] Iterator begin() noexcept {
+        const auto it = std::ranges::find_if(stored_bitset, [](u64 value) { return value != 0; });
+        if (it == stored_bitset.end()) {
+            return end();
+        }
+        const u32 word_index = static_cast<u32>(std::distance(it, stored_bitset.begin()));
+        const SlotId first_id{word_index * 64 + static_cast<u32>(std::countr_zero(*it))};
+        return Iterator(this, first_id);
+    }
+
+    [[nodiscard]] Iterator end() noexcept {
+        return Iterator(this, SlotId{SlotId::INVALID_INDEX});
+    }
+
+    [[nodiscard]] size_t size() const noexcept {
+        return values_capacity - free_list.size();
+    }
+
+private:
+    struct NonTrivialDummy {
+        NonTrivialDummy() noexcept {}
+    };
+
+    union Entry {
+        Entry() noexcept : dummy{} {}
+        ~Entry() noexcept {}
+
+        NonTrivialDummy dummy;
+        T object;
+    };
+
+    void SetStorageBit(u32 index) noexcept {
+        stored_bitset[index / 64] |= u64(1) << (index % 64);
+    }
+
+    void ResetStorageBit(u32 index) noexcept {
+        stored_bitset[index / 64] &= ~(u64(1) << (index % 64));
+    }
+
+    bool ReadStorageBit(u32 index) noexcept {
+        return ((stored_bitset[index / 64] >> (index % 64)) & 1) != 0;
+    }
+
+    void ValidateIndex(SlotId id) const noexcept {
+        DEBUG_ASSERT(id);
+        DEBUG_ASSERT(id.index / 64 < stored_bitset.size());
+        DEBUG_ASSERT(((stored_bitset[id.index / 64] >> (id.index % 64)) & 1) != 0);
+    }
+
+    [[nodiscard]] u32 FreeValueIndex() noexcept {
+        if (free_list.empty()) {
+            Reserve(values_capacity ? (values_capacity << 1) : 1);
+        }
+        const u32 free_index = free_list.back();
+        free_list.pop_back();
+        return free_index;
+    }
+
+    void Reserve(size_t new_capacity) noexcept {
+        Entry* const new_values = new Entry[new_capacity];
+        size_t index = 0;
+        for (u64 bits : stored_bitset) {
+            for (size_t bit = 0; bits; ++bit, bits >>= 1) {
+                const size_t i = index + bit;
+                if ((bits & 1) == 0) {
+                    continue;
+                }
+                T& old_value = values[i].object;
+                new (&new_values[i].object) T(std::move(old_value));
+                old_value.~T();
+            }
+            index += 64;
+        }
+
+        stored_bitset.resize((new_capacity + 63) / 64);
+
+        const size_t old_free_size = free_list.size();
+        free_list.resize(old_free_size + (new_capacity - values_capacity));
+        std::iota(free_list.begin() + old_free_size, free_list.end(),
+                  static_cast<u32>(values_capacity));
+
+        delete[] values;
+        values = new_values;
+        values_capacity = new_capacity;
+    }
+
+    Entry* values = nullptr;
+    size_t values_capacity = 0;
+
+    std::vector<u64> stored_bitset;
+    std::vector<u32> free_list;
+};
+
+} // namespace Common
+
+template <>
+struct std::hash<Common::SlotId> {
+    size_t operator()(const Common::SlotId& id) const noexcept {
+        return std::hash<u32>{}(id.index);
+    }
+};

src/core/hle/service/nvdrv/core/heap_mapper.cpp

@@ -3,110 +3,21 @@
 
 #include <mutex>
 
-#include <boost/container/small_vector.hpp>
+#include "common/range_sets.h"
-#define BOOST_NO_MT
+#include "common/range_sets.inc"
-#include <boost/pool/detail/mutex.hpp>
-#undef BOOST_NO_MT
-#include <boost/icl/interval.hpp>
-#include <boost/icl/interval_base_set.hpp>
-#include <boost/icl/interval_set.hpp>
-#include <boost/icl/split_interval_map.hpp>
-#include <boost/pool/pool.hpp>
-#include <boost/pool/pool_alloc.hpp>
-#include <boost/pool/poolfwd.hpp>
-
 #include "core/hle/service/nvdrv/core/heap_mapper.h"
 #include "video_core/host1x/host1x.h"
 
-namespace boost {
-template <typename T>
-class fast_pool_allocator<T, default_user_allocator_new_delete, details::pool::null_mutex, 4096, 0>;
-}
-
 namespace Service::Nvidia::NvCore {
 
-using IntervalCompare = std::less<DAddr>;
-using IntervalInstance = boost::icl::interval_type_default<DAddr, std::less>;
-using IntervalAllocator = boost::fast_pool_allocator<DAddr>;
-using IntervalSet = boost::icl::interval_set<DAddr>;
-using IntervalType = typename IntervalSet::interval_type;
-
-template <typename Type>
-struct counter_add_functor : public boost::icl::identity_based_inplace_combine<Type> {
-    // types
-    typedef counter_add_functor<Type> type;
-    typedef boost::icl::identity_based_inplace_combine<Type> base_type;
-
-    // public member functions
-    void operator()(Type& current, const Type& added) const {
-        current += added;
-        if (current < base_type::identity_element()) {
-            current = base_type::identity_element();
-        }
-    }
-
-    // public static functions
-    static void version(Type&){};
-};
-
-using OverlapCombine = counter_add_functor<int>;
-using OverlapSection = boost::icl::inter_section<int>;
-using OverlapCounter = boost::icl::split_interval_map<DAddr, int>;
-
 struct HeapMapper::HeapMapperInternal {
-    HeapMapperInternal(Tegra::Host1x::Host1x& host1x) : device_memory{host1x.MemoryManager()} {}
+    HeapMapperInternal(Tegra::Host1x::Host1x& host1x) : m_device_memory{host1x.MemoryManager()} {}
     ~HeapMapperInternal() = default;
 
