20 files changed, 154 insertions, 167 deletions

diff --git a/src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp b/src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp
index e6a976714..18c5324a9 100644
--- a/src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp
+++ b/src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp
@@ -40,7 +40,8 @@ void nvdisp_disp0::OnClose(DeviceFD fd) {}
 
 void nvdisp_disp0::flip(u32 buffer_handle, u32 offset, android::PixelFormat format, u32 width,
                         u32 height, u32 stride, android::BufferTransformFlags transform,
-                        const Common::Rectangle<int>& crop_rect) {
+                        const Common::Rectangle<int>& crop_rect,
+                        std::array<Service::Nvidia::NvFence, 4>& fences, u32 num_fences) {
     const VAddr addr = nvmap.GetHandleAddress(buffer_handle);
     LOG_TRACE(Service,
               "Drawing from address {:X} offset {:08X} Width {} Height {} Stride {} Format {}",
@@ -50,7 +51,7 @@ void nvdisp_disp0::flip(u32 buffer_handle, u32 offset, android::PixelFormat form
                                                stride, format, transform, crop_rect};
 
     system.GetPerfStats().EndSystemFrame();
-    system.GPU().RequestSwapBuffers(&framebuffer, nullptr, 0);
+    system.GPU().RequestSwapBuffers(&framebuffer, fences, num_fences);
     system.SpeedLimiter().DoSpeedLimiting(system.CoreTiming().GetGlobalTimeUs());
     system.GetPerfStats().BeginSystemFrame();
 }
diff --git a/src/core/hle/service/nvdrv/devices/nvdisp_disp0.h b/src/core/hle/service/nvdrv/devices/nvdisp_disp0.h
index 1ca9b2e74..04217ab12 100644
--- a/src/core/hle/service/nvdrv/devices/nvdisp_disp0.h
+++ b/src/core/hle/service/nvdrv/devices/nvdisp_disp0.h
@@ -38,7 +38,8 @@ public:
     /// Performs a screen flip, drawing the buffer pointed to by the handle.
     void flip(u32 buffer_handle, u32 offset, android::PixelFormat format, u32 width, u32 height,
               u32 stride, android::BufferTransformFlags transform,
-              const Common::Rectangle<int>& crop_rect);
+              const Common::Rectangle<int>& crop_rect,
+              std::array<Service::Nvidia::NvFence, 4>& fences, u32 num_fences);
 
     Kernel::KEvent* QueryEvent(u32 event_id) override;
 
diff --git a/src/core/hle/service/nvflinger/nvflinger.cpp b/src/core/hle/service/nvflinger/nvflinger.cpp
index aa112021d..4658f1e8b 100644
--- a/src/core/hle/service/nvflinger/nvflinger.cpp
+++ b/src/core/hle/service/nvflinger/nvflinger.cpp
@@ -269,17 +269,6 @@ void NVFlinger::Compose() {
             return; // We are likely shutting down
         }
 
-        auto& syncpoint_manager = system.Host1x().GetSyncpointManager();
-        const auto& multi_fence = buffer.fence;
-        guard->unlock();
-        for (u32 fence_id = 0; fence_id < multi_fence.num_fences; fence_id++) {
-            const auto& fence = multi_fence.fences[fence_id];
-            syncpoint_manager.WaitGuest(fence.id, fence.value);
-        }
-        guard->lock();
-
-        MicroProfileFlip();
-
         // Now send the buffer to the GPU for drawing.
         // TODO(Subv): Support more than just disp0. The display device selection is probably based
         // on which display we're drawing (Default, Internal, External, etc)
@@ -293,8 +282,10 @@ void NVFlinger::Compose() {
 
         nvdisp->flip(igbp_buffer.BufferId(), igbp_buffer.Offset(), igbp_buffer.ExternalFormat(),
                      igbp_buffer.Width(), igbp_buffer.Height(), igbp_buffer.Stride(),
-                     static_cast<android::BufferTransformFlags>(buffer.transform), crop_rect);
+                     static_cast<android::BufferTransformFlags>(buffer.transform), crop_rect,
+                     buffer.fence.fences, buffer.fence.num_fences);
 
+        MicroProfileFlip();
         guard->lock();
 
         swap_interval = buffer.swap_interval;
diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h
index 6b6764d72..e55cac0d6 100644
--- a/src/video_core/buffer_cache/buffer_cache.h
+++ b/src/video_core/buffer_cache/buffer_cache.h
@@ -826,6 +826,19 @@ void BufferCache<P>::CommitAsyncFlushesHigh() {
     const bool is_accuracy_normal =
         Settings::values.gpu_accuracy.GetValue() == Settings::GPUAccuracy::Normal;
 
