Revert "externals: Update FMT to 6.0.0"

This reverts commit ca4ca8a6dc.
externals: Update FMT to 6.0.0
2019-08-29 12:23:34 -06:00 · 2019-08-29 19:37:46 +02:00 · 2019-08-29 13:07:01 -04:00 · 2019-08-29 12:58:43 -04:00 · 2019-08-29 12:58:10 -04:00 · 2019-08-21 14:50:38 -03:00
52 changed files with 1254 additions and 498 deletions
@@ -81,6 +81,7 @@ set(HASH_FILES
    "${VIDEO_CORE}/shader/decode/register_set_predicate.cpp"
    "${VIDEO_CORE}/shader/decode/shift.cpp"
    "${VIDEO_CORE}/shader/decode/video.cpp"
+    "${VIDEO_CORE}/shader/decode/warp.cpp"
    "${VIDEO_CORE}/shader/decode/xmad.cpp"
    "${VIDEO_CORE}/shader/control_flow.cpp"
    "${VIDEO_CORE}/shader/control_flow.h"
@@ -55,6 +55,7 @@ add_custom_command(OUTPUT scm_rev.cpp
      "${VIDEO_CORE}/shader/decode/register_set_predicate.cpp"
      "${VIDEO_CORE}/shader/decode/shift.cpp"
      "${VIDEO_CORE}/shader/decode/video.cpp"
+      "${VIDEO_CORE}/shader/decode/warp.cpp"
      "${VIDEO_CORE}/shader/decode/xmad.cpp"
      "${VIDEO_CORE}/shader/control_flow.cpp"
      "${VIDEO_CORE}/shader/control_flow.h"
@@ -3,6 +3,7 @@
 #pragma once

 #include <cstddef>
+#include <memory>
 #include <type_traits>

 namespace Common {
@@ -37,4 +38,63 @@ constexpr bool IsWordAligned(T value) {
    return (value & 0b11) == 0;
 }

+template <typename T, std::size_t Align = 16>
+class AlignmentAllocator {
+public:
+    using value_type = T;
+    using size_type = std::size_t;
+    using difference_type = std::ptrdiff_t;
+
+    using pointer = T*;
+    using const_pointer = const T*;
+
+    using reference = T&;
+    using const_reference = const T&;
+
+public:
+    pointer address(reference r) noexcept {
+        return std::addressof(r);
+    }
+
+    const_pointer address(const_reference r) const noexcept {
+        return std::addressof(r);
+    }
+
+    pointer allocate(size_type n) {
+        return static_cast<pointer>(::operator new (n, std::align_val_t{Align}));
+    }
+
+    void deallocate(pointer p, size_type) {
+        ::operator delete (p, std::align_val_t{Align});
+    }
+
+    void construct(pointer p, const value_type& wert) {
+        new (p) value_type(wert);
+    }
+
+    void destroy(pointer p) {
+        p->~value_type();
+    }
+
+    size_type max_size() const noexcept {
+        return size_type(-1) / sizeof(value_type);
+    }
+
+    template <typename T2>
+    struct rebind {
+        using other = AlignmentAllocator<T2, Align>;
+    };
+
+    bool operator!=(const AlignmentAllocator<T, Align>& other) const noexcept {
+        return !(*this == other);
+    }
+
+    // Returns true if and only if storage allocated from *this
+    // can be deallocated from other, and vice versa.
+    // Always returns true for stateless allocators.
+    bool operator==(const AlignmentAllocator<T, Align>& other) const noexcept {
+        return true;
+    }
+};
+
 } // namespace Common
@@ -8,6 +8,7 @@
 #include <vector>

 #include "common/common_types.h"
+#include "core/hle/kernel/physical_memory.h"

 namespace Kernel {

@@ -77,7 +78,7 @@ struct CodeSet final {
    }

    /// The overall data that backs this code set.
-    std::vector<u8> memory;
+    Kernel::PhysicalMemory memory;

    /// The segments that comprise this code set.
    std::array<Segment, 3> segments;
@@ -0,0 +1,19 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include "common/alignment.h"
+
+namespace Kernel {
+
+// This encapsulation serves 2 purposes:
+// - First, to encapsulate host physical memory under a single type and set an
+// standard for managing it.
+// - Second to ensure all host backing memory used is aligned to 256 bytes due
+// to strict alignment restrictions on GPU memory.
+
+using PhysicalMemory = std::vector<u8, Common::AlignmentAllocator<u8, 256>>;
+
+} // namespace Kernel
@@ -247,7 +247,7 @@ VAddr Process::CreateTLSRegion() {
        ASSERT(region_address.Succeeded());

        const auto map_result = vm_manager.MapMemoryBlock(
-            *region_address, std::make_shared<std::vector<u8>>(Memory::PAGE_SIZE), 0,
+            *region_address, std::make_shared<PhysicalMemory>(Memory::PAGE_SIZE), 0,
            Memory::PAGE_SIZE, MemoryState::ThreadLocal);
        ASSERT(map_result.Succeeded());

@@ -277,7 +277,7 @@ void Process::FreeTLSRegion(VAddr tls_address) {
 }

 void Process::LoadModule(CodeSet module_, VAddr base_addr) {
-    const auto memory = std::make_shared<std::vector<u8>>(std::move(module_.memory));
+    const auto memory = std::make_shared<PhysicalMemory>(std::move(module_.memory));

    const auto MapSegment = [&](const CodeSet::Segment& segment, VMAPermission permissions,
                                MemoryState memory_state) {
@@ -327,7 +327,7 @@ void Process::AllocateMainThreadStack(u64 stack_size) {
    // Allocate and map the main thread stack
    const VAddr mapping_address = vm_manager.GetTLSIORegionEndAddress() - main_thread_stack_size;
    vm_manager
-        .MapMemoryBlock(mapping_address, std::make_shared<std::vector<u8>>(main_thread_stack_size),
+        .MapMemoryBlock(mapping_address, std::make_shared<PhysicalMemory>(main_thread_stack_size),
                        0, main_thread_stack_size, MemoryState::Stack)
        .Unwrap();
 }
@@ -28,7 +28,7 @@ SharedPtr<SharedMemory> SharedMemory::Create(KernelCore& kernel, Process* owner_
    shared_memory->other_permissions = other_permissions;

    if (address == 0) {
-        shared_memory->backing_block = std::make_shared<std::vector<u8>>(size);
+        shared_memory->backing_block = std::make_shared<Kernel::PhysicalMemory>(size);
        shared_memory->backing_block_offset = 0;

        // Refresh the address mappings for the current process.
@@ -59,8 +59,8 @@ SharedPtr<SharedMemory> SharedMemory::Create(KernelCore& kernel, Process* owner_
 }

 SharedPtr<SharedMemory> SharedMemory::CreateForApplet(
-    KernelCore& kernel, std::shared_ptr<std::vector<u8>> heap_block, std::size_t offset, u64 size,
-    MemoryPermission permissions, MemoryPermission other_permissions, std::string name) {
+    KernelCore& kernel, std::shared_ptr<Kernel::PhysicalMemory> heap_block, std::size_t offset,
+    u64 size, MemoryPermission permissions, MemoryPermission other_permissions, std::string name) {
    SharedPtr<SharedMemory> shared_memory(new SharedMemory(kernel));

    shared_memory->owner_process = nullptr;
@@ -10,6 +10,7 @@

 #include "common/common_types.h"
 #include "core/hle/kernel/object.h"
+#include "core/hle/kernel/physical_memory.h"
 #include "core/hle/kernel/process.h"
 #include "core/hle/result.h"

@@ -62,12 +63,10 @@ public:
     * block.
     * @param name Optional object name, used for debugging purposes.
     */
-    static SharedPtr<SharedMemory> CreateForApplet(KernelCore& kernel,
-                                                   std::shared_ptr<std::vector<u8>> heap_block,
-                                                   std::size_t offset, u64 size,
-                                                   MemoryPermission permissions,
-                                                   MemoryPermission other_permissions,
-                                                   std::string name = "Unknown Applet");
+    static SharedPtr<SharedMemory> CreateForApplet(
+        KernelCore& kernel, std::shared_ptr<Kernel::PhysicalMemory> heap_block, std::size_t offset,
+        u64 size, MemoryPermission permissions, MemoryPermission other_permissions,
+        std::string name = "Unknown Applet");

    std::string GetTypeName() const override {
        return "SharedMemory";
@@ -135,7 +134,7 @@ private:
    ~SharedMemory() override;

    /// Backing memory for this shared memory block.
-    std::shared_ptr<std::vector<u8>> backing_block;
+    std::shared_ptr<PhysicalMemory> backing_block;
    /// Offset into the backing block for this shared memory.
    std::size_t backing_block_offset = 0;
    /// Size of the memory block. Page-aligned.
@@ -47,7 +47,7 @@ ResultCode TransferMemory::MapMemory(VAddr address, u64 size, MemoryPermission p
        return ERR_INVALID_STATE;
    }

-    backing_block = std::make_shared<std::vector<u8>>(size);
+    backing_block = std::make_shared<PhysicalMemory>(size);

    const auto map_state = owner_permissions == MemoryPermission::None
                               ? MemoryState::TransferMemoryIsolated
@@ -8,6 +8,7 @@
 #include <vector>

 #include "core/hle/kernel/object.h"
+#include "core/hle/kernel/physical_memory.h"

 union ResultCode;

@@ -82,7 +83,7 @@ private:
    ~TransferMemory() override;

    /// Memory block backing this instance.
-    std::shared_ptr<std::vector<u8>> backing_block;
+    std::shared_ptr<PhysicalMemory> backing_block;

    /// The base address for the memory managed by this instance.
    VAddr base_address = 0;
@@ -5,6 +5,7 @@
 #include <algorithm>
 #include <iterator>
 #include <utility>
+#include "common/alignment.h"
 #include "common/assert.h"
 #include "common/logging/log.h"
 #include "common/memory_hook.h"
@@ -103,7 +104,7 @@ bool VMManager::IsValidHandle(VMAHandle handle) const {
 }

 ResultVal<VMManager::VMAHandle> VMManager::MapMemoryBlock(VAddr target,
-                                                          std::shared_ptr<std::vector<u8>> block,
+                                                          std::shared_ptr<PhysicalMemory> block,
                                                          std::size_t offset, u64 size,
                                                          MemoryState state, VMAPermission perm) {
    ASSERT(block != nullptr);
@@ -260,7 +261,7 @@ ResultVal<VAddr> VMManager::SetHeapSize(u64 size) {

    if (heap_memory == nullptr) {
        // Initialize heap
-        heap_memory = std::make_shared<std::vector<u8>>(size);
+        heap_memory = std::make_shared<PhysicalMemory>(size);
        heap_end = heap_region_base + size;
    } else {
        UnmapRange(heap_region_base, GetCurrentHeapSize());
@@ -341,7 +342,7 @@ ResultCode VMManager::MapPhysicalMemory(VAddr target, u64 size) {
            const auto map_size = std::min(end_addr - cur_addr, vma_end - cur_addr);
            if (vma.state == MemoryState::Unmapped) {
                const auto map_res =
-                    MapMemoryBlock(cur_addr, std::make_shared<std::vector<u8>>(map_size, 0), 0,
+                    MapMemoryBlock(cur_addr, std::make_shared<PhysicalMemory>(map_size, 0), 0,
                                   map_size, MemoryState::Heap, VMAPermission::ReadWrite);
                result = map_res.Code();
                if (result.IsError()) {
@@ -442,7 +443,7 @@ ResultCode VMManager::UnmapPhysicalMemory(VAddr target, u64 size) {
    if (result.IsError()) {
        for (const auto [map_address, map_size] : unmapped_regions) {
            const auto remap_res =
-                MapMemoryBlock(map_address, std::make_shared<std::vector<u8>>(map_size, 0), 0,
+                MapMemoryBlock(map_address, std::make_shared<PhysicalMemory>(map_size, 0), 0,
                               map_size, MemoryState::Heap, VMAPermission::None);
            ASSERT_MSG(remap_res.Succeeded(), "UnmapPhysicalMemory re-map on error");
        }
@@ -593,7 +594,7 @@ ResultCode VMManager::MirrorMemory(VAddr dst_addr, VAddr src_addr, u64 size, Mem
    ASSERT_MSG(vma_offset + size <= vma->second.size,
               "Shared memory exceeds bounds of mapped block");

-    const std::shared_ptr<std::vector<u8>>& backing_block = vma->second.backing_block;
+    const std::shared_ptr<PhysicalMemory>& backing_block = vma->second.backing_block;
    const std::size_t backing_block_offset = vma->second.offset + vma_offset;

    CASCADE_RESULT(auto new_vma,
@@ -606,7 +607,7 @@ ResultCode VMManager::MirrorMemory(VAddr dst_addr, VAddr src_addr, u64 size, Mem
    return RESULT_SUCCESS;
 }

-void VMManager::RefreshMemoryBlockMappings(const std::vector<u8>* block) {
+void VMManager::RefreshMemoryBlockMappings(const PhysicalMemory* block) {
    // If this ever proves to have a noticeable performance impact, allow users of the function to
    // specify a specific range of addresses to limit the scan to.
    for (const auto& p : vma_map) {
@@ -764,7 +765,7 @@ void VMManager::MergeAdjacentVMA(VirtualMemoryArea& left, const VirtualMemoryAre
                                       right.backing_block->begin() + right.offset + right.size);
        } else {
            // Slow case: make a new memory block for left and right.
-            auto new_memory = std::make_shared<std::vector<u8>>();
+            auto new_memory = std::make_shared<PhysicalMemory>();
            new_memory->insert(new_memory->end(), left.backing_block->begin() + left.offset,
                               left.backing_block->begin() + left.offset + left.size);
            new_memory->insert(new_memory->end(), right.backing_block->begin() + right.offset,
@@ -11,6 +11,7 @@
 #include "common/common_types.h"
 #include "common/memory_hook.h"
 #include "common/page_table.h"
+#include "core/hle/kernel/physical_memory.h"
 #include "core/hle/result.h"
 #include "core/memory.h"

@@ -290,7 +291,7 @@ struct VirtualMemoryArea {

    // Settings for type = AllocatedMemoryBlock
    /// Memory block backing this VMA.
-    std::shared_ptr<std::vector<u8>> backing_block = nullptr;
+    std::shared_ptr<PhysicalMemory> backing_block = nullptr;
    /// Offset into the backing_memory the mapping starts from.
    std::size_t offset = 0;

@@ -348,7 +349,7 @@ public:
     * @param size Size of the mapping.
     * @param state MemoryState tag to attach to the VMA.
     */
-    ResultVal<VMAHandle> MapMemoryBlock(VAddr target, std::shared_ptr<std::vector<u8>> block,
+    ResultVal<VMAHandle> MapMemoryBlock(VAddr target, std::shared_ptr<PhysicalMemory> block,
                                        std::size_t offset, u64 size, MemoryState state,
                                        VMAPermission perm = VMAPermission::ReadWrite);

@@ -547,7 +548,7 @@ public:
     * Scans all VMAs and updates the page table range of any that use the given vector as backing
     * memory. This should be called after any operation that causes reallocation of the vector.
     */
-    void RefreshMemoryBlockMappings(const std::vector<u8>* block);
+    void RefreshMemoryBlockMappings(const PhysicalMemory* block);