-    template <typename Func>
+    Common::RangeSet<VAddr> m_temporary_set;
-    void ForEachInOverlapCounter(OverlapCounter& current_range, VAddr cpu_addr, u64 size,
+    Common::OverlapRangeSet<VAddr> m_mapped_ranges;
-                                 Func&& func) {
+    Tegra::MaxwellDeviceMemoryManager& m_device_memory;
-        const DAddr start_address = cpu_addr;
+    std::mutex m_guard;
-        const DAddr end_address = start_address + size;
-        const IntervalType search_interval{start_address, end_address};
-        auto it = current_range.lower_bound(search_interval);
-        if (it == current_range.end()) {
-            return;
-        }
-        auto end_it = current_range.upper_bound(search_interval);
-        for (; it != end_it; it++) {
-            auto& inter = it->first;
-            DAddr inter_addr_end = inter.upper();
-            DAddr inter_addr = inter.lower();
-            if (inter_addr_end > end_address) {
-                inter_addr_end = end_address;
-            }
-            if (inter_addr < start_address) {
-                inter_addr = start_address;
-            }
-            func(inter_addr, inter_addr_end, it->second);
-        }
-    }
-
-    void RemoveEachInOverlapCounter(OverlapCounter& current_range,
-                                    const IntervalType search_interval, int subtract_value) {
-        bool any_removals = false;
-        current_range.add(std::make_pair(search_interval, subtract_value));
-        do {
-            any_removals = false;
-            auto it = current_range.lower_bound(search_interval);
-            if (it == current_range.end()) {
-                return;
-            }
-            auto end_it = current_range.upper_bound(search_interval);
-            for (; it != end_it; it++) {
-                if (it->second <= 0) {
-                    any_removals = true;
-                    current_range.erase(it);
-                    break;
-                }
-            }
-        } while (any_removals);
-    }
-
-    IntervalSet base_set;
-    OverlapCounter mapping_overlaps;
-    Tegra::MaxwellDeviceMemoryManager& device_memory;
-    std::mutex guard;
 };
 
 HeapMapper::HeapMapper(VAddr start_vaddress, DAddr start_daddress, size_t size, Core::Asid asid,
@@ -116,60 +27,48 @@ HeapMapper::HeapMapper(VAddr start_vaddress, DAddr start_daddress, size_t size,
 }
 
 HeapMapper::~HeapMapper() {
-    m_internal->device_memory.Unmap(m_daddress, m_size);
+    // Unmap whatever has been mapped.
+    m_internal->m_mapped_ranges.ForEach([this](VAddr start_addr, VAddr end_addr, s32 count) {
+        const size_t sub_size = end_addr - start_addr;
+        const size_t offset = start_addr - m_vaddress;
+        m_internal->m_device_memory.Unmap(m_daddress + offset, sub_size);
+    });
 }
 
 DAddr HeapMapper::Map(VAddr start, size_t size) {
-    std::scoped_lock lk(m_internal->guard);
+    std::scoped_lock lk(m_internal->m_guard);
-    m_internal->base_set.clear();
+    // Add the mapping range to a temporary range set.
-    const IntervalType interval{start, start + size};
+    m_internal->m_temporary_set.Clear();
-    m_internal->base_set.insert(interval);
+    m_internal->m_temporary_set.Add(start, size);
-    m_internal->ForEachInOverlapCounter(m_internal->mapping_overlaps, start, size,
+
-                                        [this](VAddr start_addr, VAddr end_addr, int) {
+    // Remove anything that's already mapped from the temporary range set.
-                                            const IntervalType other{start_addr, end_addr};
+    m_internal->m_mapped_ranges.ForEachInRange(
-                                            m_internal->base_set.subtract(other);
+        start, size, [this](VAddr start_addr, VAddr end_addr, s32) {
-                                        });
+            m_internal->m_temporary_set.Subtract(start_addr, end_addr - start_addr);
-    if (!m_internal->base_set.empty()) {
+        });
-        auto it = m_internal->base_set.begin();
+
-        auto end_it = m_internal->base_set.end();
+    // Map anything that has not been mapped yet.
-        for (; it != end_it; it++) {
+    m_internal->m_temporary_set.ForEach([this](VAddr start_addr, VAddr end_addr) {
-            const VAddr inter_addr_end = it->upper();
+        const size_t sub_size = end_addr - start_addr;
-            const VAddr inter_addr = it->lower();
+        const size_t offset = start_addr - m_vaddress;
-            const size_t offset = inter_addr - m_vaddress;
+        m_internal->m_device_memory.Map(m_daddress + offset, m_vaddress + offset, sub_size, m_asid);
-            const size_t sub_size = inter_addr_end - inter_addr;
+    });
-            m_internal->device_memory.Map(m_daddress + offset, m_vaddress + offset, sub_size,
+
-                                          m_asid);
+    // Add the mapping range to the split map, to register the map and overlaps.
-        }
+    m_internal->m_mapped_ranges.Add(start, size);
-    }
+    m_internal->m_temporary_set.Clear();
-    m_internal->mapping_overlaps += std::make_pair(interval, 1);
+    return m_daddress + static_cast<DAddr>(start - m_vaddress);
-    m_internal->base_set.clear();
-    return m_daddress + (start - m_vaddress);
 }
 
 void HeapMapper::Unmap(VAddr start, size_t size) {
-    std::scoped_lock lk(m_internal->guard);
+    std::scoped_lock lk(m_internal->m_guard);
-    m_internal->base_set.clear();
+
-    m_internal->ForEachInOverlapCounter(m_internal->mapping_overlaps, start, size,
+    // Just subtract the range and whatever is deleted, unmap it.
-                                        [this](VAddr start_addr, VAddr end_addr, int value) {
+    m_internal->m_mapped_ranges.Subtract(start, size, [this](VAddr start_addr, VAddr end_addr) {
-                                            if (value <= 1) {
+        const size_t sub_size = end_addr - start_addr;
-                                                const IntervalType other{start_addr, end_addr};
+        const size_t offset = start_addr - m_vaddress;
-                                                m_internal->base_set.insert(other);
+        m_internal->m_device_memory.Unmap(m_daddress + offset, sub_size);
-                                            }
+    });
-                                        });
-    if (!m_internal->base_set.empty()) {
-        auto it = m_internal->base_set.begin();
-        auto end_it = m_internal->base_set.end();
-        for (; it != end_it; it++) {
-            const VAddr inter_addr_end = it->upper();
-            const VAddr inter_addr = it->lower();
-            const size_t offset = inter_addr - m_vaddress;
-            const size_t sub_size = inter_addr_end - inter_addr;
-            m_internal->device_memory.Unmap(m_daddress + offset, sub_size);
-        }
-    }
-    const IntervalType to_remove{start, start + size};
-    m_internal->RemoveEachInOverlapCounter(m_internal->mapping_overlaps, to_remove, -1);
-    m_internal->base_set.clear();
 }
 