+    auto it = committed_ranges.begin();
+    while (it != committed_ranges.end()) {
+        auto& current_intervals = *it;
+        auto next_it = std::next(it);
+        while (next_it != committed_ranges.end()) {
+            for (auto& interval : *next_it) {
+                current_intervals.subtract(interval);
+            }
+            next_it++;
+        }
+        it++;
+    }
+
     boost::container::small_vector<std::pair<BufferCopy, BufferId>, 1> downloads;
     u64 total_size_bytes = 0;
     u64 largest_copy = 0;
diff --git a/src/video_core/dma_pusher.cpp b/src/video_core/dma_pusher.cpp
index b01f04d0c..9835e3ac1 100644
--- a/src/video_core/dma_pusher.cpp
+++ b/src/video_core/dma_pusher.cpp
@@ -24,8 +24,6 @@ MICROPROFILE_DEFINE(DispatchCalls, "GPU", "Execute command buffer", MP_RGB(128,
 void DmaPusher::DispatchCalls() {
     MICROPROFILE_SCOPE(DispatchCalls);
 
-    gpu.SyncGuestHost();
-
     dma_pushbuffer_subindex = 0;
 
     dma_state.is_last_call = true;
@@ -36,7 +34,6 @@ void DmaPusher::DispatchCalls() {
         }
     }
     gpu.FlushCommands();
-    gpu.SyncGuestHost();
     gpu.OnCommandListEnd();
 }
 
diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp
index 3a4646289..950c70dcd 100644
--- a/src/video_core/engines/maxwell_3d.cpp
+++ b/src/video_core/engines/maxwell_3d.cpp
@@ -242,6 +242,9 @@ void Maxwell3D::ProcessMethodCall(u32 method, u32 argument, u32 nonshadow_argume
         return;
     case MAXWELL3D_REG_INDEX(fragment_barrier):
         return rasterizer->FragmentBarrier();
+    case MAXWELL3D_REG_INDEX(invalidate_texture_data_cache):
+        rasterizer->InvalidateGPUCache();
+        return rasterizer->WaitForIdle();
     case MAXWELL3D_REG_INDEX(tiled_cache_barrier):
         return rasterizer->TiledCacheBarrier();
     }
@@ -472,10 +475,25 @@ void Maxwell3D::ProcessQueryGet() {
 
     switch (regs.query.query_get.operation) {
     case Regs::QueryOperation::Release:
-        if (regs.query.query_get.fence == 1) {
-            rasterizer->SignalSemaphore(regs.query.QueryAddress(), regs.query.query_sequence);
+        if (regs.query.query_get.fence == 1 || regs.query.query_get.short_query != 0) {
+            const GPUVAddr sequence_address{regs.query.QueryAddress()};
+            const u32 payload = regs.query.query_sequence;
+            std::function<void()> operation([this, sequence_address, payload] {
+                memory_manager.Write<u32>(sequence_address, payload);
+            });
+            rasterizer->SignalFence(std::move(operation));
         } else {
-            StampQueryResult(regs.query.query_sequence, regs.query.query_get.short_query == 0);
+            struct LongQueryResult {
+                u64_le value;
+                u64_le timestamp;
+            };
+            const GPUVAddr sequence_address{regs.query.QueryAddress()};
+            const u32 payload = regs.query.query_sequence;
+            std::function<void()> operation([this, sequence_address, payload] {
+                LongQueryResult query_result{payload, system.GPU().GetTicks()};
+                memory_manager.WriteBlock(sequence_address, &query_result, sizeof(query_result));
+            });
+            rasterizer->SignalFence(std::move(operation));
         }
         break;
     case Regs::QueryOperation::Acquire:
diff --git a/src/video_core/engines/puller.cpp b/src/video_core/engines/puller.cpp
index 8c17639e4..dd9494efa 100644
--- a/src/video_core/engines/puller.cpp
+++ b/src/video_core/engines/puller.cpp
@@ -79,12 +79,15 @@ void Puller::ProcessSemaphoreTriggerMethod() {
             u64 timestamp;
         };
 
