Merge pull request #2778 from ReinUsesLisp/nop

shader_ir: Implement NOP

.ci/templates/build-single.yml

@@ -1,20 +1,22 @@
 parameters:
   artifactSource: 'true'
+  cache: 'false'
 
 steps:
 - task: DockerInstaller@0
   displayName: 'Prepare Environment'
   inputs:
     dockerVersion: '17.09.0-ce'
-- task: CacheBeta@0
-  displayName: 'Cache Build System'
-  inputs:
-    key: yuzu-v1-$(BuildName)-$(BuildSuffix)-$(CacheSuffix)
-    path: $(System.DefaultWorkingDirectory)/ccache
-    cacheHitVar: CACHE_RESTORED
+- ${{ if eq(parameters.cache, 'true') }}:
+  - task: CacheBeta@0
+    displayName: 'Cache Build System'
+    inputs:
+      key: yuzu-v1-$(BuildName)-$(BuildSuffix)-$(CacheSuffix)
+      path: $(System.DefaultWorkingDirectory)/ccache
+      cacheHitVar: CACHE_RESTORED
 - script: chmod a+x ./.ci/scripts/$(ScriptFolder)/exec.sh && ./.ci/scripts/$(ScriptFolder)/exec.sh
   displayName: 'Build'
-- script: chmod a+x ./.ci/scripts/$(ScriptFolder)/upload.sh && ./.ci/scripts/$(ScriptFolder)/upload.sh
+- script: chmod a+x ./.ci/scripts/$(ScriptFolder)/upload.sh && RELEASE_NAME=$(BuildName) ./.ci/scripts/$(ScriptFolder)/upload.sh
   displayName: 'Package Artifacts'
 - publish: artifacts
   artifact: 'yuzu-$(BuildName)-$(BuildSuffix)'

.ci/templates/build-standard.yml

@@ -3,7 +3,7 @@ jobs:
   displayName: 'standard'
   pool:
     vmImage: ubuntu-latest
-  strategy: 
+  strategy:
     maxParallel: 10
     matrix:
       windows:
@@ -19,4 +19,5 @@ jobs:
       needSubmodules: 'true'
   - template: ./build-single.yml
     parameters:
-      artifactSource: 'false'
+      artifactSource: 'false'
+      cache: $(parameters.cache)

.ci/templates/build-testing.yml

@@ -3,19 +3,21 @@ jobs:
   displayName: 'testing'
   pool:
     vmImage: ubuntu-latest
-  strategy: 
-    maxParallel: 10
+  strategy:
+    maxParallel: 5
     matrix:
       windows:
         BuildSuffix: 'windows-testing'
         ScriptFolder: 'windows'
   steps:
+  - script: sudo apt upgrade python3-pip && pip install requests urllib3
+    displayName: 'Prepare Environment'
   - task: PythonScript@0
     condition: eq(variables['Build.Reason'], 'PullRequest')
     displayName: 'Determine Testing Status'
     inputs:
       scriptSource: 'filePath'
-      scriptPath: '../scripts/merge/check-label-presence.py'
+      scriptPath: '.ci/scripts/merge/check-label-presence.py'
       arguments: '$(System.PullRequest.PullRequestNumber) create-testing-build'
   - ${{ if eq(variables.enabletesting, 'true') }}:
     - template: ./sync-source.yml
@@ -27,4 +29,5 @@ jobs:
         matchLabel: 'testing-merge'
     - template: ./build-single.yml
       parameters:
-        artifactSource: 'false'
+        artifactSource: 'false'
+        cache: 'false'

.ci/templates/release.yml

@@ -1,29 +0,0 @@
-steps:
-  - task: DownloadPipelineArtifact@2
-    displayName: 'Download Windows Release'
-    inputs:
-      artifactName: 'yuzu-$(BuildName)-windows-mingw'
-      buildType: 'current'
-      targetPath: '$(Build.ArtifactStagingDirectory)'
-  - task: DownloadPipelineArtifact@2
-    displayName: 'Download Linux Release'
-    inputs:
-      artifactName: 'yuzu-$(BuildName)-linux'
-      buildType: 'current'
-      targetPath: '$(Build.ArtifactStagingDirectory)'
-  - task: DownloadPipelineArtifact@2
-    displayName: 'Download Release Point'
-    inputs:
-      artifactName: 'yuzu-$(BuildName)-release-point'
-      buildType: 'current'
-      targetPath: '$(Build.ArtifactStagingDirectory)'
-  - script: echo '##vso[task.setvariable variable=tagcommit]' && cat $(Build.ArtifactStagingDirectory)/tag-commit.sha
-    displayName: 'Calculate Release Point'
-  - task: GitHubRelease@0
-    inputs:
-      gitHubConnection: $(GitHubReleaseConnectionName)
-      repositoryName: '$(GitHubReleaseRepoName)'
-      action: 'create'
-      target: $(variables.tagcommit)
-      title: 'yuzu $(BuildName) #$(Build.BuildId)'
-      assets: '$(Build.ArtifactStagingDirectory)/*'

.ci/yuzu-mainline.yml

@@ -21,3 +21,5 @@ stages:
   dependsOn: format
   jobs:
   - template: ./templates/build-standard.yml
+    parameters:
+      cache: 'true'

.ci/yuzu-verify.yml

@@ -15,4 +15,6 @@ stages:
   dependsOn: format
   jobs:
   - template: ./templates/build-standard.yml
-  - template: ./templates/build-testing.yml
+    parameters:
+      cache: 'false'
+  - template: ./templates/build-testing.yml

README.md

@@ -2,6 +2,7 @@ yuzu emulator
 =============
 [![Travis CI Build Status](https://travis-ci.org/yuzu-emu/yuzu.svg?branch=master)](https://travis-ci.org/yuzu-emu/yuzu)
 [![AppVeyor CI Build Status](https://ci.appveyor.com/api/projects/status/77k97svb2usreu68?svg=true)](https://ci.appveyor.com/project/bunnei/yuzu)
+[![Azure Mainline CI Build Status](https://dev.azure.com/yuzu-emu/yuzu/_apis/build/status/yuzu%20mainline?branchName=master)](https://dev.azure.com/yuzu-emu/yuzu/)
 
 yuzu is an experimental open-source emulator for the Nintendo Switch from the creators of [Citra](https://citra-emu.org/).
 

src/core/CMakeLists.txt

@@ -111,6 +111,8 @@ add_library(core STATIC
     frontend/scope_acquire_window_context.h
     gdbstub/gdbstub.cpp
     gdbstub/gdbstub.h
+    hardware_interrupt_manager.cpp
+    hardware_interrupt_manager.h
     hle/ipc.h
     hle/ipc_helpers.h
     hle/kernel/address_arbiter.cpp
@@ -372,6 +374,7 @@ add_library(core STATIC
     hle/service/nvdrv/devices/nvmap.h
     hle/service/nvdrv/interface.cpp
     hle/service/nvdrv/interface.h
+    hle/service/nvdrv/nvdata.h
     hle/service/nvdrv/nvdrv.cpp
     hle/service/nvdrv/nvdrv.h
     hle/service/nvdrv/nvmemp.cpp

src/core/arm/unicorn/arm_unicorn.cpp

@@ -50,11 +50,14 @@ static void CodeHook(uc_engine* uc, uint64_t address, uint32_t size, void* user_
 
 static bool UnmappedMemoryHook(uc_engine* uc, uc_mem_type type, u64 addr, int size, u64 value,
                                void* user_data) {
+    auto* const system = static_cast<System*>(user_data);
+
     ARM_Interface::ThreadContext ctx{};
-    Core::CurrentArmInterface().SaveContext(ctx);
+    system->CurrentArmInterface().SaveContext(ctx);
     ASSERT_MSG(false, "Attempted to read from unmapped memory: 0x{:X}, pc=0x{:X}, lr=0x{:X}", addr,
                ctx.pc, ctx.cpu_registers[30]);
-    return {};
+
+    return false;
 }
 
 ARM_Unicorn::ARM_Unicorn(System& system) : system{system} {
@@ -65,7 +68,7 @@ ARM_Unicorn::ARM_Unicorn(System& system) : system{system} {
 
     uc_hook hook{};
     CHECKED(uc_hook_add(uc, &hook, UC_HOOK_INTR, (void*)InterruptHook, this, 0, -1));
-    CHECKED(uc_hook_add(uc, &hook, UC_HOOK_MEM_INVALID, (void*)UnmappedMemoryHook, this, 0, -1));
+    CHECKED(uc_hook_add(uc, &hook, UC_HOOK_MEM_INVALID, (void*)UnmappedMemoryHook, &system, 0, -1));
     if (GDBStub::IsServerEnabled()) {
         CHECKED(uc_hook_add(uc, &hook, UC_HOOK_CODE, (void*)CodeHook, this, 0, -1));
         last_bkpt_hit = false;

src/core/core.cpp

@@ -19,6 +19,7 @@
 #include "core/file_sys/vfs_concat.h"
 #include "core/file_sys/vfs_real.h"
 #include "core/gdbstub/gdbstub.h"
+#include "core/hardware_interrupt_manager.h"
 #include "core/hle/kernel/client_port.h"
 #include "core/hle/kernel/kernel.h"
 #include "core/hle/kernel/process.h"
@@ -151,7 +152,7 @@ struct System::Impl {
         if (!renderer->Init()) {
             return ResultStatus::ErrorVideoCore;
         }
-
+        interrupt_manager = std::make_unique<Core::Hardware::InterruptManager>(system);
         gpu_core = VideoCore::CreateGPU(system);
 
         is_powered_on = true;
@@ -298,6 +299,7 @@ struct System::Impl {
     std::unique_ptr<VideoCore::RendererBase> renderer;
     std::unique_ptr<Tegra::GPU> gpu_core;
     std::shared_ptr<Tegra::DebugContext> debug_context;
+    std::unique_ptr<Core::Hardware::InterruptManager> interrupt_manager;
     CpuCoreManager cpu_core_manager;
     bool is_powered_on = false;
 
@@ -444,6 +446,14 @@ const Tegra::GPU& System::GPU() const {
     return *impl->gpu_core;
 }
 
+Core::Hardware::InterruptManager& System::InterruptManager() {
+    return *impl->interrupt_manager;
+}
+
+const Core::Hardware::InterruptManager& System::InterruptManager() const {
+    return *impl->interrupt_manager;
+}
+
 VideoCore::RendererBase& System::Renderer() {
     return *impl->renderer;
 }

src/core/core.h

@@ -70,6 +70,10 @@ namespace Core::Timing {
 class CoreTiming;
 }
 
+namespace Core::Hardware {
+class InterruptManager;
+}
+
 namespace Core {
 
 class ARM_Interface;
@@ -234,6 +238,12 @@ public:
     /// Provides a constant reference to the core timing instance.
     const Timing::CoreTiming& CoreTiming() const;
 
+    /// Provides a reference to the interrupt manager instance.
+    Core::Hardware::InterruptManager& InterruptManager();
+
+    /// Provides a constant reference to the interrupt manager instance.
+    const Core::Hardware::InterruptManager& InterruptManager() const;
+
     /// Provides a reference to the kernel instance.
     Kernel::KernelCore& Kernel();
 
@@ -327,10 +337,6 @@ private:
     static System s_instance;
 };
 
-inline ARM_Interface& CurrentArmInterface() {
-    return System::GetInstance().CurrentArmInterface();
-}
-
 inline Kernel::Process* CurrentProcess() {
     return System::GetInstance().CurrentProcess();
 }

src/core/file_sys/program_metadata.cpp

@@ -94,6 +94,10 @@ u64 ProgramMetadata::GetFilesystemPermissions() const {
     return aci_file_access.permissions;
 }
 
+u32 ProgramMetadata::GetSystemResourceSize() const {
+    return npdm_header.system_resource_size;
+}
+
 const ProgramMetadata::KernelCapabilityDescriptors& ProgramMetadata::GetKernelCapabilities() const {
     return aci_kernel_capabilities;
 }

src/core/file_sys/program_metadata.h

@@ -58,6 +58,7 @@ public:
     u32 GetMainThreadStackSize() const;
     u64 GetTitleID() const;
     u64 GetFilesystemPermissions() const;
+    u32 GetSystemResourceSize() const;
     const KernelCapabilityDescriptors& GetKernelCapabilities() const;
 
     void Print() const;
@@ -76,7 +77,8 @@ private:
         u8 reserved_3;
         u8 main_thread_priority;
         u8 main_thread_cpu;
-        std::array<u8, 8> reserved_4;
+        std::array<u8, 4> reserved_4;
+        u32_le system_resource_size;
         u32_le process_category;
         u32_le main_stack_size;
         std::array<u8, 0x10> application_name;

src/core/hardware_interrupt_manager.cpp

@@ -0,0 +1,30 @@
+// Copyright 2019 Yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "core/core.h"
+#include "core/core_timing.h"
+#include "core/hardware_interrupt_manager.h"
+#include "core/hle/service/nvdrv/interface.h"
+#include "core/hle/service/sm/sm.h"
+
+namespace Core::Hardware {
+
+InterruptManager::InterruptManager(Core::System& system_in) : system(system_in) {
+    gpu_interrupt_event =
+        system.CoreTiming().RegisterEvent("GPUInterrupt", [this](u64 message, s64) {
+            auto nvdrv = system.ServiceManager().GetService<Service::Nvidia::NVDRV>("nvdrv");
+            const u32 syncpt = static_cast<u32>(message >> 32);
+            const u32 value = static_cast<u32>(message);
+            nvdrv->SignalGPUInterruptSyncpt(syncpt, value);
+        });
+}
+
+InterruptManager::~InterruptManager() = default;
+
+void InterruptManager::GPUInterruptSyncpt(const u32 syncpoint_id, const u32 value) {
+    const u64 msg = (static_cast<u64>(syncpoint_id) << 32ULL) | value;
+    system.CoreTiming().ScheduleEvent(10, gpu_interrupt_event, msg);
+}
+
+} // namespace Core::Hardware

src/core/hardware_interrupt_manager.h

@@ -0,0 +1,31 @@
+// 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"
+
+namespace Core {
+class System;
+}
+
+namespace Core::Timing {
+struct EventType;
+}
+
+namespace Core::Hardware {
+
+class InterruptManager {
+public:
+    explicit InterruptManager(Core::System& system);
+    ~InterruptManager();
+
+    void GPUInterruptSyncpt(u32 syncpoint_id, u32 value);
+
+private:
+    Core::System& system;
+    Core::Timing::EventType* gpu_interrupt_event{};
+};
+
+} // namespace Core::Hardware

src/core/hle/kernel/process.cpp

@@ -129,20 +129,17 @@ u64 Process::GetTotalPhysicalMemoryAvailable() const {
     return vm_manager.GetTotalPhysicalMemoryAvailable();
 }
 
-u64 Process::GetTotalPhysicalMemoryAvailableWithoutMmHeap() const {
-    // TODO: Subtract the personal heap size from this when the
-    //       personal heap is implemented.
-    return GetTotalPhysicalMemoryAvailable();
+u64 Process::GetTotalPhysicalMemoryAvailableWithoutSystemResource() const {
+    return GetTotalPhysicalMemoryAvailable() - GetSystemResourceSize();
 }
 
 u64 Process::GetTotalPhysicalMemoryUsed() const {
-    return vm_manager.GetCurrentHeapSize() + main_thread_stack_size + code_memory_size;
+    return vm_manager.GetCurrentHeapSize() + main_thread_stack_size + code_memory_size +
+           GetSystemResourceUsage();
 }
 
-u64 Process::GetTotalPhysicalMemoryUsedWithoutMmHeap() const {
-    // TODO: Subtract the personal heap size from this when the
-    //       personal heap is implemented.
-    return GetTotalPhysicalMemoryUsed();
+u64 Process::GetTotalPhysicalMemoryUsedWithoutSystemResource() const {
+    return GetTotalPhysicalMemoryUsed() - GetSystemResourceUsage();
 }
 
 void Process::RegisterThread(const Thread* thread) {
@@ -172,6 +169,7 @@ ResultCode Process::LoadFromMetadata(const FileSys::ProgramMetadata& metadata) {
     program_id = metadata.GetTitleID();
     ideal_core = metadata.GetMainThreadCore();
     is_64bit_process = metadata.Is64BitProgram();
+    system_resource_size = metadata.GetSystemResourceSize();
 
     vm_manager.Reset(metadata.GetAddressSpaceType());
 
@@ -186,19 +184,11 @@ ResultCode Process::LoadFromMetadata(const FileSys::ProgramMetadata& metadata) {
 }
 
 void Process::Run(s32 main_thread_priority, u64 stack_size) {
-    // The kernel always ensures that the given stack size is page aligned.
-    main_thread_stack_size = Common::AlignUp(stack_size, Memory::PAGE_SIZE);
-
-    // Allocate and map the main thread stack
-    // TODO(bunnei): This is heap area that should be allocated by the kernel and not mapped as part
-    // of the user address space.
-    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),
-                        0, main_thread_stack_size, MemoryState::Stack)
-        .Unwrap();
+    AllocateMainThreadStack(stack_size);
+    tls_region_address = CreateTLSRegion();
 
     vm_manager.LogLayout();
+
     ChangeStatus(ProcessStatus::Running);
 
     SetupMainThread(*this, kernel, main_thread_priority);
@@ -228,6 +218,9 @@ void Process::PrepareForTermination() {
     stop_threads(system.Scheduler(2).GetThreadList());
     stop_threads(system.Scheduler(3).GetThreadList());
 
+    FreeTLSRegion(tls_region_address);
+    tls_region_address = 0;
+
     ChangeStatus(ProcessStatus::Exited);
 }
 
@@ -327,4 +320,16 @@ void Process::ChangeStatus(ProcessStatus new_status) {
     WakeupAllWaitingThreads();
 }
 
+void Process::AllocateMainThreadStack(u64 stack_size) {
+    // The kernel always ensures that the given stack size is page aligned.
+    main_thread_stack_size = Common::AlignUp(stack_size, Memory::PAGE_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),
+                        0, main_thread_stack_size, MemoryState::Stack)
+        .Unwrap();
+}
+
 } // namespace Kernel

src/core/hle/kernel/process.h

@@ -135,6 +135,11 @@ public:
         return mutex;
     }
 
+    /// Gets the address to the process' dedicated TLS region.
+    VAddr GetTLSRegionAddress() const {
+        return tls_region_address;
+    }
+
     /// Gets the current status of the process
     ProcessStatus GetStatus() const {
         return status;
@@ -168,8 +173,24 @@ public:
         return capabilities.GetPriorityMask();
     }
 
-    u32 IsVirtualMemoryEnabled() const {
-        return is_virtual_address_memory_enabled;
+    /// Gets the amount of secure memory to allocate for memory management.
+    u32 GetSystemResourceSize() const {
+        return system_resource_size;
+    }
+
+    /// Gets the amount of secure memory currently in use for memory management.
+    u32 GetSystemResourceUsage() const {
+        // On hardware, this returns the amount of system resource memory that has
+        // been used by the kernel. This is problematic for Yuzu to emulate, because
+        // system resource memory is used for page tables -- and yuzu doesn't really
+        // have a way to calculate how much memory is required for page tables for
+        // the current process at any given time.
+        // TODO: Is this even worth implementing? Games may retrieve this value via
+        // an SDK function that gets used + available system resource size for debug
+        // or diagnostic purposes. However, it seems unlikely that a game would make
+        // decisions based on how much system memory is dedicated to its page tables.
+        // Is returning a value other than zero wise?
+        return 0;
     }
 
     /// Whether this process is an AArch64 or AArch32 process.
@@ -196,15 +217,15 @@ public:
     u64 GetTotalPhysicalMemoryAvailable() const;
 
     /// Retrieves the total physical memory available to this process in bytes,
-    /// without the size of the personal heap added to it.
-    u64 GetTotalPhysicalMemoryAvailableWithoutMmHeap() const;
+    /// without the size of the personal system resource heap added to it.
+    u64 GetTotalPhysicalMemoryAvailableWithoutSystemResource() const;
 
     /// Retrieves the total physical memory used by this process in bytes.
     u64 GetTotalPhysicalMemoryUsed() const;
 
     /// Retrieves the total physical memory used by this process in bytes,
-    /// without the size of the personal heap added to it.
-    u64 GetTotalPhysicalMemoryUsedWithoutMmHeap() const;
+    /// without the size of the personal system resource heap added to it.
+    u64 GetTotalPhysicalMemoryUsedWithoutSystemResource() const;
 
     /// Gets the list of all threads created with this process as their owner.
     const std::list<const Thread*>& GetThreadList() const {
@@ -280,6 +301,9 @@ private:
     /// a process signal.
     void ChangeStatus(ProcessStatus new_status);
 
+    /// Allocates the main thread stack for the process, given the stack size in bytes.
+    void AllocateMainThreadStack(u64 stack_size);
+
     /// Memory manager for this process.
     Kernel::VMManager vm_manager;
 
@@ -298,12 +322,16 @@ private:
     /// Title ID corresponding to the process
     u64 program_id = 0;
 
+    /// Specifies additional memory to be reserved for the process's memory management by the
+    /// system. When this is non-zero, secure memory is allocated and used for page table allocation
+    /// instead of using the normal global page tables/memory block management.
+    u32 system_resource_size = 0;
+
     /// Resource limit descriptor for this process
     SharedPtr<ResourceLimit> resource_limit;
 
     /// The ideal CPU core for this process, threads are scheduled on this core by default.
     u8 ideal_core = 0;
-    u32 is_virtual_address_memory_enabled = 0;
 
     /// The Thread Local Storage area is allocated as processes create threads,
     /// each TLS area is 0x200 bytes, so one page (0x1000) is split up in 8 parts, and each part
@@ -338,6 +366,9 @@ private:
     /// variable related facilities.
     Mutex mutex;
 
+    /// Address indicating the location of the process' dedicated TLS region.
+    VAddr tls_region_address = 0;
+
     /// Random values for svcGetInfo RandomEntropy
     std::array<u64, RANDOM_ENTROPY_SIZE> random_entropy{};
 

src/core/hle/kernel/svc.cpp

@@ -736,16 +736,16 @@ static ResultCode GetInfo(Core::System& system, u64* result, u64 info_id, u64 ha
         StackRegionBaseAddr = 14,
         StackRegionSize = 15,
         // 3.0.0+
-        IsVirtualAddressMemoryEnabled = 16,
-        PersonalMmHeapUsage = 17,
+        SystemResourceSize = 16,
+        SystemResourceUsage = 17,
         TitleId = 18,
         // 4.0.0+
         PrivilegedProcessId = 19,
         // 5.0.0+
         UserExceptionContextAddr = 20,
         // 6.0.0+
-        TotalPhysicalMemoryAvailableWithoutMmHeap = 21,
-        TotalPhysicalMemoryUsedWithoutMmHeap = 22,
+        TotalPhysicalMemoryAvailableWithoutSystemResource = 21,
+        TotalPhysicalMemoryUsedWithoutSystemResource = 22,
     };
 
     const auto info_id_type = static_cast<GetInfoType>(info_id);
@@ -763,12 +763,12 @@ static ResultCode GetInfo(Core::System& system, u64* result, u64 info_id, u64 ha
     case GetInfoType::StackRegionSize:
     case GetInfoType::TotalPhysicalMemoryAvailable:
     case GetInfoType::TotalPhysicalMemoryUsed:
-    case GetInfoType::IsVirtualAddressMemoryEnabled:
-    case GetInfoType::PersonalMmHeapUsage:
+    case GetInfoType::SystemResourceSize:
+    case GetInfoType::SystemResourceUsage:
     case GetInfoType::TitleId:
     case GetInfoType::UserExceptionContextAddr:
-    case GetInfoType::TotalPhysicalMemoryAvailableWithoutMmHeap:
-    case GetInfoType::TotalPhysicalMemoryUsedWithoutMmHeap: {
+    case GetInfoType::TotalPhysicalMemoryAvailableWithoutSystemResource:
+    case GetInfoType::TotalPhysicalMemoryUsedWithoutSystemResource: {
         if (info_sub_id != 0) {
             return ERR_INVALID_ENUM_VALUE;
         }
@@ -829,8 +829,13 @@ static ResultCode GetInfo(Core::System& system, u64* result, u64 info_id, u64 ha
             *result = process->GetTotalPhysicalMemoryUsed();
             return RESULT_SUCCESS;
 
-        case GetInfoType::IsVirtualAddressMemoryEnabled:
-            *result = process->IsVirtualMemoryEnabled();
+        case GetInfoType::SystemResourceSize:
+            *result = process->GetSystemResourceSize();
+            return RESULT_SUCCESS;
+
+        case GetInfoType::SystemResourceUsage:
+            LOG_WARNING(Kernel_SVC, "(STUBBED) Attempted to query system resource usage");
+            *result = process->GetSystemResourceUsage();
             return RESULT_SUCCESS;
 
         case GetInfoType::TitleId:
@@ -838,17 +843,15 @@ static ResultCode GetInfo(Core::System& system, u64* result, u64 info_id, u64 ha
             return RESULT_SUCCESS;
 
         case GetInfoType::UserExceptionContextAddr:
-            LOG_WARNING(Kernel_SVC,
-                        "(STUBBED) Attempted to query user exception context address, returned 0");
-            *result = 0;
+            *result = process->GetTLSRegionAddress();
             return RESULT_SUCCESS;
 
-        case GetInfoType::TotalPhysicalMemoryAvailableWithoutMmHeap:
-            *result = process->GetTotalPhysicalMemoryAvailable();
+        case GetInfoType::TotalPhysicalMemoryAvailableWithoutSystemResource:
+            *result = process->GetTotalPhysicalMemoryAvailableWithoutSystemResource();
             return RESULT_SUCCESS;
 
-        case GetInfoType::TotalPhysicalMemoryUsedWithoutMmHeap:
-            *result = process->GetTotalPhysicalMemoryUsedWithoutMmHeap();
+        case GetInfoType::TotalPhysicalMemoryUsedWithoutSystemResource:
+            *result = process->GetTotalPhysicalMemoryUsedWithoutSystemResource();
             return RESULT_SUCCESS;
 
         default:
@@ -953,6 +956,86 @@ static ResultCode GetInfo(Core::System& system, u64* result, u64 info_id, u64 ha
     }
 }
 
+/// Maps memory at a desired address
+static ResultCode MapPhysicalMemory(Core::System& system, VAddr addr, u64 size) {
+    LOG_DEBUG(Kernel_SVC, "called, addr=0x{:016X}, size=0x{:X}", addr, size);
+
+    if (!Common::Is4KBAligned(addr)) {
+        LOG_ERROR(Kernel_SVC, "Address is not aligned to 4KB, 0x{:016X}", addr);
+        return ERR_INVALID_ADDRESS;
+    }
+
+    if (!Common::Is4KBAligned(size)) {
+        LOG_ERROR(Kernel_SVC, "Size is not aligned to 4KB, 0x{:X}", size);
+        return ERR_INVALID_SIZE;
+    }
+
+    if (size == 0) {
+        LOG_ERROR(Kernel_SVC, "Size is zero");
+        return ERR_INVALID_SIZE;
+    }
+
+    if (!(addr < addr + size)) {
+        LOG_ERROR(Kernel_SVC, "Size causes 64-bit overflow of address");
+        return ERR_INVALID_MEMORY_RANGE;
+    }
+
+    Process* const current_process = system.Kernel().CurrentProcess();
+    auto& vm_manager = current_process->VMManager();
+
+    if (current_process->GetSystemResourceSize() == 0) {
+        LOG_ERROR(Kernel_SVC, "System Resource Size is zero");
+        return ERR_INVALID_STATE;
+    }
+
+    if (!vm_manager.IsWithinMapRegion(addr, size)) {
+        LOG_ERROR(Kernel_SVC, "Range not within map region");
+        return ERR_INVALID_MEMORY_RANGE;
+    }
+
+    return vm_manager.MapPhysicalMemory(addr, size);
+}
+
+/// Unmaps memory previously mapped via MapPhysicalMemory
+static ResultCode UnmapPhysicalMemory(Core::System& system, VAddr addr, u64 size) {
+    LOG_DEBUG(Kernel_SVC, "called, addr=0x{:016X}, size=0x{:X}", addr, size);
+
+    if (!Common::Is4KBAligned(addr)) {
+        LOG_ERROR(Kernel_SVC, "Address is not aligned to 4KB, 0x{:016X}", addr);
+        return ERR_INVALID_ADDRESS;
+    }
+
+    if (!Common::Is4KBAligned(size)) {
+        LOG_ERROR(Kernel_SVC, "Size is not aligned to 4KB, 0x{:X}", size);
+        return ERR_INVALID_SIZE;
+    }
+
+    if (size == 0) {
+        LOG_ERROR(Kernel_SVC, "Size is zero");
+        return ERR_INVALID_SIZE;
+    }
+
+    if (!(addr < addr + size)) {
+        LOG_ERROR(Kernel_SVC, "Size causes 64-bit overflow of address");
+        return ERR_INVALID_MEMORY_RANGE;
+    }
+
+    Process* const current_process = system.Kernel().CurrentProcess();
+    auto& vm_manager = current_process->VMManager();
+
+    if (current_process->GetSystemResourceSize() == 0) {
+        LOG_ERROR(Kernel_SVC, "System Resource Size is zero");
+        return ERR_INVALID_STATE;
+    }
+
+    if (!vm_manager.IsWithinMapRegion(addr, size)) {
+        LOG_ERROR(Kernel_SVC, "Range not within map region");
+        return ERR_INVALID_MEMORY_RANGE;
+    }
+
+    return vm_manager.UnmapPhysicalMemory(addr, size);
+}
+
 /// Sets the thread activity
 static ResultCode SetThreadActivity(Core::System& system, Handle handle, u32 activity) {
     LOG_DEBUG(Kernel_SVC, "called, handle=0x{:08X}, activity=0x{:08X}", handle, activity);
@@ -1654,8 +1737,8 @@ static ResultCode SignalProcessWideKey(Core::System& system, VAddr condition_var
 // Wait for an address (via Address Arbiter)
 static ResultCode WaitForAddress(Core::System& system, VAddr address, u32 type, s32 value,
                                  s64 timeout) {
-    LOG_WARNING(Kernel_SVC, "called, address=0x{:X}, type=0x{:X}, value=0x{:X}, timeout={}",
-                address, type, value, timeout);
+    LOG_TRACE(Kernel_SVC, "called, address=0x{:X}, type=0x{:X}, value=0x{:X}, timeout={}", address,
+              type, value, timeout);
 
     // If the passed address is a kernel virtual address, return invalid memory state.
     if (Memory::IsKernelVirtualAddress(address)) {
@@ -1677,8 +1760,8 @@ static ResultCode WaitForAddress(Core::System& system, VAddr address, u32 type,
 // Signals to an address (via Address Arbiter)
 static ResultCode SignalToAddress(Core::System& system, VAddr address, u32 type, s32 value,
                                   s32 num_to_wake) {
-    LOG_WARNING(Kernel_SVC, "called, address=0x{:X}, type=0x{:X}, value=0x{:X}, num_to_wake=0x{:X}",
-                address, type, value, num_to_wake);
+    LOG_TRACE(Kernel_SVC, "called, address=0x{:X}, type=0x{:X}, value=0x{:X}, num_to_wake=0x{:X}",
+              address, type, value, num_to_wake);
 
     // If the passed address is a kernel virtual address, return invalid memory state.
     if (Memory::IsKernelVirtualAddress(address)) {
@@ -2310,8 +2393,8 @@ static const FunctionDef SVC_Table[] = {
     {0x29, SvcWrap<GetInfo>, "GetInfo"},
     {0x2A, nullptr, "FlushEntireDataCache"},
     {0x2B, nullptr, "FlushDataCache"},
-    {0x2C, nullptr, "MapPhysicalMemory"},
-    {0x2D, nullptr, "UnmapPhysicalMemory"},
+    {0x2C, SvcWrap<MapPhysicalMemory>, "MapPhysicalMemory"},
+    {0x2D, SvcWrap<UnmapPhysicalMemory>, "UnmapPhysicalMemory"},
     {0x2E, nullptr, "GetFutureThreadInfo"},
     {0x2F, nullptr, "GetLastThreadInfo"},
     {0x30, SvcWrap<GetResourceLimitLimitValue>, "GetResourceLimitLimitValue"},

src/core/hle/kernel/svc_wrap.h

@@ -32,6 +32,11 @@ void SvcWrap(Core::System& system) {
     FuncReturn(system, func(system, Param(system, 0)).raw);
 }
 
+template <ResultCode func(Core::System&, u64, u64)>
+void SvcWrap(Core::System& system) {
+    FuncReturn(system, func(system, Param(system, 0), Param(system, 1)).raw);
+}
+
 template <ResultCode func(Core::System&, u32)>
 void SvcWrap(Core::System& system) {
     FuncReturn(system, func(system, static_cast<u32>(Param(system, 0))).raw);

src/core/hle/kernel/vm_manager.cpp

@@ -11,6 +11,8 @@
 #include "core/core.h"
 #include "core/file_sys/program_metadata.h"
 #include "core/hle/kernel/errors.h"
+#include "core/hle/kernel/process.h"
+#include "core/hle/kernel/resource_limit.h"
 #include "core/hle/kernel/vm_manager.h"
 #include "core/memory.h"
 #include "core/memory_setup.h"
@@ -48,10 +50,14 @@ bool VirtualMemoryArea::CanBeMergedWith(const VirtualMemoryArea& next) const {
         type != next.type) {
         return false;
     }
-    if (type == VMAType::AllocatedMemoryBlock &&
-        (backing_block != next.backing_block || offset + size != next.offset)) {
+    if ((attribute & MemoryAttribute::DeviceMapped) == MemoryAttribute::DeviceMapped) {
+        // TODO: Can device mapped memory be merged sanely?
+        // Not merging it may cause inaccuracies versus hardware when memory layout is queried.
         return false;
     }
+    if (type == VMAType::AllocatedMemoryBlock) {
+        return true;
+    }
     if (type == VMAType::BackingMemory && backing_memory + size != next.backing_memory) {
         return false;
     }
@@ -99,7 +105,7 @@ bool VMManager::IsValidHandle(VMAHandle handle) const {
 ResultVal<VMManager::VMAHandle> VMManager::MapMemoryBlock(VAddr target,
                                                           std::shared_ptr<std::vector<u8>> block,
                                                           std::size_t offset, u64 size,
-                                                          MemoryState state) {
+                                                          MemoryState state, VMAPermission perm) {
     ASSERT(block != nullptr);
     ASSERT(offset + size <= block->size());
 
@@ -109,7 +115,7 @@ ResultVal<VMManager::VMAHandle> VMManager::MapMemoryBlock(VAddr target,
     ASSERT(final_vma.size == size);
 
     final_vma.type = VMAType::AllocatedMemoryBlock;
-    final_vma.permissions = VMAPermission::ReadWrite;
+    final_vma.permissions = perm;
     final_vma.state = state;
     final_vma.backing_block = std::move(block);
     final_vma.offset = offset;
@@ -288,6 +294,166 @@ ResultVal<VAddr> VMManager::SetHeapSize(u64 size) {
     return MakeResult<VAddr>(heap_region_base);
 }
 