 } // namespace Service::Nvidia::NvCore

src/video_core/CMakeLists.txt

@@ -274,7 +274,6 @@ add_library(video_core STATIC
     texture_cache/image_view_info.h
     texture_cache/render_targets.h
     texture_cache/samples_helper.h
-    texture_cache/slot_vector.h
     texture_cache/texture_cache.cpp
     texture_cache/texture_cache.h
     texture_cache/texture_cache_base.h

src/video_core/buffer_cache/buffer_cache.h

@@ -7,6 +7,7 @@
 #include <memory>
 #include <numeric>
 
+#include "common/range_sets.inc"
 #include "video_core/buffer_cache/buffer_cache_base.h"
 #include "video_core/guest_memory.h"
 #include "video_core/host1x/gpu_device_memory_manager.h"
@@ -20,7 +21,7 @@ BufferCache<P>::BufferCache(Tegra::MaxwellDeviceMemoryManager& device_memory_, R
     : runtime{runtime_}, device_memory{device_memory_}, memory_tracker{device_memory} {
     // Ensure the first slot is used for the null buffer
     void(slot_buffers.insert(runtime, NullBufferParams{}));
-    common_ranges.clear();
+    gpu_modified_ranges.Clear();
     inline_buffer_id = NULL_BUFFER_ID;
 
     if (!runtime.CanReportMemoryUsage()) {
@@ -43,6 +44,9 @@ BufferCache<P>::BufferCache(Tegra::MaxwellDeviceMemoryManager& device_memory_, R
                  DEFAULT_CRITICAL_MEMORY));
 }
 
+template <class P>
+BufferCache<P>::~BufferCache() = default;
+
 template <class P>
 void BufferCache<P>::RunGarbageCollector() {
     const bool aggressive_gc = total_used_memory >= critical_memory;
@@ -96,20 +100,17 @@ void BufferCache<P>::TickFrame() {
     ++frame_tick;
     delayed_destruction_ring.Tick();
 
-    if constexpr (IMPLEMENTS_ASYNC_DOWNLOADS) {
+    for (auto& buffer : async_buffers_death_ring) {
-        for (auto& buffer : async_buffers_death_ring) {
+        runtime.FreeDeferredStagingBuffer(buffer);
-            runtime.FreeDeferredStagingBuffer(buffer);
-        }
-        async_buffers_death_ring.clear();
     }
+    async_buffers_death_ring.clear();
 }
 
 template <class P>
 void BufferCache<P>::WriteMemory(DAddr device_addr, u64 size) {
     if (memory_tracker.IsRegionGpuModified(device_addr, size)) {
-        const IntervalType subtract_interval{device_addr, device_addr + size};
+        ClearDownload(device_addr, size);
-        ClearDownload(subtract_interval);
+        gpu_modified_ranges.Subtract(device_addr, size);
-        common_ranges.subtract(subtract_interval);
     }
     memory_tracker.MarkRegionAsCpuModified(device_addr, size);
 }
@@ -174,11 +175,11 @@ void BufferCache<P>::DownloadMemory(DAddr device_addr, u64 size) {
 }
 
 template <class P>
-void BufferCache<P>::ClearDownload(IntervalType subtract_interval) {
+void BufferCache<P>::ClearDownload(DAddr device_addr, u64 size) {
-    RemoveEachInOverlapCounter(async_downloads, subtract_interval, -1024);
+    async_downloads.DeleteAll(device_addr, size);
-    uncommitted_ranges.subtract(subtract_interval);
+    uncommitted_gpu_modified_ranges.Subtract(device_addr, size);
-    for (auto& interval_set : committed_ranges) {
+    for (auto& interval_set : committed_gpu_modified_ranges) {
-        interval_set.subtract(subtract_interval);
+        interval_set.Subtract(device_addr, size);
     }
 }
 
@@ -195,8 +196,7 @@ bool BufferCache<P>::DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 am
         return false;
     }
 
-    const IntervalType subtract_interval{*cpu_dest_address, *cpu_dest_address + amount};
+    ClearDownload(*cpu_dest_address, amount);
-    ClearDownload(subtract_interval);
 
     BufferId buffer_a;
     BufferId buffer_b;
@@ -215,21 +215,20 @@ bool BufferCache<P>::DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 am
         .size = amount,
     }};
 
-    boost::container::small_vector<IntervalType, 4> tmp_intervals;
+    boost::container::small_vector<std::pair<DAddr, size_t>, 4> tmp_intervals;
     auto mirror = [&](DAddr base_address, DAddr base_address_end) {
         const u64 size = base_address_end - base_address;
         const DAddr diff = base_address - *cpu_src_address;
         const DAddr new_base_address = *cpu_dest_address + diff;
-        const IntervalType add_interval{new_base_address, new_base_address + size};
+        tmp_intervals.push_back({new_base_address, size});
-        tmp_intervals.push_back(add_interval);
+        uncommitted_gpu_modified_ranges.Add(new_base_address, size);
-        uncommitted_ranges.add(add_interval);
     };
-    ForEachInRangeSet(common_ranges, *cpu_src_address, amount, mirror);
+    gpu_modified_ranges.ForEachInRange(*cpu_src_address, amount, mirror);
     // This subtraction in this order is important for overlapping copies.
-    common_ranges.subtract(subtract_interval);
+    gpu_modified_ranges.Subtract(*cpu_dest_address, amount);
     const bool has_new_downloads = tmp_intervals.size() != 0;
-    for (const IntervalType& add_interval : tmp_intervals) {
+    for (const auto& pair : tmp_intervals) {
-        common_ranges.add(add_interval);
+        gpu_modified_ranges.Add(pair.first, pair.second);
     }
     const auto& copy = copies[0];
     src_buffer.MarkUsage(copy.src_offset, copy.size);
@@ -257,9 +256,8 @@ bool BufferCache<P>::DMAClear(GPUVAddr dst_address, u64 amount, u32 value) {
     }
 
     const size_t size = amount * sizeof(u32);
-    const IntervalType subtract_interval{*cpu_dst_address, *cpu_dst_address + size};
+    ClearDownload(*cpu_dst_address, size);
-    ClearDownload(subtract_interval);
+    gpu_modified_ranges.Subtract(*cpu_dst_address, size);
-    common_ranges.subtract(subtract_interval);
 