-        Block block{};
-        block.sequence = regs.semaphore_sequence;
-        // TODO(Kmather73): Generate a real GPU timestamp and write it here instead of
-        // CoreTiming
-        block.timestamp = gpu.GetTicks();
-        memory_manager.WriteBlock(regs.semaphore_address.SemaphoreAddress(), &block, sizeof(block));
+        const GPUVAddr sequence_address{regs.semaphore_address.SemaphoreAddress()};
+        const u32 payload = regs.semaphore_sequence;
+        std::function<void()> operation([this, sequence_address, payload] {
+            Block block{};
+            block.sequence = payload;
+            block.timestamp = gpu.GetTicks();
+            memory_manager.WriteBlock(sequence_address, &block, sizeof(block));
+        });
+        rasterizer->SignalFence(std::move(operation));
     } else {
         do {
             const u32 word{memory_manager.Read<u32>(regs.semaphore_address.SemaphoreAddress())};
@@ -94,6 +97,7 @@ void Puller::ProcessSemaphoreTriggerMethod() {
                 regs.acquire_active = true;
                 regs.acquire_mode = false;
                 if (word != regs.acquire_value) {
+                    rasterizer->ReleaseFences();
                     std::this_thread::sleep_for(std::chrono::milliseconds(1));
                     continue;
                 }
@@ -101,11 +105,13 @@ void Puller::ProcessSemaphoreTriggerMethod() {
                 regs.acquire_active = true;
                 regs.acquire_mode = true;
                 if (word < regs.acquire_value) {
+                    rasterizer->ReleaseFences();
                     std::this_thread::sleep_for(std::chrono::milliseconds(1));
                     continue;
                 }
             } else if (op == GpuSemaphoreOperation::AcquireMask) {
-                if (word & regs.semaphore_sequence == 0) {
+                if (word && regs.semaphore_sequence == 0) {
+                    rasterizer->ReleaseFences();
                     std::this_thread::sleep_for(std::chrono::milliseconds(1));
                     continue;
                 }
@@ -117,16 +123,23 @@ void Puller::ProcessSemaphoreTriggerMethod() {
 }
 
 void Puller::ProcessSemaphoreRelease() {
-    rasterizer->SignalSemaphore(regs.semaphore_address.SemaphoreAddress(), regs.semaphore_release);
+    const GPUVAddr sequence_address{regs.semaphore_address.SemaphoreAddress()};
+    const u32 payload = regs.semaphore_release;
+    std::function<void()> operation([this, sequence_address, payload] {
+        memory_manager.Write<u32>(sequence_address, payload);
+    });
+    rasterizer->SignalFence(std::move(operation));
 }
 
 void Puller::ProcessSemaphoreAcquire() {
-    const u32 word = memory_manager.Read<u32>(regs.semaphore_address.SemaphoreAddress());
+    u32 word = memory_manager.Read<u32>(regs.semaphore_address.SemaphoreAddress());
     const auto value = regs.semaphore_acquire;
-    std::this_thread::sleep_for(std::chrono::milliseconds(5));
-    if (word != value) {
+    while (word != value) {
         regs.acquire_active = true;
         regs.acquire_value = value;
+        std::this_thread::sleep_for(std::chrono::milliseconds(1));
+        rasterizer->ReleaseFences();
+        word = memory_manager.Read<u32>(regs.semaphore_address.SemaphoreAddress());
         // TODO(kemathe73) figure out how to do the acquire_timeout
         regs.acquire_mode = false;
         regs.acquire_source = false;
@@ -147,9 +160,9 @@ void Puller::CallPullerMethod(const MethodCall& method_call) {
     case BufferMethods::SemaphoreAddressHigh:
     case BufferMethods::SemaphoreAddressLow:
     case BufferMethods::SemaphoreSequencePayload:
-    case BufferMethods::WrcacheFlush:
     case BufferMethods::SyncpointPayload:
         break;
+    case BufferMethods::WrcacheFlush:
     case BufferMethods::RefCnt:
         rasterizer->SignalReference();
         break;
@@ -173,7 +186,7 @@ void Puller::CallPullerMethod(const MethodCall& method_call) {
     }
     case BufferMethods::MemOpB: {
         // Implement this better.
-        rasterizer->SyncGuestHost();
+        rasterizer->InvalidateGPUCache();
         break;
     }
     case BufferMethods::MemOpC:
diff --git a/src/video_core/fence_manager.h b/src/video_core/fence_manager.h
index 03a70e5e0..c390ac91b 100644
--- a/src/video_core/fence_manager.h
+++ b/src/video_core/fence_manager.h
@@ -5,6 +5,8 @@
 
 #include <algorithm>
 #include <cstring>
+#include <deque>
+#include <functional>
 #include <memory>
 #include <queue>
 