+ResultCode VMManager::MapPhysicalMemory(VAddr target, u64 size) {
+    const auto end_addr = target + size;
+    const auto last_addr = end_addr - 1;
+    VAddr cur_addr = target;
+
+    ResultCode result = RESULT_SUCCESS;
+
+    // Check how much memory we've already mapped.
+    const auto mapped_size_result = SizeOfAllocatedVMAsInRange(target, size);
+    if (mapped_size_result.Failed()) {
+        return mapped_size_result.Code();
+    }
+
+    // If we've already mapped the desired amount, return early.
+    const std::size_t mapped_size = *mapped_size_result;
+    if (mapped_size == size) {
+        return RESULT_SUCCESS;
+    }
+
+    // Check that we can map the memory we want.
+    const auto res_limit = system.CurrentProcess()->GetResourceLimit();
+    const u64 physmem_remaining = res_limit->GetMaxResourceValue(ResourceType::PhysicalMemory) -
+                                  res_limit->GetCurrentResourceValue(ResourceType::PhysicalMemory);
+    if (physmem_remaining < (size - mapped_size)) {
+        return ERR_RESOURCE_LIMIT_EXCEEDED;
+    }
+
+    // Keep track of the memory regions we unmap.
+    std::vector<std::pair<u64, u64>> mapped_regions;
+
+    // Iterate, trying to map memory.
+    {
+        cur_addr = target;
+
+        auto iter = FindVMA(target);
+        ASSERT_MSG(iter != vma_map.end(), "MapPhysicalMemory iter != end");
+
+        while (true) {
+            const auto& vma = iter->second;
+            const auto vma_start = vma.base;
+            const auto vma_end = vma_start + vma.size;
+            const auto vma_last = vma_end - 1;
+
+            // Map the memory block
+            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,
+                                   map_size, MemoryState::Heap, VMAPermission::ReadWrite);
+                result = map_res.Code();
+                if (result.IsError()) {
+                    break;
+                }
+
+                mapped_regions.emplace_back(cur_addr, map_size);
+            }
+
+            // Break once we hit the end of the range.
+            if (last_addr <= vma_last) {
+                break;
+            }
+
+            // Advance to the next block.
+            cur_addr = vma_end;
+            iter = FindVMA(cur_addr);
+            ASSERT_MSG(iter != vma_map.end(), "MapPhysicalMemory iter != end");
+        }
+    }
+
+    // If we failed, unmap memory.
+    if (result.IsError()) {
+        for (const auto [unmap_address, unmap_size] : mapped_regions) {
+            ASSERT_MSG(UnmapRange(unmap_address, unmap_size).IsSuccess(),
+                       "MapPhysicalMemory un-map on error");
+        }
+
+        return result;
+    }
+
+    // Update amount of mapped physical memory.
+    physical_memory_mapped += size - mapped_size;
+
+    return RESULT_SUCCESS;
+}
+
+ResultCode VMManager::UnmapPhysicalMemory(VAddr target, u64 size) {
+    const auto end_addr = target + size;
+    const auto last_addr = end_addr - 1;
+    VAddr cur_addr = target;
+
+    ResultCode result = RESULT_SUCCESS;
+
+    // Check how much memory is currently mapped.
+    const auto mapped_size_result = SizeOfUnmappablePhysicalMemoryInRange(target, size);
+    if (mapped_size_result.Failed()) {
+        return mapped_size_result.Code();
+    }
+
+    // If we've already unmapped all the memory, return early.
+    const std::size_t mapped_size = *mapped_size_result;
+    if (mapped_size == 0) {
+        return RESULT_SUCCESS;
+    }
+
+    // Keep track of the memory regions we unmap.
+    std::vector<std::pair<u64, u64>> unmapped_regions;
+
+    // Try to unmap regions.
+    {
+        cur_addr = target;
+
+        auto iter = FindVMA(target);
+        ASSERT_MSG(iter != vma_map.end(), "UnmapPhysicalMemory iter != end");
+
+        while (true) {
+            const auto& vma = iter->second;
+            const auto vma_start = vma.base;
+            const auto vma_end = vma_start + vma.size;
+            const auto vma_last = vma_end - 1;
+
+            // Unmap the memory block
+            const auto unmap_size = std::min(end_addr - cur_addr, vma_end - cur_addr);
+            if (vma.state == MemoryState::Heap) {
+                result = UnmapRange(cur_addr, unmap_size);
+                if (result.IsError()) {
+                    break;
+                }
+
+                unmapped_regions.emplace_back(cur_addr, unmap_size);
+            }
+
+            // Break once we hit the end of the range.
+            if (last_addr <= vma_last) {
+                break;
+            }
+
+            // Advance to the next block.
+            cur_addr = vma_end;
+            iter = FindVMA(cur_addr);
+            ASSERT_MSG(iter != vma_map.end(), "UnmapPhysicalMemory iter != end");
+        }
+    }
+
+    // If we failed, re-map regions.
+    // TODO: Preserve memory contents?
+    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,
+                               map_size, MemoryState::Heap, VMAPermission::None);
+            ASSERT_MSG(remap_res.Succeeded(), "UnmapPhysicalMemory re-map on error");
+        }
+    }
+
+    // Update mapped amount
+    physical_memory_mapped -= mapped_size;
+
+    return RESULT_SUCCESS;
+}
+
 ResultCode VMManager::MapCodeMemory(VAddr dst_address, VAddr src_address, u64 size) {
     constexpr auto ignore_attribute = MemoryAttribute::LockedForIPC | MemoryAttribute::DeviceMapped;
     const auto src_check_result = CheckRangeState(
@@ -435,7 +601,7 @@ ResultCode VMManager::MirrorMemory(VAddr dst_addr, VAddr src_addr, u64 size, Mem
     // Protect mirror with permissions from old region
     Reprotect(new_vma, vma->second.permissions);
     // Remove permissions from old region
-    Reprotect(vma, VMAPermission::None);
+    ReprotectRange(src_addr, size, VMAPermission::None);
 
     return RESULT_SUCCESS;
 }
@@ -568,14 +734,14 @@ VMManager::VMAIter VMManager::SplitVMA(VMAIter vma_handle, u64 offset_in_vma) {
 VMManager::VMAIter VMManager::MergeAdjacent(VMAIter iter) {
     const VMAIter next_vma = std::next(iter);
     if (next_vma != vma_map.end() && iter->second.CanBeMergedWith(next_vma->second)) {
-        iter->second.size += next_vma->second.size;
+        MergeAdjacentVMA(iter->second, next_vma->second);
         vma_map.erase(next_vma);
     }
 
     if (iter != vma_map.begin()) {
         VMAIter prev_vma = std::prev(iter);
         if (prev_vma->second.CanBeMergedWith(iter->second)) {
-            prev_vma->second.size += iter->second.size;
+            MergeAdjacentVMA(prev_vma->second, iter->second);
             vma_map.erase(iter);
             iter = prev_vma;
         }
@@ -584,6 +750,38 @@ VMManager::VMAIter VMManager::MergeAdjacent(VMAIter iter) {
     return iter;
 }
 
+void VMManager::MergeAdjacentVMA(VirtualMemoryArea& left, const VirtualMemoryArea& right) {
+    ASSERT(left.CanBeMergedWith(right));
+
+    // Always merge allocated memory blocks, even when they don't share the same backing block.
+    if (left.type == VMAType::AllocatedMemoryBlock &&
+        (left.backing_block != right.backing_block || left.offset + left.size != right.offset)) {
+        // Check if we can save work.
+        if (left.offset == 0 && left.size == left.backing_block->size()) {
+            // Fast case: left is an entire backing block.
+            left.backing_block->insert(left.backing_block->end(),
+                                       right.backing_block->begin() + right.offset,
+                                       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>>();
+            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,
+                               right.backing_block->begin() + right.offset + right.size);
+            left.backing_block = new_memory;
+            left.offset = 0;
+        }
+
+        // Page table update is needed, because backing memory changed.
+        left.size += right.size;
+        UpdatePageTableForVMA(left);
+    } else {
+        // Just update the size.
+        left.size += right.size;
+    }
+}
+
 void VMManager::UpdatePageTableForVMA(const VirtualMemoryArea& vma) {
     switch (vma.type) {
     case VMAType::Free:
@@ -758,6 +956,84 @@ VMManager::CheckResults VMManager::CheckRangeState(VAddr address, u64 size, Memo
         std::make_tuple(initial_state, initial_permissions, initial_attributes & ~ignore_mask));
 }
 
+ResultVal<std::size_t> VMManager::SizeOfAllocatedVMAsInRange(VAddr address,
+                                                             std::size_t size) const {
+    const VAddr end_addr = address + size;
+    const VAddr last_addr = end_addr - 1;
+    std::size_t mapped_size = 0;
+
+    VAddr cur_addr = address;
+    auto iter = FindVMA(cur_addr);
+    ASSERT_MSG(iter != vma_map.end(), "SizeOfAllocatedVMAsInRange iter != end");
+
+    while (true) {
+        const auto& vma = iter->second;
+        const VAddr vma_start = vma.base;
+        const VAddr vma_end = vma_start + vma.size;
+        const VAddr vma_last = vma_end - 1;
+
+        // Add size if relevant.
+        if (vma.state != MemoryState::Unmapped) {
+            mapped_size += std::min(end_addr - cur_addr, vma_end - cur_addr);
+        }
+
+        // Break once we hit the end of the range.
+        if (last_addr <= vma_last) {
+            break;
+        }
+
+        // Advance to the next block.
+        cur_addr = vma_end;
+        iter = std::next(iter);
+        ASSERT_MSG(iter != vma_map.end(), "SizeOfAllocatedVMAsInRange iter != end");
+    }
+
+    return MakeResult(mapped_size);
+}
+
+ResultVal<std::size_t> VMManager::SizeOfUnmappablePhysicalMemoryInRange(VAddr address,
+                                                                        std::size_t size) const {
+    const VAddr end_addr = address + size;
+    const VAddr last_addr = end_addr - 1;
+    std::size_t mapped_size = 0;
+
+    VAddr cur_addr = address;
+    auto iter = FindVMA(cur_addr);
+    ASSERT_MSG(iter != vma_map.end(), "SizeOfUnmappablePhysicalMemoryInRange iter != end");
+
+    while (true) {
+        const auto& vma = iter->second;
+        const auto vma_start = vma.base;
+        const auto vma_end = vma_start + vma.size;
+        const auto vma_last = vma_end - 1;
+        const auto state = vma.state;
+        const auto attr = vma.attribute;
+
+        // Memory within region must be free or mapped heap.
+        if (!((state == MemoryState::Heap && attr == MemoryAttribute::None) ||
+              (state == MemoryState::Unmapped))) {
+            return ERR_INVALID_ADDRESS_STATE;
+        }
+
+        // Add size if relevant.
+        if (state != MemoryState::Unmapped) {
+            mapped_size += std::min(end_addr - cur_addr, vma_end - cur_addr);
+        }
+
+        // Break once we hit the end of the range.
+        if (last_addr <= vma_last) {
+            break;
+        }
+
+        // Advance to the next block.
+        cur_addr = vma_end;
+        iter = std::next(iter);
+        ASSERT_MSG(iter != vma_map.end(), "SizeOfUnmappablePhysicalMemoryInRange iter != end");
+    }
+
+    return MakeResult(mapped_size);
+}
+
 u64 VMManager::GetTotalPhysicalMemoryAvailable() const {
     LOG_WARNING(Kernel, "(STUBBED) called");
     return 0xF8000000;

src/core/hle/kernel/vm_manager.h

@@ -349,7 +349,8 @@ public:
      * @param state MemoryState tag to attach to the VMA.
      */
     ResultVal<VMAHandle> MapMemoryBlock(VAddr target, std::shared_ptr<std::vector<u8>> block,
-                                        std::size_t offset, u64 size, MemoryState state);
+                                        std::size_t offset, u64 size, MemoryState state,
+                                        VMAPermission perm = VMAPermission::ReadWrite);
 
     /**
      * Maps an unmanaged host memory pointer at a given address.
@@ -450,6 +451,34 @@ public:
     ///
     ResultVal<VAddr> SetHeapSize(u64 size);
 
+    /// Maps memory at a given address.
+    ///
+    /// @param addr The virtual address to map memory at.
+    /// @param size The amount of memory to map.
+    ///
+    /// @note The destination address must lie within the Map region.
+    ///
+    /// @note This function requires that SystemResourceSize be non-zero,
+    ///       however, this is just because if it were not then the
+    ///       resulting page tables could be exploited on hardware by
+    ///       a malicious program. SystemResource usage does not need
+    ///       to be explicitly checked or updated here.
+    ResultCode MapPhysicalMemory(VAddr target, u64 size);
+
+    /// Unmaps memory at a given address.
+    ///
+    /// @param addr The virtual address to unmap memory at.
+    /// @param size The amount of memory to unmap.
+    ///
+    /// @note The destination address must lie within the Map region.
+    ///
+    /// @note This function requires that SystemResourceSize be non-zero,
+    ///       however, this is just because if it were not then the
+    ///       resulting page tables could be exploited on hardware by
+    ///       a malicious program. SystemResource usage does not need
+    ///       to be explicitly checked or updated here.
+    ResultCode UnmapPhysicalMemory(VAddr target, u64 size);
+
     /// Maps a region of memory as code memory.
     ///
     /// @param dst_address The base address of the region to create the aliasing memory region.
@@ -657,6 +686,11 @@ private:
      */
     VMAIter MergeAdjacent(VMAIter vma);
 
+    /**
+     * Merges two adjacent VMAs.
+     */
+    void MergeAdjacentVMA(VirtualMemoryArea& left, const VirtualMemoryArea& right);
+
     /// Updates the pages corresponding to this VMA so they match the VMA's attributes.
     void UpdatePageTableForVMA(const VirtualMemoryArea& vma);
 
@@ -701,6 +735,13 @@ private:
                                  MemoryAttribute attribute_mask, MemoryAttribute attribute,
                                  MemoryAttribute ignore_mask) const;
 
+    /// Gets the amount of memory currently mapped (state != Unmapped) in a range.
+    ResultVal<std::size_t> SizeOfAllocatedVMAsInRange(VAddr address, std::size_t size) const;
+
+    /// Gets the amount of memory unmappable by UnmapPhysicalMemory in a range.
+    ResultVal<std::size_t> SizeOfUnmappablePhysicalMemoryInRange(VAddr address,
+                                                                 std::size_t size) const;
+
     /**
      * A map covering the entirety of the managed address space, keyed by the `base` field of each
      * VMA. It must always be modified by splitting or merging VMAs, so that the invariant
@@ -742,6 +783,11 @@ private:
     // end of the range. This is essentially 'base_address + current_size'.
     VAddr heap_end = 0;
 
+    // The current amount of memory mapped via MapPhysicalMemory.
+    // This is used here (and in Nintendo's kernel) only for debugging, and does not impact
+    // any behavior.
+    u64 physical_memory_mapped = 0;
+
     Core::System& system;
 };
 } // namespace Kernel

src/core/hle/service/nvdrv/devices/nvdevice.h

@@ -8,6 +8,11 @@
 #include "common/bit_field.h"
 #include "common/common_types.h"
 #include "common/swap.h"
+#include "core/hle/service/nvdrv/nvdata.h"
+
+namespace Core {
+class System;
+}
 
 namespace Service::Nvidia::Devices {
 
@@ -15,7 +20,7 @@ namespace Service::Nvidia::Devices {
 /// implement the ioctl interface.
 class nvdevice {
 public:
-    nvdevice() = default;
+    explicit nvdevice(Core::System& system) : system{system} {};
     virtual ~nvdevice() = default;
     union Ioctl {
         u32_le raw;
@@ -33,7 +38,11 @@ public:
      * @param output A buffer where the output data will be written to.
      * @returns The result code of the ioctl.
      */
-    virtual u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) = 0;
+    virtual u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+                      IoctlCtrl& ctrl) = 0;
+
+protected:
+    Core::System& system;
 };
 
 } // namespace Service::Nvidia::Devices

src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp

@@ -13,10 +13,12 @@
 
 namespace Service::Nvidia::Devices {
 
-nvdisp_disp0::nvdisp_disp0(std::shared_ptr<nvmap> nvmap_dev) : nvmap_dev(std::move(nvmap_dev)) {}
+nvdisp_disp0::nvdisp_disp0(Core::System& system, std::shared_ptr<nvmap> nvmap_dev)
+    : nvdevice(system), nvmap_dev(std::move(nvmap_dev)) {}
 nvdisp_disp0 ::~nvdisp_disp0() = default;
 
-u32 nvdisp_disp0::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) {
+u32 nvdisp_disp0::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+                        IoctlCtrl& ctrl) {
     UNIMPLEMENTED_MSG("Unimplemented ioctl");
     return 0;
 }
@@ -34,9 +36,8 @@ void nvdisp_disp0::flip(u32 buffer_handle, u32 offset, u32 format, u32 width, u3
         addr,      offset,   width, height, stride, static_cast<PixelFormat>(format),
         transform, crop_rect};
 
-    auto& instance = Core::System::GetInstance();
-    instance.GetPerfStats().EndGameFrame();
-    instance.GPU().SwapBuffers(framebuffer);
+    system.GetPerfStats().EndGameFrame();
+    system.GPU().SwapBuffers(framebuffer);
 }
 
 } // namespace Service::Nvidia::Devices

src/core/hle/service/nvdrv/devices/nvdisp_disp0.h

@@ -17,10 +17,11 @@ class nvmap;
 
 class nvdisp_disp0 final : public nvdevice {
 public:
-    explicit nvdisp_disp0(std::shared_ptr<nvmap> nvmap_dev);
+    explicit nvdisp_disp0(Core::System& system, std::shared_ptr<nvmap> nvmap_dev);
     ~nvdisp_disp0() override;
 
-    u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) override;
+    u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+              IoctlCtrl& ctrl) override;
 
     /// Performs a screen flip, drawing the buffer pointed to by the handle.
     void flip(u32 buffer_handle, u32 offset, u32 format, u32 width, u32 height, u32 stride,

src/core/hle/service/nvdrv/devices/nvhost_as_gpu.cpp

@@ -22,10 +22,12 @@ enum {
 };
 }
 
-nvhost_as_gpu::nvhost_as_gpu(std::shared_ptr<nvmap> nvmap_dev) : nvmap_dev(std::move(nvmap_dev)) {}
+nvhost_as_gpu::nvhost_as_gpu(Core::System& system, std::shared_ptr<nvmap> nvmap_dev)
+    : nvdevice(system), nvmap_dev(std::move(nvmap_dev)) {}
 nvhost_as_gpu::~nvhost_as_gpu() = default;
 
-u32 nvhost_as_gpu::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) {
+u32 nvhost_as_gpu::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+                         IoctlCtrl& ctrl) {
     LOG_DEBUG(Service_NVDRV, "called, command=0x{:08X}, input_size=0x{:X}, output_size=0x{:X}",
               command.raw, input.size(), output.size());
 
@@ -65,7 +67,7 @@ u32 nvhost_as_gpu::AllocateSpace(const std::vector<u8>& input, std::vector<u8>&
     LOG_DEBUG(Service_NVDRV, "called, pages={:X}, page_size={:X}, flags={:X}", params.pages,
               params.page_size, params.flags);
 
-    auto& gpu = Core::System::GetInstance().GPU();
+    auto& gpu = system.GPU();
     const u64 size{static_cast<u64>(params.pages) * static_cast<u64>(params.page_size)};
     if (params.flags & 1) {
         params.offset = gpu.MemoryManager().AllocateSpace(params.offset, size, 1);
@@ -85,7 +87,7 @@ u32 nvhost_as_gpu::Remap(const std::vector<u8>& input, std::vector<u8>& output)
     std::vector<IoctlRemapEntry> entries(num_entries);
     std::memcpy(entries.data(), input.data(), input.size());
 
-    auto& gpu = Core::System::GetInstance().GPU();
+    auto& gpu = system.GPU();
     for (const auto& entry : entries) {
         LOG_WARNING(Service_NVDRV, "remap entry, offset=0x{:X} handle=0x{:X} pages=0x{:X}",
                     entry.offset, entry.nvmap_handle, entry.pages);
@@ -136,7 +138,7 @@ u32 nvhost_as_gpu::MapBufferEx(const std::vector<u8>& input, std::vector<u8>& ou
     // case to prevent unexpected behavior.
     ASSERT(object->id == params.nvmap_handle);
 
-    auto& gpu = Core::System::GetInstance().GPU();
+    auto& gpu = system.GPU();
 
     if (params.flags & 1) {
         params.offset = gpu.MemoryManager().MapBufferEx(object->addr, params.offset, object->size);
@@ -173,8 +175,7 @@ u32 nvhost_as_gpu::UnmapBuffer(const std::vector<u8>& input, std::vector<u8>& ou
         return 0;
     }
 
-    params.offset = Core::System::GetInstance().GPU().MemoryManager().UnmapBuffer(params.offset,
-                                                                                  itr->second.size);
+    params.offset = system.GPU().MemoryManager().UnmapBuffer(params.offset, itr->second.size);
     buffer_mappings.erase(itr->second.offset);
 
     std::memcpy(output.data(), &params, output.size());

src/core/hle/service/nvdrv/devices/nvhost_as_gpu.h

@@ -17,10 +17,11 @@ class nvmap;
 
 class nvhost_as_gpu final : public nvdevice {
 public:
-    explicit nvhost_as_gpu(std::shared_ptr<nvmap> nvmap_dev);
+    explicit nvhost_as_gpu(Core::System& system, std::shared_ptr<nvmap> nvmap_dev);
     ~nvhost_as_gpu() override;
 
-    u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) override;
+    u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+              IoctlCtrl& ctrl) override;
 
 private:
     enum class IoctlCommand : u32_le {

src/core/hle/service/nvdrv/devices/nvhost_ctrl.cpp

@@ -7,14 +7,20 @@
 
 #include "common/assert.h"
 #include "common/logging/log.h"
+#include "core/core.h"
+#include "core/hle/kernel/readable_event.h"
+#include "core/hle/kernel/writable_event.h"
 #include "core/hle/service/nvdrv/devices/nvhost_ctrl.h"
+#include "video_core/gpu.h"
 
 namespace Service::Nvidia::Devices {
 
-nvhost_ctrl::nvhost_ctrl() = default;
+nvhost_ctrl::nvhost_ctrl(Core::System& system, EventInterface& events_interface)
+    : nvdevice(system), events_interface{events_interface} {}
 nvhost_ctrl::~nvhost_ctrl() = default;
 
-u32 nvhost_ctrl::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) {
+u32 nvhost_ctrl::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+                       IoctlCtrl& ctrl) {
     LOG_DEBUG(Service_NVDRV, "called, command=0x{:08X}, input_size=0x{:X}, output_size=0x{:X}",
               command.raw, input.size(), output.size());
 
@@ -22,11 +28,15 @@ u32 nvhost_ctrl::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<
     case IoctlCommand::IocGetConfigCommand:
         return NvOsGetConfigU32(input, output);
     case IoctlCommand::IocCtrlEventWaitCommand:
-        return IocCtrlEventWait(input, output, false);
+        return IocCtrlEventWait(input, output, false, ctrl);
     case IoctlCommand::IocCtrlEventWaitAsyncCommand:
-        return IocCtrlEventWait(input, output, true);
+        return IocCtrlEventWait(input, output, true, ctrl);
     case IoctlCommand::IocCtrlEventRegisterCommand:
         return IocCtrlEventRegister(input, output);
+    case IoctlCommand::IocCtrlEventUnregisterCommand:
+        return IocCtrlEventUnregister(input, output);
+    case IoctlCommand::IocCtrlEventSignalCommand:
+        return IocCtrlEventSignal(input, output);
     }
     UNIMPLEMENTED_MSG("Unimplemented ioctl");
     return 0;
@@ -41,23 +51,137 @@ u32 nvhost_ctrl::NvOsGetConfigU32(const std::vector<u8>& input, std::vector<u8>&
 }
 
 u32 nvhost_ctrl::IocCtrlEventWait(const std::vector<u8>& input, std::vector<u8>& output,
-                                  bool is_async) {
+                                  bool is_async, IoctlCtrl& ctrl) {
     IocCtrlEventWaitParams params{};
     std::memcpy(&params, input.data(), sizeof(params));
-    LOG_WARNING(Service_NVDRV,
-                "(STUBBED) called, syncpt_id={}, threshold={}, timeout={}, is_async={}",
-                params.syncpt_id, params.threshold, params.timeout, is_async);
+    LOG_DEBUG(Service_NVDRV, "syncpt_id={}, threshold={}, timeout={}, is_async={}",
+              params.syncpt_id, params.threshold, params.timeout, is_async);
 
-    // TODO(Subv): Implement actual syncpt waiting.
-    params.value = 0;
+    if (params.syncpt_id >= MaxSyncPoints) {
+        return NvResult::BadParameter;
+    }
+
+    auto& gpu = system.GPU();
+    // This is mostly to take into account unimplemented features. As synced
+    // gpu is always synced.
+    if (!gpu.IsAsync()) {
+        return NvResult::Success;
+    }
+    auto lock = gpu.LockSync();
+    const u32 current_syncpoint_value = gpu.GetSyncpointValue(params.syncpt_id);
+    const s32 diff = current_syncpoint_value - params.threshold;
+    if (diff >= 0) {
+        params.value = current_syncpoint_value;
+        std::memcpy(output.data(), &params, sizeof(params));
+        return NvResult::Success;
+    }
+    const u32 target_value = current_syncpoint_value - diff;
+
+    if (!is_async) {
+        params.value = 0;
+    }
+
+    if (params.timeout == 0) {
+        std::memcpy(output.data(), &params, sizeof(params));
+        return NvResult::Timeout;
+    }
+
+    u32 event_id;
+    if (is_async) {
+        event_id = params.value & 0x00FF;
+        if (event_id >= MaxNvEvents) {
+            std::memcpy(output.data(), &params, sizeof(params));
+            return NvResult::BadParameter;
+        }
+    } else {
+        if (ctrl.fresh_call) {
+            const auto result = events_interface.GetFreeEvent();
+            if (result) {
+                event_id = *result;
+            } else {
+                LOG_CRITICAL(Service_NVDRV, "No Free Events available!");
+                event_id = params.value & 0x00FF;
+            }
+        } else {
+            event_id = ctrl.event_id;
+        }
+    }
+
+    EventState status = events_interface.status[event_id];
+    if (event_id < MaxNvEvents || status == EventState::Free || status == EventState::Registered) {
+        events_interface.SetEventStatus(event_id, EventState::Waiting);
+        events_interface.assigned_syncpt[event_id] = params.syncpt_id;
+        events_interface.assigned_value[event_id] = target_value;
+        if (is_async) {
+            params.value = params.syncpt_id << 4;
+        } else {
+            params.value = ((params.syncpt_id & 0xfff) << 16) | 0x10000000;
+        }
+        params.value |= event_id;
+        events_interface.events[event_id].writable->Clear();
+        gpu.RegisterSyncptInterrupt(params.syncpt_id, target_value);
+        if (!is_async && ctrl.fresh_call) {
+            ctrl.must_delay = true;
+            ctrl.timeout = params.timeout;
+            ctrl.event_id = event_id;
+            return NvResult::Timeout;
+        }
+        std::memcpy(output.data(), &params, sizeof(params));
+        return NvResult::Timeout;
+    }
     std::memcpy(output.data(), &params, sizeof(params));
-    return 0;
+    return NvResult::BadParameter;
 }
 
 u32 nvhost_ctrl::IocCtrlEventRegister(const std::vector<u8>& input, std::vector<u8>& output) {
-    LOG_WARNING(Service_NVDRV, "(STUBBED) called");
-    // TODO(bunnei): Implement this.
-    return 0;
+    IocCtrlEventRegisterParams params{};
+    std::memcpy(&params, input.data(), sizeof(params));
+    const u32 event_id = params.user_event_id & 0x00FF;
+    LOG_DEBUG(Service_NVDRV, " called, user_event_id: {:X}", event_id);
+    if (event_id >= MaxNvEvents) {
+        return NvResult::BadParameter;
+    }
+    if (events_interface.registered[event_id]) {
+        return NvResult::BadParameter;
+    }
+    events_interface.RegisterEvent(event_id);
+    return NvResult::Success;
+}
+
+u32 nvhost_ctrl::IocCtrlEventUnregister(const std::vector<u8>& input, std::vector<u8>& output) {
+    IocCtrlEventUnregisterParams params{};
+    std::memcpy(&params, input.data(), sizeof(params));
+    const u32 event_id = params.user_event_id & 0x00FF;
+    LOG_DEBUG(Service_NVDRV, " called, user_event_id: {:X}", event_id);
+    if (event_id >= MaxNvEvents) {
+        return NvResult::BadParameter;
+    }
+    if (!events_interface.registered[event_id]) {
+        return NvResult::BadParameter;
+    }
+    events_interface.UnregisterEvent(event_id);
+    return NvResult::Success;
+}
+
+u32 nvhost_ctrl::IocCtrlEventSignal(const std::vector<u8>& input, std::vector<u8>& output) {
+    IocCtrlEventSignalParams params{};
+    std::memcpy(&params, input.data(), sizeof(params));
+    // TODO(Blinkhawk): This is normally called when an NvEvents timeout on WaitSynchronization
+    // It is believed from RE to cancel the GPU Event. However, better research is required
+    u32 event_id = params.user_event_id & 0x00FF;
+    LOG_WARNING(Service_NVDRV, "(STUBBED) called, user_event_id: {:X}", event_id);
+    if (event_id >= MaxNvEvents) {
+        return NvResult::BadParameter;
+    }
+    if (events_interface.status[event_id] == EventState::Waiting) {
+        auto& gpu = system.GPU();
+        if (gpu.CancelSyncptInterrupt(events_interface.assigned_syncpt[event_id],
+                                      events_interface.assigned_value[event_id])) {
+            events_interface.LiberateEvent(event_id);
+            events_interface.events[event_id].writable->Signal();
+        }
+    }
+    return NvResult::Success;
 }
 
 } // namespace Service::Nvidia::Devices

src/core/hle/service/nvdrv/devices/nvhost_ctrl.h

@@ -8,15 +8,17 @@
 #include <vector>
 #include "common/common_types.h"
 #include "core/hle/service/nvdrv/devices/nvdevice.h"
+#include "core/hle/service/nvdrv/nvdrv.h"
 
 namespace Service::Nvidia::Devices {
 
 class nvhost_ctrl final : public nvdevice {
 public:
-    nvhost_ctrl();
+    explicit nvhost_ctrl(Core::System& system, EventInterface& events_interface);
     ~nvhost_ctrl() override;
 
-    u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) override;
+    u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+              IoctlCtrl& ctrl) override;
 
 private:
     enum class IoctlCommand : u32_le {
@@ -132,9 +134,16 @@ private:
 
     u32 NvOsGetConfigU32(const std::vector<u8>& input, std::vector<u8>& output);
 
-    u32 IocCtrlEventWait(const std::vector<u8>& input, std::vector<u8>& output, bool is_async);
+    u32 IocCtrlEventWait(const std::vector<u8>& input, std::vector<u8>& output, bool is_async,
+                         IoctlCtrl& ctrl);
 
     u32 IocCtrlEventRegister(const std::vector<u8>& input, std::vector<u8>& output);
+
+    u32 IocCtrlEventUnregister(const std::vector<u8>& input, std::vector<u8>& output);
+
+    u32 IocCtrlEventSignal(const std::vector<u8>& input, std::vector<u8>& output);
+
+    EventInterface& events_interface;
 };
 
 } // namespace Service::Nvidia::Devices

src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.cpp

@@ -12,10 +12,11 @@
 
 namespace Service::Nvidia::Devices {
 
-nvhost_ctrl_gpu::nvhost_ctrl_gpu() = default;
+nvhost_ctrl_gpu::nvhost_ctrl_gpu(Core::System& system) : nvdevice(system) {}
 nvhost_ctrl_gpu::~nvhost_ctrl_gpu() = default;
 
-u32 nvhost_ctrl_gpu::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) {
+u32 nvhost_ctrl_gpu::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+                           IoctlCtrl& ctrl) {
     LOG_DEBUG(Service_NVDRV, "called, command=0x{:08X}, input_size=0x{:X}, output_size=0x{:X}",
               command.raw, input.size(), output.size());
 
@@ -185,7 +186,7 @@ u32 nvhost_ctrl_gpu::GetGpuTime(const std::vector<u8>& input, std::vector<u8>& o
 
     IoctlGetGpuTime params{};
     std::memcpy(&params, input.data(), input.size());
-    const auto ns = Core::Timing::CyclesToNs(Core::System::GetInstance().CoreTiming().GetTicks());
+    const auto ns = Core::Timing::CyclesToNs(system.CoreTiming().GetTicks());
     params.gpu_time = static_cast<u64_le>(ns.count());
     std::memcpy(output.data(), &params, output.size());
     return 0;

src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.h

@@ -13,10 +13,11 @@ namespace Service::Nvidia::Devices {
 
 class nvhost_ctrl_gpu final : public nvdevice {
 public:
-    nvhost_ctrl_gpu();
+    explicit nvhost_ctrl_gpu(Core::System& system);
     ~nvhost_ctrl_gpu() override;
 
-    u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) override;
+    u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+              IoctlCtrl& ctrl) override;
 
 private:
     enum class IoctlCommand : u32_le {

src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp

@@ -13,10 +13,12 @@
 
 namespace Service::Nvidia::Devices {
 
-nvhost_gpu::nvhost_gpu(std::shared_ptr<nvmap> nvmap_dev) : nvmap_dev(std::move(nvmap_dev)) {}
+nvhost_gpu::nvhost_gpu(Core::System& system, std::shared_ptr<nvmap> nvmap_dev)
+    : nvdevice(system), nvmap_dev(std::move(nvmap_dev)) {}
 nvhost_gpu::~nvhost_gpu() = default;
 
-u32 nvhost_gpu::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) {
+u32 nvhost_gpu::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+                      IoctlCtrl& ctrl) {
     LOG_DEBUG(Service_NVDRV, "called, command=0x{:08X}, input_size=0x{:X}, output_size=0x{:X}",
               command.raw, input.size(), output.size());
 
@@ -119,8 +121,10 @@ u32 nvhost_gpu::AllocGPFIFOEx2(const std::vector<u8>& input, std::vector<u8>& ou
                 params.num_entries, params.flags, params.unk0, params.unk1, params.unk2,
                 params.unk3);
 
-    params.fence_out.id = 0;
-    params.fence_out.value = 0;
+    auto& gpu = system.GPU();
+    params.fence_out.id = assigned_syncpoints;
+    params.fence_out.value = gpu.GetSyncpointValue(assigned_syncpoints);
+    assigned_syncpoints++;
     std::memcpy(output.data(), &params, output.size());
     return 0;
 }
@@ -143,7 +147,7 @@ u32 nvhost_gpu::SubmitGPFIFO(const std::vector<u8>& input, std::vector<u8>& outp
     IoctlSubmitGpfifo params{};
     std::memcpy(&params, input.data(), sizeof(IoctlSubmitGpfifo));
     LOG_WARNING(Service_NVDRV, "(STUBBED) called, gpfifo={:X}, num_entries={:X}, flags={:X}",
-                params.address, params.num_entries, params.flags);
+                params.address, params.num_entries, params.flags.raw);
 
     ASSERT_MSG(input.size() == sizeof(IoctlSubmitGpfifo) +
                                    params.num_entries * sizeof(Tegra::CommandListHeader),
@@ -153,10 +157,18 @@ u32 nvhost_gpu::SubmitGPFIFO(const std::vector<u8>& input, std::vector<u8>& outp
     std::memcpy(entries.data(), &input[sizeof(IoctlSubmitGpfifo)],
                 params.num_entries * sizeof(Tegra::CommandListHeader));
 
-    Core::System::GetInstance().GPU().PushGPUEntries(std::move(entries));
+    UNIMPLEMENTED_IF(params.flags.add_wait.Value() != 0);
+    UNIMPLEMENTED_IF(params.flags.add_increment.Value() != 0);
+
+    auto& gpu = system.GPU();
+    u32 current_syncpoint_value = gpu.GetSyncpointValue(params.fence_out.id);
+    if (params.flags.increment.Value()) {
+        params.fence_out.value += current_syncpoint_value;
+    } else {
+        params.fence_out.value = current_syncpoint_value;
+    }
+    gpu.PushGPUEntries(std::move(entries));
 
-    params.fence_out.id = 0;
-    params.fence_out.value = 0;
     std::memcpy(output.data(), &params, sizeof(IoctlSubmitGpfifo));
     return 0;
 }
@@ -168,16 +180,24 @@ u32 nvhost_gpu::KickoffPB(const std::vector<u8>& input, std::vector<u8>& output)
     IoctlSubmitGpfifo params{};
     std::memcpy(&params, input.data(), sizeof(IoctlSubmitGpfifo));
     LOG_WARNING(Service_NVDRV, "(STUBBED) called, gpfifo={:X}, num_entries={:X}, flags={:X}",
-                params.address, params.num_entries, params.flags);
+                params.address, params.num_entries, params.flags.raw);
 
     Tegra::CommandList entries(params.num_entries);
     Memory::ReadBlock(params.address, entries.data(),
                       params.num_entries * sizeof(Tegra::CommandListHeader));
 
-    Core::System::GetInstance().GPU().PushGPUEntries(std::move(entries));
+    UNIMPLEMENTED_IF(params.flags.add_wait.Value() != 0);
+    UNIMPLEMENTED_IF(params.flags.add_increment.Value() != 0);
+
+    auto& gpu = system.GPU();
+    u32 current_syncpoint_value = gpu.GetSyncpointValue(params.fence_out.id);
+    if (params.flags.increment.Value()) {
+        params.fence_out.value += current_syncpoint_value;
+    } else {
+        params.fence_out.value = current_syncpoint_value;
+    }
+    gpu.PushGPUEntries(std::move(entries));
 
-    params.fence_out.id = 0;
-    params.fence_out.value = 0;
     std::memcpy(output.data(), &params, output.size());
     return 0;
 }

src/core/hle/service/nvdrv/devices/nvhost_gpu.h

@@ -10,6 +10,7 @@
 #include "common/common_types.h"
 #include "common/swap.h"
 #include "core/hle/service/nvdrv/devices/nvdevice.h"
+#include "core/hle/service/nvdrv/nvdata.h"
 
 namespace Service::Nvidia::Devices {
 
@@ -20,10 +21,11 @@ constexpr u32 NVGPU_IOCTL_CHANNEL_KICKOFF_PB(0x1b);
 
 class nvhost_gpu final : public nvdevice {
 public:
-    explicit nvhost_gpu(std::shared_ptr<nvmap> nvmap_dev);
+    explicit nvhost_gpu(Core::System& system, std::shared_ptr<nvmap> nvmap_dev);
     ~nvhost_gpu() override;
 
-    u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) override;
+    u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+              IoctlCtrl& ctrl) override;
 
 private:
     enum class IoctlCommand : u32_le {
@@ -113,11 +115,7 @@ private:
     static_assert(sizeof(IoctlGetErrorNotification) == 16,
                   "IoctlGetErrorNotification is incorrect size");
 
-    struct IoctlFence {
-        u32_le id;
-        u32_le value;
-    };
-    static_assert(sizeof(IoctlFence) == 8, "IoctlFence is incorrect size");
+    static_assert(sizeof(Fence) == 8, "Fence is incorrect size");
 
     struct IoctlAllocGpfifoEx {
         u32_le num_entries;
@@ -132,13 +130,13 @@ private:
     static_assert(sizeof(IoctlAllocGpfifoEx) == 32, "IoctlAllocGpfifoEx is incorrect size");
 
     struct IoctlAllocGpfifoEx2 {
-        u32_le num_entries;   // in
-        u32_le flags;         // in
-        u32_le unk0;          // in (1 works)
-        IoctlFence fence_out; // out
-        u32_le unk1;          // in
-        u32_le unk2;          // in
-        u32_le unk3;          // in
+        u32_le num_entries; // in
+        u32_le flags;       // in
+        u32_le unk0;        // in (1 works)
+        Fence fence_out;    // out
+        u32_le unk1;        // in
+        u32_le unk2;        // in
+        u32_le unk3;        // in
     };
     static_assert(sizeof(IoctlAllocGpfifoEx2) == 32, "IoctlAllocGpfifoEx2 is incorrect size");
 