    /// Dumps the address space layout to the log, for debugging
    void LogLayout() const;
@@ -777,7 +778,7 @@ private:
    // the entire virtual address space extents that bound the allocations, including any holes.
    // This makes deallocation and reallocation of holes fast and keeps process memory contiguous
    // in the emulator address space, allowing Memory::GetPointer to be reasonably safe.
-    std::shared_ptr<std::vector<u8>> heap_memory;
+    std::shared_ptr<PhysicalMemory> heap_memory;

    // The end of the currently allocated heap. This is not an inclusive
    // end of the range. This is essentially 'base_address + current_size'.
@@ -19,16 +19,16 @@

 namespace Service::Audio {

-void InstallInterfaces(SM::ServiceManager& service_manager) {
+void InstallInterfaces(SM::ServiceManager& service_manager, Core::System& system) {
    std::make_shared<AudCtl>()->InstallAsService(service_manager);
    std::make_shared<AudOutA>()->InstallAsService(service_manager);
-    std::make_shared<AudOutU>()->InstallAsService(service_manager);
+    std::make_shared<AudOutU>(system)->InstallAsService(service_manager);
    std::make_shared<AudInA>()->InstallAsService(service_manager);
    std::make_shared<AudInU>()->InstallAsService(service_manager);
    std::make_shared<AudRecA>()->InstallAsService(service_manager);
    std::make_shared<AudRecU>()->InstallAsService(service_manager);
    std::make_shared<AudRenA>()->InstallAsService(service_manager);
-    std::make_shared<AudRenU>()->InstallAsService(service_manager);
+    std::make_shared<AudRenU>(system)->InstallAsService(service_manager);
    std::make_shared<CodecCtl>()->InstallAsService(service_manager);
    std::make_shared<HwOpus>()->InstallAsService(service_manager);

@@ -4,6 +4,10 @@

 #pragma once

+namespace Core {
+class System;
+}
+
 namespace Service::SM {
 class ServiceManager;
 }
@@ -11,6 +15,6 @@ class ServiceManager;
 namespace Service::Audio {

 /// Registers all Audio services with the specified service manager.
-void InstallInterfaces(SM::ServiceManager& service_manager);
+void InstallInterfaces(SM::ServiceManager& service_manager, Core::System& system);

 } // namespace Service::Audio
@@ -40,8 +40,8 @@ enum class AudioState : u32 {

 class IAudioOut final : public ServiceFramework<IAudioOut> {
 public:
-    IAudioOut(AudoutParams audio_params, AudioCore::AudioOut& audio_core, std::string&& device_name,
-              std::string&& unique_name)
+    IAudioOut(Core::System& system, AudoutParams audio_params, AudioCore::AudioOut& audio_core,
+              std::string&& device_name, std::string&& unique_name)
        : ServiceFramework("IAudioOut"), audio_core(audio_core),
          device_name(std::move(device_name)), audio_params(audio_params) {
        // clang-format off
@@ -65,7 +65,6 @@ public:
        RegisterHandlers(functions);

        // This is the event handle used to check if the audio buffer was released
-        auto& system = Core::System::GetInstance();
        buffer_event = Kernel::WritableEvent::CreateEventPair(
            system.Kernel(), Kernel::ResetType::Manual, "IAudioOutBufferReleased");

@@ -212,6 +211,22 @@ private:
    Kernel::EventPair buffer_event;
 };

+AudOutU::AudOutU(Core::System& system_) : ServiceFramework("audout:u"), system{system_} {
+    // clang-format off
+    static const FunctionInfo functions[] = {
+        {0, &AudOutU::ListAudioOutsImpl, "ListAudioOuts"},
+        {1, &AudOutU::OpenAudioOutImpl, "OpenAudioOut"},
+        {2, &AudOutU::ListAudioOutsImpl, "ListAudioOutsAuto"},
+        {3, &AudOutU::OpenAudioOutImpl, "OpenAudioOutAuto"},
+    };
+    // clang-format on
+
+    RegisterHandlers(functions);
+    audio_core = std::make_unique<AudioCore::AudioOut>();
+}
+
+AudOutU::~AudOutU() = default;
+
 void AudOutU::ListAudioOutsImpl(Kernel::HLERequestContext& ctx) {
    LOG_DEBUG(Service_Audio, "called");

@@ -248,7 +263,7 @@ void AudOutU::OpenAudioOutImpl(Kernel::HLERequestContext& ctx) {

    std::string unique_name{fmt::format("{}-{}", device_name, audio_out_interfaces.size())};
    auto audio_out_interface = std::make_shared<IAudioOut>(
-        params, *audio_core, std::move(device_name), std::move(unique_name));
+        system, params, *audio_core, std::move(device_name), std::move(unique_name));

    IPC::ResponseBuilder rb{ctx, 6, 0, 1};
    rb.Push(RESULT_SUCCESS);
@@ -256,20 +271,9 @@ void AudOutU::OpenAudioOutImpl(Kernel::HLERequestContext& ctx) {
    rb.Push<u32>(params.channel_count);
    rb.Push<u32>(static_cast<u32>(AudioCore::Codec::PcmFormat::Int16));
    rb.Push<u32>(static_cast<u32>(AudioState::Stopped));
-    rb.PushIpcInterface<Audio::IAudioOut>(audio_out_interface);
+    rb.PushIpcInterface<IAudioOut>(audio_out_interface);

    audio_out_interfaces.push_back(std::move(audio_out_interface));
 }

-AudOutU::AudOutU() : ServiceFramework("audout:u") {
-    static const FunctionInfo functions[] = {{0, &AudOutU::ListAudioOutsImpl, "ListAudioOuts"},
-                                             {1, &AudOutU::OpenAudioOutImpl, "OpenAudioOut"},
-                                             {2, &AudOutU::ListAudioOutsImpl, "ListAudioOutsAuto"},
-                                             {3, &AudOutU::OpenAudioOutImpl, "OpenAudioOutAuto"}};
-    RegisterHandlers(functions);
-    audio_core = std::make_unique<AudioCore::AudioOut>();
-}
-
-AudOutU::~AudOutU() = default;
-
 } // namespace Service::Audio
@@ -11,6 +11,10 @@ namespace AudioCore {
 class AudioOut;
 }

+namespace Core {
+class System;
+}
+
 namespace Kernel {
 class HLERequestContext;
 }
@@ -21,15 +25,17 @@ class IAudioOut;

 class AudOutU final : public ServiceFramework<AudOutU> {
 public:
-    AudOutU();
+    explicit AudOutU(Core::System& system_);
    ~AudOutU() override;

 private:
+    void ListAudioOutsImpl(Kernel::HLERequestContext& ctx);
+    void OpenAudioOutImpl(Kernel::HLERequestContext& ctx);
+
    std::vector<std::shared_ptr<IAudioOut>> audio_out_interfaces;
    std::unique_ptr<AudioCore::AudioOut> audio_core;

-    void ListAudioOutsImpl(Kernel::HLERequestContext& ctx);
-    void OpenAudioOutImpl(Kernel::HLERequestContext& ctx);
+    Core::System& system;
 };

 } // namespace Service::Audio
@@ -5,6 +5,7 @@
 #include <algorithm>
 #include <array>
 #include <memory>
+#include <string_view>

 #include "audio_core/audio_renderer.h"
 #include "common/alignment.h"
@@ -25,7 +26,7 @@ namespace Service::Audio {

 class IAudioRenderer final : public ServiceFramework<IAudioRenderer> {
 public:
-    explicit IAudioRenderer(AudioCore::AudioRendererParameter audren_params,
+    explicit IAudioRenderer(Core::System& system, AudioCore::AudioRendererParameter audren_params,
                            const std::size_t instance_number)
        : ServiceFramework("IAudioRenderer") {
        // clang-format off
@@ -46,7 +47,6 @@ public:
        // clang-format on
        RegisterHandlers(functions);

-        auto& system = Core::System::GetInstance();
        system_event = Kernel::WritableEvent::CreateEventPair(
            system.Kernel(), Kernel::ResetType::Manual, "IAudioRenderer:SystemEvent");
        renderer = std::make_unique<AudioCore::AudioRenderer>(
@@ -160,7 +160,8 @@ private:

 class IAudioDevice final : public ServiceFramework<IAudioDevice> {
 public:
-    IAudioDevice() : ServiceFramework("IAudioDevice") {
+    explicit IAudioDevice(Core::System& system, u32_le revision_num)
+        : ServiceFramework("IAudioDevice"), revision{revision_num} {
        static const FunctionInfo functions[] = {
            {0, &IAudioDevice::ListAudioDeviceName, "ListAudioDeviceName"},
            {1, &IAudioDevice::SetAudioDeviceOutputVolume, "SetAudioDeviceOutputVolume"},
@@ -178,7 +179,7 @@ public:
        };
        RegisterHandlers(functions);

-        auto& kernel = Core::System::GetInstance().Kernel();
+        auto& kernel = system.Kernel();
        buffer_event = Kernel::WritableEvent::CreateEventPair(kernel, Kernel::ResetType::Automatic,
                                                              "IAudioOutBufferReleasedEvent");

@@ -189,15 +190,47 @@ public:
    }

 private:
-    void ListAudioDeviceName(Kernel::HLERequestContext& ctx) {
-        LOG_WARNING(Service_Audio, "(STUBBED) called");
+    using AudioDeviceName = std::array<char, 256>;
+    static constexpr std::array<std::string_view, 4> audio_device_names{{
+        "AudioStereoJackOutput",
+        "AudioBuiltInSpeakerOutput",
+        "AudioTvOutput",
+        "AudioUsbDeviceOutput",
+    }};
+    enum class DeviceType {
+        AHUBHeadphones,
+        AHUBSpeakers,
+        HDA,
+        USBOutput,
+    };

-        constexpr std::array<char, 15> audio_interface{{"AudioInterface"}};
-        ctx.WriteBuffer(audio_interface);
+    void ListAudioDeviceName(Kernel::HLERequestContext& ctx) {
+        LOG_DEBUG(Service_Audio, "called");
+
+        const bool usb_output_supported =
+            IsFeatureSupported(AudioFeatures::AudioUSBDeviceOutput, revision);
+        const std::size_t count = ctx.GetWriteBufferSize() / sizeof(AudioDeviceName);
+
+        std::vector<AudioDeviceName> name_buffer;
+        name_buffer.reserve(audio_device_names.size());
+
+        for (std::size_t i = 0; i < count && i < audio_device_names.size(); i++) {
+            const auto type = static_cast<DeviceType>(i);
+
+            if (!usb_output_supported && type == DeviceType::USBOutput) {
+                continue;
+            }
+
+            const auto& device_name = audio_device_names[i];
+            auto& entry = name_buffer.emplace_back();
+            device_name.copy(entry.data(), device_name.size());
+        }
+
+        ctx.WriteBuffer(name_buffer);

        IPC::ResponseBuilder rb{ctx, 3};
        rb.Push(RESULT_SUCCESS);
-        rb.Push<u32>(1);
+        rb.Push(static_cast<u32>(name_buffer.size()));
    }

    void SetAudioDeviceOutputVolume(Kernel::HLERequestContext& ctx) {
@@ -216,12 +249,16 @@ private:
    void GetActiveAudioDeviceName(Kernel::HLERequestContext& ctx) {
        LOG_WARNING(Service_Audio, "(STUBBED) called");

-        constexpr std::array<char, 12> audio_interface{{"AudioDevice"}};
-        ctx.WriteBuffer(audio_interface);
+        // Currently set to always be TV audio output.
+        const auto& device_name = audio_device_names[2];

-        IPC::ResponseBuilder rb{ctx, 3};
+        AudioDeviceName out_device_name{};
+        device_name.copy(out_device_name.data(), device_name.size());
+
+        ctx.WriteBuffer(out_device_name);
+
+        IPC::ResponseBuilder rb{ctx, 2};
        rb.Push(RESULT_SUCCESS);
-        rb.Push<u32>(1);
    }

    void QueryAudioDeviceSystemEvent(Kernel::HLERequestContext& ctx) {
@@ -250,12 +287,13 @@ private:
        rb.PushCopyObjects(audio_output_device_switch_event.readable);
    }

+    u32_le revision = 0;
    Kernel::EventPair buffer_event;
    Kernel::EventPair audio_output_device_switch_event;

 }; // namespace Audio

-AudRenU::AudRenU() : ServiceFramework("audren:u") {
+AudRenU::AudRenU(Core::System& system_) : ServiceFramework("audren:u"), system{system_} {
    // clang-format off
    static const FunctionInfo functions[] = {
        {0, &AudRenU::OpenAudioRenderer, "OpenAudioRenderer"},
@@ -328,7 +366,7 @@ void AudRenU::GetAudioRendererWorkBufferSize(Kernel::HLERequestContext& ctx) {
    };

    // Calculates the portion of the size related to the mix data (and the sorting thereof).
-    const auto calculate_mix_info_size = [this](const AudioCore::AudioRendererParameter& params) {
+    const auto calculate_mix_info_size = [](const AudioCore::AudioRendererParameter& params) {
        // The size of the mixing info data structure.
        constexpr u64 mix_info_size = 0x940;

@@ -400,7 +438,7 @@ void AudRenU::GetAudioRendererWorkBufferSize(Kernel::HLERequestContext& ctx) {

    // Calculates the part of the size related to the splitter context.
    const auto calculate_splitter_context_size =
-        [this](const AudioCore::AudioRendererParameter& params) -> u64 {
+        [](const AudioCore::AudioRendererParameter& params) -> u64 {
        if (!IsFeatureSupported(AudioFeatures::Splitter, params.revision)) {
            return 0;
        }
@@ -447,7 +485,7 @@ void AudRenU::GetAudioRendererWorkBufferSize(Kernel::HLERequestContext& ctx) {
    };

    // Calculates the part of the size related to performance statistics.
-    const auto calculate_perf_size = [this](const AudioCore::AudioRendererParameter& params) {
+    const auto calculate_perf_size = [](const AudioCore::AudioRendererParameter& params) {
        // Extra size value appended to the end of the calculation.
        constexpr u64 appended = 128;

@@ -474,78 +512,76 @@ void AudRenU::GetAudioRendererWorkBufferSize(Kernel::HLERequestContext& ctx) {
    };

    // Calculates the part of the size that relates to the audio command buffer.
-    const auto calculate_command_buffer_size =
-        [this](const AudioCore::AudioRendererParameter& params) {
-            constexpr u64 alignment = (buffer_alignment_size - 1) * 2;
+    const auto calculate_command_buffer_size = [](const AudioCore::AudioRendererParameter& params) {
+        constexpr u64 alignment = (buffer_alignment_size - 1) * 2;

-            if (!IsFeatureSupported(AudioFeatures::VariadicCommandBuffer, params.revision)) {
-                constexpr u64 command_buffer_size = 0x18000;
+        if (!IsFeatureSupported(AudioFeatures::VariadicCommandBuffer, params.revision)) {
+            constexpr u64 command_buffer_size = 0x18000;

-                return command_buffer_size + alignment;
-            }
+            return command_buffer_size + alignment;
+        }

-            // When the variadic command buffer is supported, this means
-            // the command generator for the audio renderer can issue commands
-            // that are (as one would expect), variable in size. So what we need to do
-            // is determine the maximum possible size for a few command data structures
-            // then multiply them by the amount of present commands indicated by the given
-            // respective audio parameters.
+        // When the variadic command buffer is supported, this means
+        // the command generator for the audio renderer can issue commands
+        // that are (as one would expect), variable in size. So what we need to do
+        // is determine the maximum possible size for a few command data structures
+        // then multiply them by the amount of present commands indicated by the given
+        // respective audio parameters.