@@ -19,28 +21,7 @@ namespace VideoCommon {
 
 class FenceBase {
 public:
-    explicit FenceBase(u32 payload_, bool is_stubbed_)
-        : address{}, payload{payload_}, is_semaphore{false}, is_stubbed{is_stubbed_} {}
-
-    explicit FenceBase(u8* address_, u32 payload_, bool is_stubbed_)
-        : address{address_}, payload{payload_}, is_semaphore{true}, is_stubbed{is_stubbed_} {}
-
-    u8* GetAddress() const {
-        return address;
-    }
-
-    u32 GetPayload() const {
-        return payload;
-    }
-
-    bool IsSemaphore() const {
-        return is_semaphore;
-    }
-
-private:
-    u8* address;
-    u32 payload;
-    bool is_semaphore;
+    explicit FenceBase(bool is_stubbed_) : is_stubbed{is_stubbed_} {}
 
 protected:
     bool is_stubbed;
@@ -60,31 +41,28 @@ public:
         buffer_cache.AccumulateFlushes();
     }
 
-    void SignalSemaphore(u8* addr, u32 value) {
+    void SyncOperation(std::function<void()>&& func) {
+        uncommitted_operations.emplace_back(std::move(func));
+    }
+
+    void SignalFence(std::function<void()>&& func) {
         TryReleasePendingFences();
         const bool should_flush = ShouldFlush();
         CommitAsyncFlushes();
-        TFence new_fence = CreateFence(addr, value, !should_flush);
+        uncommitted_operations.emplace_back(std::move(func));
+        CommitOperations();
+        TFence new_fence = CreateFence(!should_flush);
         fences.push(new_fence);
         QueueFence(new_fence);
         if (should_flush) {
             rasterizer.FlushCommands();
         }
-        rasterizer.SyncGuestHost();
     }
 
     void SignalSyncPoint(u32 value) {
         syncpoint_manager.IncrementGuest(value);
-        TryReleasePendingFences();
-        const bool should_flush = ShouldFlush();
-        CommitAsyncFlushes();
-        TFence new_fence = CreateFence(value, !should_flush);
-        fences.push(new_fence);
-        QueueFence(new_fence);
-        if (should_flush) {
-            rasterizer.FlushCommands();
-        }
-        rasterizer.SyncGuestHost();
+        std::function<void()> func([this, value] { syncpoint_manager.IncrementHost(value); });
+        SignalFence(std::move(func));
     }
 
     void WaitPendingFences() {
@@ -94,12 +72,10 @@ public:
                 WaitFence(current_fence);
             }
             PopAsyncFlushes();
-            if (current_fence->IsSemaphore()) {
-                char* address = reinterpret_cast<char*>(current_fence->GetAddress());
-                auto payload = current_fence->GetPayload();
-                std::memcpy(address, &payload, sizeof(payload));
-            } else {
-                syncpoint_manager.IncrementHost(current_fence->GetPayload());
+            auto operations = std::move(pending_operations.front());
+            pending_operations.pop_front();
+            for (auto& operation : operations) {
+                operation();
             }
             PopFence();
         }
@@ -114,11 +90,9 @@ protected:
 
     virtual ~FenceManager() = default;
 
-    /// Creates a Sync Point Fence Interface, does not create a backend fence if 'is_stubbed' is
+    /// Creates a Fence Interface, does not create a backend fence if 'is_stubbed' is
     /// true
-    virtual TFence CreateFence(u32 value, bool is_stubbed) = 0;
-    /// Creates a Semaphore Fence Interface, does not create a backend fence if 'is_stubbed' is true
-    virtual TFence CreateFence(u8* addr, u32 value, bool is_stubbed) = 0;
+    virtual TFence CreateFence(bool is_stubbed) = 0;
     /// Queues a fence into the backend if the fence isn't stubbed.
     virtual void QueueFence(TFence& fence) = 0;
     /// Notifies that the backend fence has been signaled/reached in host GPU.
@@ -141,12 +115,10 @@ private:
                 return;
             }
             PopAsyncFlushes();
-            if (current_fence->IsSemaphore()) {
-                char* address = reinterpret_cast<char*>(current_fence->GetAddress());
-                const auto payload = current_fence->GetPayload();
-                std::memcpy(address, &payload, sizeof(payload));
-            } else {
-                syncpoint_manager.IncrementHost(current_fence->GetPayload());
+            auto operations = std::move(pending_operations.front());
+            pending_operations.pop_front();
+            for (auto& operation : operations) {
+                operation();
             }
             PopFence();
         }
@@ -165,16 +137,20 @@ private:
     }
 
     void PopAsyncFlushes() {
-        std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
-        texture_cache.PopAsyncFlushes();
-        buffer_cache.PopAsyncFlushes();
+        {
+            std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
+            texture_cache.PopAsyncFlushes();
+            buffer_cache.PopAsyncFlushes();
+        }
         query_cache.PopAsyncFlushes();
     }
 
     void CommitAsyncFlushes() {
-        std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
-        texture_cache.CommitAsyncFlushes();
-        buffer_cache.CommitAsyncFlushes();
+        {
+            std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
+            texture_cache.CommitAsyncFlushes();
+            buffer_cache.CommitAsyncFlushes();
+        }
         query_cache.CommitAsyncFlushes();
     }
 