     const BufferId buffer = FindBuffer(*cpu_dst_address, static_cast<u32>(size));
     Buffer& dest_buffer = slot_buffers[buffer];
@@ -300,11 +298,11 @@ std::pair<typename P::Buffer*, u32> BufferCache<P>::ObtainCPUBuffer(
         MarkWrittenBuffer(buffer_id, device_addr, size);
         break;
     case ObtainBufferOperation::DiscardWrite: {
-        DAddr device_addr_start = Common::AlignDown(device_addr, 64);
+        const DAddr device_addr_start = Common::AlignDown(device_addr, 64);
-        DAddr device_addr_end = Common::AlignUp(device_addr + size, 64);
+        const DAddr device_addr_end = Common::AlignUp(device_addr + size, 64);
-        IntervalType interval{device_addr_start, device_addr_end};
+        const size_t new_size = device_addr_end - device_addr_start;
-        ClearDownload(interval);
+        ClearDownload(device_addr_start, new_size);
-        common_ranges.subtract(interval);
+        gpu_modified_ranges.Subtract(device_addr_start, new_size);
         break;
     }
     default:
@@ -504,46 +502,40 @@ void BufferCache<P>::FlushCachedWrites() {
 
 template <class P>
 bool BufferCache<P>::HasUncommittedFlushes() const noexcept {
-    return !uncommitted_ranges.empty() || !committed_ranges.empty();
+    return !uncommitted_gpu_modified_ranges.Empty() || !committed_gpu_modified_ranges.empty();
 }
 
 template <class P>
 void BufferCache<P>::AccumulateFlushes() {
-    if (uncommitted_ranges.empty()) {
+    if (uncommitted_gpu_modified_ranges.Empty()) {
         return;
     }
-    committed_ranges.emplace_back(std::move(uncommitted_ranges));
+    committed_gpu_modified_ranges.emplace_back(std::move(uncommitted_gpu_modified_ranges));
 }
 
 template <class P>
 bool BufferCache<P>::ShouldWaitAsyncFlushes() const noexcept {
-    if constexpr (IMPLEMENTS_ASYNC_DOWNLOADS) {
+    return (!async_buffers.empty() && async_buffers.front().has_value());
-        return (!async_buffers.empty() && async_buffers.front().has_value());
-    } else {
-        return false;
-    }
 }
 
 template <class P>
 void BufferCache<P>::CommitAsyncFlushesHigh() {
     AccumulateFlushes();
 
-    if (committed_ranges.empty()) {
+    if (committed_gpu_modified_ranges.empty()) {
-        if constexpr (IMPLEMENTS_ASYNC_DOWNLOADS) {
+        async_buffers.emplace_back(std::optional<Async_Buffer>{});
-            async_buffers.emplace_back(std::optional<Async_Buffer>{});
-        }
         return;
     }
     MICROPROFILE_SCOPE(GPU_DownloadMemory);
 
-    auto it = committed_ranges.begin();
+    auto it = committed_gpu_modified_ranges.begin();
-    while (it != committed_ranges.end()) {
+    while (it != committed_gpu_modified_ranges.end()) {
         auto& current_intervals = *it;
         auto next_it = std::next(it);
-        while (next_it != committed_ranges.end()) {
+        while (next_it != committed_gpu_modified_ranges.end()) {
-            for (auto& interval : *next_it) {
+            next_it->ForEach([&current_intervals](DAddr start, DAddr end) {
-                current_intervals.subtract(interval);
+                current_intervals.Subtract(start, end - start);
-            }
+            });
             next_it++;
         }
         it++;
@@ -552,10 +544,10 @@ void BufferCache<P>::CommitAsyncFlushesHigh() {
     boost::container::small_vector<std::pair<BufferCopy, BufferId>, 16> downloads;
     u64 total_size_bytes = 0;
     u64 largest_copy = 0;
-    for (const IntervalSet& intervals : committed_ranges) {
+    for (const Common::RangeSet<DAddr>& range_set : committed_gpu_modified_ranges) {
-        for (auto& interval : intervals) {
+        range_set.ForEach([&](DAddr interval_lower, DAddr interval_upper) {
-            const std::size_t size = interval.upper() - interval.lower();
+            const std::size_t size = interval_upper - interval_lower;
-            const DAddr device_addr = interval.lower();
+            const DAddr device_addr = interval_lower;
             ForEachBufferInRange(device_addr, size, [&](BufferId buffer_id, Buffer& buffer) {
                 const DAddr buffer_start = buffer.CpuAddr();
                 const DAddr buffer_end = buffer_start + buffer.SizeBytes();
@@ -583,77 +575,35 @@ void BufferCache<P>::CommitAsyncFlushesHigh() {
                             largest_copy = std::max(largest_copy, new_size);
                         };
 