-            constexpr u64 max_biquad_filters = 2;
-            constexpr u64 max_mix_buffers = 24;
+        constexpr u64 max_biquad_filters = 2;
+        constexpr u64 max_mix_buffers = 24;

-            constexpr u64 biquad_filter_command_size = 0x2C;
+        constexpr u64 biquad_filter_command_size = 0x2C;

-            constexpr u64 depop_mix_command_size = 0x24;
-            constexpr u64 depop_setup_command_size = 0x50;
+        constexpr u64 depop_mix_command_size = 0x24;
+        constexpr u64 depop_setup_command_size = 0x50;

-            constexpr u64 effect_command_max_size = 0x540;
+        constexpr u64 effect_command_max_size = 0x540;

-            constexpr u64 mix_command_size = 0x1C;
-            constexpr u64 mix_ramp_command_size = 0x24;
-            constexpr u64 mix_ramp_grouped_command_size = 0x13C;
+        constexpr u64 mix_command_size = 0x1C;
+        constexpr u64 mix_ramp_command_size = 0x24;
+        constexpr u64 mix_ramp_grouped_command_size = 0x13C;

-            constexpr u64 perf_command_size = 0x28;
+        constexpr u64 perf_command_size = 0x28;

-            constexpr u64 sink_command_size = 0x130;
+        constexpr u64 sink_command_size = 0x130;

-            constexpr u64 submix_command_max_size =
-                depop_mix_command_size + (mix_command_size * max_mix_buffers) * max_mix_buffers;
+        constexpr u64 submix_command_max_size =
+            depop_mix_command_size + (mix_command_size * max_mix_buffers) * max_mix_buffers;

-            constexpr u64 volume_command_size = 0x1C;
-            constexpr u64 volume_ramp_command_size = 0x20;
+        constexpr u64 volume_command_size = 0x1C;
+        constexpr u64 volume_ramp_command_size = 0x20;

-            constexpr u64 voice_biquad_filter_command_size =
-                biquad_filter_command_size * max_biquad_filters;
-            constexpr u64 voice_data_command_size = 0x9C;
-            const u64 voice_command_max_size =
-                (params.splitter_count * depop_setup_command_size) +
-                (voice_data_command_size + voice_biquad_filter_command_size +
-                 volume_ramp_command_size + mix_ramp_grouped_command_size);
+        constexpr u64 voice_biquad_filter_command_size =
+            biquad_filter_command_size * max_biquad_filters;
+        constexpr u64 voice_data_command_size = 0x9C;
+        const u64 voice_command_max_size =
+            (params.splitter_count * depop_setup_command_size) +
+            (voice_data_command_size + voice_biquad_filter_command_size + volume_ramp_command_size +
+             mix_ramp_grouped_command_size);

-            // Now calculate the individual elements that comprise the size and add them together.
-            const u64 effect_commands_size = params.effect_count * effect_command_max_size;
+        // Now calculate the individual elements that comprise the size and add them together.
+        const u64 effect_commands_size = params.effect_count * effect_command_max_size;

-            const u64 final_mix_commands_size =
-                depop_mix_command_size + volume_command_size * max_mix_buffers;
+        const u64 final_mix_commands_size =
+            depop_mix_command_size + volume_command_size * max_mix_buffers;

-            const u64 perf_commands_size =
-                perf_command_size *
-                (CalculateNumPerformanceEntries(params) + max_perf_detail_entries);
+        const u64 perf_commands_size =
+            perf_command_size * (CalculateNumPerformanceEntries(params) + max_perf_detail_entries);

-            const u64 sink_commands_size = params.sink_count * sink_command_size;
+        const u64 sink_commands_size = params.sink_count * sink_command_size;

-            const u64 splitter_commands_size =
-                params.num_splitter_send_channels * max_mix_buffers * mix_ramp_command_size;
+        const u64 splitter_commands_size =
+            params.num_splitter_send_channels * max_mix_buffers * mix_ramp_command_size;

-            const u64 submix_commands_size = params.submix_count * submix_command_max_size;
+        const u64 submix_commands_size = params.submix_count * submix_command_max_size;

-            const u64 voice_commands_size = params.voice_count * voice_command_max_size;
+        const u64 voice_commands_size = params.voice_count * voice_command_max_size;

-            return effect_commands_size + final_mix_commands_size + perf_commands_size +
-                   sink_commands_size + splitter_commands_size + submix_commands_size +
-                   voice_commands_size + alignment;
-        };
+        return effect_commands_size + final_mix_commands_size + perf_commands_size +
+               sink_commands_size + splitter_commands_size + submix_commands_size +
+               voice_commands_size + alignment;
+    };

    IPC::RequestParser rp{ctx};
    const auto params = rp.PopRaw<AudioCore::AudioRendererParameter>();
@@ -578,12 +614,16 @@ void AudRenU::GetAudioRendererWorkBufferSize(Kernel::HLERequestContext& ctx) {
 }

 void AudRenU::GetAudioDeviceService(Kernel::HLERequestContext& ctx) {
-    LOG_DEBUG(Service_Audio, "called");
+    IPC::RequestParser rp{ctx};
+    const u64 aruid = rp.Pop<u64>();

+    LOG_DEBUG(Service_Audio, "called. aruid={:016X}", aruid);
+
+    // Revisionless variant of GetAudioDeviceServiceWithRevisionInfo that
+    // always assumes the initial release revision (REV1).
    IPC::ResponseBuilder rb{ctx, 2, 0, 1};
-
    rb.Push(RESULT_SUCCESS);
-    rb.PushIpcInterface<Audio::IAudioDevice>();
+    rb.PushIpcInterface<IAudioDevice>(system, Common::MakeMagic('R', 'E', 'V', '1'));
 }

 void AudRenU::OpenAudioRendererAuto(Kernel::HLERequestContext& ctx) {
@@ -593,13 +633,19 @@ void AudRenU::OpenAudioRendererAuto(Kernel::HLERequestContext& ctx) {
 }

 void AudRenU::GetAudioDeviceServiceWithRevisionInfo(Kernel::HLERequestContext& ctx) {
-    LOG_WARNING(Service_Audio, "(STUBBED) called");
+    struct Parameters {
+        u32 revision;
+        u64 aruid;
+    };
+
+    IPC::RequestParser rp{ctx};
+    const auto [revision, aruid] = rp.PopRaw<Parameters>();
+
+    LOG_DEBUG(Service_Audio, "called. revision={:08X}, aruid={:016X}", revision, aruid);

    IPC::ResponseBuilder rb{ctx, 2, 0, 1};
-
    rb.Push(RESULT_SUCCESS);
-    rb.PushIpcInterface<Audio::IAudioDevice>(); // TODO(ogniK): Figure out what is different
-                                                // based on the current revision
+    rb.PushIpcInterface<IAudioDevice>(system, revision);
 }

 void AudRenU::OpenAudioRendererImpl(Kernel::HLERequestContext& ctx) {
@@ -608,14 +654,16 @@ void AudRenU::OpenAudioRendererImpl(Kernel::HLERequestContext& ctx) {
    IPC::ResponseBuilder rb{ctx, 2, 0, 1};

    rb.Push(RESULT_SUCCESS);
-    rb.PushIpcInterface<IAudioRenderer>(params, audren_instance_count++);
+    rb.PushIpcInterface<IAudioRenderer>(system, params, audren_instance_count++);
 }

-bool AudRenU::IsFeatureSupported(AudioFeatures feature, u32_le revision) const {
+bool IsFeatureSupported(AudioFeatures feature, u32_le revision) {
    // Byte swap
    const u32_be version_num = revision - Common::MakeMagic('R', 'E', 'V', '0');

    switch (feature) {
+    case AudioFeatures::AudioUSBDeviceOutput:
+        return version_num >= 4U;
    case AudioFeatures::Splitter:
        return version_num >= 2U;
    case AudioFeatures::PerformanceMetricsVersion2:
@@ -6,6 +6,10 @@

 #include "core/hle/service/service.h"

+namespace Core {
+class System;
+}
+
 namespace Kernel {
 class HLERequestContext;
 }
@@ -14,7 +18,7 @@ namespace Service::Audio {

 class AudRenU final : public ServiceFramework<AudRenU> {
 public:
-    explicit AudRenU();
+    explicit AudRenU(Core::System& system_);
    ~AudRenU() override;

 private:
@@ -26,14 +30,19 @@ private:

    void OpenAudioRendererImpl(Kernel::HLERequestContext& ctx);

-    enum class AudioFeatures : u32 {
-        Splitter,
-        PerformanceMetricsVersion2,
-        VariadicCommandBuffer,
-    };
-
-    bool IsFeatureSupported(AudioFeatures feature, u32_le revision) const;
    std::size_t audren_instance_count = 0;
+    Core::System& system;
 };

+// Describes a particular audio feature that may be supported in a particular revision.
+enum class AudioFeatures : u32 {
+    AudioUSBDeviceOutput,
+    Splitter,
+    PerformanceMetricsVersion2,
+    VariadicCommandBuffer,
+};
+
+// Tests if a particular audio feature is supported with a given audio revision.
+bool IsFeatureSupported(AudioFeatures feature, u32_le revision);
+
 } // namespace Service::Audio
@@ -77,7 +77,7 @@ enum class LoadState : u32 {
    Done = 1,
 };

-static void DecryptSharedFont(const std::vector<u32>& input, std::vector<u8>& output,
+static void DecryptSharedFont(const std::vector<u32>& input, Kernel::PhysicalMemory& output,
                              std::size_t& offset) {
    ASSERT_MSG(offset + (input.size() * sizeof(u32)) < SHARED_FONT_MEM_SIZE,
               "Shared fonts exceeds 17mb!");
@@ -94,7 +94,7 @@ static void DecryptSharedFont(const std::vector<u32>& input, std::vector<u8>& ou
    offset += transformed_font.size() * sizeof(u32);
 }

-static void EncryptSharedFont(const std::vector<u8>& input, std::vector<u8>& output,
+static void EncryptSharedFont(const std::vector<u8>& input, Kernel::PhysicalMemory& output,
                              std::size_t& offset) {
    ASSERT_MSG(offset + input.size() + 8 < SHARED_FONT_MEM_SIZE, "Shared fonts exceeds 17mb!");
    const u32 KEY = EXPECTED_MAGIC ^ EXPECTED_RESULT;
@@ -121,7 +121,7 @@ struct PL_U::Impl {
        return shared_font_regions.at(index);
    }

-    void BuildSharedFontsRawRegions(const std::vector<u8>& input) {
+    void BuildSharedFontsRawRegions(const Kernel::PhysicalMemory& input) {
        // As we can derive the xor key we can just populate the offsets
        // based on the shared memory dump
        unsigned cur_offset = 0;
@@ -144,7 +144,7 @@ struct PL_U::Impl {
    Kernel::SharedPtr<Kernel::SharedMemory> shared_font_mem;

    /// Backing memory for the shared font data
-    std::shared_ptr<std::vector<u8>> shared_font;
+    std::shared_ptr<Kernel::PhysicalMemory> shared_font;

    // Automatically populated based on shared_fonts dump or system archives.
    std::vector<FontRegion> shared_font_regions;
@@ -166,7 +166,7 @@ PL_U::PL_U() : ServiceFramework("pl:u"), impl{std::make_unique<Impl>()} {
    // Rebuild shared fonts from data ncas
    if (nand->HasEntry(static_cast<u64>(FontArchives::Standard),
                       FileSys::ContentRecordType::Data)) {
-        impl->shared_font = std::make_shared<std::vector<u8>>(SHARED_FONT_MEM_SIZE);
+        impl->shared_font = std::make_shared<Kernel::PhysicalMemory>(SHARED_FONT_MEM_SIZE);
        for (auto font : SHARED_FONTS) {
            const auto nca =
                nand->GetEntry(static_cast<u64>(font.first), FileSys::ContentRecordType::Data);
@@ -207,7 +207,7 @@ PL_U::PL_U() : ServiceFramework("pl:u"), impl{std::make_unique<Impl>()} {
        }

    } else {
-        impl->shared_font = std::make_shared<std::vector<u8>>(
+        impl->shared_font = std::make_shared<Kernel::PhysicalMemory>(
            SHARED_FONT_MEM_SIZE); // Shared memory needs to always be allocated and a fixed size

        const std::string user_path = FileUtil::GetUserPath(FileUtil::UserPath::SysDataDir);
@@ -206,7 +206,7 @@ void Init(std::shared_ptr<SM::ServiceManager>& sm, Core::System& system) {
    AM::InstallInterfaces(*sm, nv_flinger, system);
    AOC::InstallInterfaces(*sm);
    APM::InstallInterfaces(system);
-    Audio::InstallInterfaces(*sm);
+    Audio::InstallInterfaces(*sm, system);
    BCAT::InstallInterfaces(*sm);
    BPC::InstallInterfaces(*sm);
    BtDrv::InstallInterfaces(*sm);
@@ -295,7 +295,7 @@ Kernel::CodeSet ElfReader::LoadInto(VAddr vaddr) {
        }
    }

-    std::vector<u8> program_image(total_image_size);
+    Kernel::PhysicalMemory program_image(total_image_size);
    std::size_t current_image_position = 0;

    Kernel::CodeSet codeset;
@@ -69,7 +69,7 @@ AppLoader::LoadResult AppLoader_KIP::Load(Kernel::Process& process) {

    const VAddr base_address = process.VMManager().GetCodeRegionBaseAddress();
    Kernel::CodeSet codeset;
-    std::vector<u8> program_image;
+    Kernel::PhysicalMemory program_image;

    const auto load_segment = [&program_image](Kernel::CodeSet::Segment& segment,
                                               const std::vector<u8>& data, u32 offset) {
@@ -143,7 +143,7 @@ static bool LoadNroImpl(Kernel::Process& process, const std::vector<u8>& data,
    }

    // Build program image
-    std::vector<u8> program_image(PageAlignSize(nro_header.file_size));
+    Kernel::PhysicalMemory program_image(PageAlignSize(nro_header.file_size));
    std::memcpy(program_image.data(), data.data(), program_image.size());
    if (program_image.size() != PageAlignSize(nro_header.file_size)) {
        return {};
@@ -89,7 +89,7 @@ std::optional<VAddr> AppLoader_NSO::LoadModule(Kernel::Process& process,