@@ -153,10 +151,16 @@ private:
     struct IoctlSubmitGpfifo {
         u64_le address;     // pointer to gpfifo entry structs
         u32_le num_entries; // number of fence objects being submitted
-        u32_le flags;
-        IoctlFence fence_out; // returned new fence object for others to wait on
+        union {
+            u32_le raw;
+            BitField<0, 1, u32_le> add_wait;      // append a wait sync_point to the list
+            BitField<1, 1, u32_le> add_increment; // append an increment to the list
+            BitField<2, 1, u32_le> new_hw_format; // Mostly ignored
+            BitField<8, 1, u32_le> increment;     // increment the returned fence
+        } flags;
+        Fence fence_out; // returned new fence object for others to wait on
     };
-    static_assert(sizeof(IoctlSubmitGpfifo) == 16 + sizeof(IoctlFence),
+    static_assert(sizeof(IoctlSubmitGpfifo) == 16 + sizeof(Fence),
                   "IoctlSubmitGpfifo is incorrect size");
 
     struct IoctlGetWaitbase {
@@ -184,6 +188,7 @@ private:
     u32 ChannelSetTimeout(const std::vector<u8>& input, std::vector<u8>& output);
 
     std::shared_ptr<nvmap> nvmap_dev;
+    u32 assigned_syncpoints{};
 };
 
 } // namespace Service::Nvidia::Devices

src/core/hle/service/nvdrv/devices/nvhost_nvdec.cpp

@@ -10,10 +10,11 @@
 
 namespace Service::Nvidia::Devices {
 
-nvhost_nvdec::nvhost_nvdec() = default;
+nvhost_nvdec::nvhost_nvdec(Core::System& system) : nvdevice(system) {}
 nvhost_nvdec::~nvhost_nvdec() = default;
 
-u32 nvhost_nvdec::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) {
+u32 nvhost_nvdec::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+                        IoctlCtrl& ctrl) {
     LOG_DEBUG(Service_NVDRV, "called, command=0x{:08X}, input_size=0x{:X}, output_size=0x{:X}",
               command.raw, input.size(), output.size());
 

src/core/hle/service/nvdrv/devices/nvhost_nvdec.h

@@ -13,10 +13,11 @@ namespace Service::Nvidia::Devices {
 
 class nvhost_nvdec final : public nvdevice {
 public:
-    nvhost_nvdec();
+    explicit nvhost_nvdec(Core::System& system);
     ~nvhost_nvdec() override;
 
-    u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) override;
+    u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+              IoctlCtrl& ctrl) override;
 
 private:
     enum class IoctlCommand : u32_le {

src/core/hle/service/nvdrv/devices/nvhost_nvjpg.cpp

@@ -10,10 +10,11 @@
 
 namespace Service::Nvidia::Devices {
 
-nvhost_nvjpg::nvhost_nvjpg() = default;
+nvhost_nvjpg::nvhost_nvjpg(Core::System& system) : nvdevice(system) {}
 nvhost_nvjpg::~nvhost_nvjpg() = default;
 
-u32 nvhost_nvjpg::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) {
+u32 nvhost_nvjpg::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+                        IoctlCtrl& ctrl) {
     LOG_DEBUG(Service_NVDRV, "called, command=0x{:08X}, input_size=0x{:X}, output_size=0x{:X}",
               command.raw, input.size(), output.size());
 

src/core/hle/service/nvdrv/devices/nvhost_nvjpg.h

@@ -13,10 +13,11 @@ namespace Service::Nvidia::Devices {
 
 class nvhost_nvjpg final : public nvdevice {
 public:
-    nvhost_nvjpg();
+    explicit nvhost_nvjpg(Core::System& system);
     ~nvhost_nvjpg() override;
 
-    u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) override;
+    u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+              IoctlCtrl& ctrl) override;
 
 private:
     enum class IoctlCommand : u32_le {

src/core/hle/service/nvdrv/devices/nvhost_vic.cpp

@@ -10,10 +10,11 @@
 
 namespace Service::Nvidia::Devices {
 
-nvhost_vic::nvhost_vic() = default;
+nvhost_vic::nvhost_vic(Core::System& system) : nvdevice(system) {}
 nvhost_vic::~nvhost_vic() = default;
 
-u32 nvhost_vic::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) {
+u32 nvhost_vic::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+                      IoctlCtrl& ctrl) {
     LOG_DEBUG(Service_NVDRV, "called, command=0x{:08X}, input_size=0x{:X}, output_size=0x{:X}",
               command.raw, input.size(), output.size());
 

src/core/hle/service/nvdrv/devices/nvhost_vic.h

@@ -13,10 +13,11 @@ namespace Service::Nvidia::Devices {
 
 class nvhost_vic final : public nvdevice {
 public:
-    nvhost_vic();
+    explicit nvhost_vic(Core::System& system);
     ~nvhost_vic() override;
 
-    u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) override;
+    u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+              IoctlCtrl& ctrl) override;
 
 private:
     enum class IoctlCommand : u32_le {

src/core/hle/service/nvdrv/devices/nvmap.cpp

@@ -18,7 +18,7 @@ enum {
 };
 }
 
-nvmap::nvmap() = default;
+nvmap::nvmap(Core::System& system) : nvdevice(system) {}
 nvmap::~nvmap() = default;
 
 VAddr nvmap::GetObjectAddress(u32 handle) const {
@@ -28,7 +28,8 @@ VAddr nvmap::GetObjectAddress(u32 handle) const {
     return object->addr;
 }
 
-u32 nvmap::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) {
+u32 nvmap::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+                 IoctlCtrl& ctrl) {
     switch (static_cast<IoctlCommand>(command.raw)) {
     case IoctlCommand::Create:
         return IocCreate(input, output);

src/core/hle/service/nvdrv/devices/nvmap.h

@@ -16,13 +16,14 @@ namespace Service::Nvidia::Devices {
 
 class nvmap final : public nvdevice {
 public:
-    nvmap();
+    explicit nvmap(Core::System& system);
     ~nvmap() override;
 
     /// Returns the allocated address of an nvmap object given its handle.
     VAddr GetObjectAddress(u32 handle) const;
 
-    u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) override;
+    u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+              IoctlCtrl& ctrl) override;
 
     /// Represents an nvmap object.
     struct Object {

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

@@ -8,12 +8,18 @@
 #include "core/hle/ipc_helpers.h"
 #include "core/hle/kernel/kernel.h"
 #include "core/hle/kernel/readable_event.h"
+#include "core/hle/kernel/thread.h"
 #include "core/hle/kernel/writable_event.h"
 #include "core/hle/service/nvdrv/interface.h"
+#include "core/hle/service/nvdrv/nvdata.h"
 #include "core/hle/service/nvdrv/nvdrv.h"
 
 namespace Service::Nvidia {
 
+void NVDRV::SignalGPUInterruptSyncpt(const u32 syncpoint_id, const u32 value) {
+    nvdrv->SignalSyncpt(syncpoint_id, value);
+}
+
 void NVDRV::Open(Kernel::HLERequestContext& ctx) {
     LOG_DEBUG(Service_NVDRV, "called");
 
@@ -36,11 +42,31 @@ void NVDRV::Ioctl(Kernel::HLERequestContext& ctx) {
 
     std::vector<u8> output(ctx.GetWriteBufferSize());
 
+    IoctlCtrl ctrl{};
+
+    u32 result = nvdrv->Ioctl(fd, command, ctx.ReadBuffer(), output, ctrl);
+
+    if (ctrl.must_delay) {
+        ctrl.fresh_call = false;
+        ctx.SleepClientThread(
+            "NVServices::DelayedResponse", ctrl.timeout,
+            [=](Kernel::SharedPtr<Kernel::Thread> thread, Kernel::HLERequestContext& ctx,
+                Kernel::ThreadWakeupReason reason) {
+                IoctlCtrl ctrl2{ctrl};
+                std::vector<u8> output2 = output;
+                u32 result = nvdrv->Ioctl(fd, command, ctx.ReadBuffer(), output2, ctrl2);
+                ctx.WriteBuffer(output2);
+                IPC::ResponseBuilder rb{ctx, 3};
+                rb.Push(RESULT_SUCCESS);
+                rb.Push(result);
+            },
+            nvdrv->GetEventWriteable(ctrl.event_id));
+    } else {
+        ctx.WriteBuffer(output);
+    }
     IPC::ResponseBuilder rb{ctx, 3};
     rb.Push(RESULT_SUCCESS);
-    rb.Push(nvdrv->Ioctl(fd, command, ctx.ReadBuffer(), output));
-
-    ctx.WriteBuffer(output);
+    rb.Push(result);
 }
 
 void NVDRV::Close(Kernel::HLERequestContext& ctx) {
@@ -66,13 +92,19 @@ void NVDRV::Initialize(Kernel::HLERequestContext& ctx) {
 void NVDRV::QueryEvent(Kernel::HLERequestContext& ctx) {
     IPC::RequestParser rp{ctx};
     u32 fd = rp.Pop<u32>();
-    u32 event_id = rp.Pop<u32>();
+    // TODO(Blinkhawk): Figure the meaning of the flag at bit 16
+    u32 event_id = rp.Pop<u32>() & 0x000000FF;
     LOG_WARNING(Service_NVDRV, "(STUBBED) called, fd={:X}, event_id={:X}", fd, event_id);
 
     IPC::ResponseBuilder rb{ctx, 3, 1};
     rb.Push(RESULT_SUCCESS);
-    rb.PushCopyObjects(query_event.readable);
-    rb.Push<u32>(0);
+    if (event_id < MaxNvEvents) {
+        rb.PushCopyObjects(nvdrv->GetEvent(event_id));
+        rb.Push<u32>(NvResult::Success);
+    } else {
+        rb.Push<u32>(0);
+        rb.Push<u32>(NvResult::BadParameter);
+    }
 }
 
 void NVDRV::SetClientPID(Kernel::HLERequestContext& ctx) {
@@ -127,10 +159,6 @@ NVDRV::NVDRV(std::shared_ptr<Module> nvdrv, const char* name)
         {13, &NVDRV::FinishInitialize, "FinishInitialize"},
     };
     RegisterHandlers(functions);
-
-    auto& kernel = Core::System::GetInstance().Kernel();
-    query_event = Kernel::WritableEvent::CreateEventPair(kernel, Kernel::ResetType::Automatic,
-                                                         "NVDRV::query_event");
 }
 
 NVDRV::~NVDRV() = default;

src/core/hle/service/nvdrv/interface.h

@@ -19,6 +19,8 @@ public:
     NVDRV(std::shared_ptr<Module> nvdrv, const char* name);
     ~NVDRV() override;
 
+    void SignalGPUInterruptSyncpt(const u32 syncpoint_id, const u32 value);
+
 private:
     void Open(Kernel::HLERequestContext& ctx);
     void Ioctl(Kernel::HLERequestContext& ctx);
@@ -33,8 +35,6 @@ private:
     std::shared_ptr<Module> nvdrv;
 
     u64 pid{};
-
-    Kernel::EventPair query_event;
 };
 
 } // namespace Service::Nvidia

src/core/hle/service/nvdrv/nvdata.h

@@ -0,0 +1,48 @@
+#pragma once
+
+#include <array>
+#include "common/common_types.h"
+
+namespace Service::Nvidia {
+
+constexpr u32 MaxSyncPoints = 192;
+constexpr u32 MaxNvEvents = 64;
+
+struct Fence {
+    s32 id;
+    u32 value;
+};
+
+static_assert(sizeof(Fence) == 8, "Fence has wrong size");
+
+struct MultiFence {
+    u32 num_fences;
+    std::array<Fence, 4> fences;
+};
+
+enum NvResult : u32 {
+    Success = 0,
+    BadParameter = 4,
+    Timeout = 5,
+    ResourceError = 15,
+};
+
+enum class EventState {
+    Free = 0,
+    Registered = 1,
+    Waiting = 2,
+    Busy = 3,
+};
+
+struct IoctlCtrl {
+    // First call done to the servioce for services that call itself again after a call.
+    bool fresh_call{true};
+    // Tells the Ioctl Wrapper that it must delay the IPC response and send the thread to sleep
+    bool must_delay{};
+    // Timeout for the delay
+    s64 timeout{};
+    // NV Event Id
+    s32 event_id{-1};
+};
+
+} // namespace Service::Nvidia

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

@@ -4,7 +4,10 @@
 
 #include <utility>
 
+#include <fmt/format.h>
 #include "core/hle/ipc_helpers.h"
+#include "core/hle/kernel/readable_event.h"
+#include "core/hle/kernel/writable_event.h"
 #include "core/hle/service/nvdrv/devices/nvdevice.h"
 #include "core/hle/service/nvdrv/devices/nvdisp_disp0.h"
 #include "core/hle/service/nvdrv/devices/nvhost_as_gpu.h"
@@ -22,8 +25,9 @@
 
 namespace Service::Nvidia {
 
-void InstallInterfaces(SM::ServiceManager& service_manager, NVFlinger::NVFlinger& nvflinger) {
-    auto module_ = std::make_shared<Module>();
+void InstallInterfaces(SM::ServiceManager& service_manager, NVFlinger::NVFlinger& nvflinger,
+                       Core::System& system) {
+    auto module_ = std::make_shared<Module>(system);
     std::make_shared<NVDRV>(module_, "nvdrv")->InstallAsService(service_manager);
     std::make_shared<NVDRV>(module_, "nvdrv:a")->InstallAsService(service_manager);
     std::make_shared<NVDRV>(module_, "nvdrv:s")->InstallAsService(service_manager);
@@ -32,17 +36,25 @@ void InstallInterfaces(SM::ServiceManager& service_manager, NVFlinger::NVFlinger
     nvflinger.SetNVDrvInstance(module_);
 }
 
-Module::Module() {
-    auto nvmap_dev = std::make_shared<Devices::nvmap>();
-    devices["/dev/nvhost-as-gpu"] = std::make_shared<Devices::nvhost_as_gpu>(nvmap_dev);
-    devices["/dev/nvhost-gpu"] = std::make_shared<Devices::nvhost_gpu>(nvmap_dev);
-    devices["/dev/nvhost-ctrl-gpu"] = std::make_shared<Devices::nvhost_ctrl_gpu>();
+Module::Module(Core::System& system) {
+    auto& kernel = system.Kernel();
+    for (u32 i = 0; i < MaxNvEvents; i++) {
+        std::string event_label = fmt::format("NVDRV::NvEvent_{}", i);
+        events_interface.events[i] = Kernel::WritableEvent::CreateEventPair(
+            kernel, Kernel::ResetType::Automatic, event_label);
+        events_interface.status[i] = EventState::Free;
+        events_interface.registered[i] = false;
+    }
+    auto nvmap_dev = std::make_shared<Devices::nvmap>(system);
+    devices["/dev/nvhost-as-gpu"] = std::make_shared<Devices::nvhost_as_gpu>(system, nvmap_dev);
+    devices["/dev/nvhost-gpu"] = std::make_shared<Devices::nvhost_gpu>(system, nvmap_dev);
+    devices["/dev/nvhost-ctrl-gpu"] = std::make_shared<Devices::nvhost_ctrl_gpu>(system);
     devices["/dev/nvmap"] = nvmap_dev;
-    devices["/dev/nvdisp_disp0"] = std::make_shared<Devices::nvdisp_disp0>(nvmap_dev);
-    devices["/dev/nvhost-ctrl"] = std::make_shared<Devices::nvhost_ctrl>();
-    devices["/dev/nvhost-nvdec"] = std::make_shared<Devices::nvhost_nvdec>();
-    devices["/dev/nvhost-nvjpg"] = std::make_shared<Devices::nvhost_nvjpg>();
-    devices["/dev/nvhost-vic"] = std::make_shared<Devices::nvhost_vic>();
+    devices["/dev/nvdisp_disp0"] = std::make_shared<Devices::nvdisp_disp0>(system, nvmap_dev);
+    devices["/dev/nvhost-ctrl"] = std::make_shared<Devices::nvhost_ctrl>(system, events_interface);
+    devices["/dev/nvhost-nvdec"] = std::make_shared<Devices::nvhost_nvdec>(system);
+    devices["/dev/nvhost-nvjpg"] = std::make_shared<Devices::nvhost_nvjpg>(system);
+    devices["/dev/nvhost-vic"] = std::make_shared<Devices::nvhost_vic>(system);
 }
 
 Module::~Module() = default;
@@ -59,12 +71,13 @@ u32 Module::Open(const std::string& device_name) {
     return fd;
 }
 
-u32 Module::Ioctl(u32 fd, u32 command, const std::vector<u8>& input, std::vector<u8>& output) {
+u32 Module::Ioctl(u32 fd, u32 command, const std::vector<u8>& input, std::vector<u8>& output,
+                  IoctlCtrl& ctrl) {
     auto itr = open_files.find(fd);
     ASSERT_MSG(itr != open_files.end(), "Tried to talk to an invalid device");
 
     auto& device = itr->second;
-    return device->ioctl({command}, input, output);
+    return device->ioctl({command}, input, output, ctrl);
 }
 
 ResultCode Module::Close(u32 fd) {
@@ -77,4 +90,22 @@ ResultCode Module::Close(u32 fd) {
     return RESULT_SUCCESS;
 }
 
+void Module::SignalSyncpt(const u32 syncpoint_id, const u32 value) {
+    for (u32 i = 0; i < MaxNvEvents; i++) {
+        if (events_interface.assigned_syncpt[i] == syncpoint_id &&
+            events_interface.assigned_value[i] == value) {
+            events_interface.LiberateEvent(i);
+            events_interface.events[i].writable->Signal();
+        }
+    }
+}
+
+Kernel::SharedPtr<Kernel::ReadableEvent> Module::GetEvent(const u32 event_id) const {
+    return events_interface.events[event_id].readable;
+}
+
+Kernel::SharedPtr<Kernel::WritableEvent> Module::GetEventWriteable(const u32 event_id) const {
+    return events_interface.events[event_id].writable;
+}
+
 } // namespace Service::Nvidia

src/core/hle/service/nvdrv/nvdrv.h

@@ -8,8 +8,14 @@
 #include <unordered_map>
 #include <vector>
 #include "common/common_types.h"
+#include "core/hle/kernel/writable_event.h"
+#include "core/hle/service/nvdrv/nvdata.h"
 #include "core/hle/service/service.h"
 
+namespace Core {
+class System;
+}
+
 namespace Service::NVFlinger {
 class NVFlinger;
 }
@@ -20,16 +26,72 @@ namespace Devices {
 class nvdevice;
 }
 
-struct IoctlFence {
-    u32 id;
-    u32 value;
+struct EventInterface {
+    // Mask representing currently busy events
+    u64 events_mask{};
+    // Each kernel event associated to an NV event
+    std::array<Kernel::EventPair, MaxNvEvents> events;
+    // The status of the current NVEvent
+    std::array<EventState, MaxNvEvents> status{};
+    // Tells if an NVEvent is registered or not
+    std::array<bool, MaxNvEvents> registered{};
+    // When an NVEvent is waiting on GPU interrupt, this is the sync_point
+    // associated with it.
+    std::array<u32, MaxNvEvents> assigned_syncpt{};
+    // This is the value of the GPU interrupt for which the NVEvent is waiting
+    // for.
+    std::array<u32, MaxNvEvents> assigned_value{};
+    // Constant to denote an unasigned syncpoint.
+    static constexpr u32 unassigned_syncpt = 0xFFFFFFFF;
+    std::optional<u32> GetFreeEvent() const {
+        u64 mask = events_mask;
+        for (u32 i = 0; i < MaxNvEvents; i++) {
+            const bool is_free = (mask & 0x1) == 0;
+            if (is_free) {
+                if (status[i] == EventState::Registered || status[i] == EventState::Free) {
+                    return {i};
+                }
+            }
+            mask = mask >> 1;
+        }
+        return {};
+    }
+    void SetEventStatus(const u32 event_id, EventState new_status) {
+        EventState old_status = status[event_id];
+        if (old_status == new_status) {
+            return;
+        }
+        status[event_id] = new_status;
+        if (new_status == EventState::Registered) {
+            registered[event_id] = true;
+        }
+        if (new_status == EventState::Waiting || new_status == EventState::Busy) {
+            events_mask |= (1ULL << event_id);
+        }
+    }
+    void RegisterEvent(const u32 event_id) {
+        registered[event_id] = true;
+        if (status[event_id] == EventState::Free) {
+            status[event_id] = EventState::Registered;
+        }
+    }
+    void UnregisterEvent(const u32 event_id) {
+        registered[event_id] = false;
+        if (status[event_id] == EventState::Registered) {
+            status[event_id] = EventState::Free;
+        }
+    }
+    void LiberateEvent(const u32 event_id) {
+        status[event_id] = registered[event_id] ? EventState::Registered : EventState::Free;
+        events_mask &= ~(1ULL << event_id);
+        assigned_syncpt[event_id] = unassigned_syncpt;
+        assigned_value[event_id] = 0;
+    }
 };
 
-static_assert(sizeof(IoctlFence) == 8, "IoctlFence has wrong size");
-
 class Module final {
 public:
-    Module();
+    Module(Core::System& system);
     ~Module();
 
     /// Returns a pointer to one of the available devices, identified by its name.
@@ -44,10 +106,17 @@ public:
     /// Opens a device node and returns a file descriptor to it.
     u32 Open(const std::string& device_name);
     /// Sends an ioctl command to the specified file descriptor.
-    u32 Ioctl(u32 fd, u32 command, const std::vector<u8>& input, std::vector<u8>& output);
+    u32 Ioctl(u32 fd, u32 command, const std::vector<u8>& input, std::vector<u8>& output,
+              IoctlCtrl& ctrl);
     /// Closes a device file descriptor and returns operation success.
     ResultCode Close(u32 fd);
 
+    void SignalSyncpt(const u32 syncpoint_id, const u32 value);
+
+    Kernel::SharedPtr<Kernel::ReadableEvent> GetEvent(u32 event_id) const;
+
+    Kernel::SharedPtr<Kernel::WritableEvent> GetEventWriteable(u32 event_id) const;
+
 private:
     /// Id to use for the next open file descriptor.
     u32 next_fd = 1;
@@ -57,9 +126,12 @@ private:
 
     /// Mapping of device node names to their implementation.
     std::unordered_map<std::string, std::shared_ptr<Devices::nvdevice>> devices;
+
+    EventInterface events_interface;
 };
 
 /// Registers all NVDRV services with the specified service manager.
-void InstallInterfaces(SM::ServiceManager& service_manager, NVFlinger::NVFlinger& nvflinger);
+void InstallInterfaces(SM::ServiceManager& service_manager, NVFlinger::NVFlinger& nvflinger,
+                       Core::System& system);
 
 } // namespace Service::Nvidia

src/core/hle/service/nvflinger/buffer_queue.cpp

@@ -34,7 +34,8 @@ void BufferQueue::SetPreallocatedBuffer(u32 slot, const IGBPBuffer& igbp_buffer)
     buffer_wait_event.writable->Signal();
 }
 
-std::optional<u32> BufferQueue::DequeueBuffer(u32 width, u32 height) {
+std::optional<std::pair<u32, Service::Nvidia::MultiFence*>> BufferQueue::DequeueBuffer(u32 width,
+                                                                                       u32 height) {
     auto itr = std::find_if(queue.begin(), queue.end(), [&](const Buffer& buffer) {
         // Only consider free buffers. Buffers become free once again after they've been Acquired
         // and Released by the compositor, see the NVFlinger::Compose method.
@@ -51,7 +52,7 @@ std::optional<u32> BufferQueue::DequeueBuffer(u32 width, u32 height) {
     }
 
     itr->status = Buffer::Status::Dequeued;
-    return itr->slot;
+    return {{itr->slot, &itr->multi_fence}};
 }
 
 const IGBPBuffer& BufferQueue::RequestBuffer(u32 slot) const {
@@ -63,7 +64,8 @@ const IGBPBuffer& BufferQueue::RequestBuffer(u32 slot) const {
 }
 
 void BufferQueue::QueueBuffer(u32 slot, BufferTransformFlags transform,
-                              const Common::Rectangle<int>& crop_rect) {
+                              const Common::Rectangle<int>& crop_rect, u32 swap_interval,
+                              Service::Nvidia::MultiFence& multi_fence) {
     auto itr = std::find_if(queue.begin(), queue.end(),
                             [&](const Buffer& buffer) { return buffer.slot == slot; });
     ASSERT(itr != queue.end());
@@ -71,12 +73,21 @@ void BufferQueue::QueueBuffer(u32 slot, BufferTransformFlags transform,
     itr->status = Buffer::Status::Queued;
     itr->transform = transform;
     itr->crop_rect = crop_rect;
+    itr->swap_interval = swap_interval;
+    itr->multi_fence = multi_fence;
+    queue_sequence.push_back(slot);
 }
 
 std::optional<std::reference_wrapper<const BufferQueue::Buffer>> BufferQueue::AcquireBuffer() {
-    auto itr = std::find_if(queue.begin(), queue.end(), [](const Buffer& buffer) {
-        return buffer.status == Buffer::Status::Queued;
-    });
+    auto itr = queue.end();
+    // Iterate to find a queued buffer matching the requested slot.
+    while (itr == queue.end() && !queue_sequence.empty()) {
+        u32 slot = queue_sequence.front();
+        itr = std::find_if(queue.begin(), queue.end(), [&slot](const Buffer& buffer) {
+            return buffer.status == Buffer::Status::Queued && buffer.slot == slot;
+        });
+        queue_sequence.pop_front();
+    }
     if (itr == queue.end())
         return {};
     itr->status = Buffer::Status::Acquired;

src/core/hle/service/nvflinger/buffer_queue.h

@@ -4,6 +4,7 @@
 
 #pragma once
 
+#include <list>
 #include <optional>
 #include <vector>
 
@@ -12,6 +13,7 @@
 #include "common/swap.h"
 #include "core/hle/kernel/object.h"
 #include "core/hle/kernel/writable_event.h"
+#include "core/hle/service/nvdrv/nvdata.h"
 
 namespace Service::NVFlinger {
 
@@ -68,13 +70,17 @@ public:
         IGBPBuffer igbp_buffer;
         BufferTransformFlags transform;
         Common::Rectangle<int> crop_rect;
+        u32 swap_interval;
+        Service::Nvidia::MultiFence multi_fence;
     };
 
     void SetPreallocatedBuffer(u32 slot, const IGBPBuffer& igbp_buffer);
-    std::optional<u32> DequeueBuffer(u32 width, u32 height);
+    std::optional<std::pair<u32, Service::Nvidia::MultiFence*>> DequeueBuffer(u32 width,
+                                                                              u32 height);
     const IGBPBuffer& RequestBuffer(u32 slot) const;
     void QueueBuffer(u32 slot, BufferTransformFlags transform,
-                     const Common::Rectangle<int>& crop_rect);
+                     const Common::Rectangle<int>& crop_rect, u32 swap_interval,
+                     Service::Nvidia::MultiFence& multi_fence);
     std::optional<std::reference_wrapper<const Buffer>> AcquireBuffer();
     void ReleaseBuffer(u32 slot);
     u32 Query(QueryType type);
@@ -92,6 +98,7 @@ private:
     u64 layer_id;
 
     std::vector<Buffer> queue;
+    std::list<u32> queue_sequence;
     Kernel::EventPair buffer_wait_event;
 };
 

src/core/hle/service/nvflinger/nvflinger.cpp

@@ -37,15 +37,14 @@ NVFlinger::NVFlinger(Core::Timing::CoreTiming& core_timing) : core_timing{core_t
     displays.emplace_back(4, "Null");
 
     // Schedule the screen composition events
-    const auto ticks = Settings::values.force_30fps_mode ? frame_ticks_30fps : frame_ticks;
+    composition_event = core_timing.RegisterEvent("ScreenComposition", [this](u64 userdata,
+                                                                              s64 cycles_late) {
+        Compose();
+        const auto ticks = Settings::values.force_30fps_mode ? frame_ticks_30fps : GetNextTicks();
+        this->core_timing.ScheduleEvent(std::max<s64>(0LL, ticks - cycles_late), composition_event);
+    });
 
-    composition_event = core_timing.RegisterEvent(
-        "ScreenComposition", [this, ticks](u64 userdata, s64 cycles_late) {
-            Compose();
-            this->core_timing.ScheduleEvent(ticks - cycles_late, composition_event);
-        });
-
-    core_timing.ScheduleEvent(ticks, composition_event);
+    core_timing.ScheduleEvent(frame_ticks, composition_event);
 }
 
 NVFlinger::~NVFlinger() {
@@ -206,8 +205,14 @@ void NVFlinger::Compose() {
                      igbp_buffer.width, igbp_buffer.height, igbp_buffer.stride,
                      buffer->get().transform, buffer->get().crop_rect);
 
+        swap_interval = buffer->get().swap_interval;
         buffer_queue.ReleaseBuffer(buffer->get().slot);
     }
 }
 
+s64 NVFlinger::GetNextTicks() const {
+    constexpr s64 max_hertz = 120LL;
+    return (Core::Timing::BASE_CLOCK_RATE * (1LL << swap_interval)) / max_hertz;
+}
+
 } // namespace Service::NVFlinger

src/core/hle/service/nvflinger/nvflinger.h

@@ -74,6 +74,8 @@ public:
     /// finished.
     void Compose();
 
+    s64 GetNextTicks() const;
+
 private:
     /// Finds the display identified by the specified ID.
     VI::Display* FindDisplay(u64 display_id);
@@ -98,6 +100,8 @@ private:
     /// layers.
     u32 next_buffer_queue_id = 1;
 
+    u32 swap_interval = 1;
+
     /// Event that handles screen composition.
     Core::Timing::EventType* composition_event;
 

src/core/hle/service/service.cpp

@@ -236,7 +236,7 @@ void Init(std::shared_ptr<SM::ServiceManager>& sm, Core::System& system) {
     NIM::InstallInterfaces(*sm);
     NPNS::InstallInterfaces(*sm);
     NS::InstallInterfaces(*sm);
-    Nvidia::InstallInterfaces(*sm, *nv_flinger);
+    Nvidia::InstallInterfaces(*sm, *nv_flinger, system);
     PCIe::InstallInterfaces(*sm);
     PCTL::InstallInterfaces(*sm);
     PCV::InstallInterfaces(*sm);

src/core/hle/service/vi/vi.cpp

@@ -21,6 +21,7 @@
 #include "core/hle/kernel/readable_event.h"
 #include "core/hle/kernel/thread.h"
 #include "core/hle/kernel/writable_event.h"
+#include "core/hle/service/nvdrv/nvdata.h"
 #include "core/hle/service/nvdrv/nvdrv.h"
 #include "core/hle/service/nvflinger/buffer_queue.h"
 #include "core/hle/service/nvflinger/nvflinger.h"
@@ -328,32 +329,22 @@ public:
     Data data;
 };
 
-struct BufferProducerFence {
-    u32 is_valid;
-    std::array<Nvidia::IoctlFence, 4> fences;
-};
-static_assert(sizeof(BufferProducerFence) == 36, "BufferProducerFence has wrong size");
-
 class IGBPDequeueBufferResponseParcel : public Parcel {
 public:
-    explicit IGBPDequeueBufferResponseParcel(u32 slot) : slot(slot) {}
+    explicit IGBPDequeueBufferResponseParcel(u32 slot, Service::Nvidia::MultiFence& multi_fence)
+        : slot(slot), multi_fence(multi_fence) {}
     ~IGBPDequeueBufferResponseParcel() override = default;
 
 protected:
     void SerializeData() override {
-        // TODO(Subv): Find out how this Fence is used.
-        BufferProducerFence fence = {};
-        fence.is_valid = 1;
-        for (auto& fence_ : fence.fences)
-            fence_.id = -1;
-
         Write(slot);
         Write<u32_le>(1);
-        WriteObject(fence);
+        WriteObject(multi_fence);
         Write<u32_le>(0);
     }
 
     u32_le slot;
+    Service::Nvidia::MultiFence multi_fence;
 };
 
 class IGBPRequestBufferRequestParcel : public Parcel {
@@ -400,12 +391,6 @@ public:
         data = Read<Data>();
     }
 
-    struct Fence {
-        u32_le id;
-        u32_le value;
-    };
-    static_assert(sizeof(Fence) == 8, "Fence has wrong size");
-
     struct Data {
         u32_le slot;
         INSERT_PADDING_WORDS(3);
@@ -418,15 +403,15 @@ public:
         s32_le scaling_mode;
         NVFlinger::BufferQueue::BufferTransformFlags transform;
         u32_le sticky_transform;
-        INSERT_PADDING_WORDS(2);
-        u32_le fence_is_valid;
-        std::array<Fence, 2> fences;
+        INSERT_PADDING_WORDS(1);
+        u32_le swap_interval;
+        Service::Nvidia::MultiFence multi_fence;
 
         Common::Rectangle<int> GetCropRect() const {
             return {crop_left, crop_top, crop_right, crop_bottom};
         }
     };
-    static_assert(sizeof(Data) == 80, "ParcelData has wrong size");
+    static_assert(sizeof(Data) == 96, "ParcelData has wrong size");
 
     Data data;
 };
@@ -547,11 +532,11 @@ private:
             IGBPDequeueBufferRequestParcel request{ctx.ReadBuffer()};
             const u32 width{request.data.width};
             const u32 height{request.data.height};
-            std::optional<u32> slot = buffer_queue.DequeueBuffer(width, height);
+            auto result = buffer_queue.DequeueBuffer(width, height);
 
-            if (slot) {
+            if (result) {
                 // Buffer is available
-                IGBPDequeueBufferResponseParcel response{*slot};
+                IGBPDequeueBufferResponseParcel response{result->first, *result->second};
                 ctx.WriteBuffer(response.Serialize());
             } else {
                 // Wait the current thread until a buffer becomes available
@@ -561,10 +546,10 @@ private:
                         Kernel::ThreadWakeupReason reason) {
                         // Repeat TransactParcel DequeueBuffer when a buffer is available
                         auto& buffer_queue = nv_flinger->FindBufferQueue(id);
-                        std::optional<u32> slot = buffer_queue.DequeueBuffer(width, height);
-                        ASSERT_MSG(slot != std::nullopt, "Could not dequeue buffer.");
+                        auto result = buffer_queue.DequeueBuffer(width, height);
+                        ASSERT_MSG(result != std::nullopt, "Could not dequeue buffer.");
 
-                        IGBPDequeueBufferResponseParcel response{*slot};
+                        IGBPDequeueBufferResponseParcel response{result->first, *result->second};
                         ctx.WriteBuffer(response.Serialize());
                         IPC::ResponseBuilder rb{ctx, 2};
                         rb.Push(RESULT_SUCCESS);
@@ -582,7 +567,8 @@ private:
             IGBPQueueBufferRequestParcel request{ctx.ReadBuffer()};
 
             buffer_queue.QueueBuffer(request.data.slot, request.data.transform,
-                                     request.data.GetCropRect());
+                                     request.data.GetCropRect(), request.data.swap_interval,
+                                     request.data.multi_fence);
 
             IGBPQueueBufferResponseParcel response{1280, 720};
             ctx.WriteBuffer(response.Serialize());

src/video_core/CMakeLists.txt

@@ -1,4 +1,5 @@
 add_library(video_core STATIC
+    buffer_cache.h
     dma_pusher.cpp
     dma_pusher.h
     debug_utils/debug_utils.cpp
@@ -43,8 +44,6 @@ add_library(video_core STATIC
     renderer_opengl/gl_device.h
     renderer_opengl/gl_framebuffer_cache.cpp
     renderer_opengl/gl_framebuffer_cache.h
-    renderer_opengl/gl_global_cache.cpp
-    renderer_opengl/gl_global_cache.h
     renderer_opengl/gl_rasterizer.cpp
     renderer_opengl/gl_rasterizer.h
     renderer_opengl/gl_resource_manager.cpp

src/video_core/buffer_cache.h

@@ -0,0 +1,299 @@
+// 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

src/video_core/dma_pusher.cpp

@@ -22,7 +22,7 @@ void DmaPusher::DispatchCalls() {
     MICROPROFILE_SCOPE(DispatchCalls);
 
     // On entering GPU code, assume all memory may be touched by the ARM core.
-    gpu.Maxwell3D().dirty_flags.OnMemoryWrite();
+    gpu.Maxwell3D().dirty.OnMemoryWrite();
 
     dma_pushbuffer_subindex = 0;
 

src/video_core/engines/kepler_compute.cpp

@@ -37,7 +37,7 @@ void KeplerCompute::CallMethod(const GPU::MethodCall& method_call) {
         const bool is_last_call = method_call.IsLastCall();
         upload_state.ProcessData(method_call.argument, is_last_call);
         if (is_last_call) {
-            system.GPU().Maxwell3D().dirty_flags.OnMemoryWrite();
+            system.GPU().Maxwell3D().dirty.OnMemoryWrite();
         }
         break;
     }
@@ -50,13 +50,14 @@ void KeplerCompute::CallMethod(const GPU::MethodCall& method_call) {
 }
 
 void KeplerCompute::ProcessLaunch() {
-
     const GPUVAddr launch_desc_loc = regs.launch_desc_loc.Address();
     memory_manager.ReadBlockUnsafe(launch_desc_loc, &launch_description,
                                    LaunchParams::NUM_LAUNCH_PARAMETERS * sizeof(u32));
 
-    const GPUVAddr code_loc = regs.code_loc.Address() + launch_description.program_start;
-    LOG_WARNING(HW_GPU, "Compute Kernel Execute at Address 0x{:016x}, STUBBED", code_loc);
+    const GPUVAddr code_addr = regs.code_loc.Address() + launch_description.program_start;
+    LOG_TRACE(HW_GPU, "Compute invocation launched at address 0x{:016x}", code_addr);
+
+    rasterizer.DispatchCompute(code_addr);
 }
 