-                        ForEachInRangeSet(common_ranges, device_addr_out, range_size, add_download);
+                        gpu_modified_ranges.ForEachInRange(device_addr_out, range_size,
+                                                           add_download);
                     });
             });
-        }
+        });
     }
-    committed_ranges.clear();
+    committed_gpu_modified_ranges.clear();
     if (downloads.empty()) {
-        if constexpr (IMPLEMENTS_ASYNC_DOWNLOADS) {
+        async_buffers.emplace_back(std::optional<Async_Buffer>{});
-            async_buffers.emplace_back(std::optional<Async_Buffer>{});
-        }
         return;
     }
-    if constexpr (IMPLEMENTS_ASYNC_DOWNLOADS) {
+    auto download_staging = runtime.DownloadStagingBuffer(total_size_bytes, true);
-        auto download_staging = runtime.DownloadStagingBuffer(total_size_bytes, true);
+    boost::container::small_vector<BufferCopy, 4> normalized_copies;
-        boost::container::small_vector<BufferCopy, 4> normalized_copies;
+    runtime.PreCopyBarrier();
-        IntervalSet new_async_range{};
+    for (auto& [copy, buffer_id] : downloads) {
-        runtime.PreCopyBarrier();
+        copy.dst_offset += download_staging.offset;
-        for (auto& [copy, buffer_id] : downloads) {
+        const std::array copies{copy};
-            copy.dst_offset += download_staging.offset;
+        BufferCopy second_copy{copy};
-            const std::array copies{copy};
+        Buffer& buffer = slot_buffers[buffer_id];
-            BufferCopy second_copy{copy};
+        second_copy.src_offset = static_cast<size_t>(buffer.CpuAddr()) + copy.src_offset;
-            Buffer& buffer = slot_buffers[buffer_id];
+        const DAddr orig_device_addr = static_cast<DAddr>(second_copy.src_offset);
-            second_copy.src_offset = static_cast<size_t>(buffer.CpuAddr()) + copy.src_offset;
+        async_downloads.Add(orig_device_addr, copy.size);
-            DAddr orig_device_addr = static_cast<DAddr>(second_copy.src_offset);
+        buffer.MarkUsage(copy.src_offset, copy.size);
-            const IntervalType base_interval{orig_device_addr, orig_device_addr + copy.size};
+        runtime.CopyBuffer(download_staging.buffer, buffer, copies, false);
-            async_downloads += std::make_pair(base_interval, 1);
+        normalized_copies.push_back(second_copy);
-            buffer.MarkUsage(copy.src_offset, copy.size);
-            runtime.CopyBuffer(download_staging.buffer, buffer, copies, false);
-            normalized_copies.push_back(second_copy);
-        }
-        runtime.PostCopyBarrier();
-        pending_downloads.emplace_back(std::move(normalized_copies));
-        async_buffers.emplace_back(download_staging);
-    } else {
-        if (!Settings::IsGPULevelHigh()) {
-            committed_ranges.clear();
-            uncommitted_ranges.clear();
-        } else {
-            if constexpr (USE_MEMORY_MAPS) {
-                auto download_staging = runtime.DownloadStagingBuffer(total_size_bytes);
-                runtime.PreCopyBarrier();
-                for (auto& [copy, buffer_id] : downloads) {
-                    // Have in mind the staging buffer offset for the copy
-                    copy.dst_offset += download_staging.offset;
-                    const std::array copies{copy};
-                    Buffer& buffer = slot_buffers[buffer_id];
-                    buffer.MarkUsage(copy.src_offset, copy.size);
-                    runtime.CopyBuffer(download_staging.buffer, buffer, copies, false);
-                }
-                runtime.PostCopyBarrier();
-                runtime.Finish();
-                for (const auto& [copy, buffer_id] : downloads) {
-                    const Buffer& buffer = slot_buffers[buffer_id];
-                    const DAddr device_addr = buffer.CpuAddr() + copy.src_offset;
-                    // Undo the modified offset
-                    const u64 dst_offset = copy.dst_offset - download_staging.offset;
-                    const u8* read_mapped_memory = download_staging.mapped_span.data() + dst_offset;
-                    device_memory.WriteBlockUnsafe(device_addr, read_mapped_memory, copy.size);
-                }
-            } else {
-                const std::span<u8> immediate_buffer = ImmediateBuffer(largest_copy);
-                for (const auto& [copy, buffer_id] : downloads) {
-                    Buffer& buffer = slot_buffers[buffer_id];
-                    buffer.ImmediateDownload(copy.src_offset,
-                                             immediate_buffer.subspan(0, copy.size));
-                    const DAddr device_addr = buffer.CpuAddr() + copy.src_offset;
-                    device_memory.WriteBlockUnsafe(device_addr, immediate_buffer.data(), copy.size);
-                }
-            }
-        }
     }
+    runtime.PostCopyBarrier();
+    pending_downloads.emplace_back(std::move(normalized_copies));
+    async_buffers.emplace_back(download_staging);
 }
 
 template <class P>
@@ -676,37 +626,31 @@ void BufferCache<P>::PopAsyncBuffers() {
         async_buffers.pop_front();
         return;
     }
-    if constexpr (IMPLEMENTS_ASYNC_DOWNLOADS) {
+    auto& downloads = pending_downloads.front();
-        auto& downloads = pending_downloads.front();
+    auto& async_buffer = async_buffers.front();
-        auto& async_buffer = async_buffers.front();
+    u8* base = async_buffer->mapped_span.data();
-        u8* base = async_buffer->mapped_span.data();
+    const size_t base_offset = async_buffer->offset;
-        const size_t base_offset = async_buffer->offset;
+    for (const auto& copy : downloads) {
-        for (const auto& copy : downloads) {
+        const DAddr device_addr = static_cast<DAddr>(copy.src_offset);
-            const DAddr device_addr = static_cast<DAddr>(copy.src_offset);
+        const u64 dst_offset = copy.dst_offset - base_offset;
-            const u64 dst_offset = copy.dst_offset - base_offset;
+        const u8* read_mapped_memory = base + dst_offset;
-            const u8* read_mapped_memory = base + dst_offset;
+        async_downloads.ForEachInRange(device_addr, copy.size, [&](DAddr start, DAddr end, s32) {
-            ForEachInOverlapCounter(
+            device_memory.WriteBlockUnsafe(start, &read_mapped_memory[start - device_addr],
-                async_downloads, device_addr, copy.size, [&](DAddr start, DAddr end, int count) {
+                                           end - start);
-                    device_memory.WriteBlockUnsafe(start, &read_mapped_memory[start - device_addr],
+        });
-                                                   end - start);
+        async_downloads.Subtract(device_addr, copy.size, [&](DAddr start, DAddr end) {
-                    if (count == 1) {
+            gpu_modified_ranges.Subtract(start, end - start);
-                        const IntervalType base_interval{start, end};
+        });
-                        common_ranges.subtract(base_interval);
-                    }
-                });
-            const IntervalType subtract_interval{device_addr, device_addr + copy.size};
-            RemoveEachInOverlapCounter(async_downloads, subtract_interval, -1);
-        }
-        async_buffers_death_ring.emplace_back(*async_buffer);
-        async_buffers.pop_front();
-        pending_downloads.pop_front();
     }
+    async_buffers_death_ring.emplace_back(*async_buffer);
+    async_buffers.pop_front();
+    pending_downloads.pop_front();
 }
 
 template <class P>
 bool BufferCache<P>::IsRegionGpuModified(DAddr addr, size_t size) {
     bool is_dirty = false;
-    ForEachInRangeSet(common_ranges, addr, size, [&](DAddr, DAddr) { is_dirty = true; });
+    gpu_modified_ranges.ForEachInRange(addr, size, [&](DAddr, DAddr) { is_dirty = true; });
     return is_dirty;
 }
 