    // Build program image
    Kernel::CodeSet codeset;
-    std::vector<u8> program_image;
+    Kernel::PhysicalMemory program_image;
    for (std::size_t i = 0; i < nso_header.segments.size(); ++i) {
        std::vector<u8> data =
            file.ReadBytes(nso_header.segments_compressed_size[i], nso_header.segments[i].offset);
@@ -1,5 +1,7 @@
 add_library(video_core STATIC
-    buffer_cache.h
+    buffer_cache/buffer_block.h
+    buffer_cache/buffer_cache.h
+    buffer_cache/map_interval.h
    dma_pusher.cpp
    dma_pusher.h
    debug_utils/debug_utils.cpp
@@ -100,6 +102,7 @@ add_library(video_core STATIC
    shader/decode/integer_set.cpp
    shader/decode/half_set.cpp
    shader/decode/video.cpp
+    shader/decode/warp.cpp
    shader/decode/xmad.cpp
    shader/decode/other.cpp
    shader/control_flow.cpp
@@ -1,299 +0,0 @@
-// Copyright 2019 yuzu Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#pragma once
-
-#include <array>
-#include <memory>
-#include <mutex>
-#include <unordered_map>
-#include <unordered_set>
-#include <utility>
-#include <vector>
-
-#include "common/alignment.h"
-#include "common/common_types.h"
-#include "core/core.h"
-#include "video_core/memory_manager.h"
-#include "video_core/rasterizer_cache.h"
-
-namespace VideoCore {
-class RasterizerInterface;
-}
-
-namespace VideoCommon {
-
-template <typename BufferStorageType>
-class CachedBuffer final : public RasterizerCacheObject {
-public:
-    explicit CachedBuffer(VAddr cpu_addr, u8* host_ptr)
-        : RasterizerCacheObject{host_ptr}, host_ptr{host_ptr}, cpu_addr{cpu_addr} {}
-    ~CachedBuffer() override = default;
-
-    VAddr GetCpuAddr() const override {
-        return cpu_addr;
-    }
-
-    std::size_t GetSizeInBytes() const override {
-        return size;
-    }
-
-    u8* GetWritableHostPtr() const {
-        return host_ptr;
-    }
-
-    std::size_t GetSize() const {
-        return size;
-    }
-
-    std::size_t GetCapacity() const {
-        return capacity;
-    }
-
-    bool IsInternalized() const {
-        return is_internal;
-    }
-
-    const BufferStorageType& GetBuffer() const {
-        return buffer;
-    }
-
-    void SetSize(std::size_t new_size) {
-        size = new_size;
-    }
-
-    void SetInternalState(bool is_internal_) {
-        is_internal = is_internal_;
-    }
-
-    BufferStorageType ExchangeBuffer(BufferStorageType buffer_, std::size_t new_capacity) {
-        capacity = new_capacity;
-        std::swap(buffer, buffer_);
-        return buffer_;
-    }
-
-private:
-    u8* host_ptr{};
-    VAddr cpu_addr{};
-    std::size_t size{};
-    std::size_t capacity{};
-    bool is_internal{};
-    BufferStorageType buffer;
-};
-
-template <typename BufferStorageType, typename BufferType, typename StreamBuffer>
-class BufferCache : public RasterizerCache<std::shared_ptr<CachedBuffer<BufferStorageType>>> {
-public:
-    using Buffer = std::shared_ptr<CachedBuffer<BufferStorageType>>;
-    using BufferInfo = std::pair<const BufferType*, u64>;
-
-    explicit BufferCache(VideoCore::RasterizerInterface& rasterizer, Core::System& system,
-                         std::unique_ptr<StreamBuffer> stream_buffer)
-        : RasterizerCache<Buffer>{rasterizer}, system{system},
-          stream_buffer{std::move(stream_buffer)}, stream_buffer_handle{
-                                                       this->stream_buffer->GetHandle()} {}
-    ~BufferCache() = default;
-
-    void Unregister(const Buffer& entry) override {
-        std::lock_guard lock{RasterizerCache<Buffer>::mutex};
-        if (entry->IsInternalized()) {
-            internalized_entries.erase(entry->GetCacheAddr());
-        }
-        ReserveBuffer(entry);
-        RasterizerCache<Buffer>::Unregister(entry);
-    }
-
-    void TickFrame() {
-        marked_for_destruction_index =
-            (marked_for_destruction_index + 1) % marked_for_destruction_ring_buffer.size();
-        MarkedForDestruction().clear();
-    }
-
-    BufferInfo UploadMemory(GPUVAddr gpu_addr, std::size_t size, std::size_t alignment = 4,
-                            bool internalize = false, bool is_written = false) {
-        std::lock_guard lock{RasterizerCache<Buffer>::mutex};
-
-        auto& memory_manager = system.GPU().MemoryManager();
-        const auto host_ptr = memory_manager.GetPointer(gpu_addr);
-        if (!host_ptr) {
-            return {GetEmptyBuffer(size), 0};
-        }
-        const auto cache_addr = ToCacheAddr(host_ptr);
-
-        // Cache management is a big overhead, so only cache entries with a given size.
-        // TODO: Figure out which size is the best for given games.
-        constexpr std::size_t max_stream_size = 0x800;
-        if (!internalize && size < max_stream_size &&
-            internalized_entries.find(cache_addr) == internalized_entries.end()) {
-            return StreamBufferUpload(host_ptr, size, alignment);
-        }
-
-        auto entry = RasterizerCache<Buffer>::TryGet(cache_addr);
-        if (!entry) {
-            return FixedBufferUpload(gpu_addr, host_ptr, size, internalize, is_written);
-        }
-
-        if (entry->GetSize() < size) {
-            IncreaseBufferSize(entry, size);
-        }
-        if (is_written) {
-            entry->MarkAsModified(true, *this);
-        }
-        return {ToHandle(entry->GetBuffer()), 0};
-    }
-
-    /// Uploads from a host memory. Returns the OpenGL buffer where it's located and its offset.
-    BufferInfo UploadHostMemory(const void* raw_pointer, std::size_t size,
-                                std::size_t alignment = 4) {
-        std::lock_guard lock{RasterizerCache<Buffer>::mutex};
-        return StreamBufferUpload(raw_pointer, size, alignment);
-    }
-
-    void Map(std::size_t max_size) {
-        std::tie(buffer_ptr, buffer_offset_base, invalidated) = stream_buffer->Map(max_size, 4);
-        buffer_offset = buffer_offset_base;
-    }
-
-    /// Finishes the upload stream, returns true on bindings invalidation.
-    bool Unmap() {
-        stream_buffer->Unmap(buffer_offset - buffer_offset_base);
-        return std::exchange(invalidated, false);
-    }
-
-    virtual const BufferType* GetEmptyBuffer(std::size_t size) = 0;
-
-protected:
-    void FlushObjectInner(const Buffer& entry) override {
-        DownloadBufferData(entry->GetBuffer(), 0, entry->GetSize(), entry->GetWritableHostPtr());
-    }
-
-    virtual BufferStorageType CreateBuffer(std::size_t size) = 0;
-
-    virtual const BufferType* ToHandle(const BufferStorageType& storage) = 0;
-
-    virtual void UploadBufferData(const BufferStorageType& buffer, std::size_t offset,
-                                  std::size_t size, const u8* data) = 0;
-
-    virtual void DownloadBufferData(const BufferStorageType& buffer, std::size_t offset,
-                                    std::size_t size, u8* data) = 0;
-
-    virtual void CopyBufferData(const BufferStorageType& src, const BufferStorageType& dst,
-                                std::size_t src_offset, std::size_t dst_offset,
-                                std::size_t size) = 0;
-
-private:
-    BufferInfo StreamBufferUpload(const void* raw_pointer, std::size_t size,
-                                  std::size_t alignment) {
-        AlignBuffer(alignment);
-        const std::size_t uploaded_offset = buffer_offset;
-        std::memcpy(buffer_ptr, raw_pointer, size);
-
-        buffer_ptr += size;
-        buffer_offset += size;
-        return {&stream_buffer_handle, uploaded_offset};
-    }
-
-    BufferInfo FixedBufferUpload(GPUVAddr gpu_addr, u8* host_ptr, std::size_t size,
-                                 bool internalize, bool is_written) {
-        auto& memory_manager = Core::System::GetInstance().GPU().MemoryManager();
-        const auto cpu_addr = memory_manager.GpuToCpuAddress(gpu_addr);
-        ASSERT(cpu_addr);
-
-        auto entry = GetUncachedBuffer(*cpu_addr, host_ptr);
-        entry->SetSize(size);
-        entry->SetInternalState(internalize);
-        RasterizerCache<Buffer>::Register(entry);
-
-        if (internalize) {
-            internalized_entries.emplace(ToCacheAddr(host_ptr));
-        }
-        if (is_written) {
-            entry->MarkAsModified(true, *this);
-        }
-
-        if (entry->GetCapacity() < size) {
-            MarkedForDestruction().push_back(entry->ExchangeBuffer(CreateBuffer(size), size));
-        }
-
-        UploadBufferData(entry->GetBuffer(), 0, size, host_ptr);
-        return {ToHandle(entry->GetBuffer()), 0};
-    }
-
-    void IncreaseBufferSize(Buffer& entry, std::size_t new_size) {
-        const std::size_t old_size = entry->GetSize();
-        if (entry->GetCapacity() < new_size) {
-            const auto& old_buffer = entry->GetBuffer();
-            auto new_buffer = CreateBuffer(new_size);
-
-            // Copy bits from the old buffer to the new buffer.
-            CopyBufferData(old_buffer, new_buffer, 0, 0, old_size);
-            MarkedForDestruction().push_back(
-                entry->ExchangeBuffer(std::move(new_buffer), new_size));
-
-            // This buffer could have been used
-            invalidated = true;
-        }
-        // Upload the new bits.
-        const std::size_t size_diff = new_size - old_size;
-        UploadBufferData(entry->GetBuffer(), old_size, size_diff, entry->GetHostPtr() + old_size);
-
-        // Update entry's size in the object and in the cache.
-        Unregister(entry);
-
-        entry->SetSize(new_size);
-        RasterizerCache<Buffer>::Register(entry);
-    }
-
-    Buffer GetUncachedBuffer(VAddr cpu_addr, u8* host_ptr) {
-        if (auto entry = TryGetReservedBuffer(host_ptr)) {
-            return entry;
-        }
-        return std::make_shared<CachedBuffer<BufferStorageType>>(cpu_addr, host_ptr);
-    }
-
-    Buffer TryGetReservedBuffer(u8* host_ptr) {
-        const auto it = buffer_reserve.find(ToCacheAddr(host_ptr));
-        if (it == buffer_reserve.end()) {
-            return {};
-        }
-        auto& reserve = it->second;
-        auto entry = reserve.back();
-        reserve.pop_back();
-        return entry;
-    }
-
-    void ReserveBuffer(Buffer entry) {
-        buffer_reserve[entry->GetCacheAddr()].push_back(std::move(entry));
-    }
-
-    void AlignBuffer(std::size_t alignment) {
-        // Align the offset, not the mapped pointer
-        const std::size_t offset_aligned = Common::AlignUp(buffer_offset, alignment);
-        buffer_ptr += offset_aligned - buffer_offset;
-        buffer_offset = offset_aligned;
-    }
-
-    std::vector<BufferStorageType>& MarkedForDestruction() {
-        return marked_for_destruction_ring_buffer[marked_for_destruction_index];
-    }
-
-    Core::System& system;
-
-    std::unique_ptr<StreamBuffer> stream_buffer;
-    BufferType stream_buffer_handle{};
-
-    bool invalidated = false;
-
-    u8* buffer_ptr = nullptr;
-    u64 buffer_offset = 0;
-    u64 buffer_offset_base = 0;
-
-    std::size_t marked_for_destruction_index = 0;
-    std::array<std::vector<BufferStorageType>, 4> marked_for_destruction_ring_buffer;
-
-    std::unordered_set<CacheAddr> internalized_entries;
-    std::unordered_map<CacheAddr, std::vector<Buffer>> buffer_reserve;
-};
-
-} // namespace VideoCommon
@@ -0,0 +1,77 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <unordered_set>
+#include <utility>
+
+#include "common/alignment.h"
+#include "common/common_types.h"
+#include "video_core/gpu.h"
+
+namespace VideoCommon {
+
+class BufferBlock {
+public:
+    bool Overlaps(const CacheAddr start, const CacheAddr end) const {
+        return (cache_addr < end) && (cache_addr_end > start);
+    }
+
+    bool IsInside(const CacheAddr other_start, const CacheAddr other_end) const {
+        return cache_addr <= other_start && other_end <= cache_addr_end;
+    }
+
+    u8* GetWritableHostPtr() const {
+        return FromCacheAddr(cache_addr);
+    }
+
+    u8* GetWritableHostPtr(std::size_t offset) const {
+        return FromCacheAddr(cache_addr + offset);
+    }
+
+    std::size_t GetOffset(const CacheAddr in_addr) {
+        return static_cast<std::size_t>(in_addr - cache_addr);
+    }
+
+    CacheAddr GetCacheAddr() const {
+        return cache_addr;
+    }
+
+    CacheAddr GetCacheAddrEnd() const {
+        return cache_addr_end;
+    }
+
+    void SetCacheAddr(const CacheAddr new_addr) {
+        cache_addr = new_addr;
+        cache_addr_end = new_addr + size;
+    }
+
+    std::size_t GetSize() const {
+        return size;
+    }
+
+    void SetEpoch(u64 new_epoch) {
+        epoch = new_epoch;
+    }
+
+    u64 GetEpoch() {
+        return epoch;
+    }
+
+protected:
+    explicit BufferBlock(CacheAddr cache_addr, const std::size_t size) : size{size} {
+        SetCacheAddr(cache_addr);
+    }
+    ~BufferBlock() = default;
+
+private:
+    CacheAddr cache_addr{};
+    CacheAddr cache_addr_end{};
+    u64 pages{};
+    std::size_t size{};
+    u64 epoch{};
+};
+
+} // namespace VideoCommon
@@ -0,0 +1,449 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <array>
+#include <memory>
+#include <mutex>
+#include <unordered_map>
+#include <unordered_set>
+#include <utility>
+#include <vector>
+
+#include "common/alignment.h"
+#include "common/common_types.h"
+#include "core/core.h"
+#include "video_core/buffer_cache/buffer_block.h"
+#include "video_core/buffer_cache/map_interval.h"
+#include "video_core/memory_manager.h"
+
+namespace VideoCore {
+class RasterizerInterface;
+}
+
+namespace VideoCommon {
+
+using MapInterval = std::shared_ptr<MapIntervalBase>;
+
+template <typename TBuffer, typename TBufferType, typename StreamBuffer>
+class BufferCache {
+public:
+    using BufferInfo = std::pair<const TBufferType*, u64>;
+
+    BufferInfo UploadMemory(GPUVAddr gpu_addr, std::size_t size, std::size_t alignment = 4,
+                            bool is_written = false) {
+        std::lock_guard lock{mutex};
+
+        auto& memory_manager = system.GPU().MemoryManager();
+        const auto host_ptr = memory_manager.GetPointer(gpu_addr);
+        if (!host_ptr) {
+            return {GetEmptyBuffer(size), 0};
+        }
+        const auto cache_addr = ToCacheAddr(host_ptr);
+
+        // Cache management is a big overhead, so only cache entries with a given size.
+        // TODO: Figure out which size is the best for given games.
+        constexpr std::size_t max_stream_size = 0x800;
+        if (size < max_stream_size) {
+            if (!is_written && !IsRegionWritten(cache_addr, cache_addr + size - 1)) {
+                return StreamBufferUpload(host_ptr, size, alignment);
+            }
+        }
+
+        auto block = GetBlock(cache_addr, size);
+        auto map = MapAddress(block, gpu_addr, cache_addr, size);
+        if (is_written) {
+            map->MarkAsModified(true, GetModifiedTicks());
+            if (!map->IsWritten()) {
+                map->MarkAsWritten(true);
+                MarkRegionAsWritten(map->GetStart(), map->GetEnd() - 1);
+            }
+        } else {
+            if (map->IsWritten()) {
+                WriteBarrier();
+            }
+        }
+
+        const u64 offset = static_cast<u64>(block->GetOffset(cache_addr));
+
+        return {ToHandle(block), offset};
+    }
+
+    /// Uploads from a host memory. Returns the OpenGL buffer where it's located and its offset.
+    BufferInfo UploadHostMemory(const void* raw_pointer, std::size_t size,
+                                std::size_t alignment = 4) {
+        std::lock_guard lock{mutex};
+        return StreamBufferUpload(raw_pointer, size, alignment);
+    }
+
+    void Map(std::size_t max_size) {
+        std::lock_guard lock{mutex};
+
+        std::tie(buffer_ptr, buffer_offset_base, invalidated) = stream_buffer->Map(max_size, 4);
+        buffer_offset = buffer_offset_base;
+    }
+
+    /// Finishes the upload stream, returns true on bindings invalidation.
+    bool Unmap() {
+        std::lock_guard lock{mutex};
+
+        stream_buffer->Unmap(buffer_offset - buffer_offset_base);
+        return std::exchange(invalidated, false);
+    }
+
+    void TickFrame() {
+        ++epoch;