 } // namespace Tegra::Engines

src/video_core/engines/kepler_memory.cpp

@@ -34,7 +34,7 @@ void KeplerMemory::CallMethod(const GPU::MethodCall& method_call) {
         const bool is_last_call = method_call.IsLastCall();
         upload_state.ProcessData(method_call.argument, is_last_call);
         if (is_last_call) {
-            system.GPU().Maxwell3D().dirty_flags.OnMemoryWrite();
+            system.GPU().Maxwell3D().dirty.OnMemoryWrite();
         }
         break;
     }

src/video_core/engines/maxwell_3d.cpp

@@ -22,6 +22,7 @@ Maxwell3D::Maxwell3D(Core::System& system, VideoCore::RasterizerInterface& raste
                      MemoryManager& memory_manager)
     : system{system}, rasterizer{rasterizer}, memory_manager{memory_manager},
       macro_interpreter{*this}, upload_state{memory_manager, regs.upload} {
+    InitDirtySettings();
     InitializeRegisterDefaults();
 }
 
@@ -69,6 +70,10 @@ void Maxwell3D::InitializeRegisterDefaults() {
     regs.stencil_back_func_mask = 0xFFFFFFFF;
     regs.stencil_back_mask = 0xFFFFFFFF;
 
+    regs.depth_test_func = Regs::ComparisonOp::Always;
+    regs.cull.front_face = Regs::Cull::FrontFace::CounterClockWise;
+    regs.cull.cull_face = Regs::Cull::CullFace::Back;
+
     // TODO(Rodrigo): Most games do not set a point size. I think this is a case of a
     // register carrying a default value. Assume it's OpenGL's default (1).
     regs.point_size = 1.0f;
@@ -86,6 +91,159 @@ void Maxwell3D::InitializeRegisterDefaults() {
     regs.rt_separate_frag_data = 1;
 }
 
+#define DIRTY_REGS_POS(field_name) (offsetof(Maxwell3D::DirtyRegs, field_name))
+
+void Maxwell3D::InitDirtySettings() {
+    const auto set_block = [this](const u32 start, const u32 range, const u8 position) {
+        const auto start_itr = dirty_pointers.begin() + start;
+        const auto end_itr = start_itr + range;
+        std::fill(start_itr, end_itr, position);
+    };
+    dirty.regs.fill(true);
+
+    // Init Render Targets
+    constexpr u32 registers_per_rt = sizeof(regs.rt[0]) / sizeof(u32);
+    constexpr u32 rt_start_reg = MAXWELL3D_REG_INDEX(rt);
+    constexpr u32 rt_end_reg = rt_start_reg + registers_per_rt * 8;
+    u32 rt_dirty_reg = DIRTY_REGS_POS(render_target);
+    for (u32 rt_reg = rt_start_reg; rt_reg < rt_end_reg; rt_reg += registers_per_rt) {
+        set_block(rt_reg, registers_per_rt, rt_dirty_reg);
+        rt_dirty_reg++;
+    }
+    constexpr u32 depth_buffer_flag = DIRTY_REGS_POS(depth_buffer);
+    dirty_pointers[MAXWELL3D_REG_INDEX(zeta_enable)] = depth_buffer_flag;
+    dirty_pointers[MAXWELL3D_REG_INDEX(zeta_width)] = depth_buffer_flag;
+    dirty_pointers[MAXWELL3D_REG_INDEX(zeta_height)] = depth_buffer_flag;
+    constexpr u32 registers_in_zeta = sizeof(regs.zeta) / sizeof(u32);
+    constexpr u32 zeta_reg = MAXWELL3D_REG_INDEX(zeta);
+    set_block(zeta_reg, registers_in_zeta, depth_buffer_flag);
+
+    // Init Vertex Arrays
+    constexpr u32 vertex_array_start = MAXWELL3D_REG_INDEX(vertex_array);
+    constexpr u32 vertex_array_size = sizeof(regs.vertex_array[0]) / sizeof(u32);
+    constexpr u32 vertex_array_end = vertex_array_start + vertex_array_size * Regs::NumVertexArrays;
+    u32 va_reg = DIRTY_REGS_POS(vertex_array);
+    u32 vi_reg = DIRTY_REGS_POS(vertex_instance);
+    for (u32 vertex_reg = vertex_array_start; vertex_reg < vertex_array_end;
+         vertex_reg += vertex_array_size) {
+        set_block(vertex_reg, 3, va_reg);
+        // The divisor concerns vertex array instances
+        dirty_pointers[vertex_reg + 3] = vi_reg;
+        va_reg++;
+        vi_reg++;
+    }
+    constexpr u32 vertex_limit_start = MAXWELL3D_REG_INDEX(vertex_array_limit);
+    constexpr u32 vertex_limit_size = sizeof(regs.vertex_array_limit[0]) / sizeof(u32);
+    constexpr u32 vertex_limit_end = vertex_limit_start + vertex_limit_size * Regs::NumVertexArrays;
+    va_reg = DIRTY_REGS_POS(vertex_array);
+    for (u32 vertex_reg = vertex_limit_start; vertex_reg < vertex_limit_end;
+         vertex_reg += vertex_limit_size) {
+        set_block(vertex_reg, vertex_limit_size, va_reg);
+        va_reg++;
+    }
+    constexpr u32 vertex_instance_start = MAXWELL3D_REG_INDEX(instanced_arrays);
+    constexpr u32 vertex_instance_size =
+        sizeof(regs.instanced_arrays.is_instanced[0]) / sizeof(u32);
+    constexpr u32 vertex_instance_end =
+        vertex_instance_start + vertex_instance_size * Regs::NumVertexArrays;
+    vi_reg = DIRTY_REGS_POS(vertex_instance);
+    for (u32 vertex_reg = vertex_instance_start; vertex_reg < vertex_instance_end;
+         vertex_reg += vertex_instance_size) {
+        set_block(vertex_reg, vertex_instance_size, vi_reg);
+        vi_reg++;
+    }
+    set_block(MAXWELL3D_REG_INDEX(vertex_attrib_format), regs.vertex_attrib_format.size(),
+              DIRTY_REGS_POS(vertex_attrib_format));
+
+    // Init Shaders
+    constexpr u32 shader_registers_count =
+        sizeof(regs.shader_config[0]) * Regs::MaxShaderProgram / sizeof(u32);
+    set_block(MAXWELL3D_REG_INDEX(shader_config[0]), shader_registers_count,
+              DIRTY_REGS_POS(shaders));
+
+    // State
+
+    // Viewport
+    constexpr u32 viewport_dirty_reg = DIRTY_REGS_POS(viewport);
+    constexpr u32 viewport_start = MAXWELL3D_REG_INDEX(viewports);
+    constexpr u32 viewport_size = sizeof(regs.viewports) / sizeof(u32);
+    set_block(viewport_start, viewport_size, viewport_dirty_reg);
+    constexpr u32 view_volume_start = MAXWELL3D_REG_INDEX(view_volume_clip_control);
+    constexpr u32 view_volume_size = sizeof(regs.view_volume_clip_control) / sizeof(u32);
+    set_block(view_volume_start, view_volume_size, viewport_dirty_reg);
+
+    // Viewport transformation
+    constexpr u32 viewport_trans_start = MAXWELL3D_REG_INDEX(viewport_transform);
+    constexpr u32 viewport_trans_size = sizeof(regs.viewport_transform) / sizeof(u32);
+    set_block(viewport_trans_start, viewport_trans_size, DIRTY_REGS_POS(viewport_transform));
+
+    // Cullmode
+    constexpr u32 cull_mode_start = MAXWELL3D_REG_INDEX(cull);
+    constexpr u32 cull_mode_size = sizeof(regs.cull) / sizeof(u32);
+    set_block(cull_mode_start, cull_mode_size, DIRTY_REGS_POS(cull_mode));
+
+    // Screen y control
+    dirty_pointers[MAXWELL3D_REG_INDEX(screen_y_control)] = DIRTY_REGS_POS(screen_y_control);
+
+    // Primitive Restart
+    constexpr u32 primitive_restart_start = MAXWELL3D_REG_INDEX(primitive_restart);
+    constexpr u32 primitive_restart_size = sizeof(regs.primitive_restart) / sizeof(u32);
+    set_block(primitive_restart_start, primitive_restart_size, DIRTY_REGS_POS(primitive_restart));
+
+    // Depth Test
+    constexpr u32 depth_test_dirty_reg = DIRTY_REGS_POS(depth_test);
+    dirty_pointers[MAXWELL3D_REG_INDEX(depth_test_enable)] = depth_test_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(depth_write_enabled)] = depth_test_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(depth_test_func)] = depth_test_dirty_reg;
+
+    // Stencil Test
+    constexpr u32 stencil_test_dirty_reg = DIRTY_REGS_POS(stencil_test);
+    dirty_pointers[MAXWELL3D_REG_INDEX(stencil_enable)] = stencil_test_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(stencil_front_func_func)] = stencil_test_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(stencil_front_func_ref)] = stencil_test_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(stencil_front_func_mask)] = stencil_test_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(stencil_front_op_fail)] = stencil_test_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(stencil_front_op_zfail)] = stencil_test_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(stencil_front_op_zpass)] = stencil_test_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(stencil_front_mask)] = stencil_test_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(stencil_two_side_enable)] = stencil_test_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(stencil_back_func_func)] = stencil_test_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(stencil_back_func_ref)] = stencil_test_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(stencil_back_func_mask)] = stencil_test_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(stencil_back_op_fail)] = stencil_test_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(stencil_back_op_zfail)] = stencil_test_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(stencil_back_op_zpass)] = stencil_test_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(stencil_back_mask)] = stencil_test_dirty_reg;
+
+    // Color Mask
+    constexpr u32 color_mask_dirty_reg = DIRTY_REGS_POS(color_mask);
+    dirty_pointers[MAXWELL3D_REG_INDEX(color_mask_common)] = color_mask_dirty_reg;
+    set_block(MAXWELL3D_REG_INDEX(color_mask), sizeof(regs.color_mask) / sizeof(u32),
+              color_mask_dirty_reg);
+    // Blend State
+    constexpr u32 blend_state_dirty_reg = DIRTY_REGS_POS(blend_state);
+    set_block(MAXWELL3D_REG_INDEX(blend_color), sizeof(regs.blend_color) / sizeof(u32),
+              blend_state_dirty_reg);
+    dirty_pointers[MAXWELL3D_REG_INDEX(independent_blend_enable)] = blend_state_dirty_reg;
+    set_block(MAXWELL3D_REG_INDEX(blend), sizeof(regs.blend) / sizeof(u32), blend_state_dirty_reg);
+    set_block(MAXWELL3D_REG_INDEX(independent_blend), sizeof(regs.independent_blend) / sizeof(u32),
+              blend_state_dirty_reg);
+
+    // Scissor State
+    constexpr u32 scissor_test_dirty_reg = DIRTY_REGS_POS(scissor_test);
+    set_block(MAXWELL3D_REG_INDEX(scissor_test), sizeof(regs.scissor_test) / sizeof(u32),
+              scissor_test_dirty_reg);
+
+    // Polygon Offset
+    constexpr u32 polygon_offset_dirty_reg = DIRTY_REGS_POS(polygon_offset);
+    dirty_pointers[MAXWELL3D_REG_INDEX(polygon_offset_fill_enable)] = polygon_offset_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(polygon_offset_line_enable)] = polygon_offset_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(polygon_offset_point_enable)] = polygon_offset_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(polygon_offset_units)] = polygon_offset_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(polygon_offset_factor)] = polygon_offset_dirty_reg;
+    dirty_pointers[MAXWELL3D_REG_INDEX(polygon_offset_clamp)] = polygon_offset_dirty_reg;
+}
+
 void Maxwell3D::CallMacroMethod(u32 method, std::vector<u32> parameters) {
     // Reset the current macro.
     executing_macro = 0;
@@ -108,6 +266,14 @@ void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) {
 
     const u32 method = method_call.method;
 
+    if (method == cb_data_state.current) {
+        regs.reg_array[method] = method_call.argument;
+        ProcessCBData(method_call.argument);
+        return;
+    } else if (cb_data_state.current != null_cb_data) {
+        FinishCBData();
+    }
+
     // It is an error to write to a register other than the current macro's ARG register before it
     // has finished execution.
     if (executing_macro != 0) {
@@ -143,49 +309,19 @@ void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) {
 
     if (regs.reg_array[method] != method_call.argument) {
         regs.reg_array[method] = method_call.argument;
-        // Color buffers
-        constexpr u32 first_rt_reg = MAXWELL3D_REG_INDEX(rt);
-        constexpr u32 registers_per_rt = sizeof(regs.rt[0]) / sizeof(u32);
-        if (method >= first_rt_reg &&
-            method < first_rt_reg + registers_per_rt * Regs::NumRenderTargets) {
-            const std::size_t rt_index = (method - first_rt_reg) / registers_per_rt;
-            dirty_flags.color_buffer.set(rt_index);
-        }
-
-        // Zeta buffer
-        constexpr u32 registers_in_zeta = sizeof(regs.zeta) / sizeof(u32);
-        if (method == MAXWELL3D_REG_INDEX(zeta_enable) ||
-            method == MAXWELL3D_REG_INDEX(zeta_width) ||
-            method == MAXWELL3D_REG_INDEX(zeta_height) ||
-            (method >= MAXWELL3D_REG_INDEX(zeta) &&
-             method < MAXWELL3D_REG_INDEX(zeta) + registers_in_zeta)) {
-            dirty_flags.zeta_buffer = true;
-        }
-
-        // Shader
-        constexpr u32 shader_registers_count =
-            sizeof(regs.shader_config[0]) * Regs::MaxShaderProgram / sizeof(u32);
-        if (method >= MAXWELL3D_REG_INDEX(shader_config[0]) &&
-            method < MAXWELL3D_REG_INDEX(shader_config[0]) + shader_registers_count) {
-            dirty_flags.shaders = true;
-        }
-
-        // Vertex format
-        if (method >= MAXWELL3D_REG_INDEX(vertex_attrib_format) &&
-            method < MAXWELL3D_REG_INDEX(vertex_attrib_format) + regs.vertex_attrib_format.size()) {
-            dirty_flags.vertex_attrib_format = true;
-        }
-
-        // Vertex buffer
-        if (method >= MAXWELL3D_REG_INDEX(vertex_array) &&
-            method < MAXWELL3D_REG_INDEX(vertex_array) + 4 * Regs::NumVertexArrays) {
-            dirty_flags.vertex_array.set((method - MAXWELL3D_REG_INDEX(vertex_array)) >> 2);
-        } else if (method >= MAXWELL3D_REG_INDEX(vertex_array_limit) &&
-                   method < MAXWELL3D_REG_INDEX(vertex_array_limit) + 2 * Regs::NumVertexArrays) {
-            dirty_flags.vertex_array.set((method - MAXWELL3D_REG_INDEX(vertex_array_limit)) >> 1);
-        } else if (method >= MAXWELL3D_REG_INDEX(instanced_arrays) &&
-                   method < MAXWELL3D_REG_INDEX(instanced_arrays) + Regs::NumVertexArrays) {
-            dirty_flags.vertex_array.set(method - MAXWELL3D_REG_INDEX(instanced_arrays));
+        const std::size_t dirty_reg = dirty_pointers[method];
+        if (dirty_reg) {
+            dirty.regs[dirty_reg] = true;
+            if (dirty_reg >= DIRTY_REGS_POS(vertex_array) &&
+                dirty_reg < DIRTY_REGS_POS(vertex_array_buffers)) {
+                dirty.vertex_array_buffers = true;
+            } else if (dirty_reg >= DIRTY_REGS_POS(vertex_instance) &&
+                       dirty_reg < DIRTY_REGS_POS(vertex_instances)) {
+                dirty.vertex_instances = true;
+            } else if (dirty_reg >= DIRTY_REGS_POS(render_target) &&
+                       dirty_reg < DIRTY_REGS_POS(render_settings)) {
+                dirty.render_settings = true;
+            }
         }
     }
 
@@ -214,7 +350,7 @@ void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) {
     case MAXWELL3D_REG_INDEX(const_buffer.cb_data[13]):
     case MAXWELL3D_REG_INDEX(const_buffer.cb_data[14]):
     case MAXWELL3D_REG_INDEX(const_buffer.cb_data[15]): {
-        ProcessCBData(method_call.argument);
+        StartCBData(method);
         break;
     }
     case MAXWELL3D_REG_INDEX(cb_bind[0].raw_config): {
@@ -249,6 +385,10 @@ void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) {
         ProcessQueryGet();
         break;
     }
+    case MAXWELL3D_REG_INDEX(condition.mode): {
+        ProcessQueryCondition();
+        break;
+    }
     case MAXWELL3D_REG_INDEX(sync_info): {
         ProcessSyncPoint();
         break;
@@ -261,7 +401,7 @@ void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) {
         const bool is_last_call = method_call.IsLastCall();
         upload_state.ProcessData(method_call.argument, is_last_call);
         if (is_last_call) {
-            dirty_flags.OnMemoryWrite();
+            dirty.OnMemoryWrite();
         }
         break;
     }
@@ -302,6 +442,7 @@ void Maxwell3D::ProcessQueryGet() {
         result = regs.query.query_sequence;
         break;
     default:
+        result = 1;
         UNIMPLEMENTED_MSG("Unimplemented query select type {}",
                           static_cast<u32>(regs.query.query_get.select.Value()));
     }
@@ -333,7 +474,6 @@ void Maxwell3D::ProcessQueryGet() {
             query_result.timestamp = system.CoreTiming().GetTicks();
             memory_manager.WriteBlock(sequence_address, &query_result, sizeof(query_result));
         }
-        dirty_flags.OnMemoryWrite();
         break;
     }
     default:
@@ -342,12 +482,52 @@ void Maxwell3D::ProcessQueryGet() {
     }
 }
 
+void Maxwell3D::ProcessQueryCondition() {
+    const GPUVAddr condition_address{regs.condition.Address()};
+    switch (regs.condition.mode) {
+    case Regs::ConditionMode::Always: {
+        execute_on = true;
+        break;
+    }
+    case Regs::ConditionMode::Never: {
+        execute_on = false;
+        break;
+    }
+    case Regs::ConditionMode::ResNonZero: {
+        Regs::QueryCompare cmp;
+        memory_manager.ReadBlockUnsafe(condition_address, &cmp, sizeof(cmp));
+        execute_on = cmp.initial_sequence != 0U && cmp.initial_mode != 0U;
+        break;
+    }
+    case Regs::ConditionMode::Equal: {
+        Regs::QueryCompare cmp;
+        memory_manager.ReadBlockUnsafe(condition_address, &cmp, sizeof(cmp));
+        execute_on =
+            cmp.initial_sequence == cmp.current_sequence && cmp.initial_mode == cmp.current_mode;
+        break;
+    }
+    case Regs::ConditionMode::NotEqual: {
+        Regs::QueryCompare cmp;
+        memory_manager.ReadBlockUnsafe(condition_address, &cmp, sizeof(cmp));
+        execute_on =
+            cmp.initial_sequence != cmp.current_sequence || cmp.initial_mode != cmp.current_mode;
+        break;
+    }
+    default: {
+        UNIMPLEMENTED_MSG("Uninplemented Condition Mode!");
+        execute_on = true;
+        break;
+    }
+    }
+}
+
 void Maxwell3D::ProcessSyncPoint() {
     const u32 sync_point = regs.sync_info.sync_point.Value();
     const u32 increment = regs.sync_info.increment.Value();
     const u32 cache_flush = regs.sync_info.unknown.Value();
-    LOG_DEBUG(HW_GPU, "Syncpoint set {}, increment: {}, unk: {}", sync_point, increment,
-              cache_flush);
+    if (increment) {
+        system.GPU().IncrementSyncPoint(sync_point);
+    }
 }
 
 void Maxwell3D::DrawArrays() {
@@ -405,23 +585,39 @@ void Maxwell3D::ProcessCBBind(Regs::ShaderStage stage) {
 }
 
 void Maxwell3D::ProcessCBData(u32 value) {
+    const u32 id = cb_data_state.id;
+    cb_data_state.buffer[id][cb_data_state.counter] = value;
+    // Increment the current buffer position.
+    regs.const_buffer.cb_pos = regs.const_buffer.cb_pos + 4;
+    cb_data_state.counter++;
+}
+
+void Maxwell3D::StartCBData(u32 method) {
+    constexpr u32 first_cb_data = MAXWELL3D_REG_INDEX(const_buffer.cb_data[0]);
+    cb_data_state.start_pos = regs.const_buffer.cb_pos;
+    cb_data_state.id = method - first_cb_data;
+    cb_data_state.current = method;
+    cb_data_state.counter = 0;
+    ProcessCBData(regs.const_buffer.cb_data[cb_data_state.id]);
+}
+
+void Maxwell3D::FinishCBData() {
     // Write the input value to the current const buffer at the current position.
     const GPUVAddr buffer_address = regs.const_buffer.BufferAddress();
     ASSERT(buffer_address != 0);
 
     // Don't allow writing past the end of the buffer.
-    ASSERT(regs.const_buffer.cb_pos + sizeof(u32) <= regs.const_buffer.cb_size);
+    ASSERT(regs.const_buffer.cb_pos <= regs.const_buffer.cb_size);
 
-    const GPUVAddr address{buffer_address + regs.const_buffer.cb_pos};
+    const GPUVAddr address{buffer_address + cb_data_state.start_pos};
+    const std::size_t size = regs.const_buffer.cb_pos - cb_data_state.start_pos;
 
-    u8* ptr{memory_manager.GetPointer(address)};
-    rasterizer.InvalidateRegion(ToCacheAddr(ptr), sizeof(u32));
-    memory_manager.Write<u32>(address, value);
+    const u32 id = cb_data_state.id;
+    memory_manager.WriteBlock(address, cb_data_state.buffer[id].data(), size);
+    dirty.OnMemoryWrite();
 
-    dirty_flags.OnMemoryWrite();
-
-    // Increment the current buffer position.
-    regs.const_buffer.cb_pos = regs.const_buffer.cb_pos + 4;
+    cb_data_state.id = null_cb_data;
+    cb_data_state.current = null_cb_data;
 }
 
 Texture::TICEntry Maxwell3D::GetTICEntry(u32 tic_index) const {

src/video_core/engines/maxwell_3d.h

@@ -67,6 +67,7 @@ public:
         static constexpr std::size_t MaxShaderStage = 5;
         // Maximum number of const buffers per shader stage.
         static constexpr std::size_t MaxConstBuffers = 18;
+        static constexpr std::size_t MaxConstBufferSize = 0x10000;
 
         enum class QueryMode : u32 {
             Write = 0,
@@ -89,6 +90,20 @@ public:
 
         enum class QuerySelect : u32 {
             Zero = 0,
+            TimeElapsed = 2,
+            TransformFeedbackPrimitivesGenerated = 11,
+            PrimitivesGenerated = 18,
+            SamplesPassed = 21,
+            TransformFeedbackUnknown = 26,
+        };
+
+        struct QueryCompare {
+            u32 initial_sequence;
+            u32 initial_mode;
+            u32 unknown1;
+            u32 unknown2;
+            u32 current_sequence;
+            u32 current_mode;
         };
 
         enum class QuerySyncCondition : u32 {
@@ -96,6 +111,14 @@ public:
             GreaterThan = 1,
         };
 
+        enum class ConditionMode : u32 {
+            Never = 0,
+            Always = 1,
+            ResNonZero = 2,
+            Equal = 3,
+            NotEqual = 4,
+        };
+
         enum class ShaderProgram : u32 {
             VertexA = 0,
             VertexB = 1,
@@ -814,7 +837,18 @@ public:
                     BitField<4, 1, u32> alpha_to_one;
                 } multisample_control;
 
-                INSERT_PADDING_WORDS(0x7);
+                INSERT_PADDING_WORDS(0x4);
+
+                struct {
+                    u32 address_high;
+                    u32 address_low;
+                    ConditionMode mode;
+
+                    GPUVAddr Address() const {
+                        return static_cast<GPUVAddr>((static_cast<GPUVAddr>(address_high) << 32) |
+                                                     address_low);
+                    }
+                } condition;
 
                 struct {
                     u32 tsc_address_high;
@@ -1123,23 +1157,77 @@ public:
 
     State state{};
 
-    struct DirtyFlags {
-        std::bitset<8> color_buffer{0xFF};
-        std::bitset<32> vertex_array{0xFFFFFFFF};
+    struct DirtyRegs {
+        static constexpr std::size_t NUM_REGS = 256;
+        union {
+            struct {
+                bool null_dirty;
 
-        bool vertex_attrib_format = true;
-        bool zeta_buffer = true;
-        bool shaders = true;
+                // Vertex Attributes
+                bool vertex_attrib_format;
+
+                // Vertex Arrays
+                std::array<bool, 32> vertex_array;
+
+                bool vertex_array_buffers;
+
+                // Vertex Instances
+                std::array<bool, 32> vertex_instance;
+
+                bool vertex_instances;
+
+                // Render Targets
+                std::array<bool, 8> render_target;
+                bool depth_buffer;
+
+                bool render_settings;
+
+                // Shaders
+                bool shaders;
+
+                // Rasterizer State
+                bool viewport;
+                bool clip_coefficient;
+                bool cull_mode;
+                bool primitive_restart;
+                bool depth_test;
+                bool stencil_test;
+                bool blend_state;
+                bool scissor_test;
+                bool transform_feedback;
+                bool color_mask;
+                bool polygon_offset;
+
+                // Complementary
+                bool viewport_transform;
+                bool screen_y_control;
+
+                bool memory_general;
+            };
+            std::array<bool, NUM_REGS> regs;
+        };
+
+        void ResetVertexArrays() {
+            vertex_array.fill(true);
+            vertex_array_buffers = true;
+        }
+
+        void ResetRenderTargets() {
+            depth_buffer = true;
+            render_target.fill(true);
+            render_settings = true;
+        }
 
         void OnMemoryWrite() {
-            zeta_buffer = true;
             shaders = true;
-            color_buffer.set();
-            vertex_array.set();
+            memory_general = true;
+            ResetRenderTargets();
+            ResetVertexArrays();
         }
-    };
 
-    DirtyFlags dirty_flags;
+    } dirty{};
+
+    std::array<u8, Regs::NUM_REGS> dirty_pointers{};
 
     /// Reads a register value located at the input method address
     u32 GetRegisterValue(u32 method) const;
@@ -1168,6 +1256,10 @@ public:
         return macro_memory;
     }
 
+    bool ShouldExecute() const {
+        return execute_on;
+    }
+
 private:
     void InitializeRegisterDefaults();
 
@@ -1191,14 +1283,27 @@ private:
     /// Interpreter for the macro codes uploaded to the GPU.
     MacroInterpreter macro_interpreter;
 
+    static constexpr u32 null_cb_data = 0xFFFFFFFF;
+    struct {
+        std::array<std::array<u32, 0x4000>, 16> buffer;
+        u32 current{null_cb_data};
+        u32 id{null_cb_data};
+        u32 start_pos{};
+        u32 counter{};
+    } cb_data_state;
+
     Upload::State upload_state;
 
+    bool execute_on{true};
+
     /// Retrieves information about a specific TIC entry from the TIC buffer.
     Texture::TICEntry GetTICEntry(u32 tic_index) const;
 
     /// Retrieves information about a specific TSC entry from the TSC buffer.
     Texture::TSCEntry GetTSCEntry(u32 tsc_index) const;
 
+    void InitDirtySettings();
+
     /**
      * Call a macro on this engine.
      * @param method Method to call
@@ -1218,11 +1323,16 @@ private:
     /// Handles a write to the QUERY_GET register.
     void ProcessQueryGet();
 
+    // Handles Conditional Rendering
+    void ProcessQueryCondition();
+
     /// Handles writes to syncing register.
     void ProcessSyncPoint();
 
     /// Handles a write to the CB_DATA[i] register.
+    void StartCBData(u32 method);
     void ProcessCBData(u32 value);
+    void FinishCBData();
 
     /// Handles a write to the CB_BIND register.
     void ProcessCBBind(Regs::ShaderStage stage);
@@ -1289,6 +1399,7 @@ ASSERT_REG_POSITION(clip_distance_enabled, 0x544);
 ASSERT_REG_POSITION(point_size, 0x546);
 ASSERT_REG_POSITION(zeta_enable, 0x54E);
 ASSERT_REG_POSITION(multisample_control, 0x54F);
+ASSERT_REG_POSITION(condition, 0x554);
 ASSERT_REG_POSITION(tsc, 0x557);
 ASSERT_REG_POSITION(polygon_offset_factor, 0x55b);
 ASSERT_REG_POSITION(tic, 0x55D);

src/video_core/engines/maxwell_dma.cpp

@@ -38,7 +38,7 @@ void MaxwellDMA::CallMethod(const GPU::MethodCall& method_call) {
 }
 
 void MaxwellDMA::HandleCopy() {
-    LOG_WARNING(HW_GPU, "Requested a DMA copy");
+    LOG_TRACE(HW_GPU, "Requested a DMA copy");
 
     const GPUVAddr source = regs.src_address.Address();
     const GPUVAddr dest = regs.dst_address.Address();
@@ -58,7 +58,7 @@ void MaxwellDMA::HandleCopy() {
     }
 
     // All copies here update the main memory, so mark all rasterizer states as invalid.
-    system.GPU().Maxwell3D().dirty_flags.OnMemoryWrite();
+    system.GPU().Maxwell3D().dirty.OnMemoryWrite();
 
     if (regs.exec.is_dst_linear && regs.exec.is_src_linear) {
         // When the enable_2d bit is disabled, the copy is performed as if we were copying a 1D

src/video_core/engines/shader_bytecode.h

@@ -78,7 +78,7 @@ union Attribute {
     constexpr explicit Attribute(u64 value) : value(value) {}
 
     enum class Index : u64 {
-        PointSize = 6,
+        LayerViewportPointSize = 6,
         Position = 7,
         Attribute_0 = 8,
         Attribute_31 = 39,
@@ -559,6 +559,11 @@ union Instruction {
     BitField<39, 8, Register> gpr39;
     BitField<48, 16, u64> opcode;
 
+    union {
+        BitField<8, 5, ConditionCode> cc;
+        BitField<13, 1, u64> trigger;
+    } nop;
+
     union {
         BitField<8, 8, Register> gpr;
         BitField<20, 24, s64> offset;
@@ -931,8 +936,6 @@ union Instruction {
     } csetp;
 
     union {
-        BitField<35, 4, PredCondition> cond;
-        BitField<49, 1, u64> h_and;
         BitField<6, 1, u64> ftz;
         BitField<45, 2, PredOperation> op;
         BitField<3, 3, u64> pred3;
@@ -940,9 +943,21 @@ union Instruction {
         BitField<43, 1, u64> negate_a;
         BitField<44, 1, u64> abs_a;
         BitField<47, 2, HalfType> type_a;
-        BitField<31, 1, u64> negate_b;
-        BitField<30, 1, u64> abs_b;
-        BitField<28, 2, HalfType> type_b;
+        union {
+            BitField<35, 4, PredCondition> cond;
+            BitField<49, 1, u64> h_and;
+            BitField<31, 1, u64> negate_b;
+            BitField<30, 1, u64> abs_b;
+            BitField<28, 2, HalfType> type_b;
+        } reg;
+        union {
+            BitField<56, 1, u64> negate_b;
+            BitField<54, 1, u64> abs_b;
+        } cbuf;
+        union {
+            BitField<49, 4, PredCondition> cond;
+            BitField<53, 1, u64> h_and;
+        } cbuf_and_imm;
         BitField<42, 1, u64> neg_pred;
         BitField<39, 3, u64> pred39;
     } hsetp2;
@@ -1506,6 +1521,7 @@ public:
         TMML,   // Texture Mip Map Level
         SUST,   // Surface Store
         EXIT,
+        NOP,
         IPA,
         OUT_R, // Emit vertex/primitive
         ISBERD,
@@ -1548,7 +1564,9 @@ public:
         HFMA2_RC,
         HFMA2_RR,
         HFMA2_IMM_R,
+        HSETP2_C,
         HSETP2_R,
+        HSETP2_IMM,
         HSET2_R,
         POPC_C,
         POPC_R,
@@ -1783,6 +1801,7 @@ private:
             INST("110111110110----", Id::TMML_B, Type::Texture, "TMML_B"),
             INST("1101111101011---", Id::TMML, Type::Texture, "TMML"),
             INST("11101011001-----", Id::SUST, Type::Image, "SUST"),
+            INST("0101000010110---", Id::NOP, Type::Trivial, "NOP"),
             INST("11100000--------", Id::IPA, Type::Trivial, "IPA"),
             INST("1111101111100---", Id::OUT_R, Type::Trivial, "OUT_R"),
             INST("1110111111010---", Id::ISBERD, Type::Trivial, "ISBERD"),
@@ -1831,7 +1850,9 @@ private:
             INST("01100---1-------", Id::HFMA2_RC, Type::Hfma2, "HFMA2_RC"),
             INST("0101110100000---", Id::HFMA2_RR, Type::Hfma2, "HFMA2_RR"),
             INST("01110---0-------", Id::HFMA2_IMM_R, Type::Hfma2, "HFMA2_R_IMM"),
-            INST("0101110100100---", Id::HSETP2_R, Type::HalfSetPredicate, "HSETP_R"),
+            INST("0111111-1-------", Id::HSETP2_C, Type::HalfSetPredicate, "HSETP2_C"),
+            INST("0101110100100---", Id::HSETP2_R, Type::HalfSetPredicate, "HSETP2_R"),
+            INST("0111111-0-------", Id::HSETP2_IMM, Type::HalfSetPredicate, "HSETP2_IMM"),
             INST("0101110100011---", Id::HSET2_R, Type::HalfSet, "HSET2_R"),
             INST("0101000010000---", Id::MUFU, Type::Arithmetic, "MUFU"),
             INST("0100110010010---", Id::RRO_C, Type::Arithmetic, "RRO_C"),

src/video_core/gpu.cpp

@@ -29,9 +29,10 @@ u32 FramebufferConfig::BytesPerPixel(PixelFormat format) {
     UNREACHABLE();
 }
 
-GPU::GPU(Core::System& system, VideoCore::RendererBase& renderer) : renderer{renderer} {
+GPU::GPU(Core::System& system, VideoCore::RendererBase& renderer, bool is_async)
+    : system{system}, renderer{renderer}, is_async{is_async} {
     auto& rasterizer{renderer.Rasterizer()};
-    memory_manager = std::make_unique<Tegra::MemoryManager>(rasterizer);
+    memory_manager = std::make_unique<Tegra::MemoryManager>(system, rasterizer);
     dma_pusher = std::make_unique<Tegra::DmaPusher>(*this);
     maxwell_3d = std::make_unique<Engines::Maxwell3D>(system, rasterizer, *memory_manager);
     fermi_2d = std::make_unique<Engines::Fermi2D>(rasterizer, *memory_manager);
@@ -50,6 +51,14 @@ const Engines::Maxwell3D& GPU::Maxwell3D() const {
     return *maxwell_3d;
 }
 
+Engines::KeplerCompute& GPU::KeplerCompute() {
+    return *kepler_compute;
+}
+
+const Engines::KeplerCompute& GPU::KeplerCompute() const {
+    return *kepler_compute;
+}
+
 MemoryManager& GPU::MemoryManager() {
     return *memory_manager;
 }
@@ -66,6 +75,51 @@ const DmaPusher& GPU::DmaPusher() const {
     return *dma_pusher;
 }
 
+void GPU::IncrementSyncPoint(const u32 syncpoint_id) {
+    syncpoints[syncpoint_id]++;
+    std::lock_guard lock{sync_mutex};
+    if (!syncpt_interrupts[syncpoint_id].empty()) {
+        u32 value = syncpoints[syncpoint_id].load();
+        auto it = syncpt_interrupts[syncpoint_id].begin();
+        while (it != syncpt_interrupts[syncpoint_id].end()) {
+            if (value >= *it) {
+                TriggerCpuInterrupt(syncpoint_id, *it);
+                it = syncpt_interrupts[syncpoint_id].erase(it);
+                continue;
+            }
+            it++;
+        }
+    }
+}
+
+u32 GPU::GetSyncpointValue(const u32 syncpoint_id) const {
+    return syncpoints[syncpoint_id].load();
+}
+
+void GPU::RegisterSyncptInterrupt(const u32 syncpoint_id, const u32 value) {
+    auto& interrupt = syncpt_interrupts[syncpoint_id];
+    bool contains = std::any_of(interrupt.begin(), interrupt.end(),
+                                [value](u32 in_value) { return in_value == value; });
+    if (contains) {
+        return;
+    }
+    syncpt_interrupts[syncpoint_id].emplace_back(value);
+}
+
+bool GPU::CancelSyncptInterrupt(const u32 syncpoint_id, const u32 value) {
+    std::lock_guard lock{sync_mutex};
+    auto& interrupt = syncpt_interrupts[syncpoint_id];
+    const auto iter =
+        std::find_if(interrupt.begin(), interrupt.end(),
+                     [value](u32 interrupt_value) { return value == interrupt_value; });
+
+    if (iter == interrupt.end()) {
+        return false;
+    }
+    interrupt.erase(iter);
+    return true;
+}
+
 u32 RenderTargetBytesPerPixel(RenderTargetFormat format) {
     ASSERT(format != RenderTargetFormat::NONE);
 