@@ -1320,10 +1264,8 @@ void BufferCache<P>::UpdateComputeTextureBuffers() {
 template <class P>
 void BufferCache<P>::MarkWrittenBuffer(BufferId buffer_id, DAddr device_addr, u32 size) {
     memory_tracker.MarkRegionAsGpuModified(device_addr, size);
-
+    gpu_modified_ranges.Add(device_addr, size);
-    const IntervalType base_interval{device_addr, device_addr + size};
+    uncommitted_gpu_modified_ranges.Add(device_addr, size);
-    common_ranges.add(base_interval);
-    uncommitted_ranges.add(base_interval);
 }
 
 template <class P>
@@ -1603,9 +1545,8 @@ bool BufferCache<P>::InlineMemory(DAddr dest_address, size_t copy_size,
 template <class P>
 void BufferCache<P>::InlineMemoryImplementation(DAddr dest_address, size_t copy_size,
                                                 std::span<const u8> inlined_buffer) {
-    const IntervalType subtract_interval{dest_address, dest_address + copy_size};
+    ClearDownload(dest_address, copy_size);
-    ClearDownload(subtract_interval);
+    gpu_modified_ranges.Subtract(dest_address, copy_size);
-    common_ranges.subtract(subtract_interval);
 
     BufferId buffer_id = FindBuffer(dest_address, static_cast<u32>(copy_size));
     auto& buffer = slot_buffers[buffer_id];
@@ -1655,12 +1596,9 @@ void BufferCache<P>::DownloadBufferMemory(Buffer& buffer, DAddr device_addr, u64
                 largest_copy = std::max(largest_copy, new_size);
             };
 
-            const DAddr start_address = device_addr_out;
+            gpu_modified_ranges.ForEachInRange(device_addr_out, range_size, add_download);
-            const DAddr end_address = start_address + range_size;
+            ClearDownload(device_addr_out, range_size);
-            ForEachInRangeSet(common_ranges, start_address, range_size, add_download);
+            gpu_modified_ranges.Subtract(device_addr_out, range_size);
-            const IntervalType subtract_interval{start_address, end_address};
-            ClearDownload(subtract_interval);
-            common_ranges.subtract(subtract_interval);
         });
     if (total_size_bytes == 0) {
         return;

src/video_core/buffer_cache/buffer_cache_base.h

@@ -13,25 +13,15 @@
 #include <unordered_map>
 #include <vector>
 
-#include <boost/container/small_vector.hpp>
-#define BOOST_NO_MT
-#include <boost/pool/detail/mutex.hpp>
-#undef BOOST_NO_MT
-#include <boost/icl/interval.hpp>
-#include <boost/icl/interval_base_set.hpp>
-#include <boost/icl/interval_set.hpp>
-#include <boost/icl/split_interval_map.hpp>
-#include <boost/pool/pool.hpp>
-#include <boost/pool/pool_alloc.hpp>
-#include <boost/pool/poolfwd.hpp>
-
 #include "common/common_types.h"
 #include "common/div_ceil.h"
 #include "common/literals.h"
 #include "common/lru_cache.h"
 #include "common/microprofile.h"
+#include "common/range_sets.h"
 #include "common/scope_exit.h"
 #include "common/settings.h"
+#include "common/slot_vector.h"
 #include "video_core/buffer_cache/buffer_base.h"
 #include "video_core/control/channel_state_cache.h"
 #include "video_core/delayed_destruction_ring.h"
@@ -41,21 +31,15 @@
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/memory_manager.h"
 #include "video_core/surface.h"
-#include "video_core/texture_cache/slot_vector.h"
 #include "video_core/texture_cache/types.h"
 
-namespace boost {
-template <typename T>
-class fast_pool_allocator<T, default_user_allocator_new_delete, details::pool::null_mutex, 4096, 0>;
-}
-
 namespace VideoCommon {
 
 MICROPROFILE_DECLARE(GPU_PrepareBuffers);
 MICROPROFILE_DECLARE(GPU_BindUploadBuffers);
 MICROPROFILE_DECLARE(GPU_DownloadMemory);
 
-using BufferId = SlotId;
+using BufferId = Common::SlotId;
 
 using VideoCore::Surface::PixelFormat;
 using namespace Common::Literals;
@@ -184,7 +168,6 @@ class BufferCache : public VideoCommon::ChannelSetupCaches<BufferCacheChannelInf
     static constexpr bool NEEDS_BIND_STORAGE_INDEX = P::NEEDS_BIND_STORAGE_INDEX;
     static constexpr bool USE_MEMORY_MAPS = P::USE_MEMORY_MAPS;
     static constexpr bool SEPARATE_IMAGE_BUFFERS_BINDINGS = P::SEPARATE_IMAGE_BUFFER_BINDINGS;
-    static constexpr bool IMPLEMENTS_ASYNC_DOWNLOADS = P::IMPLEMENTS_ASYNC_DOWNLOADS;
     static constexpr bool USE_MEMORY_MAPS_FOR_UPLOADS = P::USE_MEMORY_MAPS_FOR_UPLOADS;
 
     static constexpr s64 DEFAULT_EXPECTED_MEMORY = 512_MiB;
@@ -202,34 +185,6 @@ class BufferCache : public VideoCommon::ChannelSetupCaches<BufferCacheChannelInf
     using Async_Buffer = typename P::Async_Buffer;
     using MemoryTracker = typename P::MemoryTracker;
 
-    using IntervalCompare = std::less<DAddr>;
-    using IntervalInstance = boost::icl::interval_type_default<DAddr, std::less>;
-    using IntervalAllocator = boost::fast_pool_allocator<DAddr>;
-    using IntervalSet = boost::icl::interval_set<DAddr>;
-    using IntervalType = typename IntervalSet::interval_type;
-
-    template <typename Type>
-    struct counter_add_functor : public boost::icl::identity_based_inplace_combine<Type> {
-        // types
-        typedef counter_add_functor<Type> type;
-        typedef boost::icl::identity_based_inplace_combine<Type> base_type;
-
-        // public member functions
-        void operator()(Type& current, const Type& added) const {
-            current += added;
-            if (current < base_type::identity_element()) {
-                current = base_type::identity_element();
-            }
-        }
-
-        // public static functions
-        static void version(Type&){};
-    };
-
-    using OverlapCombine = counter_add_functor<int>;
-    using OverlapSection = boost::icl::inter_section<int>;
-    using OverlapCounter = boost::icl::split_interval_map<DAddr, int>;
-
     struct OverlapResult {
         boost::container::small_vector<BufferId, 16> ids;
         DAddr begin;
@@ -240,6 +195,8 @@ class BufferCache : public VideoCommon::ChannelSetupCaches<BufferCacheChannelInf
 public:
     explicit BufferCache(Tegra::MaxwellDeviceMemoryManager& device_memory_, Runtime& runtime_);
 