+        while (!pending_destruction.empty()) {
+            if (pending_destruction.front()->GetEpoch() + 1 > epoch) {
+                break;
+            }
+            pending_destruction.pop_front();
+        }
+    }
+
+    /// Write any cached resources overlapping the specified region back to memory
+    void FlushRegion(CacheAddr addr, std::size_t size) {
+        std::lock_guard lock{mutex};
+
+        std::vector<MapInterval> objects = GetMapsInRange(addr, size);
+        std::sort(objects.begin(), objects.end(), [](const MapInterval& a, const MapInterval& b) {
+            return a->GetModificationTick() < b->GetModificationTick();
+        });
+        for (auto& object : objects) {
+            if (object->IsModified() && object->IsRegistered()) {
+                FlushMap(object);
+            }
+        }
+    }
+
+    /// Mark the specified region as being invalidated
+    void InvalidateRegion(CacheAddr addr, u64 size) {
+        std::lock_guard lock{mutex};
+
+        std::vector<MapInterval> objects = GetMapsInRange(addr, size);
+        for (auto& object : objects) {
+            if (object->IsRegistered()) {
+                Unregister(object);
+            }
+        }
+    }
+
+    virtual const TBufferType* GetEmptyBuffer(std::size_t size) = 0;
+
+protected:
+    explicit BufferCache(VideoCore::RasterizerInterface& rasterizer, Core::System& system,
+                         std::unique_ptr<StreamBuffer> stream_buffer)
+        : rasterizer{rasterizer}, system{system}, stream_buffer{std::move(stream_buffer)},
+          stream_buffer_handle{this->stream_buffer->GetHandle()} {}
+
+    ~BufferCache() = default;
+
+    virtual const TBufferType* ToHandle(const TBuffer& storage) = 0;
+
+    virtual void WriteBarrier() = 0;
+
+    virtual TBuffer CreateBlock(CacheAddr cache_addr, std::size_t size) = 0;
+
+    virtual void UploadBlockData(const TBuffer& buffer, std::size_t offset, std::size_t size,
+                                 const u8* data) = 0;
+
+    virtual void DownloadBlockData(const TBuffer& buffer, std::size_t offset, std::size_t size,
+                                   u8* data) = 0;
+
+    virtual void CopyBlock(const TBuffer& src, const TBuffer& dst, std::size_t src_offset,
+                           std::size_t dst_offset, std::size_t size) = 0;
+
+    /// Register an object into the cache
+    void Register(const MapInterval& new_map, bool inherit_written = false) {
+        const CacheAddr cache_ptr = new_map->GetStart();
+        const std::optional<VAddr> cpu_addr =
+            system.GPU().MemoryManager().GpuToCpuAddress(new_map->GetGpuAddress());
+        if (!cache_ptr || !cpu_addr) {
+            LOG_CRITICAL(HW_GPU, "Failed to register buffer with unmapped gpu_address 0x{:016x}",
+                         new_map->GetGpuAddress());
+            return;
+        }
+        const std::size_t size = new_map->GetEnd() - new_map->GetStart();
+        new_map->SetCpuAddress(*cpu_addr);
+        new_map->MarkAsRegistered(true);
+        const IntervalType interval{new_map->GetStart(), new_map->GetEnd()};
+        mapped_addresses.insert({interval, new_map});
+        rasterizer.UpdatePagesCachedCount(*cpu_addr, size, 1);
+        if (inherit_written) {
+            MarkRegionAsWritten(new_map->GetStart(), new_map->GetEnd() - 1);
+            new_map->MarkAsWritten(true);
+        }
+    }
+
+    /// Unregisters an object from the cache
+    void Unregister(MapInterval& map) {
+        const std::size_t size = map->GetEnd() - map->GetStart();
+        rasterizer.UpdatePagesCachedCount(map->GetCpuAddress(), size, -1);
+        map->MarkAsRegistered(false);
+        if (map->IsWritten()) {
+            UnmarkRegionAsWritten(map->GetStart(), map->GetEnd() - 1);
+        }
+        const IntervalType delete_interval{map->GetStart(), map->GetEnd()};
+        mapped_addresses.erase(delete_interval);
+    }
+
+private:
+    MapInterval CreateMap(const CacheAddr start, const CacheAddr end, const GPUVAddr gpu_addr) {
+        return std::make_shared<MapIntervalBase>(start, end, gpu_addr);
+    }
+
+    MapInterval MapAddress(const TBuffer& block, const GPUVAddr gpu_addr,
+                           const CacheAddr cache_addr, const std::size_t size) {
+
+        std::vector<MapInterval> overlaps = GetMapsInRange(cache_addr, size);
+        if (overlaps.empty()) {
+            const CacheAddr cache_addr_end = cache_addr + size;
+            MapInterval new_map = CreateMap(cache_addr, cache_addr_end, gpu_addr);
+            u8* host_ptr = FromCacheAddr(cache_addr);
+            UploadBlockData(block, block->GetOffset(cache_addr), size, host_ptr);
+            Register(new_map);
+            return new_map;
+        }
+
+        const CacheAddr cache_addr_end = cache_addr + size;
+        if (overlaps.size() == 1) {
+            MapInterval& current_map = overlaps[0];
+            if (current_map->IsInside(cache_addr, cache_addr_end)) {
+                return current_map;
+            }
+        }
+        CacheAddr new_start = cache_addr;
+        CacheAddr new_end = cache_addr_end;
+        bool write_inheritance = false;
+        bool modified_inheritance = false;
+        // Calculate new buffer parameters
+        for (auto& overlap : overlaps) {
+            new_start = std::min(overlap->GetStart(), new_start);
+            new_end = std::max(overlap->GetEnd(), new_end);
+            write_inheritance |= overlap->IsWritten();
+            modified_inheritance |= overlap->IsModified();
+        }
+        GPUVAddr new_gpu_addr = gpu_addr + new_start - cache_addr;
+        for (auto& overlap : overlaps) {
+            Unregister(overlap);
+        }
+        UpdateBlock(block, new_start, new_end, overlaps);
+        MapInterval new_map = CreateMap(new_start, new_end, new_gpu_addr);
+        if (modified_inheritance) {
+            new_map->MarkAsModified(true, GetModifiedTicks());
+        }
+        Register(new_map, write_inheritance);
+        return new_map;
+    }
+
+    void UpdateBlock(const TBuffer& block, CacheAddr start, CacheAddr end,
+                     std::vector<MapInterval>& overlaps) {
+        const IntervalType base_interval{start, end};
+        IntervalSet interval_set{};
+        interval_set.add(base_interval);
+        for (auto& overlap : overlaps) {
+            const IntervalType subtract{overlap->GetStart(), overlap->GetEnd()};
+            interval_set.subtract(subtract);
+        }
+        for (auto& interval : interval_set) {
+            std::size_t size = interval.upper() - interval.lower();
+            if (size > 0) {
+                u8* host_ptr = FromCacheAddr(interval.lower());
+                UploadBlockData(block, block->GetOffset(interval.lower()), size, host_ptr);
+            }
+        }
+    }
+
+    std::vector<MapInterval> GetMapsInRange(CacheAddr addr, std::size_t size) {
+        if (size == 0) {
+            return {};
+        }
+
+        std::vector<MapInterval> objects{};
+        const IntervalType interval{addr, addr + size};
+        for (auto& pair : boost::make_iterator_range(mapped_addresses.equal_range(interval))) {
+            objects.push_back(pair.second);
+        }
+
+        return objects;
+    }
+
+    /// Returns a ticks counter used for tracking when cached objects were last modified
+    u64 GetModifiedTicks() {
+        return ++modified_ticks;
+    }
+
+    void FlushMap(MapInterval map) {
+        std::size_t size = map->GetEnd() - map->GetStart();
+        TBuffer block = blocks[map->GetStart() >> block_page_bits];
+        u8* host_ptr = FromCacheAddr(map->GetStart());
+        DownloadBlockData(block, block->GetOffset(map->GetStart()), size, host_ptr);
+        map->MarkAsModified(false, 0);
+    }
+
+    BufferInfo StreamBufferUpload(const void* raw_pointer, std::size_t size,
+                                  std::size_t alignment) {
+        AlignBuffer(alignment);
+        const std::size_t uploaded_offset = buffer_offset;
+        std::memcpy(buffer_ptr, raw_pointer, size);
+
+        buffer_ptr += size;
+        buffer_offset += size;
+        return {&stream_buffer_handle, uploaded_offset};
+    }
+
+    void AlignBuffer(std::size_t alignment) {
+        // Align the offset, not the mapped pointer
+        const std::size_t offset_aligned = Common::AlignUp(buffer_offset, alignment);
+        buffer_ptr += offset_aligned - buffer_offset;
+        buffer_offset = offset_aligned;
+    }
+
+    TBuffer EnlargeBlock(TBuffer buffer) {
+        const std::size_t old_size = buffer->GetSize();
+        const std::size_t new_size = old_size + block_page_size;
+        const CacheAddr cache_addr = buffer->GetCacheAddr();
+        TBuffer new_buffer = CreateBlock(cache_addr, new_size);
+        CopyBlock(buffer, new_buffer, 0, 0, old_size);
+        buffer->SetEpoch(epoch);
+        pending_destruction.push_back(buffer);
+        const CacheAddr cache_addr_end = cache_addr + new_size - 1;
+        u64 page_start = cache_addr >> block_page_bits;
+        const u64 page_end = cache_addr_end >> block_page_bits;
+        while (page_start <= page_end) {
+            blocks[page_start] = new_buffer;
+            ++page_start;
+        }
+        return new_buffer;
+    }
+
+    TBuffer MergeBlocks(TBuffer first, TBuffer second) {
+        const std::size_t size_1 = first->GetSize();
+        const std::size_t size_2 = second->GetSize();
+        const CacheAddr first_addr = first->GetCacheAddr();
+        const CacheAddr second_addr = second->GetCacheAddr();
+        const CacheAddr new_addr = std::min(first_addr, second_addr);
+        const std::size_t new_size = size_1 + size_2;
+        TBuffer new_buffer = CreateBlock(new_addr, new_size);
+        CopyBlock(first, new_buffer, 0, new_buffer->GetOffset(first_addr), size_1);
+        CopyBlock(second, new_buffer, 0, new_buffer->GetOffset(second_addr), size_2);
+        first->SetEpoch(epoch);
+        second->SetEpoch(epoch);
+        pending_destruction.push_back(first);
+        pending_destruction.push_back(second);
+        const CacheAddr cache_addr_end = new_addr + new_size - 1;
+        u64 page_start = new_addr >> block_page_bits;
+        const u64 page_end = cache_addr_end >> block_page_bits;
+        while (page_start <= page_end) {
+            blocks[page_start] = new_buffer;
+            ++page_start;
+        }
+        return new_buffer;
+    }
+
+    TBuffer GetBlock(const CacheAddr cache_addr, const std::size_t size) {
+        TBuffer found{};
+        const CacheAddr cache_addr_end = cache_addr + size - 1;
+        u64 page_start = cache_addr >> block_page_bits;
+        const u64 page_end = cache_addr_end >> block_page_bits;
+        const u64 num_pages = page_end - page_start + 1;
+        while (page_start <= page_end) {
+            auto it = blocks.find(page_start);
+            if (it == blocks.end()) {
+                if (found) {
+                    found = EnlargeBlock(found);
+                } else {
+                    const CacheAddr start_addr = (page_start << block_page_bits);
+                    found = CreateBlock(start_addr, block_page_size);
+                    blocks[page_start] = found;
+                }
+            } else {
+                if (found) {
+                    if (found == it->second) {
+                        ++page_start;
+                        continue;
+                    }
+                    found = MergeBlocks(found, it->second);
+                } else {
+                    found = it->second;
+                }
+            }
+            ++page_start;
+        }
+        return found;
+    }
+
+    void MarkRegionAsWritten(const CacheAddr start, const CacheAddr end) {
+        u64 page_start = start >> write_page_bit;
+        const u64 page_end = end >> write_page_bit;
+        while (page_start <= page_end) {
+            auto it = written_pages.find(page_start);
+            if (it != written_pages.end()) {
+                it->second = it->second + 1;
+            } else {
+                written_pages[page_start] = 1;
+            }
+            page_start++;
+        }
+    }
+
+    void UnmarkRegionAsWritten(const CacheAddr start, const CacheAddr end) {
+        u64 page_start = start >> write_page_bit;
+        const u64 page_end = end >> write_page_bit;
+        while (page_start <= page_end) {
+            auto it = written_pages.find(page_start);
+            if (it != written_pages.end()) {
+                if (it->second > 1) {
+                    it->second = it->second - 1;
+                } else {
+                    written_pages.erase(it);
+                }
+            }
+            page_start++;
+        }
+    }
+
+    bool IsRegionWritten(const CacheAddr start, const CacheAddr end) const {
+        u64 page_start = start >> write_page_bit;
+        const u64 page_end = end >> write_page_bit;
+        while (page_start <= page_end) {
+            if (written_pages.count(page_start) > 0) {
+                return true;
+            }
+            page_start++;
+        }
+        return false;
+    }
+
+    std::unique_ptr<StreamBuffer> stream_buffer;
+    TBufferType stream_buffer_handle{};
+
+    bool invalidated = false;
+
+    u8* buffer_ptr = nullptr;
+    u64 buffer_offset = 0;
+    u64 buffer_offset_base = 0;
+
+    using IntervalSet = boost::icl::interval_set<CacheAddr>;
+    using IntervalCache = boost::icl::interval_map<CacheAddr, MapInterval>;
+    using IntervalType = typename IntervalCache::interval_type;
+    IntervalCache mapped_addresses{};
+
+    static constexpr u64 write_page_bit{11};
+    std::unordered_map<u64, u32> written_pages{};
+
+    static constexpr u64 block_page_bits{21};
+    static constexpr u64 block_page_size{1 << block_page_bits};
+    std::unordered_map<u64, TBuffer> blocks{};
+
+    std::list<TBuffer> pending_destruction{};
+    u64 epoch{};
+    u64 modified_ticks{};
+    VideoCore::RasterizerInterface& rasterizer;
+    Core::System& system;
+    std::recursive_mutex mutex;
+};
+
+} // namespace VideoCommon
@@ -0,0 +1,89 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include "common/common_types.h"
+#include "video_core/gpu.h"
+
+namespace VideoCommon {
+
+class MapIntervalBase {
+public:
+    MapIntervalBase(const CacheAddr start, const CacheAddr end, const GPUVAddr gpu_addr)
+        : start{start}, end{end}, gpu_addr{gpu_addr} {}
+
+    void SetCpuAddress(VAddr new_cpu_addr) {
+        cpu_addr = new_cpu_addr;
+    }
+
+    VAddr GetCpuAddress() const {
+        return cpu_addr;
+    }
+
+    GPUVAddr GetGpuAddress() const {
+        return gpu_addr;
+    }
+
+    bool IsInside(const CacheAddr other_start, const CacheAddr other_end) const {
+        return (start <= other_start && other_end <= end);
+    }
+
+    bool operator==(const MapIntervalBase& rhs) const {
+        return std::tie(start, end) == std::tie(rhs.start, rhs.end);
+    }
+
+    bool operator!=(const MapIntervalBase& rhs) const {
+        return !operator==(rhs);
+    }
+
+    void MarkAsRegistered(const bool registered) {
+        is_registered = registered;
+    }
+
+    bool IsRegistered() const {
+        return is_registered;
+    }
+
+    CacheAddr GetStart() const {
+        return start;
+    }
+
+    CacheAddr GetEnd() const {
+        return end;
+    }
+
+    void MarkAsModified(const bool is_modified_, const u64 tick) {
+        is_modified = is_modified_;
+        ticks = tick;
+    }
+
+    bool IsModified() const {
+        return is_modified;
+    }
+
+    u64 GetModificationTick() const {
+        return ticks;
+    }
+
+    void MarkAsWritten(const bool is_written_) {
+        is_written = is_written_;
+    }
+
+    bool IsWritten() const {
+        return is_written;
+    }
+
+private:
+    CacheAddr start;
+    CacheAddr end;
+    GPUVAddr gpu_addr;
+    VAddr cpu_addr{};
+    bool is_written{};
+    bool is_modified{};
+    bool is_registered{};
+    u64 ticks{};
+};
+
+} // namespace VideoCommon
@@ -31,6 +31,7 @@ void DmaPusher::DispatchCalls() {
            break;
        }
    }
+    gpu.FlushCommands();
 }