@@ -143,12 +197,12 @@ enum class BufferMethods {
     NotifyIntr = 0x8,
     WrcacheFlush = 0x9,
     Unk28 = 0xA,
-    Unk2c = 0xB,
+    UnkCacheFlush = 0xB,
     RefCnt = 0x14,
     SemaphoreAcquire = 0x1A,
     SemaphoreRelease = 0x1B,
-    Unk70 = 0x1C,
-    Unk74 = 0x1D,
+    FenceValue = 0x1C,
+    FenceAction = 0x1D,
     Unk78 = 0x1E,
     Unk7c = 0x1F,
     Yield = 0x20,
@@ -194,6 +248,10 @@ void GPU::CallPullerMethod(const MethodCall& method_call) {
     case BufferMethods::SemaphoreAddressLow:
     case BufferMethods::SemaphoreSequence:
     case BufferMethods::RefCnt:
+    case BufferMethods::UnkCacheFlush:
+    case BufferMethods::WrcacheFlush:
+    case BufferMethods::FenceValue:
+    case BufferMethods::FenceAction:
         break;
     case BufferMethods::SemaphoreTrigger: {
         ProcessSemaphoreTriggerMethod();
@@ -204,21 +262,11 @@ void GPU::CallPullerMethod(const MethodCall& method_call) {
         LOG_ERROR(HW_GPU, "Special puller engine method NotifyIntr not implemented");
         break;
     }
-    case BufferMethods::WrcacheFlush: {
-        // TODO(Kmather73): Research and implement this method.
-        LOG_ERROR(HW_GPU, "Special puller engine method WrcacheFlush not implemented");
-        break;
-    }
     case BufferMethods::Unk28: {
         // TODO(Kmather73): Research and implement this method.
         LOG_ERROR(HW_GPU, "Special puller engine method Unk28 not implemented");
         break;
     }
-    case BufferMethods::Unk2c: {
-        // TODO(Kmather73): Research and implement this method.
-        LOG_ERROR(HW_GPU, "Special puller engine method Unk2c not implemented");
-        break;
-    }
     case BufferMethods::SemaphoreAcquire: {
         ProcessSemaphoreAcquire();
         break;

src/video_core/gpu.h

@@ -5,8 +5,12 @@
 #pragma once
 
 #include <array>
+#include <atomic>
+#include <list>
 #include <memory>
+#include <mutex>
 #include "common/common_types.h"
+#include "core/hle/service/nvdrv/nvdata.h"
 #include "core/hle/service/nvflinger/buffer_queue.h"
 #include "video_core/dma_pusher.h"
 
@@ -127,7 +131,7 @@ class MemoryManager;
 
 class GPU {
 public:
-    explicit GPU(Core::System& system, VideoCore::RendererBase& renderer);
+    explicit GPU(Core::System& system, VideoCore::RendererBase& renderer, bool is_async);
 
     virtual ~GPU();
 
@@ -155,6 +159,12 @@ public:
     /// Returns a const reference to the Maxwell3D GPU engine.
     const Engines::Maxwell3D& Maxwell3D() const;
 
+    /// Returns a reference to the KeplerCompute GPU engine.
+    Engines::KeplerCompute& KeplerCompute();
+
+    /// Returns a reference to the KeplerCompute GPU engine.
+    const Engines::KeplerCompute& KeplerCompute() const;
+
     /// Returns a reference to the GPU memory manager.
     Tegra::MemoryManager& MemoryManager();
 
@@ -164,6 +174,22 @@ public:
     /// Returns a reference to the GPU DMA pusher.
     Tegra::DmaPusher& DmaPusher();
 
+    void IncrementSyncPoint(u32 syncpoint_id);
+
+    u32 GetSyncpointValue(u32 syncpoint_id) const;
+
+    void RegisterSyncptInterrupt(u32 syncpoint_id, u32 value);
+
+    bool CancelSyncptInterrupt(u32 syncpoint_id, u32 value);
+
+    std::unique_lock<std::mutex> LockSync() {
+        return std::unique_lock{sync_mutex};
+    }
+
+    bool IsAsync() const {
+        return is_async;
+    }
+
     /// Returns a const reference to the GPU DMA pusher.
     const Tegra::DmaPusher& DmaPusher() const;
 
@@ -194,7 +220,12 @@ public:
 
                 u32 semaphore_acquire;
                 u32 semaphore_release;
-                INSERT_PADDING_WORDS(0xE4);
+                u32 fence_value;
+                union {
+                    BitField<4, 4, u32> operation;
+                    BitField<8, 8, u32> id;
+                } fence_action;
+                INSERT_PADDING_WORDS(0xE2);
 
                 // Puller state
                 u32 acquire_mode;
@@ -228,6 +259,9 @@ public:
     /// Notify rasterizer that any caches of the specified region should be flushed and invalidated
     virtual void FlushAndInvalidateRegion(CacheAddr addr, u64 size) = 0;
 
+protected:
+    virtual void TriggerCpuInterrupt(u32 syncpoint_id, u32 value) const = 0;
+
 private:
     void ProcessBindMethod(const MethodCall& method_call);
     void ProcessSemaphoreTriggerMethod();
@@ -246,6 +280,7 @@ private:
 protected:
     std::unique_ptr<Tegra::DmaPusher> dma_pusher;
     VideoCore::RendererBase& renderer;
+    Core::System& system;
 
 private:
     std::unique_ptr<Tegra::MemoryManager> memory_manager;
@@ -262,6 +297,14 @@ private:
     std::unique_ptr<Engines::MaxwellDMA> maxwell_dma;
     /// Inline memory engine
     std::unique_ptr<Engines::KeplerMemory> kepler_memory;
+
+    std::array<std::atomic<u32>, Service::Nvidia::MaxSyncPoints> syncpoints{};
+
+    std::array<std::list<u32>, Service::Nvidia::MaxSyncPoints> syncpt_interrupts;
+
+    std::mutex sync_mutex;
+
+    const bool is_async;
 };
 
 #define ASSERT_REG_POSITION(field_name, position)                                                  \
@@ -274,6 +317,8 @@ ASSERT_REG_POSITION(semaphore_trigger, 0x7);
 ASSERT_REG_POSITION(reference_count, 0x14);
 ASSERT_REG_POSITION(semaphore_acquire, 0x1A);
 ASSERT_REG_POSITION(semaphore_release, 0x1B);
+ASSERT_REG_POSITION(fence_value, 0x1C);
+ASSERT_REG_POSITION(fence_action, 0x1D);
 
 ASSERT_REG_POSITION(acquire_mode, 0x100);
 ASSERT_REG_POSITION(acquire_source, 0x101);

src/video_core/gpu_asynch.cpp

@@ -2,6 +2,8 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
+#include "core/core.h"
+#include "core/hardware_interrupt_manager.h"
 #include "video_core/gpu_asynch.h"
 #include "video_core/gpu_thread.h"
 #include "video_core/renderer_base.h"
@@ -9,7 +11,7 @@
 namespace VideoCommon {
 
 GPUAsynch::GPUAsynch(Core::System& system, VideoCore::RendererBase& renderer)
-    : GPU(system, renderer), gpu_thread{system} {}
+    : GPU(system, renderer, true), gpu_thread{system} {}
 
 GPUAsynch::~GPUAsynch() = default;
 
@@ -38,4 +40,9 @@ void GPUAsynch::FlushAndInvalidateRegion(CacheAddr addr, u64 size) {
     gpu_thread.FlushAndInvalidateRegion(addr, size);
 }
 
+void GPUAsynch::TriggerCpuInterrupt(const u32 syncpoint_id, const u32 value) const {
+    auto& interrupt_manager = system.InterruptManager();
+    interrupt_manager.GPUInterruptSyncpt(syncpoint_id, value);
+}
+
 } // namespace VideoCommon

src/video_core/gpu_asynch.h

@@ -27,6 +27,9 @@ public:
     void InvalidateRegion(CacheAddr addr, u64 size) override;
     void FlushAndInvalidateRegion(CacheAddr addr, u64 size) override;
 
+protected:
+    void TriggerCpuInterrupt(u32 syncpoint_id, u32 value) const override;
+
 private:
     GPUThread::ThreadManager gpu_thread;
 };

src/video_core/gpu_synch.cpp

@@ -8,7 +8,7 @@
 namespace VideoCommon {
 
 GPUSynch::GPUSynch(Core::System& system, VideoCore::RendererBase& renderer)
-    : GPU(system, renderer) {}
+    : GPU(system, renderer, false) {}
 
 GPUSynch::~GPUSynch() = default;
 

src/video_core/gpu_synch.h

@@ -25,6 +25,10 @@ public:
     void FlushRegion(CacheAddr addr, u64 size) override;
     void InvalidateRegion(CacheAddr addr, u64 size) override;
     void FlushAndInvalidateRegion(CacheAddr addr, u64 size) override;
+
+protected:
+    void TriggerCpuInterrupt([[maybe_unused]] u32 syncpoint_id,
+                             [[maybe_unused]] u32 value) const override {}
 };
 
 } // namespace VideoCommon

src/video_core/gpu_thread.cpp

@@ -21,7 +21,8 @@ static void RunThread(VideoCore::RendererBase& renderer, Tegra::DmaPusher& dma_p
     MicroProfileOnThreadCreate("GpuThread");
 
     // Wait for first GPU command before acquiring the window context
-    state.WaitForCommands();
+    while (state.queue.Empty())
+        ;
 
     // If emulation was stopped during disk shader loading, abort before trying to acquire context
     if (!state.is_running) {
@@ -32,7 +33,6 @@ static void RunThread(VideoCore::RendererBase& renderer, Tegra::DmaPusher& dma_p
 
     CommandDataContainer next;
     while (state.is_running) {
-        state.WaitForCommands();
         while (!state.queue.Empty()) {
             state.queue.Pop(next);
             if (const auto submit_list = std::get_if<SubmitListCommand>(&next.data)) {
@@ -49,8 +49,7 @@ static void RunThread(VideoCore::RendererBase& renderer, Tegra::DmaPusher& dma_p
             } else {
                 UNREACHABLE();
             }
-            state.signaled_fence = next.fence;
-            state.TrySynchronize();
+            state.signaled_fence.store(next.fence);
         }
     }
 }
@@ -89,12 +88,7 @@ void ThreadManager::FlushRegion(CacheAddr addr, u64 size) {
 }
 
 void ThreadManager::InvalidateRegion(CacheAddr addr, u64 size) {
-    if (state.queue.Empty()) {
-        // It's quicker to invalidate a single region on the CPU if the queue is already empty
-        system.Renderer().Rasterizer().InvalidateRegion(addr, size);
-    } else {
-        PushCommand(InvalidateRegionCommand(addr, size));
-    }
+    system.Renderer().Rasterizer().InvalidateRegion(addr, size);
 }
 
 void ThreadManager::FlushAndInvalidateRegion(CacheAddr addr, u64 size) {
@@ -105,22 +99,13 @@ void ThreadManager::FlushAndInvalidateRegion(CacheAddr addr, u64 size) {
 u64 ThreadManager::PushCommand(CommandData&& command_data) {
     const u64 fence{++state.last_fence};
     state.queue.Push(CommandDataContainer(std::move(command_data), fence));
-    state.SignalCommands();
     return fence;
 }
 
 MICROPROFILE_DEFINE(GPU_wait, "GPU", "Wait for the GPU", MP_RGB(128, 128, 192));
 void SynchState::WaitForSynchronization(u64 fence) {
-    if (signaled_fence >= fence) {
-        return;
-    }
-
-    // Wait for the GPU to be idle (all commands to be executed)
-    {
-        MICROPROFILE_SCOPE(GPU_wait);
-        std::unique_lock lock{synchronization_mutex};
-        synchronization_condition.wait(lock, [this, fence] { return signaled_fence >= fence; });
-    }
+    while (signaled_fence.load() < fence)
+        ;
 }
 
 } // namespace VideoCommon::GPUThread

src/video_core/gpu_thread.h

@@ -88,41 +88,9 @@ struct CommandDataContainer {
 /// Struct used to synchronize the GPU thread
 struct SynchState final {
     std::atomic_bool is_running{true};
-    std::atomic_int queued_frame_count{};
-    std::mutex synchronization_mutex;
-    std::mutex commands_mutex;
-    std::condition_variable commands_condition;
-    std::condition_variable synchronization_condition;
-
-    /// Returns true if the gap in GPU commands is small enough that we can consider the CPU and GPU
-    /// synchronized. This is entirely empirical.
-    bool IsSynchronized() const {
-        constexpr std::size_t max_queue_gap{5};
-        return queue.Size() <= max_queue_gap;
-    }
-
-    void TrySynchronize() {
-        if (IsSynchronized()) {
-            std::lock_guard lock{synchronization_mutex};
-            synchronization_condition.notify_one();
-        }
-    }
 
     void WaitForSynchronization(u64 fence);
 
-    void SignalCommands() {
-        if (queue.Empty()) {
-            return;
-        }
-
-        commands_condition.notify_one();
-    }
-
-    void WaitForCommands() {
-        std::unique_lock lock{commands_mutex};
-        commands_condition.wait(lock, [this] { return !queue.Empty(); });
-    }
-
     using CommandQueue = Common::SPSCQueue<CommandDataContainer>;
     CommandQueue queue;
     u64 last_fence{};

src/video_core/macro_interpreter.cpp

@@ -4,14 +4,18 @@
 
 #include "common/assert.h"
 #include "common/logging/log.h"
+#include "common/microprofile.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/macro_interpreter.h"
 
+MICROPROFILE_DEFINE(MacroInterp, "GPU", "Execute macro interpreter", MP_RGB(128, 128, 192));
+
 namespace Tegra {
 
 MacroInterpreter::MacroInterpreter(Engines::Maxwell3D& maxwell3d) : maxwell3d(maxwell3d) {}
 
 void MacroInterpreter::Execute(u32 offset, std::vector<u32> parameters) {
+    MICROPROFILE_SCOPE(MacroInterp);
     Reset();
     registers[1] = parameters[0];
     this->parameters = std::move(parameters);

src/video_core/memory_manager.cpp

@@ -5,13 +5,17 @@
 #include "common/alignment.h"
 #include "common/assert.h"
 #include "common/logging/log.h"
+#include "core/core.h"
+#include "core/hle/kernel/process.h"
+#include "core/hle/kernel/vm_manager.h"
 #include "core/memory.h"
 #include "video_core/memory_manager.h"
 #include "video_core/rasterizer_interface.h"
 
 namespace Tegra {
 
-MemoryManager::MemoryManager(VideoCore::RasterizerInterface& rasterizer) : rasterizer{rasterizer} {
+MemoryManager::MemoryManager(Core::System& system, VideoCore::RasterizerInterface& rasterizer)
+    : rasterizer{rasterizer}, system{system} {
     std::fill(page_table.pointers.begin(), page_table.pointers.end(), nullptr);
     std::fill(page_table.attributes.begin(), page_table.attributes.end(),
               Common::PageType::Unmapped);
@@ -49,6 +53,11 @@ GPUVAddr MemoryManager::MapBufferEx(VAddr cpu_addr, u64 size) {
     const GPUVAddr gpu_addr{FindFreeRegion(address_space_base, aligned_size)};
 
     MapBackingMemory(gpu_addr, Memory::GetPointer(cpu_addr), aligned_size, cpu_addr);
+    ASSERT(system.CurrentProcess()
+               ->VMManager()
+               .SetMemoryAttribute(cpu_addr, size, Kernel::MemoryAttribute::DeviceMapped,
+                                   Kernel::MemoryAttribute::DeviceMapped)
+               .IsSuccess());
 
     return gpu_addr;
 }
@@ -59,7 +68,11 @@ GPUVAddr MemoryManager::MapBufferEx(VAddr cpu_addr, GPUVAddr gpu_addr, u64 size)
     const u64 aligned_size{Common::AlignUp(size, page_size)};
 
     MapBackingMemory(gpu_addr, Memory::GetPointer(cpu_addr), aligned_size, cpu_addr);
-
+    ASSERT(system.CurrentProcess()
+               ->VMManager()
+               .SetMemoryAttribute(cpu_addr, size, Kernel::MemoryAttribute::DeviceMapped,
+                                   Kernel::MemoryAttribute::DeviceMapped)
+               .IsSuccess());
     return gpu_addr;
 }
 
@@ -68,9 +81,16 @@ GPUVAddr MemoryManager::UnmapBuffer(GPUVAddr gpu_addr, u64 size) {
 
     const u64 aligned_size{Common::AlignUp(size, page_size)};
     const CacheAddr cache_addr{ToCacheAddr(GetPointer(gpu_addr))};
+    const auto cpu_addr = GpuToCpuAddress(gpu_addr);
+    ASSERT(cpu_addr);
 
     rasterizer.FlushAndInvalidateRegion(cache_addr, aligned_size);
     UnmapRange(gpu_addr, aligned_size);
+    ASSERT(system.CurrentProcess()
+               ->VMManager()
+               .SetMemoryAttribute(cpu_addr.value(), size, Kernel::MemoryAttribute::DeviceMapped,
+                                   Kernel::MemoryAttribute::None)
+               .IsSuccess());
 
     return gpu_addr;
 }

src/video_core/memory_manager.h

@@ -14,6 +14,10 @@ namespace VideoCore {
 class RasterizerInterface;
 }
 
+namespace Core {
+class System;
+}
+
 namespace Tegra {
 
 /**
@@ -47,7 +51,7 @@ struct VirtualMemoryArea {
 
 class MemoryManager final {
 public:
-    explicit MemoryManager(VideoCore::RasterizerInterface& rasterizer);
+    explicit MemoryManager(Core::System& system, VideoCore::RasterizerInterface& rasterizer);
     ~MemoryManager();
 
     GPUVAddr AllocateSpace(u64 size, u64 align);
@@ -173,6 +177,8 @@ private:
     Common::PageTable page_table{page_bits};
     VMAMap vma_map;
     VideoCore::RasterizerInterface& rasterizer;
+
+    Core::System& system;
 };
 
 } // namespace Tegra

src/video_core/rasterizer_interface.h

@@ -34,6 +34,9 @@ public:
     /// Clear the current framebuffer
     virtual void Clear() = 0;
 
+    /// Dispatches a compute shader invocation
+    virtual void DispatchCompute(GPUVAddr code_addr) = 0;
+
     /// Notify rasterizer that all caches should be flushed to Switch memory
     virtual void FlushAll() = 0;
 
@@ -47,6 +50,9 @@ public:
     /// and invalidated
     virtual void FlushAndInvalidateRegion(CacheAddr addr, u64 size) = 0;
 
+    /// Notify rasterizer that a frame is about to finish
+    virtual void TickFrame() = 0;
+
     /// Attempt to use a faster method to perform a surface copy
     virtual bool AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Regs::Surface& src,
                                        const Tegra::Engines::Fermi2D::Regs::Surface& dst,

src/video_core/renderer_opengl/gl_buffer_cache.cpp

@@ -2,103 +2,57 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
-#include <cstring>
 #include <memory>
 
-#include "common/alignment.h"
-#include "core/core.h"
-#include "video_core/memory_manager.h"
+#include <glad/glad.h>
+
+#include "common/assert.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 {
 
-CachedBufferEntry::CachedBufferEntry(VAddr cpu_addr, std::size_t size, GLintptr offset,
-                                     std::size_t alignment, u8* host_ptr)
-    : RasterizerCacheObject{host_ptr}, cpu_addr{cpu_addr}, size{size}, offset{offset},
-      alignment{alignment} {}
+OGLBufferCache::OGLBufferCache(RasterizerOpenGL& rasterizer, Core::System& system,
+                               std::size_t stream_size)
+    : VideoCommon::BufferCache<OGLBuffer, GLuint, OGLStreamBuffer>{
+          rasterizer, system, std::make_unique<OGLStreamBuffer>(stream_size, true)} {}
 
-OGLBufferCache::OGLBufferCache(RasterizerOpenGL& rasterizer, std::size_t size)
-    : RasterizerCache{rasterizer}, stream_buffer(size, true) {}
+OGLBufferCache::~OGLBufferCache() = default;
 
-GLintptr OGLBufferCache::UploadMemory(GPUVAddr gpu_addr, std::size_t size, std::size_t alignment,
-                                      bool cache) {
-    std::lock_guard lock{mutex};
-    auto& memory_manager = Core::System::GetInstance().GPU().MemoryManager();
-
-    // 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.
-    cache &= size >= 2048;
-
-    const auto& host_ptr{memory_manager.GetPointer(gpu_addr)};
-    if (cache) {
-        auto entry = TryGet(host_ptr);
-        if (entry) {
-            if (entry->GetSize() >= size && entry->GetAlignment() == alignment) {
-                return entry->GetOffset();
-            }
-            Unregister(entry);
-        }
-    }
-
-    AlignBuffer(alignment);
-    const GLintptr uploaded_offset = buffer_offset;
-
-    if (!host_ptr) {
-        return uploaded_offset;
-    }
-
-    std::memcpy(buffer_ptr, host_ptr, size);
-    buffer_ptr += size;
-    buffer_offset += size;
-
-    if (cache) {
-        auto entry = std::make_shared<CachedBufferEntry>(
-            *memory_manager.GpuToCpuAddress(gpu_addr), size, uploaded_offset, alignment, host_ptr);
-        Register(entry);
-    }
-
-    return uploaded_offset;
+OGLBuffer OGLBufferCache::CreateBuffer(std::size_t size) {
+    OGLBuffer buffer;
+    buffer.Create();
+    glNamedBufferData(buffer.handle, static_cast<GLsizeiptr>(size), nullptr, GL_DYNAMIC_DRAW);
+    return buffer;
 }
 
-GLintptr OGLBufferCache::UploadHostMemory(const void* raw_pointer, std::size_t size,
-                                          std::size_t alignment) {
-    std::lock_guard lock{mutex};
-    AlignBuffer(alignment);
-    std::memcpy(buffer_ptr, raw_pointer, size);
-    const GLintptr uploaded_offset = buffer_offset;
-
-    buffer_ptr += size;
-    buffer_offset += size;
-    return uploaded_offset;
+const GLuint* OGLBufferCache::ToHandle(const OGLBuffer& buffer) {
+    return &buffer.handle;
 }
 
-bool OGLBufferCache::Map(std::size_t max_size) {
-    bool invalidate;
-    std::tie(buffer_ptr, buffer_offset_base, invalidate) =
-        stream_buffer.Map(static_cast<GLsizeiptr>(max_size), 4);
-    buffer_offset = buffer_offset_base;
-
-    if (invalidate) {
-        InvalidateAll();
-    }
-    return invalidate;
+const GLuint* OGLBufferCache::GetEmptyBuffer(std::size_t) {
+    static const GLuint null_buffer = 0;
+    return &null_buffer;
 }
 
-void OGLBufferCache::Unmap() {
-    stream_buffer.Unmap(buffer_offset - buffer_offset_base);
+void OGLBufferCache::UploadBufferData(const OGLBuffer& buffer, std::size_t offset, std::size_t size,
+                                      const u8* data) {
+    glNamedBufferSubData(buffer.handle, static_cast<GLintptr>(offset),
+                         static_cast<GLsizeiptr>(size), data);
 }
 
-GLuint OGLBufferCache::GetHandle() const {
-    return stream_buffer.GetHandle();
+void OGLBufferCache::DownloadBufferData(const OGLBuffer& buffer, std::size_t offset,
+                                        std::size_t size, u8* data) {
+    glGetNamedBufferSubData(buffer.handle, static_cast<GLintptr>(offset),
+                            static_cast<GLsizeiptr>(size), data);
 }
 
-void OGLBufferCache::AlignBuffer(std::size_t alignment) {
-    // Align the offset, not the mapped pointer
-    const GLintptr offset_aligned =
-        static_cast<GLintptr>(Common::AlignUp(static_cast<std::size_t>(buffer_offset), alignment));
-    buffer_ptr += offset_aligned - buffer_offset;
-    buffer_offset = offset_aligned;
+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));
 }
 
 } // namespace OpenGL

src/video_core/renderer_opengl/gl_buffer_cache.h

@@ -4,80 +4,44 @@
 
 #pragma once
 
-#include <cstddef>
 #include <memory>
-#include <tuple>
 
 #include "common/common_types.h"
+#include "video_core/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"
 
+namespace Core {
+class System;
+}
+
 namespace OpenGL {
 
+class OGLStreamBuffer;
 class RasterizerOpenGL;
 
-class CachedBufferEntry final : public RasterizerCacheObject {
+class OGLBufferCache final : public VideoCommon::BufferCache<OGLBuffer, GLuint, OGLStreamBuffer> {
 public:
-    explicit CachedBufferEntry(VAddr cpu_addr, std::size_t size, GLintptr offset,
-                               std::size_t alignment, u8* host_ptr);
+    explicit OGLBufferCache(RasterizerOpenGL& rasterizer, Core::System& system,
+                            std::size_t stream_size);
+    ~OGLBufferCache();
 
-    VAddr GetCpuAddr() const override {
-        return cpu_addr;
-    }
-
-    std::size_t GetSizeInBytes() const override {
-        return size;
-    }
-
-    std::size_t GetSize() const {
-        return size;
-    }
-
-    GLintptr GetOffset() const {
-        return offset;
-    }
-
-    std::size_t GetAlignment() const {
-        return alignment;
-    }
-
-private:
-    VAddr cpu_addr{};
-    std::size_t size{};
-    GLintptr offset{};
-    std::size_t alignment{};
-};
-
-class OGLBufferCache final : public RasterizerCache<std::shared_ptr<CachedBufferEntry>> {
-public:
-    explicit OGLBufferCache(RasterizerOpenGL& rasterizer, std::size_t size);
-
-    /// Uploads data from a guest GPU address. Returns host's buffer offset where it's been
-    /// allocated.
-    GLintptr UploadMemory(GPUVAddr gpu_addr, std::size_t size, std::size_t alignment = 4,
-                          bool cache = true);
-
-    /// Uploads from a host memory. Returns host's buffer offset where it's been allocated.
-    GLintptr UploadHostMemory(const void* raw_pointer, std::size_t size, std::size_t alignment = 4);
-
-    bool Map(std::size_t max_size);
-    void Unmap();
-
-    GLuint GetHandle() const;
+    const GLuint* GetEmptyBuffer(std::size_t) override;
 
 protected:
-    void AlignBuffer(std::size_t alignment);
+    OGLBuffer CreateBuffer(std::size_t size) override;
 
-    // We do not have to flush this cache as things in it are never modified by us.
-    void FlushObjectInner(const std::shared_ptr<CachedBufferEntry>& object) override {}
+    const GLuint* ToHandle(const OGLBuffer& buffer) override;
 
-private:
-    OGLStreamBuffer stream_buffer;
+    void UploadBufferData(const OGLBuffer& buffer, std::size_t offset, std::size_t size,
+                          const u8* data) override;
 
-    u8* buffer_ptr = nullptr;
-    GLintptr buffer_offset = 0;
-    GLintptr buffer_offset_base = 0;
+    void DownloadBufferData(const OGLBuffer& buffer, std::size_t offset, std::size_t size,
+                            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;
 };
 
 } // namespace OpenGL

src/video_core/renderer_opengl/gl_device.cpp

@@ -24,8 +24,10 @@ T GetInteger(GLenum pname) {
 
 Device::Device() {
     uniform_buffer_alignment = GetInteger<std::size_t>(GL_UNIFORM_BUFFER_OFFSET_ALIGNMENT);
+    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_vertex_viewport_layer = GLAD_GL_ARB_shader_viewport_layer_array;
     has_variable_aoffi = TestVariableAoffi();
     has_component_indexing_bug = TestComponentIndexingBug();
 }
@@ -34,6 +36,7 @@ Device::Device(std::nullptr_t) {
     uniform_buffer_alignment = 0;
     max_vertex_attributes = 16;
     max_varyings = 15;
+    has_vertex_viewport_layer = true;
     has_variable_aoffi = true;
     has_component_indexing_bug = false;
 }

src/video_core/renderer_opengl/gl_device.h

@@ -18,6 +18,10 @@ public:
         return uniform_buffer_alignment;
     }
 
+    std::size_t GetShaderStorageBufferAlignment() const {
+        return shader_storage_alignment;
+    }
+
     u32 GetMaxVertexAttributes() const {
         return max_vertex_attributes;
     }
@@ -26,6 +30,10 @@ public:
         return max_varyings;
     }
 
+    bool HasVertexViewportLayer() const {
+        return has_vertex_viewport_layer;
+    }
+
     bool HasVariableAoffi() const {
         return has_variable_aoffi;
     }
@@ -39,8 +47,10 @@ private:
     static bool TestComponentIndexingBug();
 
     std::size_t uniform_buffer_alignment{};
+    std::size_t shader_storage_alignment{};
     u32 max_vertex_attributes{};
     u32 max_varyings{};
+    bool has_vertex_viewport_layer{};
     bool has_variable_aoffi{};
     bool has_component_indexing_bug{};
 };

src/video_core/renderer_opengl/gl_global_cache.cpp

@@ -1,102 +0,0 @@
-// Copyright 2018 yuzu Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#include <glad/glad.h>
-
-#include "common/logging/log.h"
-#include "core/core.h"
-#include "video_core/memory_manager.h"
-#include "video_core/renderer_opengl/gl_global_cache.h"
-#include "video_core/renderer_opengl/gl_rasterizer.h"
-#include "video_core/renderer_opengl/gl_shader_decompiler.h"
-#include "video_core/renderer_opengl/utils.h"
-
-namespace OpenGL {
-
-CachedGlobalRegion::CachedGlobalRegion(VAddr cpu_addr, u8* host_ptr, u32 size, u32 max_size)
-    : RasterizerCacheObject{host_ptr}, cpu_addr{cpu_addr}, host_ptr{host_ptr}, size{size},
-      max_size{max_size} {
-    buffer.Create();
-    LabelGLObject(GL_BUFFER, buffer.handle, cpu_addr, "GlobalMemory");
-}
-
-CachedGlobalRegion::~CachedGlobalRegion() = default;
-
-void CachedGlobalRegion::Reload(u32 size_) {
-    size = size_;
-    if (size > max_size) {
-        size = max_size;
-        LOG_CRITICAL(HW_GPU, "Global region size {} exceeded the supported size {}!", size_,
-                     max_size);
-    }
-    glNamedBufferData(buffer.handle, size, host_ptr, GL_STREAM_DRAW);
-}
-
-void CachedGlobalRegion::Flush() {
-    LOG_DEBUG(Render_OpenGL, "Flushing {} bytes to CPU memory address 0x{:16}", size, cpu_addr);
-    glGetNamedBufferSubData(buffer.handle, 0, static_cast<GLsizeiptr>(size), host_ptr);
-}
-
-GlobalRegion GlobalRegionCacheOpenGL::TryGetReservedGlobalRegion(CacheAddr addr, u32 size) const {
-    const auto search{reserve.find(addr)};
-    if (search == reserve.end()) {
-        return {};
-    }
-    return search->second;
-}
-
-GlobalRegion GlobalRegionCacheOpenGL::GetUncachedGlobalRegion(GPUVAddr addr, u8* host_ptr,
-                                                              u32 size) {
-    GlobalRegion region{TryGetReservedGlobalRegion(ToCacheAddr(host_ptr), size)};
-    if (!region) {
-        // No reserved surface available, create a new one and reserve it
-        auto& memory_manager{Core::System::GetInstance().GPU().MemoryManager()};
-        const auto cpu_addr{memory_manager.GpuToCpuAddress(addr)};
-        ASSERT(cpu_addr);
-
-        region = std::make_shared<CachedGlobalRegion>(*cpu_addr, host_ptr, size, max_ssbo_size);
-        ReserveGlobalRegion(region);
-    }
-    region->Reload(size);
-    return region;
-}
-
-void GlobalRegionCacheOpenGL::ReserveGlobalRegion(GlobalRegion region) {
-    reserve.insert_or_assign(region->GetCacheAddr(), std::move(region));
-}
-
-GlobalRegionCacheOpenGL::GlobalRegionCacheOpenGL(RasterizerOpenGL& rasterizer)
-    : RasterizerCache{rasterizer} {
-    GLint max_ssbo_size_;
-    glGetIntegerv(GL_MAX_SHADER_STORAGE_BLOCK_SIZE, &max_ssbo_size_);
-    max_ssbo_size = static_cast<u32>(max_ssbo_size_);
-}
-
-GlobalRegion GlobalRegionCacheOpenGL::GetGlobalRegion(
-    const GLShader::GlobalMemoryEntry& global_region,
-    Tegra::Engines::Maxwell3D::Regs::ShaderStage stage) {
-    std::lock_guard lock{mutex};
-
-    auto& gpu{Core::System::GetInstance().GPU()};
-    auto& memory_manager{gpu.MemoryManager()};
-    const auto cbufs{gpu.Maxwell3D().state.shader_stages[static_cast<std::size_t>(stage)]};
-    const auto addr{cbufs.const_buffers[global_region.GetCbufIndex()].address +
-                    global_region.GetCbufOffset()};
-    const auto actual_addr{memory_manager.Read<u64>(addr)};
-    const auto size{memory_manager.Read<u32>(addr + 8)};
-
-    // Look up global region in the cache based on address
-    const auto& host_ptr{memory_manager.GetPointer(actual_addr)};
-    GlobalRegion region{TryGet(host_ptr)};
-
-    if (!region) {
-        // No global region found - create a new one
-        region = GetUncachedGlobalRegion(actual_addr, host_ptr, size);
-        Register(region);
-    }
-
-    return region;
-}
-
-} // namespace OpenGL

src/video_core/renderer_opengl/gl_global_cache.h

@@ -1,82 +0,0 @@
-// Copyright 2018 yuzu Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#pragma once
-
-#include <memory>
-#include <unordered_map>
-
-#include <glad/glad.h>
-
-#include "common/assert.h"
-#include "common/common_types.h"
-#include "video_core/engines/maxwell_3d.h"
-#include "video_core/rasterizer_cache.h"
-#include "video_core/renderer_opengl/gl_resource_manager.h"
-
-namespace OpenGL {
-
-namespace GLShader {
-class GlobalMemoryEntry;
-}
-
-class RasterizerOpenGL;
-class CachedGlobalRegion;
-using GlobalRegion = std::shared_ptr<CachedGlobalRegion>;
-
-class CachedGlobalRegion final : public RasterizerCacheObject {
-public:
-    explicit CachedGlobalRegion(VAddr cpu_addr, u8* host_ptr, u32 size, u32 max_size);
-    ~CachedGlobalRegion();
-
-    VAddr GetCpuAddr() const override {
-        return cpu_addr;
-    }
-
-    std::size_t GetSizeInBytes() const override {
-        return size;
-    }
-
-    /// Gets the GL program handle for the buffer
-    GLuint GetBufferHandle() const {
-        return buffer.handle;
-    }
-
-    /// Reloads the global region from guest memory
-    void Reload(u32 size_);
-
-    void Flush();
-
-private:
-    VAddr cpu_addr{};
-    u8* host_ptr{};
-    u32 size{};
-    u32 max_size{};
-
-    OGLBuffer buffer;
-};
-
-class GlobalRegionCacheOpenGL final : public RasterizerCache<GlobalRegion> {
-public:
-    explicit GlobalRegionCacheOpenGL(RasterizerOpenGL& rasterizer);
-
-    /// Gets the current specified shader stage program
-    GlobalRegion GetGlobalRegion(const GLShader::GlobalMemoryEntry& descriptor,
-                                 Tegra::Engines::Maxwell3D::Regs::ShaderStage stage);
-
-protected:
-    void FlushObjectInner(const GlobalRegion& object) override {
-        object->Flush();
-    }
-
-private:
-    GlobalRegion TryGetReservedGlobalRegion(CacheAddr addr, u32 size) const;
-    GlobalRegion GetUncachedGlobalRegion(GPUVAddr addr, u8* host_ptr, u32 size);
-    void ReserveGlobalRegion(GlobalRegion region);
-
-    std::unordered_map<CacheAddr, GlobalRegion> reserve;
-    u32 max_ssbo_size{};
-};
-
-} // namespace OpenGL

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -4,6 +4,7 @@
 
 #include <algorithm>
 #include <array>
+#include <bitset>
 #include <memory>
 #include <string>
 #include <string_view>
@@ -19,7 +20,9 @@
 #include "core/core.h"
 #include "core/hle/kernel/process.h"
 #include "core/settings.h"
+#include "video_core/engines/kepler_compute.h"
 #include "video_core/engines/maxwell_3d.h"
+#include "video_core/memory_manager.h"
 #include "video_core/renderer_opengl/gl_rasterizer.h"
 #include "video_core/renderer_opengl/gl_shader_cache.h"
 #include "video_core/renderer_opengl/gl_shader_gen.h"
@@ -80,16 +83,31 @@ struct DrawParameters {
     }
 };
 
+static std::size_t GetConstBufferSize(const Tegra::Engines::ConstBufferInfo& buffer,
+                                      const GLShader::ConstBufferEntry& entry) {
+    if (!entry.IsIndirect()) {
+        return entry.GetSize();
+    }
+
+    if (buffer.size > Maxwell::MaxConstBufferSize) {
+        LOG_WARNING(Render_OpenGL, "Indirect constbuffer size {} exceeds maximum {}", buffer.size,
+                    Maxwell::MaxConstBufferSize);
+        return Maxwell::MaxConstBufferSize;
+    }
+
+    return buffer.size;
+}
+
 RasterizerOpenGL::RasterizerOpenGL(Core::System& system, Core::Frontend::EmuWindow& emu_window,
                                    ScreenInfo& info)
     : texture_cache{system, *this, device}, shader_cache{*this, system, emu_window, device},
-      global_cache{*this}, system{system}, screen_info{info},
-      buffer_cache(*this, STREAM_BUFFER_SIZE) {
+      system{system}, screen_info{info}, buffer_cache{*this, system, STREAM_BUFFER_SIZE} {
     OpenGLState::ApplyDefaultState();
 
     shader_program_manager = std::make_unique<GLShader::ProgramManager>();
     state.draw.shader_program = 0;
     state.Apply();
+    clear_framebuffer.Create();
 