+    ~BufferCache();
+
     void TickFrame();
 
     void WriteMemory(DAddr device_addr, u64 size);
@@ -379,75 +336,6 @@ private:
         }
     }
 
-    template <typename Func>
-    void ForEachInRangeSet(IntervalSet& current_range, DAddr device_addr, u64 size, Func&& func) {
-        const DAddr start_address = device_addr;
-        const DAddr end_address = start_address + size;
-        const IntervalType search_interval{start_address, end_address};
-        auto it = current_range.lower_bound(search_interval);
-        if (it == current_range.end()) {
-            return;
-        }
-        auto end_it = current_range.upper_bound(search_interval);
-        for (; it != end_it; it++) {
-            DAddr inter_addr_end = it->upper();
-            DAddr inter_addr = it->lower();
-            if (inter_addr_end > end_address) {
-                inter_addr_end = end_address;
-            }
-            if (inter_addr < start_address) {
-                inter_addr = start_address;
-            }
-            func(inter_addr, inter_addr_end);
-        }
-    }
-
-    template <typename Func>
-    void ForEachInOverlapCounter(OverlapCounter& current_range, DAddr device_addr, u64 size,
-                                 Func&& func) {
-        const DAddr start_address = device_addr;
-        const DAddr end_address = start_address + size;
-        const IntervalType search_interval{start_address, end_address};
-        auto it = current_range.lower_bound(search_interval);
-        if (it == current_range.end()) {
-            return;
-        }
-        auto end_it = current_range.upper_bound(search_interval);
-        for (; it != end_it; it++) {
-            auto& inter = it->first;
-            DAddr inter_addr_end = inter.upper();
-            DAddr inter_addr = inter.lower();
-            if (inter_addr_end > end_address) {
-                inter_addr_end = end_address;
-            }
-            if (inter_addr < start_address) {
-                inter_addr = start_address;
-            }
-            func(inter_addr, inter_addr_end, it->second);
-        }
-    }
-
-    void RemoveEachInOverlapCounter(OverlapCounter& current_range,
-                                    const IntervalType search_interval, int subtract_value) {
-        bool any_removals = false;
-        current_range.add(std::make_pair(search_interval, subtract_value));
-        do {
-            any_removals = false;
-            auto it = current_range.lower_bound(search_interval);
-            if (it == current_range.end()) {
-                return;
-            }
-            auto end_it = current_range.upper_bound(search_interval);
-            for (; it != end_it; it++) {
-                if (it->second <= 0) {
-                    any_removals = true;
-                    current_range.erase(it);
-                    break;
-                }
-            }
-        } while (any_removals);
-    }
-
     static bool IsRangeGranular(DAddr device_addr, size_t size) {
         return (device_addr & ~Core::DEVICE_PAGEMASK) ==
                ((device_addr + size) & ~Core::DEVICE_PAGEMASK);
@@ -552,14 +440,14 @@ private:
 
     [[nodiscard]] bool HasFastUniformBufferBound(size_t stage, u32 binding_index) const noexcept;
 
-    void ClearDownload(IntervalType subtract_interval);
+    void ClearDownload(DAddr base_addr, u64 size);
 
     void InlineMemoryImplementation(DAddr dest_address, size_t copy_size,
                                     std::span<const u8> inlined_buffer);
 
     Tegra::MaxwellDeviceMemoryManager& device_memory;
 
-    SlotVector<Buffer> slot_buffers;
+    Common::SlotVector<Buffer> slot_buffers;
     DelayedDestructionRing<Buffer, 8> delayed_destruction_ring;
 
     const Tegra::Engines::DrawManager::IndirectParams* current_draw_indirect{};
@@ -567,13 +455,12 @@ private:
     u32 last_index_count = 0;
 
     MemoryTracker memory_tracker;
-    IntervalSet uncommitted_ranges;
+    Common::RangeSet<DAddr> uncommitted_gpu_modified_ranges;
-    IntervalSet common_ranges;
+    Common::RangeSet<DAddr> gpu_modified_ranges;
-    IntervalSet cached_ranges;
+    std::deque<Common::RangeSet<DAddr>> committed_gpu_modified_ranges;
-    std::deque<IntervalSet> committed_ranges;
 
     // Async Buffers
-    OverlapCounter async_downloads;
+    Common::OverlapRangeSet<DAddr> async_downloads;
     std::deque<std::optional<Async_Buffer>> async_buffers;
     std::deque<boost::container::small_vector<BufferCopy, 4>> pending_downloads;
     std::optional<Async_Buffer> current_buffer;

src/video_core/control/scheduler.cpp

@@ -50,6 +50,7 @@ void Scheduler::Resume() {
 void Scheduler::Yield() {
     ASSERT(current_fifo != nullptr);
     Common::Fiber::YieldTo(current_fifo->context, *master_control);
+    gpu.BindChannel(current_fifo->bind_id);
 }
 
 void Scheduler::Push(s32 channel, CommandList&& entries) {

src/video_core/query_cache.h

@@ -18,12 +18,12 @@
 
 #include "common/assert.h"
 #include "common/settings.h"
+#include "common/slot_vector.h"
 #include "video_core/control/channel_state_cache.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/host1x/gpu_device_memory_manager.h"
 #include "video_core/memory_manager.h"
 #include "video_core/rasterizer_interface.h"
-#include "video_core/texture_cache/slot_vector.h"
 
 namespace VideoCore {
 enum class QueryType {
@@ -37,7 +37,7 @@ constexpr std::size_t NumQueryTypes = static_cast<size_t>(QueryType::Count);
 
 namespace VideoCommon {
 
-using AsyncJobId = SlotId;
+using AsyncJobId = Common::SlotId;
 
 static constexpr AsyncJobId NULL_ASYNC_JOB_ID{0};
 
@@ -341,7 +341,7 @@ private:
     static constexpr std::uintptr_t YUZU_PAGESIZE = 4096;
     static constexpr unsigned YUZU_PAGEBITS = 12;
 
-    SlotVector<AsyncJob> slot_async_jobs;
+    Common::SlotVector<AsyncJob> slot_async_jobs;
 
     VideoCore::RasterizerInterface& rasterizer;
     Tegra::MaxwellDeviceMemoryManager& device_memory;

src/video_core/renderer_opengl/gl_buffer_cache.h

@@ -90,7 +90,7 @@ public:
     void PostCopyBarrier();
     void Finish();
 