 bool DmaPusher::Step() {
@@ -538,6 +538,12 @@ enum class PhysicalAttributeDirection : u64 {
    Output = 1,
 };

+enum class VoteOperation : u64 {
+    All = 0, // allThreadsNV
+    Any = 1, // anyThreadNV
+    Eq = 2,  // allThreadsEqualNV
+};
+
 union Instruction {
    Instruction& operator=(const Instruction& instr) {
        value = instr.value;
@@ -564,6 +570,13 @@ union Instruction {
        BitField<13, 1, u64> trigger;
    } nop;

+    union {
+        BitField<48, 2, VoteOperation> operation;
+        BitField<45, 3, u64> dest_pred;
+        BitField<39, 3, u64> value;
+        BitField<42, 1, u64> negate_value;
+    } vote;
+
    union {
        BitField<8, 8, Register> gpr;
        BitField<20, 24, s64> offset;
@@ -1023,8 +1036,6 @@ union Instruction {
        } f2i;

        union {
-            BitField<8, 2, Register::Size> src_size;
-            BitField<10, 2, Register::Size> dst_size;
            BitField<39, 4, u64> rounding;
            // H0, H1 extract for F16 missing
            BitField<41, 1, u64> selector; // Guessed as some games set it, TODO: reverse this value
@@ -1489,6 +1500,7 @@ public:
        SYNC,
        BRK,
        DEPBAR,
+        VOTE,
        BFE_C,
        BFE_R,
        BFE_IMM,
@@ -1651,6 +1663,7 @@ public:
        Hfma2,
        Flow,
        Synch,
+        Warp,
        Memory,
        Texture,
        Image,
@@ -1777,6 +1790,7 @@ private:
            INST("111000110100---", Id::BRK, Type::Flow, "BRK"),
            INST("111000110000----", Id::EXIT, Type::Flow, "EXIT"),
            INST("1111000011110---", Id::DEPBAR, Type::Synch, "DEPBAR"),
+            INST("0101000011011---", Id::VOTE, Type::Warp, "VOTE"),
            INST("1110111111011---", Id::LD_A, Type::Memory, "LD_A"),
            INST("1110111101001---", Id::LD_S, Type::Memory, "LD_S"),
            INST("1110111101000---", Id::LD_L, Type::Memory, "LD_L"),
@@ -120,6 +120,10 @@ bool GPU::CancelSyncptInterrupt(const u32 syncpoint_id, const u32 value) {
    return true;
 }

+void GPU::FlushCommands() {
+    renderer.Rasterizer().FlushCommands();
+}
+
 u32 RenderTargetBytesPerPixel(RenderTargetFormat format) {
    ASSERT(format != RenderTargetFormat::NONE);

@@ -19,6 +19,10 @@ inline CacheAddr ToCacheAddr(const void* host_ptr) {
    return reinterpret_cast<CacheAddr>(host_ptr);
 }

+inline u8* FromCacheAddr(CacheAddr cache_addr) {
+    return reinterpret_cast<u8*>(cache_addr);
+}
+
 namespace Core {
 class System;
 }
@@ -153,6 +157,8 @@ public:
    /// Calls a GPU method.
    void CallMethod(const MethodCall& method_call);

+    void FlushCommands();
+
    /// Returns a reference to the Maxwell3D GPU engine.
    Engines::Maxwell3D& Maxwell3D();

@@ -50,6 +50,9 @@ public:
    /// and invalidated
    virtual void FlushAndInvalidateRegion(CacheAddr addr, u64 size) = 0;

+    // Notify the rasterizer to send all written commands to the host GPU.
+    virtual void FlushCommands() = 0;
+
    /// Notify rasterizer that a frame is about to finish
    virtual void TickFrame() = 0;

@@ -7,28 +7,40 @@
 #include <glad/glad.h>

 #include "common/assert.h"
+#include "common/microprofile.h"
 #include "video_core/renderer_opengl/gl_buffer_cache.h"
 #include "video_core/renderer_opengl/gl_rasterizer.h"
 #include "video_core/renderer_opengl/gl_resource_manager.h"

 namespace OpenGL {

+MICROPROFILE_DEFINE(OpenGL_Buffer_Download, "OpenGL", "Buffer Download", MP_RGB(192, 192, 128));
+
+CachedBufferBlock::CachedBufferBlock(CacheAddr cache_addr, const std::size_t size)
+    : VideoCommon::BufferBlock{cache_addr, size} {
+    gl_buffer.Create();
+    glNamedBufferData(gl_buffer.handle, static_cast<GLsizeiptr>(size), nullptr, GL_DYNAMIC_DRAW);
+}
+
+CachedBufferBlock::~CachedBufferBlock() = default;
+
 OGLBufferCache::OGLBufferCache(RasterizerOpenGL& rasterizer, Core::System& system,
                               std::size_t stream_size)
-    : VideoCommon::BufferCache<OGLBuffer, GLuint, OGLStreamBuffer>{
+    : VideoCommon::BufferCache<Buffer, GLuint, OGLStreamBuffer>{
          rasterizer, system, std::make_unique<OGLStreamBuffer>(stream_size, true)} {}

 OGLBufferCache::~OGLBufferCache() = default;

-OGLBuffer OGLBufferCache::CreateBuffer(std::size_t size) {
-    OGLBuffer buffer;
-    buffer.Create();
-    glNamedBufferData(buffer.handle, static_cast<GLsizeiptr>(size), nullptr, GL_DYNAMIC_DRAW);
-    return buffer;
+Buffer OGLBufferCache::CreateBlock(CacheAddr cache_addr, std::size_t size) {
+    return std::make_shared<CachedBufferBlock>(cache_addr, size);
 }

-const GLuint* OGLBufferCache::ToHandle(const OGLBuffer& buffer) {
-    return &buffer.handle;
+void OGLBufferCache::WriteBarrier() {
+    glMemoryBarrier(GL_ALL_BARRIER_BITS);
+}
+
+const GLuint* OGLBufferCache::ToHandle(const Buffer& buffer) {
+    return buffer->GetHandle();
 }

 const GLuint* OGLBufferCache::GetEmptyBuffer(std::size_t) {
@@ -36,23 +48,24 @@ const GLuint* OGLBufferCache::GetEmptyBuffer(std::size_t) {
    return &null_buffer;
 }

-void OGLBufferCache::UploadBufferData(const OGLBuffer& buffer, std::size_t offset, std::size_t size,
-                                      const u8* data) {
-    glNamedBufferSubData(buffer.handle, static_cast<GLintptr>(offset),
+void OGLBufferCache::UploadBlockData(const Buffer& buffer, std::size_t offset, std::size_t size,
+                                     const u8* data) {
+    glNamedBufferSubData(*buffer->GetHandle(), static_cast<GLintptr>(offset),
                         static_cast<GLsizeiptr>(size), data);
 }

-void OGLBufferCache::DownloadBufferData(const OGLBuffer& buffer, std::size_t offset,
-                                        std::size_t size, u8* data) {
-    glGetNamedBufferSubData(buffer.handle, static_cast<GLintptr>(offset),
+void OGLBufferCache::DownloadBlockData(const Buffer& buffer, std::size_t offset, std::size_t size,
+                                       u8* data) {
+    MICROPROFILE_SCOPE(OpenGL_Buffer_Download);
+    glGetNamedBufferSubData(*buffer->GetHandle(), static_cast<GLintptr>(offset),
                            static_cast<GLsizeiptr>(size), data);
 }

-void OGLBufferCache::CopyBufferData(const OGLBuffer& src, const OGLBuffer& dst,
-                                    std::size_t src_offset, std::size_t dst_offset,
-                                    std::size_t size) {
-    glCopyNamedBufferSubData(src.handle, dst.handle, static_cast<GLintptr>(src_offset),
-                             static_cast<GLintptr>(dst_offset), static_cast<GLsizeiptr>(size));
+void OGLBufferCache::CopyBlock(const Buffer& src, const Buffer& dst, std::size_t src_offset,
+                               std::size_t dst_offset, std::size_t size) {
+    glCopyNamedBufferSubData(*src->GetHandle(), *dst->GetHandle(),
+                             static_cast<GLintptr>(src_offset), static_cast<GLintptr>(dst_offset),
+                             static_cast<GLsizeiptr>(size));
 }

 } // namespace OpenGL
@@ -7,7 +7,7 @@
 #include <memory>

 #include "common/common_types.h"
-#include "video_core/buffer_cache.h"
+#include "video_core/buffer_cache/buffer_cache.h"
 #include "video_core/rasterizer_cache.h"
 #include "video_core/renderer_opengl/gl_resource_manager.h"
 #include "video_core/renderer_opengl/gl_stream_buffer.h"
@@ -21,7 +21,24 @@ namespace OpenGL {
 class OGLStreamBuffer;
 class RasterizerOpenGL;

-class OGLBufferCache final : public VideoCommon::BufferCache<OGLBuffer, GLuint, OGLStreamBuffer> {
+class CachedBufferBlock;
+
+using Buffer = std::shared_ptr<CachedBufferBlock>;
+
+class CachedBufferBlock : public VideoCommon::BufferBlock {
+public:
+    explicit CachedBufferBlock(CacheAddr cache_addr, const std::size_t size);
+    ~CachedBufferBlock();
+
+    const GLuint* GetHandle() const {
+        return &gl_buffer.handle;
+    }
+
+private:
+    OGLBuffer gl_buffer{};
+};
+
+class OGLBufferCache final : public VideoCommon::BufferCache<Buffer, GLuint, OGLStreamBuffer> {
 public:
    explicit OGLBufferCache(RasterizerOpenGL& rasterizer, Core::System& system,
                            std::size_t stream_size);
@@ -30,18 +47,20 @@ public:
    const GLuint* GetEmptyBuffer(std::size_t) override;

 protected:
-    OGLBuffer CreateBuffer(std::size_t size) override;
+    Buffer CreateBlock(CacheAddr cache_addr, std::size_t size) override;

-    const GLuint* ToHandle(const OGLBuffer& buffer) override;
+    void WriteBarrier() override;

-    void UploadBufferData(const OGLBuffer& buffer, std::size_t offset, std::size_t size,
-                          const u8* data) override;
+    const GLuint* ToHandle(const Buffer& buffer) override;

-    void DownloadBufferData(const OGLBuffer& buffer, std::size_t offset, std::size_t size,
-                            u8* data) override;
+    void UploadBlockData(const Buffer& buffer, std::size_t offset, std::size_t size,
+                         const u8* data) override;

-    void CopyBufferData(const OGLBuffer& src, const OGLBuffer& dst, std::size_t src_offset,
-                        std::size_t dst_offset, std::size_t size) override;
+    void DownloadBlockData(const Buffer& buffer, std::size_t offset, std::size_t size,
+                           u8* data) override;
+
+    void CopyBlock(const Buffer& src, const Buffer& dst, std::size_t src_offset,
+                   std::size_t dst_offset, std::size_t size) override;
 };

 } // namespace OpenGL
@@ -27,6 +27,8 @@ Device::Device() {
    shader_storage_alignment = GetInteger<std::size_t>(GL_SHADER_STORAGE_BUFFER_OFFSET_ALIGNMENT);
    max_vertex_attributes = GetInteger<u32>(GL_MAX_VERTEX_ATTRIBS);
    max_varyings = GetInteger<u32>(GL_MAX_VARYING_VECTORS);
+    has_warp_intrinsics = GLAD_GL_NV_gpu_shader5 && GLAD_GL_NV_shader_thread_group &&
+                          GLAD_GL_NV_shader_thread_shuffle;
    has_vertex_viewport_layer = GLAD_GL_ARB_shader_viewport_layer_array;
    has_variable_aoffi = TestVariableAoffi();
    has_component_indexing_bug = TestComponentIndexingBug();
@@ -36,6 +38,7 @@ Device::Device(std::nullptr_t) {
    uniform_buffer_alignment = 0;
    max_vertex_attributes = 16;
    max_varyings = 15;
+    has_warp_intrinsics = true;
    has_vertex_viewport_layer = true;
    has_variable_aoffi = true;
    has_component_indexing_bug = false;
@@ -30,6 +30,10 @@ public:
        return max_varyings;
    }

+    bool HasWarpIntrinsics() const {
+        return has_warp_intrinsics;
+    }
+
    bool HasVertexViewportLayer() const {
        return has_vertex_viewport_layer;
    }
@@ -50,6 +54,7 @@ private:
    std::size_t shader_storage_alignment{};
    u32 max_vertex_attributes{};
    u32 max_varyings{};
+    bool has_warp_intrinsics{};
    bool has_vertex_viewport_layer{};
    bool has_variable_aoffi{};
    bool has_component_indexing_bug{};
@@ -863,6 +863,10 @@ void RasterizerOpenGL::FlushAndInvalidateRegion(CacheAddr addr, u64 size) {
    InvalidateRegion(addr, size);
 }