     LOG_DEBUG(Render_OpenGL, "Sync fixed function OpenGL state here");
     CheckExtensions();
@@ -109,10 +127,10 @@ GLuint RasterizerOpenGL::SetupVertexFormat() {
     auto& gpu = system.GPU().Maxwell3D();
     const auto& regs = gpu.regs;
 
-    if (!gpu.dirty_flags.vertex_attrib_format) {
+    if (!gpu.dirty.vertex_attrib_format) {
         return state.draw.vertex_array;
     }
-    gpu.dirty_flags.vertex_attrib_format = false;
+    gpu.dirty.vertex_attrib_format = false;
 
     MICROPROFILE_SCOPE(OpenGL_VAO);
 
@@ -129,8 +147,6 @@ GLuint RasterizerOpenGL::SetupVertexFormat() {
         state.draw.vertex_array = vao;
         state.ApplyVertexArrayState();
 
-        glVertexArrayElementBuffer(vao, buffer_cache.GetHandle());
-
         // Use the vertex array as-is, assumes that the data is formatted correctly for OpenGL.
         // Enables the first 16 vertex attributes always, as we don't know which ones are actually
         // used until shader time. Note, Tegra technically supports 32, but we're capping this to 16
@@ -168,7 +184,7 @@ GLuint RasterizerOpenGL::SetupVertexFormat() {
     }
 
     // Rebinding the VAO invalidates the vertex buffer bindings.
-    gpu.dirty_flags.vertex_array.set();
+    gpu.dirty.ResetVertexArrays();
 
     state.draw.vertex_array = vao_entry.handle;
     return vao_entry.handle;
@@ -176,17 +192,20 @@ GLuint RasterizerOpenGL::SetupVertexFormat() {
 
 void RasterizerOpenGL::SetupVertexBuffer(GLuint vao) {
     auto& gpu = system.GPU().Maxwell3D();
-    const auto& regs = gpu.regs;
-
-    if (gpu.dirty_flags.vertex_array.none())
+    if (!gpu.dirty.vertex_array_buffers)
         return;
+    gpu.dirty.vertex_array_buffers = false;
+
+    const auto& regs = gpu.regs;
 
     MICROPROFILE_SCOPE(OpenGL_VB);
 
     // Upload all guest vertex arrays sequentially to our buffer
     for (u32 index = 0; index < Maxwell::NumVertexArrays; ++index) {
-        if (!gpu.dirty_flags.vertex_array[index])
+        if (!gpu.dirty.vertex_array[index])
             continue;
+        gpu.dirty.vertex_array[index] = false;
+        gpu.dirty.vertex_instance[index] = false;
 
         const auto& vertex_array = regs.vertex_array[index];
         if (!vertex_array.IsEnabled())
@@ -197,11 +216,11 @@ void RasterizerOpenGL::SetupVertexBuffer(GLuint vao) {
 
         ASSERT(end > start);
         const u64 size = end - start + 1;
-        const GLintptr vertex_buffer_offset = buffer_cache.UploadMemory(start, size);
+        const auto [vertex_buffer, vertex_buffer_offset] = buffer_cache.UploadMemory(start, size);
 
         // Bind the vertex array to the buffer at the current offset.
-        glVertexArrayVertexBuffer(vao, index, buffer_cache.GetHandle(), vertex_buffer_offset,
-                                  vertex_array.stride);
+        vertex_array_pushbuffer.SetVertexBuffer(index, vertex_buffer, vertex_buffer_offset,
+                                                vertex_array.stride);
 
         if (regs.instanced_arrays.IsInstancingEnabled(index) && vertex_array.divisor != 0) {
             // Enable vertex buffer instancing with the specified divisor.
@@ -211,11 +230,47 @@ void RasterizerOpenGL::SetupVertexBuffer(GLuint vao) {
             glVertexArrayBindingDivisor(vao, index, 0);
         }
     }
-
-    gpu.dirty_flags.vertex_array.reset();
 }
 
-DrawParameters RasterizerOpenGL::SetupDraw() {
+void RasterizerOpenGL::SetupVertexInstances(GLuint vao) {
+    auto& gpu = system.GPU().Maxwell3D();
+
+    if (!gpu.dirty.vertex_instances)
+        return;
+    gpu.dirty.vertex_instances = false;
+
+    const auto& regs = gpu.regs;
+    // Upload all guest vertex arrays sequentially to our buffer
+    for (u32 index = 0; index < Maxwell::NumVertexArrays; ++index) {
+        if (!gpu.dirty.vertex_instance[index])
+            continue;
+
+        gpu.dirty.vertex_instance[index] = false;
+
+        if (regs.instanced_arrays.IsInstancingEnabled(index) &&
+            regs.vertex_array[index].divisor != 0) {
+            // Enable vertex buffer instancing with the specified divisor.
+            glVertexArrayBindingDivisor(vao, index, regs.vertex_array[index].divisor);
+        } else {
+            // Disable the vertex buffer instancing.
+            glVertexArrayBindingDivisor(vao, index, 0);
+        }
+    }
+}
+
+GLintptr RasterizerOpenGL::SetupIndexBuffer() {
+    if (accelerate_draw != AccelDraw::Indexed) {
+        return 0;
+    }
+    MICROPROFILE_SCOPE(OpenGL_Index);
+    const auto& regs = system.GPU().Maxwell3D().regs;
+    const std::size_t size = CalculateIndexBufferSize();
+    const auto [buffer, offset] = buffer_cache.UploadMemory(regs.index_array.IndexStart(), size);
+    vertex_array_pushbuffer.SetIndexBuffer(buffer);
+    return offset;
+}
+
+DrawParameters RasterizerOpenGL::SetupDraw(GLintptr index_buffer_offset) {
     const auto& gpu = system.GPU().Maxwell3D();
     const auto& regs = gpu.regs;
     const bool is_indexed = accelerate_draw == AccelDraw::Indexed;
@@ -227,11 +282,9 @@ DrawParameters RasterizerOpenGL::SetupDraw() {
     params.primitive_mode = MaxwellToGL::PrimitiveTopology(regs.draw.topology);
 
     if (is_indexed) {
-        MICROPROFILE_SCOPE(OpenGL_Index);
         params.index_format = MaxwellToGL::IndexFormat(regs.index_array.format);
         params.count = regs.index_array.count;
-        params.index_buffer_offset =
-            buffer_cache.UploadMemory(regs.index_array.IndexStart(), CalculateIndexBufferSize());
+        params.index_buffer_offset = index_buffer_offset;
         params.base_vertex = static_cast<GLint>(regs.vb_element_base);
     } else {
         params.count = regs.vertex_buffer.count;
@@ -247,10 +300,6 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) {
     BaseBindings base_bindings;
     std::array<bool, Maxwell::NumClipDistances> clip_distances{};
 
-    // Prepare packed bindings
-    bind_ubo_pushbuffer.Setup(base_bindings.cbuf);
-    bind_ssbo_pushbuffer.Setup(base_bindings.gmem);
-
     for (std::size_t index = 0; index < Maxwell::MaxShaderProgram; ++index) {
         const auto& shader_config = gpu.regs.shader_config[index];
         const Maxwell::ShaderProgram program{static_cast<Maxwell::ShaderProgram>(index)};
@@ -271,18 +320,17 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) {
 
         GLShader::MaxwellUniformData ubo{};
         ubo.SetFromRegs(gpu, stage);
-        const GLintptr offset =
+        const auto [buffer, offset] =
             buffer_cache.UploadHostMemory(&ubo, sizeof(ubo), device.GetUniformBufferAlignment());
 
         // Bind the emulation info buffer
-        bind_ubo_pushbuffer.Push(buffer_cache.GetHandle(), offset,
-                                 static_cast<GLsizeiptr>(sizeof(ubo)));
+        bind_ubo_pushbuffer.Push(buffer, offset, static_cast<GLsizeiptr>(sizeof(ubo)));
 
         Shader shader{shader_cache.GetStageProgram(program)};
 
-        const auto stage_enum{static_cast<Maxwell::ShaderStage>(stage)};
+        const auto stage_enum = static_cast<Maxwell::ShaderStage>(stage);
         SetupDrawConstBuffers(stage_enum, shader);
-        SetupGlobalRegions(stage_enum, shader);
+        SetupDrawGlobalMemory(stage_enum, shader);
         const auto texture_buffer_usage{SetupTextures(stage_enum, shader, base_bindings)};
 
         const ProgramVariant variant{base_bindings, primitive_mode, texture_buffer_usage};
@@ -321,12 +369,9 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) {
         base_bindings = next_bindings;
     }
 
-    bind_ubo_pushbuffer.Bind();
-    bind_ssbo_pushbuffer.Bind();
-
     SyncClipEnabled(clip_distances);
 
-    gpu.dirty_flags.shaders = false;
+    gpu.dirty.shaders = false;
 }
 
 std::size_t RasterizerOpenGL::CalculateVertexArraysSize() const {
@@ -409,13 +454,13 @@ std::pair<bool, bool> RasterizerOpenGL::ConfigureFramebuffers(
 
     const FramebufferConfigState fb_config_state{using_color_fb, using_depth_fb, preserve_contents,
                                                  single_color_target};
-    if (fb_config_state == current_framebuffer_config_state &&
-        gpu.dirty_flags.color_buffer.none() && !gpu.dirty_flags.zeta_buffer) {
+    if (fb_config_state == current_framebuffer_config_state && !gpu.dirty.render_settings) {
         // Only skip if the previous ConfigureFramebuffers call was from the same kind (multiple or
         // single color targets). This is done because the guest registers may not change but the
         // host framebuffer may contain different attachments
         return current_depth_stencil_usage;
     }
+    gpu.dirty.render_settings = false;
     current_framebuffer_config_state = fb_config_state;
 
     texture_cache.GuardRenderTargets(true);
@@ -504,13 +549,71 @@ std::pair<bool, bool> RasterizerOpenGL::ConfigureFramebuffers(
     return current_depth_stencil_usage = {static_cast<bool>(depth_surface), fbkey.stencil_enable};
 }
 
+void RasterizerOpenGL::ConfigureClearFramebuffer(OpenGLState& current_state, bool using_color_fb,
+                                                 bool using_depth_fb, bool using_stencil_fb) {
+    auto& gpu = system.GPU().Maxwell3D();
+    const auto& regs = gpu.regs;
+
+    texture_cache.GuardRenderTargets(true);
+    View color_surface{};
+    if (using_color_fb) {
+        color_surface = texture_cache.GetColorBufferSurface(regs.clear_buffers.RT, false);
+    }
+    View depth_surface{};
+    if (using_depth_fb || using_stencil_fb) {
+        depth_surface = texture_cache.GetDepthBufferSurface(false);
+    }
+    texture_cache.GuardRenderTargets(false);
+
+    current_state.draw.draw_framebuffer = clear_framebuffer.handle;
+    current_state.ApplyFramebufferState();
+
+    if (color_surface) {
+        color_surface->Attach(GL_COLOR_ATTACHMENT0, GL_DRAW_FRAMEBUFFER);
+    } else {
+        glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0);
+    }
+
+    if (depth_surface) {
+        const auto& params = depth_surface->GetSurfaceParams();
+        switch (params.type) {
+        case VideoCore::Surface::SurfaceType::Depth: {
+            depth_surface->Attach(GL_DEPTH_ATTACHMENT, GL_DRAW_FRAMEBUFFER);
+            glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, 0);
+            break;
+        }
+        case VideoCore::Surface::SurfaceType::DepthStencil: {
+            depth_surface->Attach(GL_DEPTH_ATTACHMENT, GL_DRAW_FRAMEBUFFER);
+            break;
+        }
+        default: { UNIMPLEMENTED(); }
+        }
+    } else {
+        glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0,
+                               0);
+    }
+}
+
 void RasterizerOpenGL::Clear() {
-    const auto& regs = system.GPU().Maxwell3D().regs;
+    const auto& maxwell3d = system.GPU().Maxwell3D();
+
+    if (!maxwell3d.ShouldExecute()) {
+        return;
+    }
+
+    const auto& regs = maxwell3d.regs;
     bool use_color{};
     bool use_depth{};
     bool use_stencil{};
 
-    OpenGLState clear_state;
+    OpenGLState prev_state{OpenGLState::GetCurState()};
+    SCOPE_EXIT({
+        prev_state.AllDirty();
+        prev_state.Apply();
+    });
+
+    OpenGLState clear_state{OpenGLState::GetCurState()};
+    clear_state.SetDefaultViewports();
     if (regs.clear_buffers.R || regs.clear_buffers.G || regs.clear_buffers.B ||
         regs.clear_buffers.A) {
         use_color = true;
@@ -530,6 +633,7 @@ void RasterizerOpenGL::Clear() {
         // true.
         clear_state.depth.test_enabled = true;
         clear_state.depth.test_func = GL_ALWAYS;
+        clear_state.depth.write_mask = GL_TRUE;
     }
     if (regs.clear_buffers.S) {
         ASSERT_MSG(regs.zeta_enable != 0, "Tried to clear stencil but buffer is not enabled!");
@@ -566,8 +670,9 @@ void RasterizerOpenGL::Clear() {
         return;
     }
 
-    const auto [clear_depth, clear_stencil] = ConfigureFramebuffers(
-        clear_state, use_color, use_depth || use_stencil, false, regs.clear_buffers.RT.Value());
+    ConfigureClearFramebuffer(clear_state, use_color, use_depth, use_stencil);
+
+    SyncViewport(clear_state);
     if (regs.clear_flags.scissor) {
         SyncScissorTest(clear_state);
     }
@@ -576,21 +681,18 @@ void RasterizerOpenGL::Clear() {
         clear_state.EmulateViewportWithScissor();
     }
 
-    clear_state.ApplyColorMask();
-    clear_state.ApplyDepth();
-    clear_state.ApplyStencilTest();
-    clear_state.ApplyViewport();
-    clear_state.ApplyFramebufferState();
+    clear_state.AllDirty();
+    clear_state.Apply();
 
     if (use_color) {
-        glClearBufferfv(GL_COLOR, regs.clear_buffers.RT, regs.clear_color);
+        glClearBufferfv(GL_COLOR, 0, regs.clear_color);
     }
 
-    if (clear_depth && clear_stencil) {
+    if (use_depth && use_stencil) {
         glClearBufferfi(GL_DEPTH_STENCIL, 0, regs.clear_depth, regs.clear_stencil);
-    } else if (clear_depth) {
+    } else if (use_depth) {
         glClearBufferfv(GL_DEPTH, 0, &regs.clear_depth);
-    } else if (clear_stencil) {
+    } else if (use_stencil) {
         glClearBufferiv(GL_STENCIL, 0, &regs.clear_stencil);
     }
 }
@@ -601,6 +703,11 @@ void RasterizerOpenGL::DrawArrays() {
 
     MICROPROFILE_SCOPE(OpenGL_Drawing);
     auto& gpu = system.GPU().Maxwell3D();
+
+    if (!gpu.ShouldExecute()) {
+        return;
+    }
+
     const auto& regs = gpu.regs;
 
     SyncColorMask();
@@ -634,26 +741,47 @@ void RasterizerOpenGL::DrawArrays() {
                       Maxwell::MaxShaderStage;
 
     // Add space for at least 18 constant buffers
-    buffer_size +=
-        Maxwell::MaxConstBuffers * (MaxConstbufferSize + device.GetUniformBufferAlignment());
+    buffer_size += Maxwell::MaxConstBuffers *
+                   (Maxwell::MaxConstBufferSize + device.GetUniformBufferAlignment());
 
-    const bool invalidate = buffer_cache.Map(buffer_size);
-    if (invalidate) {
-        // As all cached buffers are invalidated, we need to recheck their state.
-        gpu.dirty_flags.vertex_array.set();
-    }
+    // Prepare the vertex array.
+    buffer_cache.Map(buffer_size);
 
+    // Prepare vertex array format.
     const GLuint vao = SetupVertexFormat();
-    SetupVertexBuffer(vao);
+    vertex_array_pushbuffer.Setup(vao);
 
-    DrawParameters params = SetupDraw();
+    // Upload vertex and index data.
+    SetupVertexBuffer(vao);
+    SetupVertexInstances(vao);
+    const GLintptr index_buffer_offset = SetupIndexBuffer();
+
+    // Setup draw parameters. It will automatically choose what glDraw* method to use.
+    const DrawParameters params = SetupDraw(index_buffer_offset);
+
+    // Prepare packed bindings.
+    bind_ubo_pushbuffer.Setup(0);
+    bind_ssbo_pushbuffer.Setup(0);
+
+    // Setup shaders and their used resources.
     texture_cache.GuardSamplers(true);
     SetupShaders(params.primitive_mode);
     texture_cache.GuardSamplers(false);
 
     ConfigureFramebuffers(state);
 
-    buffer_cache.Unmap();
+    // Signal the buffer cache that we are not going to upload more things.
+    const bool invalidate = buffer_cache.Unmap();
+
+    // Now that we are no longer uploading data, we can safely bind the buffers to OpenGL.
+    vertex_array_pushbuffer.Bind();
+    bind_ubo_pushbuffer.Bind();
+    bind_ssbo_pushbuffer.Bind();
+
+    if (invalidate) {
+        // As all cached buffers are invalidated, we need to recheck their state.
+        gpu.dirty.ResetVertexArrays();
+    }
 
     shader_program_manager->ApplyTo(state);
     state.Apply();
@@ -665,6 +793,46 @@ void RasterizerOpenGL::DrawArrays() {
     params.DispatchDraw();
 
     accelerate_draw = AccelDraw::Disabled;
+    gpu.dirty.memory_general = false;
+}
+
+void RasterizerOpenGL::DispatchCompute(GPUVAddr code_addr) {
+    if (!GLAD_GL_ARB_compute_variable_group_size) {
+        LOG_ERROR(Render_OpenGL, "Compute is currently not supported on this device due to the "
+                                 "lack of GL_ARB_compute_variable_group_size");
+        return;
+    }
+
+    auto kernel = shader_cache.GetComputeKernel(code_addr);
+    const auto [program, next_bindings] = kernel->GetProgramHandle({});
+    state.draw.shader_program = program;
+    state.draw.program_pipeline = 0;
+
+    const std::size_t buffer_size =
+        Tegra::Engines::KeplerCompute::NumConstBuffers *
+        (Maxwell::MaxConstBufferSize + device.GetUniformBufferAlignment());
+    buffer_cache.Map(buffer_size);
+
+    bind_ubo_pushbuffer.Setup(0);
+    bind_ssbo_pushbuffer.Setup(0);
+
+    SetupComputeConstBuffers(kernel);
+    SetupComputeGlobalMemory(kernel);
+
+    // TODO(Rodrigo): Bind images and samplers
+
+    buffer_cache.Unmap();
+
+    bind_ubo_pushbuffer.Bind();
+    bind_ssbo_pushbuffer.Bind();
+
+    state.ApplyShaderProgram();
+    state.ApplyProgramPipeline();
+
+    const auto& launch_desc = system.GPU().KeplerCompute().launch_description;
+    glDispatchComputeGroupSizeARB(launch_desc.grid_dim_x, launch_desc.grid_dim_y,
+                                  launch_desc.grid_dim_z, launch_desc.block_dim_x,
+                                  launch_desc.block_dim_y, launch_desc.block_dim_z);
 }
 
 void RasterizerOpenGL::FlushAll() {}
@@ -675,7 +843,7 @@ void RasterizerOpenGL::FlushRegion(CacheAddr addr, u64 size) {
         return;
     }
     texture_cache.FlushRegion(addr, size);
-    global_cache.FlushRegion(addr, size);
+    buffer_cache.FlushRegion(addr, size);
 }
 
 void RasterizerOpenGL::InvalidateRegion(CacheAddr addr, u64 size) {
@@ -685,7 +853,6 @@ void RasterizerOpenGL::InvalidateRegion(CacheAddr addr, u64 size) {
     }
     texture_cache.InvalidateRegion(addr, size);
     shader_cache.InvalidateRegion(addr, size);
-    global_cache.InvalidateRegion(addr, size);
     buffer_cache.InvalidateRegion(addr, size);
 }
 
@@ -696,6 +863,10 @@ void RasterizerOpenGL::FlushAndInvalidateRegion(CacheAddr addr, u64 size) {
     InvalidateRegion(addr, size);
 }
 
+void RasterizerOpenGL::TickFrame() {
+    buffer_cache.TickFrame();
+}
+
 bool RasterizerOpenGL::AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Regs::Surface& src,
                                              const Tegra::Engines::Fermi2D::Regs::Surface& dst,
                                              const Tegra::Engines::Fermi2D::Config& copy_config) {
@@ -737,14 +908,25 @@ bool RasterizerOpenGL::AccelerateDisplay(const Tegra::FramebufferConfig& config,
 void RasterizerOpenGL::SetupDrawConstBuffers(Tegra::Engines::Maxwell3D::Regs::ShaderStage stage,
                                              const Shader& shader) {
     MICROPROFILE_SCOPE(OpenGL_UBO);
-    const auto stage_index = static_cast<std::size_t>(stage);
-    const auto& shader_stage = system.GPU().Maxwell3D().state.shader_stages[stage_index];
-    const auto& entries = shader->GetShaderEntries().const_buffers;
+    const auto& stages = system.GPU().Maxwell3D().state.shader_stages;
+    const auto& shader_stage = stages[static_cast<std::size_t>(stage)];
+    for (const auto& entry : shader->GetShaderEntries().const_buffers) {
+        const auto& buffer = shader_stage.const_buffers[entry.GetIndex()];
+        SetupConstBuffer(buffer, entry);
+    }
+}
 
-    // Upload only the enabled buffers from the 16 constbuffers of each shader stage
-    for (u32 bindpoint = 0; bindpoint < entries.size(); ++bindpoint) {
-        const auto& entry = entries[bindpoint];
-        SetupConstBuffer(shader_stage.const_buffers[entry.GetIndex()], entry);
+void RasterizerOpenGL::SetupComputeConstBuffers(const Shader& kernel) {
+    MICROPROFILE_SCOPE(OpenGL_UBO);
+    const auto& launch_desc = system.GPU().KeplerCompute().launch_description;
+    for (const auto& entry : kernel->GetShaderEntries().const_buffers) {
+        const auto& config = launch_desc.const_buffer_config[entry.GetIndex()];
+        const std::bitset<8> mask = launch_desc.memory_config.const_buffer_enable_mask.Value();
+        Tegra::Engines::ConstBufferInfo buffer;
+        buffer.address = config.Address();
+        buffer.size = config.size;
+        buffer.enabled = mask[entry.GetIndex()];
+        SetupConstBuffer(buffer, entry);
     }
 }
 
@@ -752,49 +934,52 @@ void RasterizerOpenGL::SetupConstBuffer(const Tegra::Engines::ConstBufferInfo& b
                                         const GLShader::ConstBufferEntry& entry) {
     if (!buffer.enabled) {
         // Set values to zero to unbind buffers
-        bind_ubo_pushbuffer.Push(0, 0, 0);
+        bind_ubo_pushbuffer.Push(buffer_cache.GetEmptyBuffer(sizeof(float)), 0, sizeof(float));
         return;
     }
 
-    std::size_t size;
-    if (entry.IsIndirect()) {
-        // Buffer is accessed indirectly, so upload the entire thing
-        size = buffer.size;
-
-        if (size > MaxConstbufferSize) {
-            LOG_WARNING(Render_OpenGL, "Indirect constbuffer size {} exceeds maximum {}", size,
-                        MaxConstbufferSize);
-            size = MaxConstbufferSize;
-        }
-    } else {
-        // Buffer is accessed directly, upload just what we use
-        size = entry.GetSize();
-    }
-
     // Align the actual size so it ends up being a multiple of vec4 to meet the OpenGL std140
     // UBO alignment requirements.
-    size = Common::AlignUp(size, sizeof(GLvec4));
-    ASSERT_MSG(size <= MaxConstbufferSize, "Constant buffer is too big");
+    const std::size_t size = Common::AlignUp(GetConstBufferSize(buffer, entry), sizeof(GLvec4));
 
-    const std::size_t alignment = device.GetUniformBufferAlignment();
-    const GLintptr offset = buffer_cache.UploadMemory(buffer.address, size, alignment);
-    bind_ubo_pushbuffer.Push(buffer_cache.GetHandle(), offset, size);
+    const auto alignment = device.GetUniformBufferAlignment();
+    const auto [cbuf, offset] = buffer_cache.UploadMemory(buffer.address, size, alignment);
+    bind_ubo_pushbuffer.Push(cbuf, offset, size);
 }
 
-void RasterizerOpenGL::SetupGlobalRegions(Tegra::Engines::Maxwell3D::Regs::ShaderStage stage,
-                                          const Shader& shader) {
-    const auto& entries = shader->GetShaderEntries().global_memory_entries;
-    for (std::size_t bindpoint = 0; bindpoint < entries.size(); ++bindpoint) {
-        const auto& entry{entries[bindpoint]};
-        const auto& region{global_cache.GetGlobalRegion(entry, stage)};
-        if (entry.IsWritten()) {
-            region->MarkAsModified(true, global_cache);
-        }
-        bind_ssbo_pushbuffer.Push(region->GetBufferHandle(), 0,
-                                  static_cast<GLsizeiptr>(region->GetSizeInBytes()));
+void RasterizerOpenGL::SetupDrawGlobalMemory(Tegra::Engines::Maxwell3D::Regs::ShaderStage stage,
+                                             const Shader& shader) {
+    auto& gpu{system.GPU()};
+    auto& memory_manager{gpu.MemoryManager()};
+    const auto cbufs{gpu.Maxwell3D().state.shader_stages[static_cast<std::size_t>(stage)]};
+    for (const auto& entry : shader->GetShaderEntries().global_memory_entries) {
+        const auto addr{cbufs.const_buffers[entry.GetCbufIndex()].address + entry.GetCbufOffset()};
+        const auto gpu_addr{memory_manager.Read<u64>(addr)};
+        const auto size{memory_manager.Read<u32>(addr + 8)};
+        SetupGlobalMemory(entry, gpu_addr, size);
     }
 }
 
+void RasterizerOpenGL::SetupComputeGlobalMemory(const Shader& kernel) {
+    auto& gpu{system.GPU()};
+    auto& memory_manager{gpu.MemoryManager()};
+    const auto cbufs{gpu.KeplerCompute().launch_description.const_buffer_config};
+    for (const auto& entry : kernel->GetShaderEntries().global_memory_entries) {
+        const auto addr{cbufs[entry.GetCbufIndex()].Address() + entry.GetCbufOffset()};
+        const auto gpu_addr{memory_manager.Read<u64>(addr)};
+        const auto size{memory_manager.Read<u32>(addr + 8)};
+        SetupGlobalMemory(entry, gpu_addr, size);
+    }
+}
+
+void RasterizerOpenGL::SetupGlobalMemory(const GLShader::GlobalMemoryEntry& entry,
+                                         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());
+    bind_ssbo_pushbuffer.Push(ssbo, buffer_offset, static_cast<GLsizeiptr>(size));
+}
+
 TextureBufferUsage RasterizerOpenGL::SetupTextures(Maxwell::ShaderStage stage, const Shader& shader,
                                                    BaseBindings base_bindings) {
     MICROPROFILE_SCOPE(OpenGL_Texture);
@@ -883,10 +1068,11 @@ void RasterizerOpenGL::SyncClipCoef() {
 }
 
 void RasterizerOpenGL::SyncCullMode() {
-    const auto& regs = system.GPU().Maxwell3D().regs;
+    auto& maxwell3d = system.GPU().Maxwell3D();
+
+    const auto& regs = maxwell3d.regs;
 
     state.cull.enabled = regs.cull.enabled != 0;
-
     if (state.cull.enabled) {
         state.cull.front_face = MaxwellToGL::FrontFace(regs.cull.front_face);
         state.cull.mode = MaxwellToGL::CullFace(regs.cull.cull_face);
@@ -919,16 +1105,21 @@ void RasterizerOpenGL::SyncDepthTestState() {
     state.depth.test_enabled = regs.depth_test_enable != 0;
     state.depth.write_mask = regs.depth_write_enabled ? GL_TRUE : GL_FALSE;
 
-    if (!state.depth.test_enabled)
+    if (!state.depth.test_enabled) {
         return;
+    }
 
     state.depth.test_func = MaxwellToGL::ComparisonOp(regs.depth_test_func);
 }
 
 void RasterizerOpenGL::SyncStencilTestState() {
-    const auto& regs = system.GPU().Maxwell3D().regs;
-    state.stencil.test_enabled = regs.stencil_enable != 0;
+    auto& maxwell3d = system.GPU().Maxwell3D();
+    if (!maxwell3d.dirty.stencil_test) {
+        return;
+    }
+    const auto& regs = maxwell3d.regs;
 
+    state.stencil.test_enabled = regs.stencil_enable != 0;
     if (!regs.stencil_enable) {
         return;
     }
@@ -957,10 +1148,17 @@ void RasterizerOpenGL::SyncStencilTestState() {
         state.stencil.back.action_depth_fail = GL_KEEP;
         state.stencil.back.action_depth_pass = GL_KEEP;
     }
+    state.MarkDirtyStencilState();
+    maxwell3d.dirty.stencil_test = false;
 }
 
 void RasterizerOpenGL::SyncColorMask() {
-    const auto& regs = system.GPU().Maxwell3D().regs;
+    auto& maxwell3d = system.GPU().Maxwell3D();
+    if (!maxwell3d.dirty.color_mask) {
+        return;
+    }
+    const auto& regs = maxwell3d.regs;
+
     const std::size_t count =
         regs.independent_blend_enable ? Tegra::Engines::Maxwell3D::Regs::NumRenderTargets : 1;
     for (std::size_t i = 0; i < count; i++) {
@@ -971,6 +1169,9 @@ void RasterizerOpenGL::SyncColorMask() {
         dest.blue_enabled = (source.B == 0) ? GL_FALSE : GL_TRUE;
         dest.alpha_enabled = (source.A == 0) ? GL_FALSE : GL_TRUE;
     }
+
+    state.MarkDirtyColorMask();
+    maxwell3d.dirty.color_mask = false;
 }
 
 void RasterizerOpenGL::SyncMultiSampleState() {
@@ -985,7 +1186,11 @@ void RasterizerOpenGL::SyncFragmentColorClampState() {
 }
 
 void RasterizerOpenGL::SyncBlendState() {
-    const auto& regs = system.GPU().Maxwell3D().regs;
+    auto& maxwell3d = system.GPU().Maxwell3D();
+    if (!maxwell3d.dirty.blend_state) {
+        return;
+    }
+    const auto& regs = maxwell3d.regs;
 
     state.blend_color.red = regs.blend_color.r;
     state.blend_color.green = regs.blend_color.g;
@@ -1008,6 +1213,8 @@ void RasterizerOpenGL::SyncBlendState() {
         for (std::size_t i = 1; i < Tegra::Engines::Maxwell3D::Regs::NumRenderTargets; i++) {
             state.blend[i].enabled = false;
         }
+        maxwell3d.dirty.blend_state = false;
+        state.MarkDirtyBlendState();
         return;
     }
 
@@ -1024,6 +1231,9 @@ void RasterizerOpenGL::SyncBlendState() {
         blend.src_a_func = MaxwellToGL::BlendFunc(src.factor_source_a);
         blend.dst_a_func = MaxwellToGL::BlendFunc(src.factor_dest_a);
     }
+
+    state.MarkDirtyBlendState();
+    maxwell3d.dirty.blend_state = false;
 }
 
 void RasterizerOpenGL::SyncLogicOpState() {
@@ -1075,13 +1285,21 @@ void RasterizerOpenGL::SyncPointState() {
 }
 
 void RasterizerOpenGL::SyncPolygonOffset() {
-    const auto& regs = system.GPU().Maxwell3D().regs;
+    auto& maxwell3d = system.GPU().Maxwell3D();
+    if (!maxwell3d.dirty.polygon_offset) {
+        return;
+    }
+    const auto& regs = maxwell3d.regs;
+
     state.polygon_offset.fill_enable = regs.polygon_offset_fill_enable != 0;
     state.polygon_offset.line_enable = regs.polygon_offset_line_enable != 0;
     state.polygon_offset.point_enable = regs.polygon_offset_point_enable != 0;
     state.polygon_offset.units = regs.polygon_offset_units;
     state.polygon_offset.factor = regs.polygon_offset_factor;
     state.polygon_offset.clamp = regs.polygon_offset_clamp;
+
+    state.MarkDirtyPolygonOffset();
+    maxwell3d.dirty.polygon_offset = false;
 }
 
 void RasterizerOpenGL::SyncAlphaTest() {

src/video_core/renderer_opengl/gl_rasterizer.h

@@ -24,7 +24,6 @@
 #include "video_core/renderer_opengl/gl_buffer_cache.h"
 #include "video_core/renderer_opengl/gl_device.h"
 #include "video_core/renderer_opengl/gl_framebuffer_cache.h"
-#include "video_core/renderer_opengl/gl_global_cache.h"
 #include "video_core/renderer_opengl/gl_resource_manager.h"
 #include "video_core/renderer_opengl/gl_sampler_cache.h"
 #include "video_core/renderer_opengl/gl_shader_cache.h"
@@ -59,10 +58,12 @@ public:
 
     void DrawArrays() override;
     void Clear() override;
+    void DispatchCompute(GPUVAddr code_addr) override;
     void FlushAll() override;
     void FlushRegion(CacheAddr addr, u64 size) override;
     void InvalidateRegion(CacheAddr addr, u64 size) override;
     void FlushAndInvalidateRegion(CacheAddr addr, u64 size) override;
+    void TickFrame() override;
     bool AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Regs::Surface& src,
                                const Tegra::Engines::Fermi2D::Regs::Surface& dst,
                                const Tegra::Engines::Fermi2D::Config& copy_config) override;
@@ -73,11 +74,6 @@ public:
     void LoadDiskResources(const std::atomic_bool& stop_loading,
                            const VideoCore::DiskResourceLoadCallback& callback) override;
 
-    /// Maximum supported size that a constbuffer can have in bytes.
-    static constexpr std::size_t MaxConstbufferSize = 0x10000;
-    static_assert(MaxConstbufferSize % sizeof(GLvec4) == 0,
-                  "The maximum size of a constbuffer must be a multiple of the size of GLvec4");
-
 private:
     struct FramebufferConfigState {
         bool using_color_fb{};
@@ -113,17 +109,30 @@ private:
         OpenGLState& current_state, bool using_color_fb = true, bool using_depth_fb = true,
         bool preserve_contents = true, std::optional<std::size_t> single_color_target = {});
 
+    void ConfigureClearFramebuffer(OpenGLState& current_state, bool using_color_fb,
+                                   bool using_depth_fb, bool using_stencil_fb);
+
     /// Configures the current constbuffers to use for the draw command.
     void SetupDrawConstBuffers(Tegra::Engines::Maxwell3D::Regs::ShaderStage stage,
                                const Shader& shader);
 
+    /// Configures the current constbuffers to use for the kernel invocation.
+    void SetupComputeConstBuffers(const Shader& kernel);
+
     /// Configures a constant buffer.
     void SetupConstBuffer(const Tegra::Engines::ConstBufferInfo& buffer,
                           const GLShader::ConstBufferEntry& entry);
 
     /// Configures the current global memory entries to use for the draw command.
-    void SetupGlobalRegions(Tegra::Engines::Maxwell3D::Regs::ShaderStage stage,
-                            const Shader& shader);
+    void SetupDrawGlobalMemory(Tegra::Engines::Maxwell3D::Regs::ShaderStage stage,
+                               const Shader& shader);
+
+    /// Configures the current global memory entries to use for the kernel invocation.
+    void SetupComputeGlobalMemory(const Shader& kernel);
+
+    /// Configures a constant buffer.
+    void SetupGlobalMemory(const GLShader::GlobalMemoryEntry& entry, GPUVAddr gpu_addr,
+                           std::size_t size);
 
     /// Configures the current textures to use for the draw command. Returns shaders texture buffer
     /// usage.
@@ -191,7 +200,6 @@ private:
 
     TextureCacheOpenGL texture_cache;
     ShaderCacheOpenGL shader_cache;
-    GlobalRegionCacheOpenGL global_cache;
     SamplerCacheOpenGL sampler_cache;
     FramebufferCacheOpenGL framebuffer_cache;
 
@@ -210,6 +218,7 @@ private:
     static constexpr std::size_t STREAM_BUFFER_SIZE = 128 * 1024 * 1024;
     OGLBufferCache buffer_cache;
 
+    VertexArrayPushBuffer vertex_array_pushbuffer;
     BindBuffersRangePushBuffer bind_ubo_pushbuffer{GL_UNIFORM_BUFFER};
     BindBuffersRangePushBuffer bind_ssbo_pushbuffer{GL_SHADER_STORAGE_BUFFER};
 
@@ -221,14 +230,19 @@ private:
     GLuint SetupVertexFormat();
 
     void SetupVertexBuffer(GLuint vao);
+    void SetupVertexInstances(GLuint vao);
 
-    DrawParameters SetupDraw();
+    GLintptr SetupIndexBuffer();
+
+    DrawParameters SetupDraw(GLintptr index_buffer_offset);
 
     void SetupShaders(GLenum primitive_mode);
 
     enum class AccelDraw { Disabled, Arrays, Indexed };
     AccelDraw accelerate_draw = AccelDraw::Disabled;
 
+    OGLFramebuffer clear_framebuffer;
+
     using CachedPageMap = boost::icl::interval_map<u64, int>;
     CachedPageMap cached_pages;
 };

src/video_core/renderer_opengl/gl_shader_cache.cpp

@@ -23,13 +23,13 @@ namespace OpenGL {
 
 using VideoCommon::Shader::ProgramCode;
 