@@ -183,7 +159,13 @@ private:
         fences.pop();
     }
 
+    void CommitOperations() {
+        pending_operations.emplace_back(std::move(uncommitted_operations));
+    }
+
     std::queue<TFence> fences;
+    std::deque<std::function<void()>> uncommitted_operations;
+    std::deque<std::deque<std::function<void()>>> pending_operations;
 
     DelayedDestructionRing<TFence, 6> delayed_destruction_ring;
 };
diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp
index a1d19b1c8..d7a3dd96b 100644
--- a/src/video_core/gpu.cpp
+++ b/src/video_core/gpu.cpp
@@ -93,16 +93,13 @@ struct GPU::Impl {
     }
 
     /// Synchronizes CPU writes with Host GPU memory.
-    void SyncGuestHost() {
-        rasterizer->SyncGuestHost();
+    void InvalidateGPUCache() {
+        rasterizer->InvalidateGPUCache();
     }
 
     /// Signal the ending of command list.
     void OnCommandListEnd() {
-        if (is_async) {
-            // This command only applies to asynchronous GPU mode
-            gpu_thread.OnCommandListEnd();
-        }
+        gpu_thread.OnCommandListEnd();
     }
 
     /// Request a host GPU memory flush from the CPU.
@@ -296,7 +293,7 @@ struct GPU::Impl {
     }
 
     void RequestSwapBuffers(const Tegra::FramebufferConfig* framebuffer,
-                            Service::Nvidia::NvFence* fences, size_t num_fences) {
+                            std::array<Service::Nvidia::NvFence, 4>& fences, size_t num_fences) {
         size_t current_request_counter{};
         {
             std::unique_lock<std::mutex> lk(request_swap_mutex);
@@ -412,8 +409,8 @@ void GPU::FlushCommands() {
     impl->FlushCommands();
 }
 
-void GPU::SyncGuestHost() {
-    impl->SyncGuestHost();
+void GPU::InvalidateGPUCache() {
+    impl->InvalidateGPUCache();
 }
 
 void GPU::OnCommandListEnd() {
@@ -488,7 +485,7 @@ const VideoCore::ShaderNotify& GPU::ShaderNotify() const {
 }
 
 void GPU::RequestSwapBuffers(const Tegra::FramebufferConfig* framebuffer,
-                             Service::Nvidia::NvFence* fences, size_t num_fences) {
+                             std::array<Service::Nvidia::NvFence, 4>& fences, size_t num_fences) {
     impl->RequestSwapBuffers(framebuffer, fences, num_fences);
 }
 
diff --git a/src/video_core/gpu.h b/src/video_core/gpu.h
index 655373b33..0a4a8b14f 100644
--- a/src/video_core/gpu.h
+++ b/src/video_core/gpu.h
@@ -110,7 +110,7 @@ public:
     /// Flush all current written commands into the host GPU for execution.
     void FlushCommands();
     /// Synchronizes CPU writes with Host GPU memory.
-    void SyncGuestHost();
+    void InvalidateGPUCache();
     /// Signal the ending of command list.
     void OnCommandListEnd();
 
@@ -180,7 +180,7 @@ public:
     void RendererFrameEndNotify();
 
     void RequestSwapBuffers(const Tegra::FramebufferConfig* framebuffer,
-                            Service::Nvidia::NvFence* fences, size_t num_fences);
+                            std::array<Service::Nvidia::NvFence, 4>& fences, size_t num_fences);
 
     /// Performs any additional setup necessary in order to begin GPU emulation.
     /// This can be used to launch any necessary threads and register any necessary
diff --git a/src/video_core/gpu_thread.cpp b/src/video_core/gpu_thread.cpp
index 2c03545bf..1bd477011 100644
--- a/src/video_core/gpu_thread.cpp
+++ b/src/video_core/gpu_thread.cpp
@@ -98,7 +98,7 @@ void ThreadManager::FlushRegion(VAddr addr, u64 size) {
 }
 
 void ThreadManager::TickGPU() {
-    PushCommand(GPUTickCommand(), true);
+    PushCommand(GPUTickCommand());
 }
 
 void ThreadManager::InvalidateRegion(VAddr addr, u64 size) {
diff --git a/src/video_core/rasterizer_interface.h b/src/video_core/rasterizer_interface.h
index 5362aafb6..cb07f3d38 100644
--- a/src/video_core/rasterizer_interface.h
+++ b/src/video_core/rasterizer_interface.h
@@ -62,7 +62,10 @@ public:
     virtual void DisableGraphicsUniformBuffer(size_t stage, u32 index) = 0;
 
