Use spans over guest memory where possible instead of copying data.

5 files changed, 285 insertions, 18 deletions

diff --git a/src/core/core_timing.cpp b/src/core/core_timing.cpp
index e6112a3c9..b98a0cb33 100644
--- a/src/core/core_timing.cpp
+++ b/src/core/core_timing.cpp
@@ -70,7 +70,7 @@ void CoreTiming::Initialize(std::function<void()>&& on_thread_init_) {
         -> std::optional<std::chrono::nanoseconds> { return std::nullopt; };
     ev_lost = CreateEvent("_lost_event", empty_timed_callback);
     if (is_multicore) {
-        timer_thread = std::make_unique<std::thread>(ThreadEntry, std::ref(*this));
+        timer_thread = std::make_unique<std::jthread>(ThreadEntry, std::ref(*this));
     }
 }
 
@@ -255,7 +255,6 @@ void CoreTiming::ThreadLoop() {
 #ifdef _WIN32
                     while (!paused && !event.IsSet() && wait_time > 0) {
                         wait_time = *next_time - GetGlobalTimeNs().count();
-
                         if (wait_time >= timer_resolution_ns) {
                             Common::Windows::SleepForOneTick();
                         } else {
diff --git a/src/core/core_timing.h b/src/core/core_timing.h
index 5bca1c78d..c20e906fb 100644
--- a/src/core/core_timing.h
+++ b/src/core/core_timing.h
@@ -163,7 +163,7 @@ private:
     Common::Event pause_event{};
     std::mutex basic_lock;
     std::mutex advance_lock;
-    std::unique_ptr<std::thread> timer_thread;
+    std::unique_ptr<std::jthread> timer_thread;
     std::atomic<bool> paused{};
     std::atomic<bool> paused_set{};
     std::atomic<bool> wait_set{};
diff --git a/src/core/hle/service/hle_ipc.cpp b/src/core/hle/service/hle_ipc.cpp
index 2290df705..f6a1e54f2 100644
--- a/src/core/hle/service/hle_ipc.cpp
+++ b/src/core/hle/service/hle_ipc.cpp
@@ -329,8 +329,22 @@ std::vector<u8> HLERequestContext::ReadBufferCopy(std::size_t buffer_index) cons
 }
 
 std::span<const u8> HLERequestContext::ReadBuffer(std::size_t buffer_index) const {
-    static thread_local std::array<Common::ScratchBuffer<u8>, 2> read_buffer_a;
-    static thread_local std::array<Common::ScratchBuffer<u8>, 2> read_buffer_x;
+    static thread_local std::array read_buffer_a{
+        Core::Memory::CpuGuestMemory<u8, Core::Memory::GuestMemoryFlags::SafeRead>(memory, 0, 0),
+        Core::Memory::CpuGuestMemory<u8, Core::Memory::GuestMemoryFlags::SafeRead>(memory, 0, 0),
+    };
+    static thread_local std::array read_buffer_data_a{
+        Common::ScratchBuffer<u8>(),
+        Common::ScratchBuffer<u8>(),
+    };
+    static thread_local std::array read_buffer_x{
+        Core::Memory::CpuGuestMemory<u8, Core::Memory::GuestMemoryFlags::SafeRead>(memory, 0, 0),
+        Core::Memory::CpuGuestMemory<u8, Core::Memory::GuestMemoryFlags::SafeRead>(memory, 0, 0),
+    };
+    static thread_local std::array read_buffer_data_x{
+        Common::ScratchBuffer<u8>(),
+        Common::ScratchBuffer<u8>(),
+    };
 
     const bool is_buffer_a{BufferDescriptorA().size() > buffer_index &&
                            BufferDescriptorA()[buffer_index].Size()};
@@ -339,19 +353,17 @@ std::span<const u8> HLERequestContext::ReadBuffer(std::size_t buffer_index) cons
             BufferDescriptorA().size() > buffer_index, { return {}; },
             "BufferDescriptorA invalid buffer_index {}", buffer_index);
         auto& read_buffer = read_buffer_a[buffer_index];
-        read_buffer.resize_destructive(BufferDescriptorA()[buffer_index].Size());
-        memory.ReadBlock(BufferDescriptorA()[buffer_index].Address(), read_buffer.data(),
-                         read_buffer.size());
-        return read_buffer;
+        return read_buffer.Read(BufferDescriptorA()[buffer_index].Address(),
+                                BufferDescriptorA()[buffer_index].Size(),
+                                &read_buffer_data_a[buffer_index]);
     } else {
         ASSERT_OR_EXECUTE_MSG(
             BufferDescriptorX().size() > buffer_index, { return {}; },
             "BufferDescriptorX invalid buffer_index {}", buffer_index);
         auto& read_buffer = read_buffer_x[buffer_index];
-        read_buffer.resize_destructive(BufferDescriptorX()[buffer_index].Size());
-        memory.ReadBlock(BufferDescriptorX()[buffer_index].Address(), read_buffer.data(),
-                         read_buffer.size());
-        return read_buffer;
+        return read_buffer.Read(BufferDescriptorX()[buffer_index].Address(),
+                                BufferDescriptorX()[buffer_index].Size(),
+                                &read_buffer_data_x[buffer_index]);
     }
 }
 