-// One UBO is always reserved for emulation values
-constexpr u32 RESERVED_UBOS = 1;
+// One UBO is always reserved for emulation values on staged shaders
+constexpr u32 STAGE_RESERVED_UBOS = 1;
 
 struct UnspecializedShader {
     std::string code;
     GLShader::ShaderEntries entries;
-    Maxwell::ShaderProgram program_type;
+    ProgramType program_type;
 };
 
 namespace {
@@ -55,15 +55,17 @@ ProgramCode GetShaderCode(Tegra::MemoryManager& memory_manager, const GPUVAddr g
 }
 
 /// Gets the shader type from a Maxwell program type
-constexpr GLenum GetShaderType(Maxwell::ShaderProgram program_type) {
+constexpr GLenum GetShaderType(ProgramType program_type) {
     switch (program_type) {
-    case Maxwell::ShaderProgram::VertexA:
-    case Maxwell::ShaderProgram::VertexB:
+    case ProgramType::VertexA:
+    case ProgramType::VertexB:
         return GL_VERTEX_SHADER;
-    case Maxwell::ShaderProgram::Geometry:
+    case ProgramType::Geometry:
         return GL_GEOMETRY_SHADER;
-    case Maxwell::ShaderProgram::Fragment:
+    case ProgramType::Fragment:
         return GL_FRAGMENT_SHADER;
+    case ProgramType::Compute:
+        return GL_COMPUTE_SHADER;
     default:
         return GL_NONE;
     }
@@ -100,6 +102,25 @@ constexpr std::tuple<const char*, const char*, u32> GetPrimitiveDescription(GLen
     }
 }
 
+ProgramType GetProgramType(Maxwell::ShaderProgram program) {
+    switch (program) {
+    case Maxwell::ShaderProgram::VertexA:
+        return ProgramType::VertexA;
+    case Maxwell::ShaderProgram::VertexB:
+        return ProgramType::VertexB;
+    case Maxwell::ShaderProgram::TesselationControl:
+        return ProgramType::TessellationControl;
+    case Maxwell::ShaderProgram::TesselationEval:
+        return ProgramType::TessellationEval;
+    case Maxwell::ShaderProgram::Geometry:
+        return ProgramType::Geometry;
+    case Maxwell::ShaderProgram::Fragment:
+        return ProgramType::Fragment;
+    }
+    UNREACHABLE();
+    return {};
+}
+
 /// Calculates the size of a program stream
 std::size_t CalculateProgramSize(const GLShader::ProgramCode& program) {
     constexpr std::size_t start_offset = 10;
@@ -128,13 +149,13 @@ std::size_t CalculateProgramSize(const GLShader::ProgramCode& program) {
 }
 
 /// Hashes one (or two) program streams
-u64 GetUniqueIdentifier(Maxwell::ShaderProgram program_type, const ProgramCode& code,
+u64 GetUniqueIdentifier(ProgramType program_type, const ProgramCode& code,
                         const ProgramCode& code_b, std::size_t size_a = 0, std::size_t size_b = 0) {
     if (size_a == 0) {
         size_a = CalculateProgramSize(code);
     }
     u64 unique_identifier = Common::CityHash64(reinterpret_cast<const char*>(code.data()), size_a);
-    if (program_type != Maxwell::ShaderProgram::VertexA) {
+    if (program_type != ProgramType::VertexA) {
         return unique_identifier;
     }
     // VertexA programs include two programs
@@ -152,12 +173,12 @@ u64 GetUniqueIdentifier(Maxwell::ShaderProgram program_type, const ProgramCode&
 }
 
 /// Creates an unspecialized program from code streams
-GLShader::ProgramResult CreateProgram(const Device& device, Maxwell::ShaderProgram program_type,
+GLShader::ProgramResult CreateProgram(const Device& device, ProgramType program_type,
                                       ProgramCode program_code, ProgramCode program_code_b) {
     GLShader::ShaderSetup setup(program_code);
     setup.program.size_a = CalculateProgramSize(program_code);
     setup.program.size_b = 0;
-    if (program_type == Maxwell::ShaderProgram::VertexA) {
+    if (program_type == ProgramType::VertexA) {
         // VertexB is always enabled, so when VertexA is enabled, we have two vertex shaders.
         // Conventional HW does not support this, so we combine VertexA and VertexB into one
         // stage here.
@@ -168,30 +189,41 @@ GLShader::ProgramResult CreateProgram(const Device& device, Maxwell::ShaderProgr
         program_type, program_code, program_code_b, setup.program.size_a, setup.program.size_b);
 
     switch (program_type) {
-    case Maxwell::ShaderProgram::VertexA:
-    case Maxwell::ShaderProgram::VertexB:
+    case ProgramType::VertexA:
+    case ProgramType::VertexB:
         return GLShader::GenerateVertexShader(device, setup);
-    case Maxwell::ShaderProgram::Geometry:
+    case ProgramType::Geometry:
         return GLShader::GenerateGeometryShader(device, setup);
-    case Maxwell::ShaderProgram::Fragment:
+    case ProgramType::Fragment:
         return GLShader::GenerateFragmentShader(device, setup);
+    case ProgramType::Compute:
+        return GLShader::GenerateComputeShader(device, setup);
     default:
-        LOG_CRITICAL(HW_GPU, "Unimplemented program_type={}", static_cast<u32>(program_type));
-        UNREACHABLE();
+        UNIMPLEMENTED_MSG("Unimplemented program_type={}", static_cast<u32>(program_type));
         return {};
     }
 }
 
 CachedProgram SpecializeShader(const std::string& code, const GLShader::ShaderEntries& entries,
-                               Maxwell::ShaderProgram program_type, const ProgramVariant& variant,
+                               ProgramType program_type, const ProgramVariant& variant,
                                bool hint_retrievable = false) {
     auto base_bindings{variant.base_bindings};
     const auto primitive_mode{variant.primitive_mode};
     const auto texture_buffer_usage{variant.texture_buffer_usage};
 
     std::string source = "#version 430 core\n"
-                         "#extension GL_ARB_separate_shader_objects : enable\n\n";
-    source += fmt::format("#define EMULATION_UBO_BINDING {}\n", base_bindings.cbuf++);
+                         "#extension GL_ARB_separate_shader_objects : enable\n";
+    if (entries.shader_viewport_layer_array) {
+        source += "#extension GL_ARB_shader_viewport_layer_array : enable\n";
+    }
+    if (program_type == ProgramType::Compute) {
+        source += "#extension GL_ARB_compute_variable_group_size : require\n";
+    }
+    source += '\n';
+
+    if (program_type != ProgramType::Compute) {
+        source += fmt::format("#define EMULATION_UBO_BINDING {}\n", base_bindings.cbuf++);
+    }
 
     for (const auto& cbuf : entries.const_buffers) {
         source +=
@@ -218,13 +250,16 @@ CachedProgram SpecializeShader(const std::string& code, const GLShader::ShaderEn
         source += fmt::format("#define SAMPLER_{}_IS_BUFFER", i);
     }
 
-    if (program_type == Maxwell::ShaderProgram::Geometry) {
+    if (program_type == ProgramType::Geometry) {
         const auto [glsl_topology, debug_name, max_vertices] =
             GetPrimitiveDescription(primitive_mode);
 
         source += "layout (" + std::string(glsl_topology) + ") in;\n";
         source += "#define MAX_VERTEX_INPUT " + std::to_string(max_vertices) + '\n';
     }
+    if (program_type == ProgramType::Compute) {
+        source += "layout (local_size_variable) in;\n";
+    }
 
     source += code;
 
@@ -252,7 +287,7 @@ std::set<GLenum> GetSupportedFormats() {
 
 } // Anonymous namespace
 
-CachedShader::CachedShader(const ShaderParameters& params, Maxwell::ShaderProgram program_type,
+CachedShader::CachedShader(const ShaderParameters& params, ProgramType program_type,
                            GLShader::ProgramResult result)
     : RasterizerCacheObject{params.host_ptr}, host_ptr{params.host_ptr}, cpu_addr{params.cpu_addr},
       unique_identifier{params.unique_identifier}, program_type{program_type},
@@ -265,29 +300,50 @@ Shader CachedShader::CreateStageFromMemory(const ShaderParameters& params,
                                            ProgramCode&& program_code_b) {
     const auto code_size{CalculateProgramSize(program_code)};
     const auto code_size_b{CalculateProgramSize(program_code_b)};
-    auto result{CreateProgram(params.device, program_type, program_code, program_code_b)};
+    auto result{
+        CreateProgram(params.device, GetProgramType(program_type), program_code, program_code_b)};
     if (result.first.empty()) {
         // TODO(Rodrigo): Unimplemented shader stages hit here, avoid using these for now
         return {};
     }
 
     params.disk_cache.SaveRaw(ShaderDiskCacheRaw(
-        params.unique_identifier, program_type, static_cast<u32>(code_size / sizeof(u64)),
-        static_cast<u32>(code_size_b / sizeof(u64)), std::move(program_code),
-        std::move(program_code_b)));
+        params.unique_identifier, GetProgramType(program_type),
+        static_cast<u32>(code_size / sizeof(u64)), static_cast<u32>(code_size_b / sizeof(u64)),
+        std::move(program_code), std::move(program_code_b)));
 
-    return std::shared_ptr<CachedShader>(new CachedShader(params, program_type, std::move(result)));
+    return std::shared_ptr<CachedShader>(
+        new CachedShader(params, GetProgramType(program_type), std::move(result)));
 }
 
 Shader CachedShader::CreateStageFromCache(const ShaderParameters& params,
                                           Maxwell::ShaderProgram program_type,
                                           GLShader::ProgramResult result) {
-    return std::shared_ptr<CachedShader>(new CachedShader(params, program_type, std::move(result)));
+    return std::shared_ptr<CachedShader>(
+        new CachedShader(params, GetProgramType(program_type), std::move(result)));
+}
+
+Shader CachedShader::CreateKernelFromMemory(const ShaderParameters& params, ProgramCode&& code) {
+    auto result{CreateProgram(params.device, ProgramType::Compute, code, {})};
+
+    const auto code_size{CalculateProgramSize(code)};
+    params.disk_cache.SaveRaw(ShaderDiskCacheRaw(params.unique_identifier, ProgramType::Compute,
+                                                 static_cast<u32>(code_size / sizeof(u64)), 0,
+                                                 std::move(code), {}));
+
+    return std::shared_ptr<CachedShader>(
+        new CachedShader(params, ProgramType::Compute, std::move(result)));
+}
+
+Shader CachedShader::CreateKernelFromCache(const ShaderParameters& params,
+                                           GLShader::ProgramResult result) {
+    return std::shared_ptr<CachedShader>(
+        new CachedShader(params, ProgramType::Compute, std::move(result)));
 }
 
 std::tuple<GLuint, BaseBindings> CachedShader::GetProgramHandle(const ProgramVariant& variant) {
     GLuint handle{};
-    if (program_type == Maxwell::ShaderProgram::Geometry) {
+    if (program_type == ProgramType::Geometry) {
         handle = GetGeometryShader(variant);
     } else {
         const auto [entry, is_cache_miss] = programs.try_emplace(variant);
@@ -305,8 +361,11 @@ std::tuple<GLuint, BaseBindings> CachedShader::GetProgramHandle(const ProgramVar
         handle = program->handle;
     }
 
-    auto base_bindings{variant.base_bindings};
-    base_bindings.cbuf += static_cast<u32>(entries.const_buffers.size()) + RESERVED_UBOS;
+    auto base_bindings = variant.base_bindings;
+    base_bindings.cbuf += static_cast<u32>(entries.const_buffers.size());
+    if (program_type != ProgramType::Compute) {
+        base_bindings.cbuf += STAGE_RESERVED_UBOS;
+    }
     base_bindings.gmem += static_cast<u32>(entries.global_memory_entries.size());
     base_bindings.sampler += static_cast<u32>(entries.samplers.size());
 
@@ -569,7 +628,7 @@ std::unordered_map<u64, UnspecializedShader> ShaderCacheOpenGL::GenerateUnspecia
 }
 
 Shader ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) {
-    if (!system.GPU().Maxwell3D().dirty_flags.shaders) {
+    if (!system.GPU().Maxwell3D().dirty.shaders) {
         return last_shaders[static_cast<std::size_t>(program)];
     }
 
@@ -586,13 +645,15 @@ Shader ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) {
     // No shader found - create a new one
     ProgramCode program_code{GetShaderCode(memory_manager, program_addr, host_ptr)};
     ProgramCode program_code_b;
-    if (program == Maxwell::ShaderProgram::VertexA) {
+    const bool is_program_a{program == Maxwell::ShaderProgram::VertexA};
+    if (is_program_a) {
         const GPUVAddr program_addr_b{GetShaderAddress(system, Maxwell::ShaderProgram::VertexB)};
         program_code_b = GetShaderCode(memory_manager, program_addr_b,
                                        memory_manager.GetPointer(program_addr_b));
     }
 
-    const auto unique_identifier = GetUniqueIdentifier(program, program_code, program_code_b);
+    const auto unique_identifier =
+        GetUniqueIdentifier(GetProgramType(program), program_code, program_code_b);
     const auto cpu_addr{*memory_manager.GpuToCpuAddress(program_addr)};
     const ShaderParameters params{disk_cache, precompiled_programs, device, cpu_addr,
                                   host_ptr,   unique_identifier};
@@ -609,4 +670,30 @@ Shader ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) {
     return last_shaders[static_cast<std::size_t>(program)] = shader;
 }
 
+Shader ShaderCacheOpenGL::GetComputeKernel(GPUVAddr code_addr) {
+    auto& memory_manager{system.GPU().MemoryManager()};
+    const auto host_ptr{memory_manager.GetPointer(code_addr)};
+    auto kernel = TryGet(host_ptr);
+    if (kernel) {
+        return kernel;
+    }
+
+    // No kernel found - create a new one
+    auto code{GetShaderCode(memory_manager, code_addr, host_ptr)};
+    const auto unique_identifier{GetUniqueIdentifier(ProgramType::Compute, code, {})};
+    const auto cpu_addr{*memory_manager.GpuToCpuAddress(code_addr)};
+    const ShaderParameters params{disk_cache, precompiled_programs, device, cpu_addr,
+                                  host_ptr,   unique_identifier};
+
+    const auto found = precompiled_shaders.find(unique_identifier);
+    if (found == precompiled_shaders.end()) {
+        kernel = CachedShader::CreateKernelFromMemory(params, std::move(code));
+    } else {
+        kernel = CachedShader::CreateKernelFromCache(params, found->second);
+    }
+
+    Register(kernel);
+    return kernel;
+}
+
 } // namespace OpenGL

src/video_core/renderer_opengl/gl_shader_cache.h

@@ -61,6 +61,11 @@ public:
                                        Maxwell::ShaderProgram program_type,
                                        GLShader::ProgramResult result);
 
+    static Shader CreateKernelFromMemory(const ShaderParameters& params, ProgramCode&& code);
+
+    static Shader CreateKernelFromCache(const ShaderParameters& params,
+                                        GLShader::ProgramResult result);
+
     VAddr GetCpuAddr() const override {
         return cpu_addr;
     }
@@ -78,7 +83,7 @@ public:
     std::tuple<GLuint, BaseBindings> GetProgramHandle(const ProgramVariant& variant);
 
 private:
-    explicit CachedShader(const ShaderParameters& params, Maxwell::ShaderProgram program_type,
+    explicit CachedShader(const ShaderParameters& params, ProgramType program_type,
                           GLShader::ProgramResult result);
 
     // Geometry programs. These are needed because GLSL needs an input topology but it's not
@@ -104,7 +109,7 @@ private:
     u8* host_ptr{};
     VAddr cpu_addr{};
     u64 unique_identifier{};
-    Maxwell::ShaderProgram program_type{};
+    ProgramType program_type{};
     ShaderDiskCacheOpenGL& disk_cache;
     const PrecompiledPrograms& precompiled_programs;
 
@@ -132,6 +137,9 @@ public:
     /// Gets the current specified shader stage program
     Shader GetStageProgram(Maxwell::ShaderProgram program);
 
+    /// Gets a compute kernel in the passed address
+    Shader GetComputeKernel(GPUVAddr code_addr);
+
 protected:
     // We do not have to flush this cache as things in it are never modified by us.
     void FlushObjectInner(const Shader& object) override {}

src/video_core/renderer_opengl/gl_shader_decompiler.cpp

@@ -14,6 +14,7 @@
 #include "common/alignment.h"
 #include "common/assert.h"
 #include "common/common_types.h"
+#include "common/logging/log.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/renderer_opengl/gl_device.h"
 #include "video_core/renderer_opengl/gl_rasterizer.h"
@@ -36,7 +37,6 @@ using namespace std::string_literals;
 using namespace VideoCommon::Shader;
 
 using Maxwell = Tegra::Engines::Maxwell3D::Regs;
-using ShaderStage = Tegra::Engines::Maxwell3D::Regs::ShaderStage;
 using Operation = const OperationNode&;
 
 enum class Type { Bool, Bool2, Float, Int, Uint, HalfFloat };
@@ -46,7 +46,7 @@ using TextureArgument = std::pair<Type, Node>;
 using TextureIR = std::variant<TextureAoffi, TextureArgument>;
 
 constexpr u32 MAX_CONSTBUFFER_ELEMENTS =
-    static_cast<u32>(RasterizerOpenGL::MaxConstbufferSize) / (4 * sizeof(float));
+    static_cast<u32>(Maxwell::MaxConstBufferSize) / (4 * sizeof(float));
 
 class ShaderWriter {
 public:
@@ -161,9 +161,13 @@ std::string FlowStackTopName(MetaStackClass stack) {
     return fmt::format("{}_flow_stack_top", GetFlowStackPrefix(stack));
 }
 
+constexpr bool IsVertexShader(ProgramType stage) {
+    return stage == ProgramType::VertexA || stage == ProgramType::VertexB;
+}
+
 class GLSLDecompiler final {
 public:
-    explicit GLSLDecompiler(const Device& device, const ShaderIR& ir, ShaderStage stage,
+    explicit GLSLDecompiler(const Device& device, const ShaderIR& ir, ProgramType stage,
                             std::string suffix)
         : device{device}, ir{ir}, stage{stage}, suffix{suffix}, header{ir.GetHeader()} {}
 
@@ -246,24 +250,22 @@ public:
                                                        usage.is_read, usage.is_written);
         }
         entries.clip_distances = ir.GetClipDistances();
+        entries.shader_viewport_layer_array =
+            IsVertexShader(stage) && (ir.UsesLayer() || ir.UsesViewportIndex());
         entries.shader_length = ir.GetLength();
         return entries;
     }
 
 private:
-    using OperationDecompilerFn = std::string (GLSLDecompiler::*)(Operation);
-    using OperationDecompilersArray =
-        std::array<OperationDecompilerFn, static_cast<std::size_t>(OperationCode::Amount)>;
-
     void DeclareVertex() {
-        if (stage != ShaderStage::Vertex)
+        if (!IsVertexShader(stage))
             return;
 
         DeclareVertexRedeclarations();
     }
 
     void DeclareGeometry() {
-        if (stage != ShaderStage::Geometry) {
+        if (stage != ProgramType::Geometry) {
             return;
         }
 
@@ -282,22 +284,35 @@ private:
     }
 
     void DeclareVertexRedeclarations() {
-        bool clip_distances_declared = false;
-
         code.AddLine("out gl_PerVertex {{");
         ++code.scope;
 
         code.AddLine("vec4 gl_Position;");
 
-        for (const auto o : ir.GetOutputAttributes()) {
-            if (o == Attribute::Index::PointSize)
-                code.AddLine("float gl_PointSize;");
-            if (!clip_distances_declared && (o == Attribute::Index::ClipDistances0123 ||
-                                             o == Attribute::Index::ClipDistances4567)) {
+        for (const auto attribute : ir.GetOutputAttributes()) {
+            if (attribute == Attribute::Index::ClipDistances0123 ||
+                attribute == Attribute::Index::ClipDistances4567) {
                 code.AddLine("float gl_ClipDistance[];");
-                clip_distances_declared = true;
+                break;
             }
         }
+        if (!IsVertexShader(stage) || device.HasVertexViewportLayer()) {
+            if (ir.UsesLayer()) {
+                code.AddLine("int gl_Layer;");
+            }
+            if (ir.UsesViewportIndex()) {
+                code.AddLine("int gl_ViewportIndex;");
+            }
+        } else if ((ir.UsesLayer() || ir.UsesViewportIndex()) && IsVertexShader(stage) &&
+                   !device.HasVertexViewportLayer()) {
+            LOG_ERROR(
+                Render_OpenGL,
+                "GL_ARB_shader_viewport_layer_array is not available and its required by a shader");
+        }
+
+        if (ir.UsesPointSize()) {
+            code.AddLine("float gl_PointSize;");
+        }
 
         --code.scope;
         code.AddLine("}};");
@@ -325,11 +340,16 @@ private:
     }
 
     void DeclareLocalMemory() {
-        if (const u64 local_memory_size = header.GetLocalMemorySize(); local_memory_size > 0) {
-            const auto element_count = Common::AlignUp(local_memory_size, 4) / 4;
-            code.AddLine("float {}[{}];", GetLocalMemory(), element_count);
-            code.AddNewLine();
+        // TODO(Rodrigo): Unstub kernel local memory size and pass it from a register at
+        // specialization time.
+        const u64 local_memory_size =
+            stage == ProgramType::Compute ? 0x400 : header.GetLocalMemorySize();
+        if (local_memory_size == 0) {
+            return;
         }
+        const auto element_count = Common::AlignUp(local_memory_size, 4) / 4;
+        code.AddLine("float {}[{}];", GetLocalMemory(), element_count);
+        code.AddNewLine();
     }
 
     void DeclareInternalFlags() {
@@ -383,12 +403,12 @@ private:
         const u32 location{GetGenericAttributeIndex(index)};
 
         std::string name{GetInputAttribute(index)};
-        if (stage == ShaderStage::Geometry) {
+        if (stage == ProgramType::Geometry) {
             name = "gs_" + name + "[]";
         }
 
         std::string suffix;
-        if (stage == ShaderStage::Fragment) {
+        if (stage == ProgramType::Fragment) {
             const auto input_mode{header.ps.GetAttributeUse(location)};
             if (skip_unused && input_mode == AttributeUse::Unused) {
                 return;
@@ -400,7 +420,7 @@ private:
     }
 
     void DeclareOutputAttributes() {
-        if (ir.HasPhysicalAttributes() && stage != ShaderStage::Fragment) {
+        if (ir.HasPhysicalAttributes() && stage != ProgramType::Fragment) {
             for (u32 i = 0; i < GetNumPhysicalVaryings(); ++i) {
                 DeclareOutputAttribute(ToGenericAttribute(i));
             }
@@ -522,7 +542,7 @@ private:
                 constexpr u32 element_stride{4};
                 const u32 address{generic_base + index * generic_stride + element * element_stride};
 
-                const bool declared{stage != ShaderStage::Fragment ||
+                const bool declared{stage != ProgramType::Fragment ||
                                     header.ps.GetAttributeUse(index) != AttributeUse::Unused};
                 const std::string value{declared ? ReadAttribute(attribute, element) : "0"};
                 code.AddLine("case 0x{:x}: return {};", address, value);
@@ -626,7 +646,7 @@ private:
         }
 
         if (const auto abuf = std::get_if<AbufNode>(&*node)) {
-            UNIMPLEMENTED_IF_MSG(abuf->IsPhysicalBuffer() && stage == ShaderStage::Geometry,
+            UNIMPLEMENTED_IF_MSG(abuf->IsPhysicalBuffer() && stage == ProgramType::Geometry,
                                  "Physical attributes in geometry shaders are not implemented");
             if (abuf->IsPhysicalBuffer()) {
                 return fmt::format("readPhysicalAttribute(ftou({}))",
@@ -681,6 +701,9 @@ private:
         }
 
         if (const auto lmem = std::get_if<LmemNode>(&*node)) {
+            if (stage == ProgramType::Compute) {
+                LOG_WARNING(Render_OpenGL, "Local memory is stubbed on compute shaders");
+            }
             return fmt::format("{}[ftou({}) / 4]", GetLocalMemory(), Visit(lmem->GetAddress()));
         }
 
@@ -710,7 +733,7 @@ private:
 
     std::string ReadAttribute(Attribute::Index attribute, u32 element, const Node& buffer = {}) {
         const auto GeometryPass = [&](std::string_view name) {
-            if (stage == ShaderStage::Geometry && buffer) {
+            if (stage == ProgramType::Geometry && buffer) {
                 // TODO(Rodrigo): Guard geometry inputs against out of bound reads. Some games
                 // set an 0x80000000 index for those and the shader fails to build. Find out why
                 // this happens and what's its intent.
@@ -722,10 +745,10 @@ private:
         switch (attribute) {
         case Attribute::Index::Position:
             switch (stage) {
-            case ShaderStage::Geometry:
+            case ProgramType::Geometry:
                 return fmt::format("gl_in[ftou({})].gl_Position{}", Visit(buffer),
                                    GetSwizzle(element));
-            case ShaderStage::Fragment:
+            case ProgramType::Fragment:
                 return element == 3 ? "1.0f" : ("gl_FragCoord"s + GetSwizzle(element));
             default:
                 UNREACHABLE();
@@ -746,7 +769,7 @@ private:
             // TODO(Subv): Find out what the values are for the first two elements when inside a
             // vertex shader, and what's the value of the fourth element when inside a Tess Eval
             // shader.
-            ASSERT(stage == ShaderStage::Vertex);
+            ASSERT(IsVertexShader(stage));
             switch (element) {
             case 2:
                 // Config pack's first value is instance_id.
@@ -758,7 +781,7 @@ private:
             return "0";
         case Attribute::Index::FrontFacing:
             // TODO(Subv): Find out what the values are for the other elements.
-            ASSERT(stage == ShaderStage::Fragment);
+            ASSERT(stage == ProgramType::Fragment);
             switch (element) {
             case 3:
                 return "itof(gl_FrontFacing ? -1 : 0)";
@@ -780,7 +803,7 @@ private:
             return value;
         }
         // There's a bug in NVidia's proprietary drivers that makes precise fail on fragment shaders
-        const std::string precise = stage != ShaderStage::Fragment ? "precise " : "";
+        const std::string precise = stage != ProgramType::Fragment ? "precise " : "";
 
         const std::string temporary = code.GenerateTemporary();
         code.AddLine("{}float {} = {};", precise, temporary, value);
@@ -805,6 +828,45 @@ private:
         return CastOperand(VisitOperand(operation, operand_index), type);
     }
 
+    std::optional<std::pair<std::string, bool>> GetOutputAttribute(const AbufNode* abuf) {
+        switch (const auto attribute = abuf->GetIndex()) {
+        case Attribute::Index::Position:
+            return std::make_pair("gl_Position"s + GetSwizzle(abuf->GetElement()), false);
+        case Attribute::Index::LayerViewportPointSize:
+            switch (abuf->GetElement()) {
+            case 0:
+                UNIMPLEMENTED();
+                return {};
+            case 1:
+                if (IsVertexShader(stage) && !device.HasVertexViewportLayer()) {
+                    return {};
+                }
+                return std::make_pair("gl_Layer", true);
+            case 2:
+                if (IsVertexShader(stage) && !device.HasVertexViewportLayer()) {
+                    return {};
+                }
+                return std::make_pair("gl_ViewportIndex", true);
+            case 3:
+                UNIMPLEMENTED_MSG("Requires some state changes for gl_PointSize to work in shader");
+                return std::make_pair("gl_PointSize", false);
+            }
+            return {};
+        case Attribute::Index::ClipDistances0123:
+            return std::make_pair(fmt::format("gl_ClipDistance[{}]", abuf->GetElement()), false);
+        case Attribute::Index::ClipDistances4567:
+            return std::make_pair(fmt::format("gl_ClipDistance[{}]", abuf->GetElement() + 4),
+                                  false);
+        default:
+            if (IsGenericAttribute(attribute)) {
+                return std::make_pair(
+                    GetOutputAttribute(attribute) + GetSwizzle(abuf->GetElement()), false);
+            }
+            UNIMPLEMENTED_MSG("Unhandled output attribute: {}", static_cast<u32>(attribute));
+            return {};
+        }
+    }
+
     std::string CastOperand(const std::string& value, Type type) const {
         switch (type) {
         case Type::Bool:
@@ -1001,6 +1063,8 @@ private:
         const Node& src = operation[1];
 
         std::string target;
+        bool is_integer = false;
+
         if (const auto gpr = std::get_if<GprNode>(&*dest)) {
             if (gpr->GetIndex() == Register::ZeroIndex) {
                 // Writing to Register::ZeroIndex is a no op
@@ -1009,27 +1073,16 @@ private:
             target = GetRegister(gpr->GetIndex());
         } else if (const auto abuf = std::get_if<AbufNode>(&*dest)) {
             UNIMPLEMENTED_IF(abuf->IsPhysicalBuffer());
-
-            target = [&]() -> std::string {
-                switch (const auto attribute = abuf->GetIndex(); abuf->GetIndex()) {
-                case Attribute::Index::Position:
-                    return "gl_Position"s + GetSwizzle(abuf->GetElement());
-                case Attribute::Index::PointSize:
-                    return "gl_PointSize";
-                case Attribute::Index::ClipDistances0123:
-                    return fmt::format("gl_ClipDistance[{}]", abuf->GetElement());
-                case Attribute::Index::ClipDistances4567:
-                    return fmt::format("gl_ClipDistance[{}]", abuf->GetElement() + 4);
-                default:
-                    if (IsGenericAttribute(attribute)) {
-                        return GetOutputAttribute(attribute) + GetSwizzle(abuf->GetElement());
-                    }
-                    UNIMPLEMENTED_MSG("Unhandled output attribute: {}",
-                                      static_cast<u32>(attribute));
-                    return "0";
-                }
-            }();
+            const auto result = GetOutputAttribute(abuf);
+            if (!result) {
+                return {};
+            }
+            target = result->first;
+            is_integer = result->second;
         } else if (const auto lmem = std::get_if<LmemNode>(&*dest)) {
+            if (stage == ProgramType::Compute) {
+                LOG_WARNING(Render_OpenGL, "Local memory is stubbed on compute shaders");
+            }
             target = fmt::format("{}[ftou({}) / 4]", GetLocalMemory(), Visit(lmem->GetAddress()));
         } else if (const auto gmem = std::get_if<GmemNode>(&*dest)) {
             const std::string real = Visit(gmem->GetRealAddress());
@@ -1040,7 +1093,11 @@ private:
             UNREACHABLE_MSG("Assign called without a proper target");
         }
 
-        code.AddLine("{} = {};", target, Visit(src));
+        if (is_integer) {
+            code.AddLine("{} = ftoi({});", target, Visit(src));
+        } else {
+            code.AddLine("{} = {};", target, Visit(src));
+        }
         return {};
     }
 
@@ -1353,14 +1410,10 @@ private:
         return fmt::format("{}[{}]", pair, VisitOperand(operation, 1, Type::Uint));
     }
 
-    std::string LogicalAll2(Operation operation) {
+    std::string LogicalAnd2(Operation operation) {
         return GenerateUnary(operation, "all", Type::Bool, Type::Bool2);
     }
 
-    std::string LogicalAny2(Operation operation) {
-        return GenerateUnary(operation, "any", Type::Bool, Type::Bool2);
-    }
-
     template <bool with_nan>
     std::string GenerateHalfComparison(Operation operation, const std::string& compare_op) {
         const std::string comparison{GenerateBinaryCall(operation, compare_op, Type::Bool2,
@@ -1583,7 +1636,7 @@ private:
     }
 
     std::string Exit(Operation operation) {
-        if (stage != ShaderStage::Fragment) {
+        if (stage != ProgramType::Fragment) {
             code.AddLine("return;");
             return {};
         }
@@ -1634,7 +1687,7 @@ private:
     }
 
     std::string EmitVertex(Operation operation) {
-        ASSERT_MSG(stage == ShaderStage::Geometry,
+        ASSERT_MSG(stage == ProgramType::Geometry,
                    "EmitVertex is expected to be used in a geometry shader.");
 
         // If a geometry shader is attached, it will always flip (it's the last stage before
@@ -1645,7 +1698,7 @@ private:
     }
 
     std::string EndPrimitive(Operation operation) {
-        ASSERT_MSG(stage == ShaderStage::Geometry,
+        ASSERT_MSG(stage == ProgramType::Geometry,
                    "EndPrimitive is expected to be used in a geometry shader.");
 
         code.AddLine("EndPrimitive();");
@@ -1667,7 +1720,7 @@ private:
         return "utof(gl_WorkGroupID"s + GetSwizzle(element) + ')';
     }
 
-    static constexpr OperationDecompilersArray operation_decompilers = {
+    static constexpr std::array operation_decompilers = {
         &GLSLDecompiler::Assign,
 
         &GLSLDecompiler::Select,
@@ -1751,8 +1804,7 @@ private:
         &GLSLDecompiler::LogicalXor,
         &GLSLDecompiler::LogicalNegate,
         &GLSLDecompiler::LogicalPick2,
-        &GLSLDecompiler::LogicalAll2,
-        &GLSLDecompiler::LogicalAny2,
+        &GLSLDecompiler::LogicalAnd2,
 
         &GLSLDecompiler::LogicalLessThan<Type::Float>,
         &GLSLDecompiler::LogicalEqual<Type::Float>,
@@ -1816,6 +1868,7 @@ private:
         &GLSLDecompiler::WorkGroupId<1>,
         &GLSLDecompiler::WorkGroupId<2>,
     };
+    static_assert(operation_decompilers.size() == static_cast<std::size_t>(OperationCode::Amount));
 
     std::string GetRegister(u32 index) const {
         return GetDeclarationWithSuffix(index, "gpr");
@@ -1880,7 +1933,7 @@ private:
     }
 
     u32 GetNumPhysicalInputAttributes() const {
-        return stage == ShaderStage::Vertex ? GetNumPhysicalAttributes() : GetNumPhysicalVaryings();
+        return IsVertexShader(stage) ? GetNumPhysicalAttributes() : GetNumPhysicalVaryings();
     }
 
     u32 GetNumPhysicalAttributes() const {
@@ -1893,7 +1946,7 @@ private:
 
     const Device& device;
     const ShaderIR& ir;
-    const ShaderStage stage;
+    const ProgramType stage;
     const std::string suffix;
     const Header header;
 
@@ -1924,7 +1977,7 @@ std::string GetCommonDeclarations() {
         MAX_CONSTBUFFER_ELEMENTS);
 }
 
-ProgramResult Decompile(const Device& device, const ShaderIR& ir, Maxwell::ShaderStage stage,
+ProgramResult Decompile(const Device& device, const ShaderIR& ir, ProgramType stage,
                         const std::string& suffix) {
     GLSLDecompiler decompiler(device, ir, stage, suffix);
     decompiler.Decompile();

src/video_core/renderer_opengl/gl_shader_decompiler.h

@@ -12,14 +12,26 @@
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/shader/shader_ir.h"
 
-namespace OpenGL {
-class Device;
-}
-
 namespace VideoCommon::Shader {
 class ShaderIR;
 }
 
+namespace OpenGL {
+
+class Device;
+
+enum class ProgramType : u32 {
+    VertexA = 0,
+    VertexB = 1,
+    TessellationControl = 2,
+    TessellationEval = 3,
+    Geometry = 4,
+    Fragment = 5,
+    Compute = 6
+};
+
+} // namespace OpenGL
+
 namespace OpenGL::GLShader {
 
 struct ShaderEntries;
@@ -78,12 +90,13 @@ struct ShaderEntries {
     std::vector<ImageEntry> images;
     std::vector<GlobalMemoryEntry> global_memory_entries;
     std::array<bool, Maxwell::NumClipDistances> clip_distances{};
+    bool shader_viewport_layer_array{};
     std::size_t shader_length{};
 };
 
 std::string GetCommonDeclarations();
 
 ProgramResult Decompile(const Device& device, const VideoCommon::Shader::ShaderIR& ir,
-                        Maxwell::ShaderStage stage, const std::string& suffix);
+                        ProgramType stage, const std::string& suffix);
 
 } // namespace OpenGL::GLShader

src/video_core/renderer_opengl/gl_shader_disk_cache.cpp

@@ -51,7 +51,7 @@ ShaderCacheVersionHash GetShaderCacheVersionHash() {
 
 } // namespace
 
-ShaderDiskCacheRaw::ShaderDiskCacheRaw(u64 unique_identifier, Maxwell::ShaderProgram program_type,
+ShaderDiskCacheRaw::ShaderDiskCacheRaw(u64 unique_identifier, ProgramType program_type,
                                        u32 program_code_size, u32 program_code_size_b,
                                        ProgramCode program_code, ProgramCode program_code_b)
     : unique_identifier{unique_identifier}, program_type{program_type},
@@ -373,6 +373,12 @@ std::optional<ShaderDiskCacheDecompiled> ShaderDiskCacheOpenGL::LoadDecompiledEn
         }
     }
 
+    bool shader_viewport_layer_array{};
+    if (!LoadObjectFromPrecompiled(shader_viewport_layer_array)) {
+        return {};
+    }
+    entry.entries.shader_viewport_layer_array = shader_viewport_layer_array;
+
     u64 shader_length{};
     if (!LoadObjectFromPrecompiled(shader_length)) {
         return {};
@@ -445,6 +451,10 @@ bool ShaderDiskCacheOpenGL::SaveDecompiledFile(u64 unique_identifier, const std:
         }
     }
 
+    if (!SaveObjectToPrecompiled(entries.shader_viewport_layer_array)) {
+        return false;
+    }
+
     if (!SaveObjectToPrecompiled(static_cast<u64>(entries.shader_length))) {
         return false;
     }

src/video_core/renderer_opengl/gl_shader_disk_cache.h

@@ -18,7 +18,6 @@
 #include "common/assert.h"
 #include "common/common_types.h"
 #include "core/file_sys/vfs_vector.h"
-#include "video_core/engines/maxwell_3d.h"
 #include "video_core/renderer_opengl/gl_shader_gen.h"
 
 namespace Core {
@@ -34,14 +33,11 @@ namespace OpenGL {
 struct ShaderDiskCacheUsage;
 struct ShaderDiskCacheDump;
 