+void RasterizerOpenGL::FlushCommands() {
+    glFlush();
+}
+
 void RasterizerOpenGL::TickFrame() {
    buffer_cache.TickFrame();
 }
@@ -976,7 +980,7 @@ void RasterizerOpenGL::SetupGlobalMemory(const GLShader::GlobalMemoryEntry& entr
                                         GPUVAddr gpu_addr, std::size_t size) {
    const auto alignment{device.GetShaderStorageBufferAlignment()};
    const auto [ssbo, buffer_offset] =
-        buffer_cache.UploadMemory(gpu_addr, size, alignment, true, entry.IsWritten());
+        buffer_cache.UploadMemory(gpu_addr, size, alignment, entry.IsWritten());
    bind_ssbo_pushbuffer.Push(ssbo, buffer_offset, static_cast<GLsizeiptr>(size));
 }

@@ -63,6 +63,7 @@ public:
    void FlushRegion(CacheAddr addr, u64 size) override;
    void InvalidateRegion(CacheAddr addr, u64 size) override;
    void FlushAndInvalidateRegion(CacheAddr addr, u64 size) override;
+    void FlushCommands() override;
    void TickFrame() override;
    bool AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Regs::Surface& src,
                               const Tegra::Engines::Fermi2D::Regs::Surface& dst,
@@ -212,7 +212,9 @@ CachedProgram SpecializeShader(const std::string& code, const GLShader::ShaderEn
    const auto texture_buffer_usage{variant.texture_buffer_usage};

    std::string source = "#version 430 core\n"
-                         "#extension GL_ARB_separate_shader_objects : enable\n";
+                         "#extension GL_ARB_separate_shader_objects : enable\n"
+                         "#extension GL_NV_gpu_shader5 : enable\n"
+                         "#extension GL_NV_shader_thread_group : enable\n";
    if (entries.shader_viewport_layer_array) {
        source += "#extension GL_ARB_shader_viewport_layer_array : enable\n";
    }
@@ -1136,6 +1136,16 @@ private:
                               Type::Float);
    }

+    std::string FCastHalf0(Operation operation) {
+        const std::string op_a = VisitOperand(operation, 0, Type::HalfFloat);
+        return fmt::format("({})[0]", op_a);
+    }
+
+    std::string FCastHalf1(Operation operation) {
+        const std::string op_a = VisitOperand(operation, 0, Type::HalfFloat);
+        return fmt::format("({})[1]", op_a);
+    }
+
    template <Type type>
    std::string Min(Operation operation) {
        return GenerateBinaryCall(operation, "min", type, type, type);
@@ -1292,6 +1302,11 @@ private:
        return ApplyPrecise(operation, BitwiseCastResult(clamped, Type::HalfFloat));
    }

+    std::string HCastFloat(Operation operation) {
+        const std::string op_a = VisitOperand(operation, 0, Type::Float);
+        return fmt::format("fromHalf2(vec2({}, 0.0f))", op_a);
+    }
+
    std::string HUnpack(Operation operation) {
        const std::string operand{VisitOperand(operation, 0, Type::HalfFloat)};
        const auto value = [&]() -> std::string {
@@ -1720,6 +1735,48 @@ private:
        return "utof(gl_WorkGroupID"s + GetSwizzle(element) + ')';
    }

+    std::string BallotThread(Operation operation) {
+        const std::string value = VisitOperand(operation, 0, Type::Bool);
+        if (!device.HasWarpIntrinsics()) {
+            LOG_ERROR(Render_OpenGL,
+                      "Nvidia warp intrinsics are not available and its required by a shader");
+            // Stub on non-Nvidia devices by simulating all threads voting the same as the active
+            // one.
+            return fmt::format("utof({} ? 0xFFFFFFFFU : 0U)", value);
+        }
+        return fmt::format("utof(ballotThreadNV({}))", value);
+    }
+
+    std::string Vote(Operation operation, const char* func) {
+        const std::string value = VisitOperand(operation, 0, Type::Bool);
+        if (!device.HasWarpIntrinsics()) {
+            LOG_ERROR(Render_OpenGL,
+                      "Nvidia vote intrinsics are not available and its required by a shader");
+            // Stub with a warp size of one.
+            return value;
+        }
+        return fmt::format("{}({})", func, value);
+    }
+
+    std::string VoteAll(Operation operation) {
+        return Vote(operation, "allThreadsNV");
+    }
+
+    std::string VoteAny(Operation operation) {
+        return Vote(operation, "anyThreadNV");
+    }
+
+    std::string VoteEqual(Operation operation) {
+        if (!device.HasWarpIntrinsics()) {
+            LOG_ERROR(Render_OpenGL,
+                      "Nvidia vote intrinsics are not available and its required by a shader");
+            // We must return true here since a stub for a theoretical warp size of 1 will always
+            // return an equal result for all its votes.
+            return "true";
+        }
+        return Vote(operation, "allThreadsEqualNV");
+    }
+
    static constexpr std::array operation_decompilers = {
        &GLSLDecompiler::Assign,

@@ -1732,6 +1789,8 @@ private:
        &GLSLDecompiler::Negate<Type::Float>,
        &GLSLDecompiler::Absolute<Type::Float>,
        &GLSLDecompiler::FClamp,
+        &GLSLDecompiler::FCastHalf0,
+        &GLSLDecompiler::FCastHalf1,
        &GLSLDecompiler::Min<Type::Float>,
        &GLSLDecompiler::Max<Type::Float>,
        &GLSLDecompiler::FCos,
@@ -1792,6 +1851,7 @@ private:
        &GLSLDecompiler::Absolute<Type::HalfFloat>,
        &GLSLDecompiler::HNegate,
        &GLSLDecompiler::HClamp,
+        &GLSLDecompiler::HCastFloat,
        &GLSLDecompiler::HUnpack,
        &GLSLDecompiler::HMergeF32,
        &GLSLDecompiler::HMergeH0,
@@ -1867,6 +1927,11 @@ private:
        &GLSLDecompiler::WorkGroupId<0>,
        &GLSLDecompiler::WorkGroupId<1>,
        &GLSLDecompiler::WorkGroupId<2>,
+
+        &GLSLDecompiler::BallotThread,
+        &GLSLDecompiler::VoteAll,
+        &GLSLDecompiler::VoteAny,
+        &GLSLDecompiler::VoteEqual,
    };
    static_assert(operation_decompilers.size() == static_cast<std::size_t>(OperationCode::Amount));

@@ -735,6 +735,16 @@ private:
        return {};
    }

+    Id FCastHalf0(Operation operation) {
+        UNIMPLEMENTED();
+        return {};
+    }
+
+    Id FCastHalf1(Operation operation) {
+        UNIMPLEMENTED();
+        return {};
+    }
+
    Id HNegate(Operation operation) {
        UNIMPLEMENTED();
        return {};
@@ -745,6 +755,11 @@ private:
        return {};
    }

+    Id HCastFloat(Operation operation) {
+        UNIMPLEMENTED();
+        return {};
+    }
+
    Id HUnpack(Operation operation) {
        UNIMPLEMENTED();
        return {};
@@ -1057,6 +1072,26 @@ private:
        return {};
    }

+    Id BallotThread(Operation) {
+        UNIMPLEMENTED();
+        return {};
+    }
+
+    Id VoteAll(Operation) {
+        UNIMPLEMENTED();
+        return {};
+    }
+
+    Id VoteAny(Operation) {
+        UNIMPLEMENTED();
+        return {};
+    }
+
+    Id VoteEqual(Operation) {
+        UNIMPLEMENTED();
+        return {};
+    }
+
    Id DeclareBuiltIn(spv::BuiltIn builtin, spv::StorageClass storage, Id type,
                      const std::string& name) {
        const Id id = OpVariable(type, storage);
@@ -1210,6 +1245,8 @@ private:
        &SPIRVDecompiler::Unary<&Module::OpFNegate, Type::Float>,
        &SPIRVDecompiler::Unary<&Module::OpFAbs, Type::Float>,
        &SPIRVDecompiler::Ternary<&Module::OpFClamp, Type::Float>,
+        &SPIRVDecompiler::FCastHalf0,
+        &SPIRVDecompiler::FCastHalf1,
        &SPIRVDecompiler::Binary<&Module::OpFMin, Type::Float>,
        &SPIRVDecompiler::Binary<&Module::OpFMax, Type::Float>,
        &SPIRVDecompiler::Unary<&Module::OpCos, Type::Float>,
@@ -1270,6 +1307,7 @@ private:
        &SPIRVDecompiler::Unary<&Module::OpFAbs, Type::HalfFloat>,
        &SPIRVDecompiler::HNegate,
        &SPIRVDecompiler::HClamp,
+        &SPIRVDecompiler::HCastFloat,
        &SPIRVDecompiler::HUnpack,
        &SPIRVDecompiler::HMergeF32,
        &SPIRVDecompiler::HMergeH0,
@@ -1346,6 +1384,11 @@ private:
        &SPIRVDecompiler::WorkGroupId<0>,
        &SPIRVDecompiler::WorkGroupId<1>,
        &SPIRVDecompiler::WorkGroupId<2>,
+
+        &SPIRVDecompiler::BallotThread,
+        &SPIRVDecompiler::VoteAll,
+        &SPIRVDecompiler::VoteAny,
+        &SPIRVDecompiler::VoteEqual,
    };
    static_assert(operation_decompilers.size() == static_cast<std::size_t>(OperationCode::Amount));

@@ -176,6 +176,7 @@ u32 ShaderIR::DecodeInstr(NodeBlock& bb, u32 pc) {
        {OpCode::Type::Ffma, &ShaderIR::DecodeFfma},
        {OpCode::Type::Hfma2, &ShaderIR::DecodeHfma2},
        {OpCode::Type::Conversion, &ShaderIR::DecodeConversion},
+        {OpCode::Type::Warp, &ShaderIR::DecodeWarp},
        {OpCode::Type::Memory, &ShaderIR::DecodeMemory},
        {OpCode::Type::Texture, &ShaderIR::DecodeTexture},
        {OpCode::Type::Image, &ShaderIR::DecodeImage},
@@ -57,7 +57,7 @@ u32 ShaderIR::DecodeConversion(NodeBlock& bb, u32 pc) {
    case OpCode::Id::I2F_R:
    case OpCode::Id::I2F_C:
    case OpCode::Id::I2F_IMM: {
-        UNIMPLEMENTED_IF(instr.conversion.dst_size != Register::Size::Word);
+        UNIMPLEMENTED_IF(instr.conversion.dst_size == Register::Size::Long);
        UNIMPLEMENTED_IF(instr.conversion.selector);
        UNIMPLEMENTED_IF_MSG(instr.generates_cc,
                             "Condition codes generation in I2F is not implemented");
@@ -82,14 +82,19 @@ u32 ShaderIR::DecodeConversion(NodeBlock& bb, u32 pc) {
        value = GetOperandAbsNegFloat(value, false, instr.conversion.negate_a);

        SetInternalFlagsFromFloat(bb, value, instr.generates_cc);
+
+        if (instr.conversion.dst_size == Register::Size::Short) {
+            value = Operation(OperationCode::HCastFloat, PRECISE, value);
+        }
+
        SetRegister(bb, instr.gpr0, value);
        break;
    }
    case OpCode::Id::F2F_R:
    case OpCode::Id::F2F_C:
    case OpCode::Id::F2F_IMM: {
-        UNIMPLEMENTED_IF(instr.conversion.f2f.dst_size != Register::Size::Word);
-        UNIMPLEMENTED_IF(instr.conversion.f2f.src_size != Register::Size::Word);
+        UNIMPLEMENTED_IF(instr.conversion.dst_size == Register::Size::Long);
+        UNIMPLEMENTED_IF(instr.conversion.src_size == Register::Size::Long);
        UNIMPLEMENTED_IF_MSG(instr.generates_cc,
                             "Condition codes generation in F2F is not implemented");

@@ -107,6 +112,11 @@ u32 ShaderIR::DecodeConversion(NodeBlock& bb, u32 pc) {
            }
        }();

+        if (instr.conversion.src_size == Register::Size::Short) {
+            // TODO: figure where extract is sey in the encoding
+            value = Operation(OperationCode::FCastHalf0, PRECISE, value);
+        }
+
        value = GetOperandAbsNegFloat(value, instr.conversion.abs_a, instr.conversion.negate_a);

        value = [&]() {
@@ -124,19 +134,24 @@ u32 ShaderIR::DecodeConversion(NodeBlock& bb, u32 pc) {
            default:
                UNIMPLEMENTED_MSG("Unimplemented F2F rounding mode {}",
                                  static_cast<u32>(instr.conversion.f2f.rounding.Value()));
-                return Immediate(0);
+                return value;
            }
        }();
        value = GetSaturatedFloat(value, instr.alu.saturate_d);

        SetInternalFlagsFromFloat(bb, value, instr.generates_cc);
+
+        if (instr.conversion.dst_size == Register::Size::Short) {
+            value = Operation(OperationCode::HCastFloat, PRECISE, value);
+        }
+
        SetRegister(bb, instr.gpr0, value);
        break;
    }
    case OpCode::Id::F2I_R:
    case OpCode::Id::F2I_C:
    case OpCode::Id::F2I_IMM: {
-        UNIMPLEMENTED_IF(instr.conversion.src_size != Register::Size::Word);
+        UNIMPLEMENTED_IF(instr.conversion.src_size == Register::Size::Long);
        UNIMPLEMENTED_IF_MSG(instr.generates_cc,
                             "Condition codes generation in F2I is not implemented");
        Node value = [&]() {
@@ -153,6 +168,11 @@ u32 ShaderIR::DecodeConversion(NodeBlock& bb, u32 pc) {
            }
        }();