diff --git a/src/core/memory.cpp b/src/core/memory.cpp
index 257406f09..805963178 100644
--- a/src/core/memory.cpp
+++ b/src/core/memory.cpp
@@ -266,6 +266,22 @@ struct Memory::Impl {
         ReadBlockImpl<true>(*system.ApplicationProcess(), src_addr, dest_buffer, size);
     }
 
+    const u8* GetSpan(const VAddr src_addr, const std::size_t size) const {
+        if (current_page_table->blocks[src_addr >> YUZU_PAGEBITS] ==
+            current_page_table->blocks[(src_addr + size) >> YUZU_PAGEBITS]) {
+            return GetPointerSilent(src_addr);
+        }
+        return nullptr;
+    }
+
+    u8* GetSpan(const VAddr src_addr, const std::size_t size) {
+        if (current_page_table->blocks[src_addr >> YUZU_PAGEBITS] ==
+            current_page_table->blocks[(src_addr + size) >> YUZU_PAGEBITS]) {
+            return GetPointerSilent(src_addr);
+        }
+        return nullptr;
+    }
+
     template <bool UNSAFE>
     void WriteBlockImpl(const Kernel::KProcess& process, const Common::ProcessAddress dest_addr,
                         const void* src_buffer, const std::size_t size) {
@@ -559,7 +575,7 @@ struct Memory::Impl {
             }
         }
 
-        const Common::ProcessAddress end = base + size;
+        const auto end = base + size;
         ASSERT_MSG(end <= page_table.pointers.size(), "out of range mapping at {:016X}",
                    base + page_table.pointers.size());
 
@@ -570,14 +586,18 @@ struct Memory::Impl {
             while (base != end) {
                 page_table.pointers[base].Store(nullptr, type);
                 page_table.backing_addr[base] = 0;
-
+                page_table.blocks[base] = 0;
                 base += 1;
             }
         } else {
+            auto orig_base = base;
             while (base != end) {
-                page_table.pointers[base].Store(
-                    system.DeviceMemory().GetPointer<u8>(target) - (base << YUZU_PAGEBITS), type);
-                page_table.backing_addr[base] = GetInteger(target) - (base << YUZU_PAGEBITS);
+                auto host_ptr =
+                    system.DeviceMemory().GetPointer<u8>(target) - (base << YUZU_PAGEBITS);
+                auto backing = GetInteger(target) - (base << YUZU_PAGEBITS);
+                page_table.pointers[base].Store(host_ptr, type);
+                page_table.backing_addr[base] = backing;
+                page_table.blocks[base] = orig_base << YUZU_PAGEBITS;
 
                 ASSERT_MSG(page_table.pointers[base].Pointer(),
                            "memory mapping base yield a nullptr within the table");
@@ -747,6 +767,14 @@ struct Memory::Impl {
         VAddr last_address;
     };
 
+    void InvalidateRegion(Common::ProcessAddress dest_addr, size_t size) {
+        system.GPU().InvalidateRegion(GetInteger(dest_addr), size);
+    }
+
+    void FlushRegion(Common::ProcessAddress dest_addr, size_t size) {
+        system.GPU().FlushRegion(GetInteger(dest_addr), size);
+    }
+
     Core::System& system;
     Common::PageTable* current_page_table = nullptr;
     std::array<VideoCore::RasterizerDownloadArea, Core::Hardware::NUM_CPU_CORES>
@@ -881,6 +909,14 @@ void Memory::ReadBlockUnsafe(const Common::ProcessAddress src_addr, void* dest_b
     impl->ReadBlockUnsafe(src_addr, dest_buffer, size);
 }
 
+const u8* Memory::GetSpan(const VAddr src_addr, const std::size_t size) const {
+    return impl->GetSpan(src_addr, size);
+}
+
+u8* Memory::GetSpan(const VAddr src_addr, const std::size_t size) {
+    return impl->GetSpan(src_addr, size);
+}
+
 void Memory::WriteBlock(const Common::ProcessAddress dest_addr, const void* src_buffer,
                         const std::size_t size) {
     impl->WriteBlock(dest_addr, src_buffer, size);
@@ -924,4 +960,12 @@ void Memory::MarkRegionDebug(Common::ProcessAddress vaddr, u64 size, bool debug)
     impl->MarkRegionDebug(GetInteger(vaddr), size, debug);
 }
 