-    void TickFrame(VideoCommon::SlotVector<Buffer>&) noexcept {}
+    void TickFrame(Common::SlotVector<Buffer>&) noexcept {}
 
     void ClearBuffer(Buffer& dest_buffer, u32 offset, size_t size, u32 value);
 
@@ -251,7 +251,6 @@ struct BufferCacheParams {
     static constexpr bool NEEDS_BIND_STORAGE_INDEX = true;
     static constexpr bool USE_MEMORY_MAPS = true;
     static constexpr bool SEPARATE_IMAGE_BUFFER_BINDINGS = true;
-    static constexpr bool IMPLEMENTS_ASYNC_DOWNLOADS = true;
 
     // TODO: Investigate why OpenGL seems to perform worse with persistently mapped buffer uploads
     static constexpr bool USE_MEMORY_MAPS_FOR_UPLOADS = false;

src/video_core/renderer_opengl/gl_texture_cache.h

@@ -30,13 +30,13 @@ class Image;
 class ImageView;
 class Sampler;
 
+using Common::SlotVector;
 using VideoCommon::ImageId;
 using VideoCommon::ImageViewId;
 using VideoCommon::ImageViewType;
 using VideoCommon::NUM_RT;
 using VideoCommon::Region2D;
 using VideoCommon::RenderTargets;
-using VideoCommon::SlotVector;
 
 struct FormatProperties {
     GLenum compatibility_class;

src/video_core/renderer_vulkan/vk_buffer_cache.cpp

@@ -368,7 +368,7 @@ u32 BufferCacheRuntime::GetStorageBufferAlignment() const {
     return static_cast<u32>(device.GetStorageBufferAlignment());
 }
 
-void BufferCacheRuntime::TickFrame(VideoCommon::SlotVector<Buffer>& slot_buffers) noexcept {
+void BufferCacheRuntime::TickFrame(Common::SlotVector<Buffer>& slot_buffers) noexcept {
     for (auto it = slot_buffers.begin(); it != slot_buffers.end(); it++) {
         it->ResetUsageTracking();
     }

src/video_core/renderer_vulkan/vk_buffer_cache.h

@@ -81,7 +81,7 @@ public:
                                 ComputePassDescriptorQueue& compute_pass_descriptor_queue,
                                 DescriptorPool& descriptor_pool);
 
-    void TickFrame(VideoCommon::SlotVector<Buffer>& slot_buffers) noexcept;
+    void TickFrame(Common::SlotVector<Buffer>& slot_buffers) noexcept;
 
     void Finish();
 
@@ -181,7 +181,6 @@ struct BufferCacheParams {
     static constexpr bool NEEDS_BIND_STORAGE_INDEX = false;
     static constexpr bool USE_MEMORY_MAPS = true;
     static constexpr bool SEPARATE_IMAGE_BUFFER_BINDINGS = false;
-    static constexpr bool IMPLEMENTS_ASYNC_DOWNLOADS = true;
     static constexpr bool USE_MEMORY_MAPS_FOR_UPLOADS = true;
 };
 

src/video_core/renderer_vulkan/vk_texture_cache.h

@@ -20,11 +20,11 @@ struct ResolutionScalingInfo;
 
 namespace Vulkan {
 
+using Common::SlotVector;
 using VideoCommon::ImageId;
 using VideoCommon::NUM_RT;
 using VideoCommon::Region2D;
 using VideoCommon::RenderTargets;
-using VideoCommon::SlotVector;
 using VideoCore::Surface::PixelFormat;
 
 class BlitImageHelper;

src/video_core/texture_cache/texture_cache_base.h

@@ -21,6 +21,7 @@
 #include "common/lru_cache.h"
 #include "common/polyfill_ranges.h"
 #include "common/scratch_buffer.h"
+#include "common/slot_vector.h"
 #include "common/thread_worker.h"
 #include "video_core/compatible_formats.h"
 #include "video_core/control/channel_state_cache.h"
@@ -32,7 +33,6 @@
 #include "video_core/texture_cache/image_info.h"
 #include "video_core/texture_cache/image_view_base.h"
 #include "video_core/texture_cache/render_targets.h"
-#include "video_core/texture_cache/slot_vector.h"
 #include "video_core/texture_cache/types.h"
 #include "video_core/textures/texture.h"
 
@@ -451,16 +451,16 @@ private:
     struct PendingDownload {
         bool is_swizzle;
         size_t async_buffer_id;
-        SlotId object_id;
+        Common::SlotId object_id;
     };
 
-    SlotVector<Image> slot_images;
+    Common::SlotVector<Image> slot_images;
-    SlotVector<ImageMapView> slot_map_views;
+    Common::SlotVector<ImageMapView> slot_map_views;
-    SlotVector<ImageView> slot_image_views;
+    Common::SlotVector<ImageView> slot_image_views;
-    SlotVector<ImageAlloc> slot_image_allocs;
+    Common::SlotVector<ImageAlloc> slot_image_allocs;
-    SlotVector<Sampler> slot_samplers;
+    Common::SlotVector<Sampler> slot_samplers;
-    SlotVector<Framebuffer> slot_framebuffers;
+    Common::SlotVector<Framebuffer> slot_framebuffers;
-    SlotVector<BufferDownload> slot_buffer_downloads;
+    Common::SlotVector<BufferDownload> slot_buffer_downloads;
 
     // TODO: This data structure is not optimal and it should be reworked
 

src/video_core/texture_cache/types.h

@@ -5,21 +5,21 @@
 
 #include "common/common_funcs.h"
 #include "common/common_types.h"
-#include "video_core/texture_cache/slot_vector.h"
+#include "common/slot_vector.h"
 
 namespace VideoCommon {
 
 constexpr size_t NUM_RT = 8;
 constexpr size_t MAX_MIP_LEVELS = 14;
 
-constexpr SlotId CORRUPT_ID{0xfffffffe};
+constexpr Common::SlotId CORRUPT_ID{0xfffffffe};
 
-using ImageId = SlotId;
+using ImageId = Common::SlotId;
-using ImageMapId = SlotId;
+using ImageMapId = Common::SlotId;
-using ImageViewId = SlotId;
+using ImageViewId = Common::SlotId;
-using ImageAllocId = SlotId;
+using ImageAllocId = Common::SlotId;
-using SamplerId = SlotId;
+using SamplerId = Common::SlotId;
-using FramebufferId = SlotId;
+using FramebufferId = Common::SlotId;
 
 /// Fake image ID for null image views
 constexpr ImageId NULL_IMAGE_ID{0};