     /// Signal a GPU based semaphore as a fence
-    virtual void SignalSemaphore(GPUVAddr addr, u32 value) = 0;
+    virtual void SignalFence(std::function<void()>&& func) = 0;
+
+    /// Send an operation to be done after a certain amount of flushes.
+    virtual void SyncOperation(std::function<void()>&& func) = 0;
 
     /// Signal a GPU based syncpoint as a fence
     virtual void SignalSyncPoint(u32 value) = 0;
@@ -89,7 +92,7 @@ public:
     virtual void OnCPUWrite(VAddr addr, u64 size) = 0;
 
     /// Sync memory between guest and host.
-    virtual void SyncGuestHost() = 0;
+    virtual void InvalidateGPUCache() = 0;
 
     /// Unmap memory range
     virtual void UnmapMemory(VAddr addr, u64 size) = 0;
diff --git a/src/video_core/renderer_opengl/gl_fence_manager.cpp b/src/video_core/renderer_opengl/gl_fence_manager.cpp
index c76446b60..91463f854 100644
--- a/src/video_core/renderer_opengl/gl_fence_manager.cpp
+++ b/src/video_core/renderer_opengl/gl_fence_manager.cpp
@@ -10,10 +10,7 @@
 
 namespace OpenGL {
 
-GLInnerFence::GLInnerFence(u32 payload_, bool is_stubbed_) : FenceBase{payload_, is_stubbed_} {}
-
-GLInnerFence::GLInnerFence(u8* address_, u32 payload_, bool is_stubbed_)
-    : FenceBase{address_, payload_, is_stubbed_} {}
+GLInnerFence::GLInnerFence(bool is_stubbed_) : FenceBase{is_stubbed_} {}
 
 GLInnerFence::~GLInnerFence() = default;
 
@@ -48,12 +45,8 @@ FenceManagerOpenGL::FenceManagerOpenGL(VideoCore::RasterizerInterface& rasterize
                                        BufferCache& buffer_cache_, QueryCache& query_cache_)
     : GenericFenceManager{rasterizer_, gpu_, texture_cache_, buffer_cache_, query_cache_} {}
 
-Fence FenceManagerOpenGL::CreateFence(u32 value, bool is_stubbed) {
-    return std::make_shared<GLInnerFence>(value, is_stubbed);
-}
-
-Fence FenceManagerOpenGL::CreateFence(u8* addr, u32 value, bool is_stubbed) {
-    return std::make_shared<GLInnerFence>(addr, value, is_stubbed);
+Fence FenceManagerOpenGL::CreateFence(bool is_stubbed) {
+    return std::make_shared<GLInnerFence>(is_stubbed);
 }
 
 void FenceManagerOpenGL::QueueFence(Fence& fence) {
diff --git a/src/video_core/renderer_opengl/gl_fence_manager.h b/src/video_core/renderer_opengl/gl_fence_manager.h
index fced8d002..f1446e732 100644
--- a/src/video_core/renderer_opengl/gl_fence_manager.h
+++ b/src/video_core/renderer_opengl/gl_fence_manager.h
@@ -16,8 +16,7 @@ namespace OpenGL {
 
 class GLInnerFence : public VideoCommon::FenceBase {
 public:
-    explicit GLInnerFence(u32 payload_, bool is_stubbed_);
-    explicit GLInnerFence(u8* address_, u32 payload_, bool is_stubbed_);
+    explicit GLInnerFence(bool is_stubbed_);
     ~GLInnerFence();
 
     void Queue();
@@ -40,8 +39,7 @@ public:
                                 QueryCache& query_cache);
 
 protected:
-    Fence CreateFence(u32 value, bool is_stubbed) override;
-    Fence CreateFence(u8* addr, u32 value, bool is_stubbed) override;
+    Fence CreateFence(bool is_stubbed) override;
     void QueueFence(Fence& fence) override;
     bool IsFenceSignaled(Fence& fence) const override;
     void WaitFence(Fence& fence) override;
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp
index b572950a6..6ebd6cff9 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.cpp
+++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp
@@ -358,7 +358,7 @@ void RasterizerOpenGL::OnCPUWrite(VAddr addr, u64 size) {
     }
 }
 