+void Memory::InvalidateRegion(Common::ProcessAddress dest_addr, size_t size) {
+    impl->InvalidateRegion(dest_addr, size);
+}
+
+void Memory::FlushRegion(Common::ProcessAddress dest_addr, size_t size) {
+    impl->FlushRegion(dest_addr, size);
+}
+
 } // namespace Core::Memory
diff --git a/src/core/memory.h b/src/core/memory.h
index ea01824f8..ea33c769c 100644
--- a/src/core/memory.h
+++ b/src/core/memory.h
@@ -5,8 +5,12 @@
 
 #include <cstddef>
 #include <memory>
+#include <optional>
 #include <span>
 #include <string>
+#include <vector>
+
+#include "common/scratch_buffer.h"
 #include "common/typed_address.h"
 #include "core/hle/result.h"
 
@@ -24,6 +28,10 @@ class PhysicalMemory;
 class KProcess;
 } // namespace Kernel
 
+namespace Tegra {
+class MemoryManager;
+}
+
 namespace Core::Memory {
 
 /**
@@ -343,6 +351,9 @@ public:
      */
     void ReadBlockUnsafe(Common::ProcessAddress src_addr, void* dest_buffer, std::size_t size);
 
+    const u8* GetSpan(const VAddr src_addr, const std::size_t size) const;
+    u8* GetSpan(const VAddr src_addr, const std::size_t size);
+
     /**
      * Writes a range of bytes into the current process' address space at the specified
      * virtual address.
@@ -461,6 +472,8 @@ public:
     void MarkRegionDebug(Common::ProcessAddress vaddr, u64 size, bool debug);
 
     void SetGPUDirtyManagers(std::span<Core::GPUDirtyMemoryManager> managers);
+    void InvalidateRegion(Common::ProcessAddress dest_addr, size_t size);
+    void FlushRegion(Common::ProcessAddress dest_addr, size_t size);
 
 private:
     Core::System& system;
@@ -469,4 +482,203 @@ private:
     std::unique_ptr<Impl> impl;
 };
 