+        if (instr.conversion.src_size == Register::Size::Short) {
+            // TODO: figure where extract is sey in the encoding
+            value = Operation(OperationCode::FCastHalf0, PRECISE, value);
+        }
+
        value = GetOperandAbsNegFloat(value, instr.conversion.abs_a, instr.conversion.negate_a);

        value = [&]() {
@@ -30,7 +30,7 @@ u32 ShaderIR::DecodeHalfSetPredicate(NodeBlock& bb, u32 pc) {
    case OpCode::Id::HSETP2_C:
        cond = instr.hsetp2.cbuf_and_imm.cond;
        h_and = instr.hsetp2.cbuf_and_imm.h_and;
-        op_b = GetOperandAbsNegHalf(GetConstBuffer(instr.cbuf34.index, instr.cbuf34.offset),
+        op_b = GetOperandAbsNegHalf(GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset()),
                                    instr.hsetp2.cbuf.abs_b, instr.hsetp2.cbuf.negate_b);
        break;
    case OpCode::Id::HSETP2_IMM:
@@ -74,6 +74,13 @@ u32 ShaderIR::DecodeOther(NodeBlock& bb, u32 pc) {
            case SystemVariable::InvocationInfo:
                LOG_WARNING(HW_GPU, "MOV_SYS instruction with InvocationInfo is incomplete");
                return Immediate(0u);
+            case SystemVariable::Tid: {
+                Node value = Immediate(0);
+                value = BitfieldInsert(value, Operation(OperationCode::LocalInvocationIdX), 0, 9);
+                value = BitfieldInsert(value, Operation(OperationCode::LocalInvocationIdY), 16, 9);
+                value = BitfieldInsert(value, Operation(OperationCode::LocalInvocationIdZ), 26, 5);
+                return value;
+            }
            case SystemVariable::TidX:
                return Operation(OperationCode::LocalInvocationIdX);
            case SystemVariable::TidY:
@@ -0,0 +1,55 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/assert.h"
+#include "common/common_types.h"
+#include "video_core/engines/shader_bytecode.h"
+#include "video_core/shader/node_helper.h"
+#include "video_core/shader/shader_ir.h"
+
+namespace VideoCommon::Shader {
+
+using Tegra::Shader::Instruction;
+using Tegra::Shader::OpCode;
+using Tegra::Shader::Pred;
+using Tegra::Shader::VoteOperation;
+
+namespace {
+OperationCode GetOperationCode(VoteOperation vote_op) {
+    switch (vote_op) {
+    case VoteOperation::All:
+        return OperationCode::VoteAll;
+    case VoteOperation::Any:
+        return OperationCode::VoteAny;
+    case VoteOperation::Eq:
+        return OperationCode::VoteEqual;
+    default:
+        UNREACHABLE_MSG("Invalid vote operation={}", static_cast<u64>(vote_op));
+        return OperationCode::VoteAll;
+    }
+}
+} // Anonymous namespace
+
+u32 ShaderIR::DecodeWarp(NodeBlock& bb, u32 pc) {
+    const Instruction instr = {program_code[pc]};
+    const auto opcode = OpCode::Decode(instr);
+
+    switch (opcode->get().GetId()) {
+    case OpCode::Id::VOTE: {
+        const Node value = GetPredicate(instr.vote.value, instr.vote.negate_value != 0);
+        const Node active = Operation(OperationCode::BallotThread, value);
+        const Node vote = Operation(GetOperationCode(instr.vote.operation), value);
+        SetRegister(bb, instr.gpr0, active);
+        SetPredicate(bb, instr.vote.dest_pred, vote);
+        break;
+    }
+    default:
+        UNIMPLEMENTED_MSG("Unhandled warp instruction: {}", opcode->get().GetName());
+        break;
+    }
+
+    return pc;
+}
+
+} // namespace VideoCommon::Shader
@@ -30,6 +30,8 @@ enum class OperationCode {
    FNegate,       /// (MetaArithmetic, float a) -> float
    FAbsolute,     /// (MetaArithmetic, float a) -> float
    FClamp,        /// (MetaArithmetic, float value, float min, float max) -> float
+    FCastHalf0,    /// (MetaArithmetic, f16vec2 a) -> float
+    FCastHalf1,    /// (MetaArithmetic, f16vec2 a) -> float
    FMin,          /// (MetaArithmetic, float a, float b) -> float
    FMax,          /// (MetaArithmetic, float a, float b) -> float
    FCos,          /// (MetaArithmetic, float a) -> float
@@ -83,17 +85,18 @@ enum class OperationCode {
    UBitfieldExtract, /// (MetaArithmetic, uint value, int offset, int offset) -> uint
    UBitCount,        /// (MetaArithmetic, uint) -> uint

-    HAdd,      /// (MetaArithmetic, f16vec2 a, f16vec2 b) -> f16vec2
-    HMul,      /// (MetaArithmetic, f16vec2 a, f16vec2 b) -> f16vec2
-    HFma,      /// (MetaArithmetic, f16vec2 a, f16vec2 b, f16vec2 c) -> f16vec2
-    HAbsolute, /// (f16vec2 a) -> f16vec2
-    HNegate,   /// (f16vec2 a, bool first, bool second) -> f16vec2
-    HClamp,    /// (f16vec2 src, float min, float max) -> f16vec2
-    HUnpack,   /// (Tegra::Shader::HalfType, T value) -> f16vec2
-    HMergeF32, /// (f16vec2 src) -> float
-    HMergeH0,  /// (f16vec2 dest, f16vec2 src) -> f16vec2
-    HMergeH1,  /// (f16vec2 dest, f16vec2 src) -> f16vec2
-    HPack2,    /// (float a, float b) -> f16vec2
+    HAdd,       /// (MetaArithmetic, f16vec2 a, f16vec2 b) -> f16vec2
+    HMul,       /// (MetaArithmetic, f16vec2 a, f16vec2 b) -> f16vec2
+    HFma,       /// (MetaArithmetic, f16vec2 a, f16vec2 b, f16vec2 c) -> f16vec2
+    HAbsolute,  /// (f16vec2 a) -> f16vec2
+    HNegate,    /// (f16vec2 a, bool first, bool second) -> f16vec2
+    HClamp,     /// (f16vec2 src, float min, float max) -> f16vec2
+    HCastFloat, /// (MetaArithmetic, float a) -> f16vec2
+    HUnpack,    /// (Tegra::Shader::HalfType, T value) -> f16vec2
+    HMergeF32,  /// (f16vec2 src) -> float
+    HMergeH0,   /// (f16vec2 dest, f16vec2 src) -> f16vec2
+    HMergeH1,   /// (f16vec2 dest, f16vec2 src) -> f16vec2
+    HPack2,     /// (float a, float b) -> f16vec2

    LogicalAssign, /// (bool& dst, bool src) -> void
    LogicalAnd,    /// (bool a, bool b) -> bool
@@ -165,6 +168,11 @@ enum class OperationCode {
    WorkGroupIdY,       /// () -> uint
    WorkGroupIdZ,       /// () -> uint

+    BallotThread, /// (bool) -> uint
+    VoteAll,      /// (bool) -> bool
+    VoteAny,      /// (bool) -> bool
+    VoteEqual,    /// (bool) -> bool
+
    Amount,
 };

@@ -405,4 +405,9 @@ Node ShaderIR::BitfieldExtract(Node value, u32 offset, u32 bits) {
                     Immediate(offset), Immediate(bits));
 }

+Node ShaderIR::BitfieldInsert(Node base, Node insert, u32 offset, u32 bits) {
+    return Operation(OperationCode::UBitfieldInsert, NO_PRECISE, base, insert, Immediate(offset),
+                     Immediate(bits));
+}
+
 } // namespace VideoCommon::Shader
@@ -167,6 +167,7 @@ private:
    u32 DecodeFfma(NodeBlock& bb, u32 pc);
    u32 DecodeHfma2(NodeBlock& bb, u32 pc);
    u32 DecodeConversion(NodeBlock& bb, u32 pc);
+    u32 DecodeWarp(NodeBlock& bb, u32 pc);
    u32 DecodeMemory(NodeBlock& bb, u32 pc);
    u32 DecodeTexture(NodeBlock& bb, u32 pc);
    u32 DecodeImage(NodeBlock& bb, u32 pc);
@@ -279,6 +280,9 @@ private:
    /// Extracts a sequence of bits from a node
    Node BitfieldExtract(Node value, u32 offset, u32 bits);

+    /// Inserts a sequence of bits from a node
+    Node BitfieldInsert(Node base, Node insert, u32 offset, u32 bits);
+
    void WriteTexInstructionFloat(NodeBlock& bb, Tegra::Shader::Instruction instr,
                                  const Node4& components);

@@ -92,7 +92,6 @@ int main(int argc, char** argv) {

    int option_index = 0;

-    char* endarg;
 #ifdef _WIN32
    int argc_w;
    auto argv_w = CommandLineToArgvW(GetCommandLineW(), &argc_w);
@@ -226,7 +225,7 @@ int main(int argc, char** argv) {

    switch (load_result) {
    case Core::System::ResultStatus::ErrorGetLoader:
-        LOG_CRITICAL(Frontend, "Failed to obtain loader for %s!", filepath.c_str());
+        LOG_CRITICAL(Frontend, "Failed to obtain loader for {}!", filepath);
        return -1;
    case Core::System::ResultStatus::ErrorLoader:
        LOG_CRITICAL(Frontend, "Failed to load ROM!");
Author	SHA1	Message	Date
James Rowe	f0c75573b1	Revert "externals: Update FMT to 6.0.0" This reverts commit `ca4ca8a6dc`.	2019-08-29 12:23:34 -06:00
Ethan	ca4ca8a6dc	externals: Update FMT to 6.0.0	2019-08-29 19:37:46 +02:00
bunnei	e424615839	Merge pull request #2783 from FernandoS27/new-buffer-cache Implement a New LLE Buffer Cache	2019-08-29 13:07:01 -04:00
bunnei	f8cc5668f8	Merge pull request #2758 from ReinUsesLisp/packed-tid shader/decode: Implement S2R Tic	2019-08-29 12:58:43 -04:00
bunnei	680ab61327	Merge pull request #2786 from ReinUsesLisp/vote shader_ir: Implement VOTE on Nvidia drivers	2019-08-29 12:58:10 -04:00
ReinUsesLisp	4e35177e23	shader_ir: Implement VOTE Implement VOTE using Nvidia's intrinsics. Documentation about these can be found here https://developer.nvidia.com/reading-between-threads-shader-intrinsics Instead of using portable ARB instructions I opted to use Nvidia intrinsics because these are the closest we have to how Tegra X1 hardware renders. To stub VOTE on non-Nvidia drivers (including nouveau) this commit simulates a GPU with a warp size of one, returning what is meaningful for the instruction being emulated: * anyThreadNV(value) -> value * allThreadsNV(value) -> value * allThreadsEqualNV(value) -> true ballotARB, also known as "uint64_t(activeThreadsNV())", emits VOTE.ANY Rd, PT, PT; on nouveau's compiler. This doesn't match exactly to Nvidia's code VOTE.ALL Rd, PT, PT; Which is emulated with activeThreadsNV() by this commit. In theory this shouldn't really matter since .ANY, .ALL and .EQ affect the predicates (set to PT on those cases) and not the registers.	2019-08-21 14:50:38 -03:00
Fernando Sahmkow	83ec2091c1	Buffer Cache: Adress Feedback.	2019-08-21 12:14:27 -04:00
Fernando Sahmkow	6ce2c85047	Buffer_Cache: Implement flushing.	2019-08-21 12:14:26 -04:00
Fernando Sahmkow	de8ff8a1c6	Buffer_Cache: Implement barriers.	2019-08-21 12:14:25 -04:00
Fernando Sahmkow	286f4c446a	Buffer_Cache: Optimize and track written areas.	2019-08-21 12:14:25 -04:00
Fernando Sahmkow	5f4b746a1e	BufferCache: Rework mapping caching.	2019-08-21 12:14:24 -04:00
Fernando Sahmkow	86d8563314	Buffer_Cache: Fixes and optimizations.	2019-08-21 12:14:23 -04:00
Fernando Sahmkow	862bec001b	Video_Core: Implement a new Buffer Cache	2019-08-21 12:14:22 -04:00
bunnei	b4a8cfbd00	Merge pull request #2748 from FernandoS27/align-memory VM_Manager: Align allocated host physical memory to 256bytes	2019-08-21 12:10:10 -04:00
bunnei	d654b3d82e	Merge pull request #2769 from FernandoS27/commands-flush GPU: Flush commands on every dma pusher step.	2019-08-21 10:29:56 -04:00
bunnei	dfdd20142e	Merge pull request #2777 from ReinUsesLisp/hsetp2-fe3h-fix half_set_predicate: Fix HSETP2_C constant buffer offset	2019-08-21 10:29:17 -04:00
bunnei	cedc1aab4a	Merge pull request #2753 from FernandoS27/float-convert Shader_Ir: Implement F16 Variants of F2F, F2I, I2F.	2019-08-21 10:27:57 -04:00
bunnei	74a7ce1df7	Merge pull request #2773 from lioncash/test-unused yuzu-tester/yuzu: Remove unused variable	2019-08-21 10:27:29 -04:00
bunnei	ef584f1a3a	Merge pull request #2747 from lioncash/audio service/audren_u: Unstub ListAudioDeviceName	2019-08-18 09:08:25 -04:00
ReinUsesLisp	ec0da3ef64	half_set_predicate: Fix HSETP2_C constant buffer offset	2019-08-04 02:50:55 -03:00
Lioncash	6e11cfcdf0	yuzu-tester/yuzu: Correct format string Prevents an invalid formatting exception from being thrown.	2019-07-29 20:55:48 -04:00
Lioncash	a0ee10b114	yuzu-tester/yuzu: Remove unused variable Gets rid of a compilation warning.	2019-07-29 20:50:33 -04:00
Fernando Sahmkow	e52c895559	GPU: Flush commands on every dma pusher step. This commit ensures that the host gpu is constantly fed with commands to work with, while the guest gpu keeps producing the rest of the commands. This reduces syncing time between host and guest gpu.	2019-07-26 16:54:22 -04:00
ReinUsesLisp	104641db07	shader/decode: Implement S2R Tic	2019-07-22 16:16:10 -03:00
Fernando Sahmkow	11f4e739bd	Shader_Ir: Implement F16 Variants of F2F, F2I, I2F. This commit takes care of implementing the F16 Variants of the conversion instructions and makes sure conversions are done.	2019-07-20 17:38:25 -04:00
Fernando Sahmkow	febb88efc4	Common/Alignment: Add noexcept where required.	2019-07-19 21:49:54 -04:00
Fernando Sahmkow	024b5fe91a	Kernel: Address Feedback	2019-07-19 11:28:57 -04:00
Fernando Sahmkow	0901c33753	Common: Correct alignment allocator to work on C++14 or higher.	2019-07-19 11:11:42 -04:00
Fernando Sahmkow	9bede4eeed	VM_Manager: Align allocated memory to 256bytes This commit ensures that all backing memory allocated for the Guest CPU is aligned to 256 bytes. This due to how gpu memory works and the heavy constraints it has in the alignment of physical memory.	2019-07-19 10:06:08 -04:00
Lioncash	16730c4c43	service/audren_u: Handle audio USB output revision queries in ListAudioDeviceName() Audio devices use the supplied revision information in order to determine if USB audio output is able to be supported. In this case, we can only really handle using this revision information in ListAudioDeviceName(), where it checks if USB audio output is supported before supplying it as a device name. A few other scenarios exist where the revision info is checked, such as: - Early exiting from SetAudioDeviceOutputVolume if USB audio is attempted to be set when that device is unsupported. - Early exiting and returning 0.0f in GetAudioDeviceOutputVolume when USB output volume is queried and it's an unsupported device. - Falling back to AHUB headphones in GetActiveAudioDeviceName when the device type is USB output, but is unsupported based off the revision info. In order for these changes to also be implemented, a few other changes to the interface need to be made. Given we now properly handle everything about ListAudioDeviceName(), we no longer need to describe it as a stubbed function.	2019-07-19 07:55:27 -04:00
Lioncash	b9ebab71be	service/audren_u: Move revision testing code out of AudRenU The revision querying facilities are used by more than just audren. e.g. audio devices can use this to test whether or not USB audio output is supported. This will be used within the following change.	2019-07-19 07:55:23 -04:00
Lioncash	ed0485c599	service/audio: Remove global system accessors Trims out the lingering reliance on global state out of the audio code.	2019-07-19 07:29:36 -04:00
Lioncash	7653e4babc	service/audren_u: Remove unnecessary return value from GetActiveAudioDeviceName() This service function only ever returns a result and nothing more.	2019-07-19 06:57:31 -04:00
Lioncash	6ecbc6c557	service/audren_u: Report proper device names AudioDevice and AudioInterface aren't valid device names on the Switch. We should also be returning consistent names in GetActiveAudioDeviceName(). While we're at it, we can also handle proper name output in ListAudioDeviceName, by returning all the available devices on the Switch.	2019-07-19 06:57:30 -04:00