-void RasterizerOpenGL::SyncGuestHost() {
+void RasterizerOpenGL::InvalidateGPUCache() {
     MICROPROFILE_SCOPE(OpenGL_CacheManagement);
     shader_cache.SyncGuestHost();
     {
@@ -386,13 +386,12 @@ void RasterizerOpenGL::ModifyGPUMemory(size_t as_id, GPUVAddr addr, u64 size) {
     }
 }
 
-void RasterizerOpenGL::SignalSemaphore(GPUVAddr addr, u32 value) {
-    if (!gpu.IsAsync()) {
-        gpu_memory->Write<u32>(addr, value);
-        return;
-    }
-    auto paddr = gpu_memory->GetPointer(addr);
-    fence_manager.SignalSemaphore(paddr, value);
+void RasterizerOpenGL::SignalFence(std::function<void()>&& func) {
+    fence_manager.SignalFence(std::move(func));
+}
+
+void RasterizerOpenGL::SyncOperation(std::function<void()>&& func) {
+    fence_manager.SyncOperation(std::move(func));
 }
 
 void RasterizerOpenGL::SignalSyncPoint(u32 value) {
@@ -400,16 +399,10 @@ void RasterizerOpenGL::SignalSyncPoint(u32 value) {
 }
 
 void RasterizerOpenGL::SignalReference() {
-    if (!gpu.IsAsync()) {
-        return;
-    }
     fence_manager.SignalOrdering();
 }
 
 void RasterizerOpenGL::ReleaseFences() {
-    if (!gpu.IsAsync()) {
-        return;
-    }
     fence_manager.WaitPendingFences();
 }
 
@@ -426,6 +419,7 @@ void RasterizerOpenGL::WaitForIdle() {
 }
 
 void RasterizerOpenGL::FragmentBarrier() {
+    glTextureBarrier();
     glMemoryBarrier(GL_FRAMEBUFFER_BARRIER_BIT | GL_TEXTURE_FETCH_BARRIER_BIT);
 }
 
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h
index d469075a1..fe0ba979a 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.h
+++ b/src/video_core/renderer_opengl/gl_rasterizer.h
@@ -80,10 +80,11 @@ public:
     bool MustFlushRegion(VAddr addr, u64 size) override;
     void InvalidateRegion(VAddr addr, u64 size) override;
     void OnCPUWrite(VAddr addr, u64 size) override;
-    void SyncGuestHost() override;
+    void InvalidateGPUCache() override;
     void UnmapMemory(VAddr addr, u64 size) override;
     void ModifyGPUMemory(size_t as_id, GPUVAddr addr, u64 size) override;
-    void SignalSemaphore(GPUVAddr addr, u32 value) override;
+    void SignalFence(std::function<void()>&& func) override;
+    void SyncOperation(std::function<void()>&& func) override;
     void SignalSyncPoint(u32 value) override;
     void SignalReference() override;
     void ReleaseFences() override;
diff --git a/src/video_core/renderer_vulkan/vk_fence_manager.cpp b/src/video_core/renderer_vulkan/vk_fence_manager.cpp
index 301cbbabe..0214b103a 100644
--- a/src/video_core/renderer_vulkan/vk_fence_manager.cpp
+++ b/src/video_core/renderer_vulkan/vk_fence_manager.cpp
@@ -11,11 +11,8 @@
 
 namespace Vulkan {
 
-InnerFence::InnerFence(Scheduler& scheduler_, u32 payload_, bool is_stubbed_)
-    : FenceBase{payload_, is_stubbed_}, scheduler{scheduler_} {}
-
-InnerFence::InnerFence(Scheduler& scheduler_, u8* address_, u32 payload_, bool is_stubbed_)
-    : FenceBase{address_, payload_, is_stubbed_}, scheduler{scheduler_} {}
+InnerFence::InnerFence(Scheduler& scheduler_, bool is_stubbed_)
+    : FenceBase{is_stubbed_}, scheduler{scheduler_} {}
 
 InnerFence::~InnerFence() = default;
 
@@ -48,12 +45,8 @@ FenceManager::FenceManager(VideoCore::RasterizerInterface& rasterizer_, Tegra::G
     : GenericFenceManager{rasterizer_, gpu_, texture_cache_, buffer_cache_, query_cache_},
       scheduler{scheduler_} {}
 