+enum GuestMemoryFlags : u32 {
+    Read = 1 << 0,
+    Write = 1 << 1,
+    Safe = 1 << 2,
+    Cached = 1 << 3,
+
+    SafeRead = Read | Safe,
+    SafeWrite = Write | Safe,
+    SafeReadWrite = SafeRead | SafeWrite,
+    SafeReadCachedWrite = SafeReadWrite | Cached,
+
+    UnsafeRead = Read,
+    UnsafeWrite = Write,
+    UnsafeReadWrite = UnsafeRead | UnsafeWrite,
+    UnsafeReadCachedWrite = UnsafeReadWrite | Cached,
+};
+
+namespace {
+template <typename M, typename T, GuestMemoryFlags FLAGS>
+class GuestMemory {
+    using iterator = T*;
+    using const_iterator = const T*;
+    using value_type = T;
+    using element_type = T;
+    using iterator_category = std::contiguous_iterator_tag;
+
+public:
+    GuestMemory() = delete;
+    explicit GuestMemory(M& memory_, u64 addr_, std::size_t size_,
+                         Common::ScratchBuffer<T>* backup = nullptr)
+        : memory{memory_}, addr{addr_}, size{size_} {
+        static_assert(FLAGS & GuestMemoryFlags::Read || FLAGS & GuestMemoryFlags::Write);
+        if constexpr (FLAGS & GuestMemoryFlags::Read) {
+            Read(addr, size, backup);
+        }
+    }
+
+    ~GuestMemory() = default;
+
+    T* data() noexcept {
+        return data_span.data();
+    }
+
+    const T* data() const noexcept {
+        return data_span.data();
+    }
+
+    [[nodiscard]] T* begin() noexcept {
+        return data();
+    }
+
+    [[nodiscard]] const T* begin() const noexcept {
+        return data();
+    }
+
+    [[nodiscard]] T* end() noexcept {
+        return data() + size;
+    }
+
+    [[nodiscard]] const T* end() const noexcept {
+        return data() + size;
+    }
+
+    T& operator[](size_t index) noexcept {
+        return data_span[index];
+    }
+
+    const T& operator[](size_t index) const noexcept {
+        return data_span[index];
+    }
+
+    void SetAddressAndSize(u64 addr_, std::size_t size_) noexcept {
+        addr = addr_;
+        size = size_;
+        addr_changed = true;
+    }
+
+    std::span<T> Read(u64 addr_, std::size_t size_,
+                      Common::ScratchBuffer<T>* backup = nullptr) noexcept {
+        addr = addr_;
+        size = size_;
+        if (size == 0) {
+            is_data_copy = true;
+            return {};
+        }
+
+        if (TrySetSpan()) {
+            if constexpr (FLAGS & GuestMemoryFlags::Safe) {
+                memory.FlushRegion(addr, size * sizeof(T));
+            }
+        } else {
+            if (backup) {
+                backup->resize_destructive(size);
+                data_span = *backup;
+            } else {
+                data_copy.resize(size);
+                data_span = std::span(data_copy);
+            }
+            is_data_copy = true;
+            span_valid = true;
+            if constexpr (FLAGS & GuestMemoryFlags::Safe) {
+                memory.ReadBlock(addr, data_span.data(), size * sizeof(T));
+            } else {
+                memory.ReadBlockUnsafe(addr, data_span.data(), size * sizeof(T));
+            }
+        }
+        return data_span;
+    }
+
+    void Write(std::span<T> write_data) noexcept {
+        if constexpr (FLAGS & GuestMemoryFlags::Cached) {
+            memory.WriteBlockCached(addr, write_data.data(), size * sizeof(T));
+        } else if constexpr (FLAGS & GuestMemoryFlags::Safe) {
+            memory.WriteBlock(addr, write_data.data(), size * sizeof(T));
+        } else {
+            memory.WriteBlockUnsafe(addr, write_data.data(), size * sizeof(T));
+        }
+    }
+
+    bool TrySetSpan() noexcept {
+        if (u8* ptr = memory.GetSpan(addr, size * sizeof(T)); ptr) {
+            data_span = {reinterpret_cast<T*>(ptr), size};
+            span_valid = true;
+            return true;
+        }
+        return false;
+    }
+
+protected:
+    bool IsDataCopy() const noexcept {
+        return is_data_copy;
+    }
+
+    bool AddressChanged() const noexcept {
+        return addr_changed;
+    }
+
+    M& memory;
+    u64 addr;
+    size_t size;
+    std::span<T> data_span{};
+    std::vector<T> data_copy;
+    bool span_valid{false};
+    bool is_data_copy{false};
+    bool addr_changed{false};
+};
+
+template <typename M, typename T, GuestMemoryFlags FLAGS>
+class GuestMemoryScoped : public GuestMemory<M, T, FLAGS> {
+public:
+    GuestMemoryScoped() = delete;
+    explicit GuestMemoryScoped(M& memory_, u64 addr_, std::size_t size_,
+                               Common::ScratchBuffer<T>* backup = nullptr)
+        : GuestMemory<M, T, FLAGS>(memory_, addr_, size_, backup) {
+        if constexpr (!(FLAGS & GuestMemoryFlags::Read)) {
+            if (!this->TrySetSpan()) {
+                if (backup) {
+                    this->data_span = *backup;
+                    this->span_valid = true;
+                    this->is_data_copy = true;
+                }
+            }
+        }
+    }
+
+    ~GuestMemoryScoped() {
+        if constexpr (FLAGS & GuestMemoryFlags::Write) {
+            if (this->size == 0) [[unlikely]] {
+                return;
+            }
+
+            if (this->AddressChanged() || this->IsDataCopy()) {
+                ASSERT(this->span_valid);
+                if constexpr (FLAGS & GuestMemoryFlags::Cached) {
+                    this->memory.WriteBlockCached(this->addr, this->data_span.data(),
+                                                  this->size * sizeof(T));
+                } else if constexpr (FLAGS & GuestMemoryFlags::Safe) {
+                    this->memory.WriteBlock(this->addr, this->data_span.data(),
+                                            this->size * sizeof(T));
+                } else {
+                    this->memory.WriteBlockUnsafe(this->addr, this->data_span.data(),
+                                                  this->size * sizeof(T));
+                }
+            } else if constexpr (FLAGS & GuestMemoryFlags::Safe) {
+                this->memory.InvalidateRegion(this->addr, this->size * sizeof(T));
+            }
+        }
+    }
+};
+} // namespace
+
+template <typename T, GuestMemoryFlags FLAGS>
+using CpuGuestMemory = GuestMemory<Memory, T, FLAGS>;
+template <typename T, GuestMemoryFlags FLAGS>
+using CpuGuestMemoryScoped = GuestMemoryScoped<Memory, T, FLAGS>;
+template <typename T, GuestMemoryFlags FLAGS>
+using GpuGuestMemory = GuestMemory<Tegra::MemoryManager, T, FLAGS>;
+template <typename T, GuestMemoryFlags FLAGS>
+using GpuGuestMemoryScoped = GuestMemoryScoped<Tegra::MemoryManager, T, FLAGS>;
 } // namespace Core::Memory