-using ShaderDumpsMap = std::unordered_map<ShaderDiskCacheUsage, ShaderDiskCacheDump>;
-
 using ProgramCode = std::vector<u64>;
-using Maxwell = Tegra::Engines::Maxwell3D::Regs;
-
+using ShaderDumpsMap = std::unordered_map<ShaderDiskCacheUsage, ShaderDiskCacheDump>;
 using TextureBufferUsage = std::bitset<64>;
 
-/// Allocated bindings used by an OpenGL shader program.
+/// Allocated bindings used by an OpenGL shader program
 struct BaseBindings {
     u32 cbuf{};
     u32 gmem{};
@@ -126,7 +122,7 @@ namespace OpenGL {
 /// Describes a shader how it's used by the guest GPU
 class ShaderDiskCacheRaw {
 public:
-    explicit ShaderDiskCacheRaw(u64 unique_identifier, Maxwell::ShaderProgram program_type,
+    explicit ShaderDiskCacheRaw(u64 unique_identifier, ProgramType program_type,
                                 u32 program_code_size, u32 program_code_size_b,
                                 ProgramCode program_code, ProgramCode program_code_b);
     ShaderDiskCacheRaw();
@@ -141,30 +137,13 @@ public:
     }
 
     bool HasProgramA() const {
-        return program_type == Maxwell::ShaderProgram::VertexA;
+        return program_type == ProgramType::VertexA;
     }
 
-    Maxwell::ShaderProgram GetProgramType() const {
+    ProgramType GetProgramType() const {
         return program_type;
     }
 
-    Maxwell::ShaderStage GetProgramStage() const {
-        switch (program_type) {
-        case Maxwell::ShaderProgram::VertexA:
-        case Maxwell::ShaderProgram::VertexB:
-            return Maxwell::ShaderStage::Vertex;
-        case Maxwell::ShaderProgram::TesselationControl:
-            return Maxwell::ShaderStage::TesselationControl;
-        case Maxwell::ShaderProgram::TesselationEval:
-            return Maxwell::ShaderStage::TesselationEval;
-        case Maxwell::ShaderProgram::Geometry:
-            return Maxwell::ShaderStage::Geometry;
-        case Maxwell::ShaderProgram::Fragment:
-            return Maxwell::ShaderStage::Fragment;
-        }
-        UNREACHABLE();
-    }
-
     const ProgramCode& GetProgramCode() const {
         return program_code;
     }
@@ -175,7 +154,7 @@ public:
 
 private:
     u64 unique_identifier{};
-    Maxwell::ShaderProgram program_type{};
+    ProgramType program_type{};
     u32 program_code_size{};
     u32 program_code_size_b{};
 

src/video_core/renderer_opengl/gl_shader_gen.cpp

@@ -14,7 +14,8 @@ using Tegra::Engines::Maxwell3D;
 using VideoCommon::Shader::ProgramCode;
 using VideoCommon::Shader::ShaderIR;
 
-static constexpr u32 PROGRAM_OFFSET{10};
+static constexpr u32 PROGRAM_OFFSET = 10;
+static constexpr u32 COMPUTE_OFFSET = 0;
 
 ProgramResult GenerateVertexShader(const Device& device, const ShaderSetup& setup) {
     const std::string id = fmt::format("{:016x}", setup.program.unique_identifier);
@@ -29,17 +30,15 @@ layout (std140, binding = EMULATION_UBO_BINDING) uniform vs_config {
 };
 
 )";
-    const ShaderIR program_ir(setup.program.code, PROGRAM_OFFSET, setup.program.size_a);
-    ProgramResult program =
-        Decompile(device, program_ir, Maxwell3D::Regs::ShaderStage::Vertex, "vertex");
 
+    const ShaderIR program_ir(setup.program.code, PROGRAM_OFFSET, setup.program.size_a);
+    const auto stage = setup.IsDualProgram() ? ProgramType::VertexA : ProgramType::VertexB;
+    ProgramResult program = Decompile(device, program_ir, stage, "vertex");
     out += program.first;
 
     if (setup.IsDualProgram()) {
         const ShaderIR program_ir_b(setup.program.code_b, PROGRAM_OFFSET, setup.program.size_b);
-        ProgramResult program_b =
-            Decompile(device, program_ir_b, Maxwell3D::Regs::ShaderStage::Vertex, "vertex_b");
-
+        ProgramResult program_b = Decompile(device, program_ir_b, ProgramType::VertexB, "vertex_b");
         out += program_b.first;
     }
 
@@ -80,9 +79,9 @@ layout (std140, binding = EMULATION_UBO_BINDING) uniform gs_config {
 };
 
 )";
+
     const ShaderIR program_ir(setup.program.code, PROGRAM_OFFSET, setup.program.size_a);
-    ProgramResult program =
-        Decompile(device, program_ir, Maxwell3D::Regs::ShaderStage::Geometry, "geometry");
+    ProgramResult program = Decompile(device, program_ir, ProgramType::Geometry, "geometry");
     out += program.first;
 
     out += R"(
@@ -116,9 +115,7 @@ layout (std140, binding = EMULATION_UBO_BINDING) uniform fs_config {
 
 )";
     const ShaderIR program_ir(setup.program.code, PROGRAM_OFFSET, setup.program.size_a);
-    ProgramResult program =
-        Decompile(device, program_ir, Maxwell3D::Regs::ShaderStage::Fragment, "fragment");
-
+    ProgramResult program = Decompile(device, program_ir, ProgramType::Fragment, "fragment");
     out += program.first;
 
     out += R"(
@@ -130,4 +127,22 @@ void main() {
     return {std::move(out), std::move(program.second)};
 }
 
+ProgramResult GenerateComputeShader(const Device& device, const ShaderSetup& setup) {
+    const std::string id = fmt::format("{:016x}", setup.program.unique_identifier);
+
+    std::string out = "// Shader Unique Id: CS" + id + "\n\n";
+    out += GetCommonDeclarations();
+
+    const ShaderIR program_ir(setup.program.code, COMPUTE_OFFSET, setup.program.size_a);
+    ProgramResult program = Decompile(device, program_ir, ProgramType::Compute, "compute");
+    out += program.first;
+
+    out += R"(
+void main() {
+    execute_compute();
+}
+)";
+    return {std::move(out), std::move(program.second)};
+}
+
 } // namespace OpenGL::GLShader

src/video_core/renderer_opengl/gl_shader_gen.h

@@ -54,4 +54,7 @@ ProgramResult GenerateGeometryShader(const Device& device, const ShaderSetup& se
 /// Generates the GLSL fragment shader program source code for the given FS program
 ProgramResult GenerateFragmentShader(const Device& device, const ShaderSetup& setup);
 
+/// Generates the GLSL compute shader program source code for the given CS program
+ProgramResult GenerateComputeShader(const Device& device, const ShaderSetup& setup);
+
 } // namespace OpenGL::GLShader

src/video_core/renderer_opengl/gl_shader_util.cpp

@@ -10,21 +10,25 @@
 
 namespace OpenGL::GLShader {
 
-GLuint LoadShader(const char* source, GLenum type) {
-    const char* debug_type;
+namespace {
+const char* GetStageDebugName(GLenum type) {
     switch (type) {
     case GL_VERTEX_SHADER:
-        debug_type = "vertex";
-        break;
+        return "vertex";
     case GL_GEOMETRY_SHADER:
-        debug_type = "geometry";
-        break;
+        return "geometry";
     case GL_FRAGMENT_SHADER:
-        debug_type = "fragment";
-        break;
-    default:
-        UNREACHABLE();
+        return "fragment";
+    case GL_COMPUTE_SHADER:
+        return "compute";
     }
+    UNIMPLEMENTED();
+    return "unknown";
+}
+} // Anonymous namespace
+
+GLuint LoadShader(const char* source, GLenum type) {
+    const char* debug_type = GetStageDebugName(type);
     const GLuint shader_id = glCreateShader(type);
     glShaderSource(shader_id, 1, &source, nullptr);
     LOG_DEBUG(Render_OpenGL, "Compiling {} shader...", debug_type);

src/video_core/renderer_opengl/gl_state.cpp

@@ -6,8 +6,11 @@
 #include <glad/glad.h>
 #include "common/assert.h"
 #include "common/logging/log.h"
+#include "common/microprofile.h"
 #include "video_core/renderer_opengl/gl_state.h"
 
+MICROPROFILE_DEFINE(OpenGL_State, "OpenGL", "State Change", MP_RGB(192, 128, 128));
+
 namespace OpenGL {
 
 using Maxwell = Tegra::Engines::Maxwell3D::Regs;
@@ -162,6 +165,25 @@ OpenGLState::OpenGLState() {
     alpha_test.ref = 0.0f;
 }
 
+void OpenGLState::SetDefaultViewports() {
+    for (auto& item : viewports) {
+        item.x = 0;
+        item.y = 0;
+        item.width = 0;
+        item.height = 0;
+        item.depth_range_near = 0.0f;
+        item.depth_range_far = 1.0f;
+        item.scissor.enabled = false;
+        item.scissor.x = 0;
+        item.scissor.y = 0;
+        item.scissor.width = 0;
+        item.scissor.height = 0;
+    }
+
+    depth_clamp.far_plane = false;
+    depth_clamp.near_plane = false;
+}
+
 void OpenGLState::ApplyDefaultState() {
     glEnable(GL_BLEND);
     glDisable(GL_FRAMEBUFFER_SRGB);
@@ -523,7 +545,8 @@ void OpenGLState::ApplySamplers() const {
     }
 }
 
-void OpenGLState::Apply() const {
+void OpenGLState::Apply() {
+    MICROPROFILE_SCOPE(OpenGL_State);
     ApplyFramebufferState();
     ApplyVertexArrayState();
     ApplyShaderProgram();
@@ -532,19 +555,31 @@ void OpenGLState::Apply() const {
     ApplyPointSize();
     ApplyFragmentColorClamp();
     ApplyMultisample();
+    if (dirty.color_mask) {
+        ApplyColorMask();
+        dirty.color_mask = false;
+    }
     ApplyDepthClamp();
-    ApplyColorMask();
     ApplyViewport();
-    ApplyStencilTest();
+    if (dirty.stencil_state) {
+        ApplyStencilTest();
+        dirty.stencil_state = false;
+    }
     ApplySRgb();
     ApplyCulling();
     ApplyDepth();
     ApplyPrimitiveRestart();
-    ApplyBlending();
+    if (dirty.blend_state) {
+        ApplyBlending();
+        dirty.blend_state = false;
+    }
     ApplyLogicOp();
     ApplyTextures();
     ApplySamplers();
-    ApplyPolygonOffset();
+    if (dirty.polygon_offset) {
+        ApplyPolygonOffset();
+        dirty.polygon_offset = false;
+    }
     ApplyAlphaTest();
 }
 

src/video_core/renderer_opengl/gl_state.h

@@ -195,8 +195,9 @@ public:
         s_rgb_used = false;
     }
 
+    void SetDefaultViewports();
     /// Apply this state as the current OpenGL state
-    void Apply() const;
+    void Apply();
 
     void ApplyFramebufferState() const;
     void ApplyVertexArrayState() const;
@@ -237,11 +238,41 @@ public:
     /// Viewport does not affects glClearBuffer so emulate viewport using scissor test
     void EmulateViewportWithScissor();
 
+    void MarkDirtyBlendState() {
+        dirty.blend_state = true;
+    }
+
+    void MarkDirtyStencilState() {
+        dirty.stencil_state = true;
+    }
+
+    void MarkDirtyPolygonOffset() {
+        dirty.polygon_offset = true;
+    }
+
+    void MarkDirtyColorMask() {
+        dirty.color_mask = true;
+    }
+
+    void AllDirty() {
+        dirty.blend_state = true;
+        dirty.stencil_state = true;
+        dirty.polygon_offset = true;
+        dirty.color_mask = true;
+    }
+
 private:
     static OpenGLState cur_state;
 
     // Workaround for sRGB problems caused by QT not supporting srgb output
     static bool s_rgb_used;
+    struct {
+        bool blend_state;
+        bool stencil_state;
+        bool viewport_state;
+        bool polygon_offset;
+        bool color_mask;
+    } dirty{};
 };
 
 } // namespace OpenGL

src/video_core/renderer_opengl/gl_texture_cache.cpp

@@ -31,6 +31,8 @@ using VideoCore::Surface::SurfaceType;
 
 MICROPROFILE_DEFINE(OpenGL_Texture_Upload, "OpenGL", "Texture Upload", MP_RGB(128, 192, 128));
 MICROPROFILE_DEFINE(OpenGL_Texture_Download, "OpenGL", "Texture Download", MP_RGB(128, 192, 128));
+MICROPROFILE_DEFINE(OpenGL_Texture_Buffer_Copy, "OpenGL", "Texture Buffer Copy",
+                    MP_RGB(128, 192, 128));
 
 namespace {
 
@@ -135,7 +137,6 @@ constexpr std::array<FormatTuple, VideoCore::Surface::MaxPixelFormat> tex_format
 const FormatTuple& GetFormatTuple(PixelFormat pixel_format, ComponentType component_type) {
     ASSERT(static_cast<std::size_t>(pixel_format) < tex_format_tuples.size());
     const auto& format{tex_format_tuples[static_cast<std::size_t>(pixel_format)]};
-    ASSERT(component_type == format.component_type);
     return format;
 }
 
@@ -483,11 +484,15 @@ void TextureCacheOpenGL::ImageBlit(View& src_view, View& dst_view,
     const auto& dst_params{dst_view->GetSurfaceParams()};
 
     OpenGLState prev_state{OpenGLState::GetCurState()};
-    SCOPE_EXIT({ prev_state.Apply(); });
+    SCOPE_EXIT({
+        prev_state.AllDirty();
+        prev_state.Apply();
+    });
 
     OpenGLState state;
     state.draw.read_framebuffer = src_framebuffer.handle;
     state.draw.draw_framebuffer = dst_framebuffer.handle;
+    state.AllDirty();
     state.Apply();
 
     u32 buffers{};
@@ -535,6 +540,7 @@ void TextureCacheOpenGL::ImageBlit(View& src_view, View& dst_view,
 }
 
 void TextureCacheOpenGL::BufferCopy(Surface& src_surface, Surface& dst_surface) {
+    MICROPROFILE_SCOPE(OpenGL_Texture_Buffer_Copy);
     const auto& src_params = src_surface->GetSurfaceParams();
     const auto& dst_params = dst_surface->GetSurfaceParams();
     UNIMPLEMENTED_IF(src_params.num_levels > 1 || dst_params.num_levels > 1);

src/video_core/renderer_opengl/renderer_opengl.cpp

@@ -101,7 +101,6 @@ RendererOpenGL::RendererOpenGL(Core::Frontend::EmuWindow& emu_window, Core::Syst
 
 RendererOpenGL::~RendererOpenGL() = default;
 
-/// Swap buffers (render frame)
 void RendererOpenGL::SwapBuffers(
     std::optional<std::reference_wrapper<const Tegra::FramebufferConfig>> framebuffer) {
 
@@ -109,6 +108,7 @@ void RendererOpenGL::SwapBuffers(
 
     // Maintain the rasterizer's state as a priority
     OpenGLState prev_state = OpenGLState::GetCurState();
+    state.AllDirty();
     state.Apply();
 
     if (framebuffer) {
@@ -130,6 +130,8 @@ void RendererOpenGL::SwapBuffers(
 
         DrawScreen(render_window.GetFramebufferLayout());
 
+        rasterizer->TickFrame();
+
         render_window.SwapBuffers();
     }
 
@@ -139,6 +141,7 @@ void RendererOpenGL::SwapBuffers(
     system.GetPerfStats().BeginSystemFrame();
 
     // Restore the rasterizer state
+    prev_state.AllDirty();
     prev_state.Apply();
 }
 
@@ -205,6 +208,7 @@ void RendererOpenGL::InitOpenGLObjects() {
     // Link shaders and get variable locations
     shader.CreateFromSource(vertex_shader, nullptr, fragment_shader);
     state.draw.shader_program = shader.handle;
+    state.AllDirty();
     state.Apply();
     uniform_modelview_matrix = glGetUniformLocation(shader.handle, "modelview_matrix");
     uniform_color_texture = glGetUniformLocation(shader.handle, "color_texture");
@@ -262,7 +266,6 @@ void RendererOpenGL::CreateRasterizer() {
     if (rasterizer) {
         return;
     }
-    // Initialize sRGB Usage
     OpenGLState::ClearsRGBUsed();
     rasterizer = std::make_unique<RasterizerOpenGL>(system, emu_window, screen_info);
 }
@@ -338,12 +341,14 @@ void RendererOpenGL::DrawScreenTriangles(const ScreenInfo& screen_info, float x,
     // Workaround brigthness problems in SMO by enabling sRGB in the final output
     // if it has been used in the frame. Needed because of this bug in QT: QTBUG-50987
     state.framebuffer_srgb.enabled = OpenGLState::GetsRGBUsed();
+    state.AllDirty();
     state.Apply();
     glNamedBufferSubData(vertex_buffer.handle, 0, sizeof(vertices), vertices.data());
     glDrawArrays(GL_TRIANGLE_STRIP, 0, 4);
     // Restore default state
     state.framebuffer_srgb.enabled = false;
     state.texture_units[0].texture = 0;
+    state.AllDirty();
     state.Apply();
     // Clear sRGB state for the next frame
     OpenGLState::ClearsRGBUsed();
@@ -388,6 +393,7 @@ void RendererOpenGL::CaptureScreenshot() {
     GLuint old_read_fb = state.draw.read_framebuffer;
     GLuint old_draw_fb = state.draw.draw_framebuffer;
     state.draw.read_framebuffer = state.draw.draw_framebuffer = screenshot_framebuffer.handle;
+    state.AllDirty();
     state.Apply();
 
     Layout::FramebufferLayout layout{renderer_settings.screenshot_framebuffer_layout};
@@ -407,6 +413,7 @@ void RendererOpenGL::CaptureScreenshot() {
     screenshot_framebuffer.Release();
     state.draw.read_framebuffer = old_read_fb;
     state.draw.draw_framebuffer = old_draw_fb;
+    state.AllDirty();
     state.Apply();
     glDeleteRenderbuffers(1, &renderbuffer);
 

src/video_core/renderer_opengl/utils.cpp

@@ -13,29 +13,67 @@
 
 namespace OpenGL {
 
+VertexArrayPushBuffer::VertexArrayPushBuffer() = default;
+
+VertexArrayPushBuffer::~VertexArrayPushBuffer() = default;
+
+void VertexArrayPushBuffer::Setup(GLuint vao_) {
+    vao = vao_;
+    index_buffer = nullptr;
+    vertex_buffers.clear();
+}
+
+void VertexArrayPushBuffer::SetIndexBuffer(const GLuint* buffer) {
+    index_buffer = buffer;
+}
+
+void VertexArrayPushBuffer::SetVertexBuffer(GLuint binding_index, const GLuint* buffer,
+                                            GLintptr offset, GLsizei stride) {
+    vertex_buffers.push_back(Entry{binding_index, buffer, offset, stride});
+}
+
+void VertexArrayPushBuffer::Bind() {
+    if (index_buffer) {
+        glVertexArrayElementBuffer(vao, *index_buffer);
+    }
+
+    // TODO(Rodrigo): Find a way to ARB_multi_bind this
+    for (const auto& entry : vertex_buffers) {
+        glVertexArrayVertexBuffer(vao, entry.binding_index, *entry.buffer, entry.offset,
+                                  entry.stride);
+    }
+}
+
 BindBuffersRangePushBuffer::BindBuffersRangePushBuffer(GLenum target) : target{target} {}
 
 BindBuffersRangePushBuffer::~BindBuffersRangePushBuffer() = default;
 
 void BindBuffersRangePushBuffer::Setup(GLuint first_) {
     first = first_;
-    buffers.clear();
+    buffer_pointers.clear();
     offsets.clear();
     sizes.clear();
 }
 
-void BindBuffersRangePushBuffer::Push(GLuint buffer, GLintptr offset, GLsizeiptr size) {
-    buffers.push_back(buffer);
+void BindBuffersRangePushBuffer::Push(const GLuint* buffer, GLintptr offset, GLsizeiptr size) {
+    buffer_pointers.push_back(buffer);
     offsets.push_back(offset);
     sizes.push_back(size);
 }
 
-void BindBuffersRangePushBuffer::Bind() const {
-    const std::size_t count{buffers.size()};
+void BindBuffersRangePushBuffer::Bind() {
+    // Ensure sizes are valid.
+    const std::size_t count{buffer_pointers.size()};
     DEBUG_ASSERT(count == offsets.size() && count == sizes.size());
     if (count == 0) {
         return;
     }
+
+    // Dereference buffers.
+    buffers.resize(count);
+    std::transform(buffer_pointers.begin(), buffer_pointers.end(), buffers.begin(),
+                   [](const GLuint* pointer) { return *pointer; });
+
     glBindBuffersRange(target, first, static_cast<GLsizei>(count), buffers.data(), offsets.data(),
                        sizes.data());
 }

src/video_core/renderer_opengl/utils.h

@@ -11,20 +11,49 @@
 
 namespace OpenGL {
 
-class BindBuffersRangePushBuffer {
+class VertexArrayPushBuffer final {
 public:
-    BindBuffersRangePushBuffer(GLenum target);
+    explicit VertexArrayPushBuffer();
+    ~VertexArrayPushBuffer();
+
+    void Setup(GLuint vao_);
+
+    void SetIndexBuffer(const GLuint* buffer);
+
+    void SetVertexBuffer(GLuint binding_index, const GLuint* buffer, GLintptr offset,
+                         GLsizei stride);
+
+    void Bind();
+
+private:
+    struct Entry {
+        GLuint binding_index{};
+        const GLuint* buffer{};
+        GLintptr offset{};
+        GLsizei stride{};
+    };
+
+    GLuint vao{};
+    const GLuint* index_buffer{};
+    std::vector<Entry> vertex_buffers;
+};
+
+class BindBuffersRangePushBuffer final {
+public:
+    explicit BindBuffersRangePushBuffer(GLenum target);
     ~BindBuffersRangePushBuffer();
 
     void Setup(GLuint first_);
 
-    void Push(GLuint buffer, GLintptr offset, GLsizeiptr size);
+    void Push(const GLuint* buffer, GLintptr offset, GLsizeiptr size);
 
-    void Bind() const;
+    void Bind();
 
 private:
-    GLenum target;
-    GLuint first;
+    GLenum target{};
+    GLuint first{};
+    std::vector<const GLuint*> buffer_pointers;
+
     std::vector<GLuint> buffers;
     std::vector<GLintptr> offsets;
     std::vector<GLsizeiptr> sizes;

src/video_core/renderer_vulkan/vk_shader_decompiler.cpp

@@ -205,10 +205,6 @@ public:
     }
 
 private:
-    using OperationDecompilerFn = Id (SPIRVDecompiler::*)(Operation);
-    using OperationDecompilersArray =
-        std::array<OperationDecompilerFn, static_cast<std::size_t>(OperationCode::Amount)>;
-
     static constexpr auto INTERNAL_FLAGS_COUNT = static_cast<std::size_t>(InternalFlag::Amount);
 
     void AllocateBindings() {
@@ -430,20 +426,17 @@ private:
         instance_index = DeclareBuiltIn(spv::BuiltIn::InstanceIndex, spv::StorageClass::Input,
                                         t_in_uint, "instance_index");
 
-        bool is_point_size_declared = false;
         bool is_clip_distances_declared = false;
         for (const auto index : ir.GetOutputAttributes()) {
-            if (index == Attribute::Index::PointSize) {
-                is_point_size_declared = true;
-            } else if (index == Attribute::Index::ClipDistances0123 ||
-                       index == Attribute::Index::ClipDistances4567) {
+            if (index == Attribute::Index::ClipDistances0123 ||
+                index == Attribute::Index::ClipDistances4567) {
                 is_clip_distances_declared = true;
             }
         }
 
         std::vector<Id> members;
         members.push_back(t_float4);
-        if (is_point_size_declared) {
+        if (ir.UsesPointSize()) {
             members.push_back(t_float);
         }
         if (is_clip_distances_declared) {
@@ -466,7 +459,7 @@ private:
 
         position_index = MemberDecorateBuiltIn(spv::BuiltIn::Position, "position", true);
         point_size_index =
-            MemberDecorateBuiltIn(spv::BuiltIn::PointSize, "point_size", is_point_size_declared);
+            MemberDecorateBuiltIn(spv::BuiltIn::PointSize, "point_size", ir.UsesPointSize());
         clip_distances_index = MemberDecorateBuiltIn(spv::BuiltIn::ClipDistance, "clip_distances",
                                                      is_clip_distances_declared);
 
@@ -712,7 +705,8 @@ private:
                 case Attribute::Index::Position:
                     return AccessElement(t_out_float, per_vertex, position_index,
                                          abuf->GetElement());
-                case Attribute::Index::PointSize:
+                case Attribute::Index::LayerViewportPointSize:
+                    UNIMPLEMENTED_IF(abuf->GetElement() != 3);
                     return AccessElement(t_out_float, per_vertex, point_size_index);
                 case Attribute::Index::ClipDistances0123:
                     return AccessElement(t_out_float, per_vertex, clip_distances_index,
@@ -806,12 +800,7 @@ private:
         return {};
     }
 
-    Id LogicalAll2(Operation operation) {
-        UNIMPLEMENTED();
-        return {};
-    }
-
-    Id LogicalAny2(Operation operation) {
+    Id LogicalAnd2(Operation operation) {
         UNIMPLEMENTED();
         return {};
     }
@@ -1208,7 +1197,7 @@ private:
         return {};
     }
 
-    static constexpr OperationDecompilersArray operation_decompilers = {
+    static constexpr std::array operation_decompilers = {
         &SPIRVDecompiler::Assign,
 
         &SPIRVDecompiler::Ternary<&Module::OpSelect, Type::Float, Type::Bool, Type::Float,
@@ -1293,8 +1282,7 @@ private:
         &SPIRVDecompiler::Binary<&Module::OpLogicalNotEqual, Type::Bool>,
         &SPIRVDecompiler::Unary<&Module::OpLogicalNot, Type::Bool>,
         &SPIRVDecompiler::LogicalPick2,
-        &SPIRVDecompiler::LogicalAll2,
-        &SPIRVDecompiler::LogicalAny2,
+        &SPIRVDecompiler::LogicalAnd2,
 
         &SPIRVDecompiler::Binary<&Module::OpFOrdLessThan, Type::Bool, Type::Float>,
         &SPIRVDecompiler::Binary<&Module::OpFOrdEqual, Type::Bool, Type::Float>,
@@ -1359,6 +1347,7 @@ private:
         &SPIRVDecompiler::WorkGroupId<1>,
         &SPIRVDecompiler::WorkGroupId<2>,
     };
+    static_assert(operation_decompilers.size() == static_cast<std::size_t>(OperationCode::Amount));
 
     const VKDevice& device;
     const ShaderIR& ir;

src/video_core/shader/control_flow.cpp

@@ -15,7 +15,7 @@
 #include "video_core/shader/shader_ir.h"
 
 namespace VideoCommon::Shader {
-
+namespace {
 using Tegra::Shader::Instruction;
 using Tegra::Shader::OpCode;
 
@@ -29,8 +29,7 @@ struct Query {
 
 struct BlockStack {
     BlockStack() = default;
-    BlockStack(const BlockStack& b) = default;
-    BlockStack(const Query& q) : ssy_stack{q.ssy_stack}, pbk_stack{q.pbk_stack} {}
+    explicit BlockStack(const Query& q) : ssy_stack{q.ssy_stack}, pbk_stack{q.pbk_stack} {}
     std::stack<u32> ssy_stack{};
     std::stack<u32> pbk_stack{};
 };
@@ -58,7 +57,7 @@ struct BlockInfo {
 struct CFGRebuildState {
     explicit CFGRebuildState(const ProgramCode& program_code, const std::size_t program_size,
                              const u32 start)
-        : program_code{program_code}, program_size{program_size}, start{start} {}
+        : start{start}, program_code{program_code}, program_size{program_size} {}
 
     u32 start{};
     std::vector<BlockInfo> block_info{};
@@ -85,7 +84,7 @@ std::pair<BlockCollision, u32> TryGetBlock(CFGRebuildState& state, u32 address)
             return {BlockCollision::Inside, index};
         }
     }
-    return {BlockCollision::None, -1};
+    return {BlockCollision::None, 0xFFFFFFFF};
 }
 
 struct ParseInfo {
@@ -365,27 +364,29 @@ bool TryQuery(CFGRebuildState& state) {
         const auto gather_end = labels.upper_bound(block.end);
         while (gather_start != gather_end) {
             cc.push(gather_start->second);
-            gather_start++;
+            ++gather_start;
         }
     };
     if (state.queries.empty()) {
         return false;
     }
+
     Query& q = state.queries.front();
     const u32 block_index = state.registered[q.address];
     BlockInfo& block = state.block_info[block_index];
-    // If the block is visted, check if the stacks match, else gather the ssy/pbk
+    // If the block is visited, check if the stacks match, else gather the ssy/pbk
     // labels into the current stack and look if the branch at the end of the block
     // consumes a label. Schedule new queries accordingly
     if (block.visited) {
         BlockStack& stack = state.stacks[q.address];
-        const bool all_okay = (stack.ssy_stack.size() == 0 || q.ssy_stack == stack.ssy_stack) &&
-                              (stack.pbk_stack.size() == 0 || q.pbk_stack == stack.pbk_stack);
+        const bool all_okay = (stack.ssy_stack.empty() || q.ssy_stack == stack.ssy_stack) &&
+                              (stack.pbk_stack.empty() || q.pbk_stack == stack.pbk_stack);
         state.queries.pop_front();
         return all_okay;
     }
     block.visited = true;
-    state.stacks[q.address] = BlockStack{q};
+    state.stacks.insert_or_assign(q.address, BlockStack{q});
+
     Query q2(q);
     state.queries.pop_front();
     gather_labels(q2.ssy_stack, state.ssy_labels, block);
@@ -394,6 +395,7 @@ bool TryQuery(CFGRebuildState& state) {
         q2.address = block.end + 1;
         state.queries.push_back(q2);
     }
+
     Query conditional_query{q2};
     if (block.branch.is_sync) {
         if (block.branch.address == unassigned_branch) {
@@ -408,13 +410,15 @@ bool TryQuery(CFGRebuildState& state) {
         conditional_query.pbk_stack.pop();
     }
     conditional_query.address = block.branch.address;
-    state.queries.push_back(conditional_query);
+    state.queries.push_back(std::move(conditional_query));
     return true;
 }
+} // Anonymous namespace
 
-std::optional<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code, u32 program_size,
-                                              u32 start_address) {
+std::optional<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code,
+                                              std::size_t program_size, u32 start_address) {
     CFGRebuildState state{program_code, program_size, start_address};
+
     // Inspect Code and generate blocks
     state.labels.clear();
     state.labels.emplace(start_address);
@@ -424,10 +428,9 @@ std::optional<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code, u
             return {};
         }
     }
+
     // Decompile Stacks
-    Query start_query{};
-    start_query.address = state.start;
-    state.queries.push_back(start_query);
+    state.queries.push_back(Query{state.start, {}, {}});
     bool decompiled = true;
     while (!state.queries.empty()) {
         if (!TryQuery(state)) {
@@ -435,14 +438,15 @@ std::optional<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code, u
             break;
         }
     }
+
     // Sort and organize results
     std::sort(state.block_info.begin(), state.block_info.end(),
-              [](const BlockInfo& a, const BlockInfo& b) -> bool { return a.start < b.start; });
+              [](const BlockInfo& a, const BlockInfo& b) { return a.start < b.start; });
     ShaderCharacteristics result_out{};
     result_out.decompilable = decompiled;
     result_out.start = start_address;
     result_out.end = start_address;
-    for (auto& block : state.block_info) {
+    for (const auto& block : state.block_info) {
         ShaderBlock new_block{};
         new_block.start = block.start;
         new_block.end = block.end;
@@ -457,8 +461,9 @@ std::optional<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code, u
     }
     if (result_out.decompilable) {
         result_out.labels = std::move(state.labels);
-        return {result_out};
+        return {std::move(result_out)};
     }
+
     // If it's not decompilable, merge the unlabelled blocks together
     auto back = result_out.blocks.begin();
     auto next = std::next(back);
@@ -469,8 +474,8 @@ std::optional<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code, u
             continue;
         }
         back = next;
-        next++;
+        ++next;
     }
-    return {result_out};
+    return {std::move(result_out)};
 }
 } // namespace VideoCommon::Shader

src/video_core/shader/control_flow.h

@@ -4,7 +4,6 @@
 
 #pragma once
 
-#include <cstring>
 #include <list>
 #include <optional>
 #include <unordered_set>
@@ -26,27 +25,44 @@ struct Condition {
     bool IsUnconditional() const {
         return predicate == Pred::UnusedIndex && cc == ConditionCode::T;
     }
+
     bool operator==(const Condition& other) const {
         return std::tie(predicate, cc) == std::tie(other.predicate, other.cc);
     }
+
+    bool operator!=(const Condition& other) const {
+        return !operator==(other);
+    }
 };
 
 struct ShaderBlock {
-    u32 start{};
-    u32 end{};
-    bool ignore_branch{};
     struct Branch {
         Condition cond{};
         bool kills{};
         s32 address{};
+
         bool operator==(const Branch& b) const {
             return std::tie(cond, kills, address) == std::tie(b.cond, b.kills, b.address);
         }
-    } branch{};
+
+        bool operator!=(const Branch& b) const {
+            return !operator==(b);
+        }
+    };
+
+    u32 start{};
+    u32 end{};
+    bool ignore_branch{};
+    Branch branch{};
+
     bool operator==(const ShaderBlock& sb) const {
         return std::tie(start, end, ignore_branch, branch) ==
                std::tie(sb.start, sb.end, sb.ignore_branch, sb.branch);
     }
+
+    bool operator!=(const ShaderBlock& sb) const {
+        return !operator==(sb);
+    }
 };
 
 struct ShaderCharacteristics {
@@ -57,7 +73,7 @@ struct ShaderCharacteristics {
     std::unordered_set<u32> labels{};
 };
 
-std::optional<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code, u32 program_size,
-                                              u32 start_address);
+std::optional<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code,
+                                              std::size_t program_size, u32 start_address);
 
 } // namespace VideoCommon::Shader

src/video_core/shader/decode.cpp

@@ -46,15 +46,15 @@ void ShaderIR::Decode() {
         coverage_end = shader_info.end;
         if (shader_info.decompilable) {
             disable_flow_stack = true;
-            const auto insert_block = ([this](NodeBlock& nodes, u32 label) {
-                if (label == exit_branch) {
+            const auto insert_block = [this](NodeBlock& nodes, u32 label) {
+                if (label == static_cast<u32>(exit_branch)) {
                     return;
                 }
                 basic_blocks.insert({label, nodes});
-            });
+            };
             const auto& blocks = shader_info.blocks;
             NodeBlock current_block;
-            u32 current_label = exit_branch;
+            u32 current_label = static_cast<u32>(exit_branch);
             for (auto& block : blocks) {
                 if (shader_info.labels.count(block.start) != 0) {
                     insert_block(current_block, current_label);
@@ -103,7 +103,7 @@ void ShaderIR::DecodeRangeInner(NodeBlock& bb, u32 begin, u32 end) {
 }
 
 void ShaderIR::InsertControlFlow(NodeBlock& bb, const ShaderBlock& block) {
-    const auto apply_conditions = ([&](const Condition& cond, Node n) -> Node {
+    const auto apply_conditions = [&](const Condition& cond, Node n) -> Node {
         Node result = n;
         if (cond.cc != ConditionCode::T) {
             result = Conditional(GetConditionCode(cond.cc), {result});
@@ -117,7 +117,7 @@ void ShaderIR::InsertControlFlow(NodeBlock& bb, const ShaderBlock& block) {
             result = Conditional(GetPredicate(pred, is_neg), {result});
         }
         return result;
-    });
+    };
     if (block.branch.address < 0) {
         if (block.branch.kills) {
             Node n = Operation(OperationCode::Discard);

src/video_core/shader/decode/arithmetic.cpp

@@ -42,11 +42,14 @@ u32 ShaderIR::DecodeArithmetic(NodeBlock& bb, u32 pc) {
     case OpCode::Id::FMUL_R:
     case OpCode::Id::FMUL_IMM: {
         // FMUL does not have 'abs' bits and only the second operand has a 'neg' bit.
-        UNIMPLEMENTED_IF_MSG(instr.fmul.tab5cb8_2 != 0, "FMUL tab5cb8_2({}) is not implemented",
-                             instr.fmul.tab5cb8_2.Value());
-        UNIMPLEMENTED_IF_MSG(
-            instr.fmul.tab5c68_0 != 1, "FMUL tab5cb8_0({}) is not implemented",
-            instr.fmul.tab5c68_0.Value()); // SMO typical sends 1 here which seems to be the default
+        if (instr.fmul.tab5cb8_2 != 0) {
+            LOG_WARNING(HW_GPU, "FMUL tab5cb8_2({}) is not implemented",
+                        instr.fmul.tab5cb8_2.Value());
+        }
+        if (instr.fmul.tab5c68_0 != 1) {
+            LOG_WARNING(HW_GPU, "FMUL tab5cb8_0({}) is not implemented",
+                        instr.fmul.tab5c68_0.Value());
+        }
 
         op_b = GetOperandAbsNegFloat(op_b, false, instr.fmul.negate_b);