-Fence FenceManager::CreateFence(u32 value, bool is_stubbed) {
-    return std::make_shared<InnerFence>(scheduler, value, is_stubbed);
-}
-
-Fence FenceManager::CreateFence(u8* addr, u32 value, bool is_stubbed) {
-    return std::make_shared<InnerFence>(scheduler, addr, value, is_stubbed);
+Fence FenceManager::CreateFence(bool is_stubbed) {
+    return std::make_shared<InnerFence>(scheduler, is_stubbed);
 }
 
 void FenceManager::QueueFence(Fence& fence) {
diff --git a/src/video_core/renderer_vulkan/vk_fence_manager.h b/src/video_core/renderer_vulkan/vk_fence_manager.h
index ea9e88052..7fe2afcd9 100644
--- a/src/video_core/renderer_vulkan/vk_fence_manager.h
+++ b/src/video_core/renderer_vulkan/vk_fence_manager.h
@@ -25,8 +25,7 @@ class Scheduler;
 
 class InnerFence : public VideoCommon::FenceBase {
 public:
-    explicit InnerFence(Scheduler& scheduler_, u32 payload_, bool is_stubbed_);
-    explicit InnerFence(Scheduler& scheduler_, u8* address_, u32 payload_, bool is_stubbed_);
+    explicit InnerFence(Scheduler& scheduler_, bool is_stubbed_);
     ~InnerFence();
 
     void Queue();
@@ -50,8 +49,7 @@ public:
                           QueryCache& query_cache, const Device& device, Scheduler& scheduler);
 
 protected:
-    Fence CreateFence(u32 value, bool is_stubbed) override;
-    Fence CreateFence(u8* addr, u32 value, bool is_stubbed) override;
+    Fence CreateFence(bool is_stubbed) override;
     void QueueFence(Fence& fence) override;
     bool IsFenceSignaled(Fence& fence) const override;
     void WaitFence(Fence& fence) override;
diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp
index d7b57e0f3..a35e41199 100644
--- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp
+++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp
@@ -428,7 +428,7 @@ void RasterizerVulkan::OnCPUWrite(VAddr addr, u64 size) {
     }
 }
 
-void RasterizerVulkan::SyncGuestHost() {
+void RasterizerVulkan::InvalidateGPUCache() {
     pipeline_cache.SyncGuestHost();
     {
         std::scoped_lock lock{buffer_cache.mutex};
@@ -455,13 +455,12 @@ void RasterizerVulkan::ModifyGPUMemory(size_t as_id, GPUVAddr addr, u64 size) {
     }
 }
 
-void RasterizerVulkan::SignalSemaphore(GPUVAddr addr, u32 value) {
-    if (!gpu.IsAsync()) {
-        gpu_memory->Write<u32>(addr, value);
-        return;
-    }
-    auto paddr = gpu_memory->GetPointer(addr);
-    fence_manager.SignalSemaphore(paddr, value);
+void RasterizerVulkan::SignalFence(std::function<void()>&& func) {
+    fence_manager.SignalFence(std::move(func));
+}
+
+void RasterizerVulkan::SyncOperation(std::function<void()>&& func) {
+    fence_manager.SyncOperation(std::move(func));
 }
 
 void RasterizerVulkan::SignalSyncPoint(u32 value) {
@@ -469,16 +468,10 @@ void RasterizerVulkan::SignalSyncPoint(u32 value) {
 }
 
 void RasterizerVulkan::SignalReference() {
-    if (!gpu.IsAsync()) {
-        return;
-    }
     fence_manager.SignalOrdering();
 }
 
 void RasterizerVulkan::ReleaseFences() {
-    if (!gpu.IsAsync()) {
-        return;
-    }
     fence_manager.WaitPendingFences();
 }
 
diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.h b/src/video_core/renderer_vulkan/vk_rasterizer.h
index c836158b8..fb9e83e8f 100644
--- a/src/video_core/renderer_vulkan/vk_rasterizer.h
+++ b/src/video_core/renderer_vulkan/vk_rasterizer.h
@@ -76,10 +76,11 @@ public:
     bool MustFlushRegion(VAddr addr, u64 size) override;
     void InvalidateRegion(VAddr addr, u64 size) override;
     void OnCPUWrite(VAddr addr, u64 size) override;
-    void SyncGuestHost() override;
+    void InvalidateGPUCache() override;
     void UnmapMemory(VAddr addr, u64 size) override;
     void ModifyGPUMemory(size_t as_id, GPUVAddr addr, u64 size) override;
-    void SignalSemaphore(GPUVAddr addr, u32 value) override;
+    void SignalFence(std::function<void()>&& func) override;
+    void SyncOperation(std::function<void()>&& func) override;
     void SignalSyncPoint(u32 value) override;
     void SignalReference() override;
     void ReleaseFences() override;