13 files changed, 74 insertions, 101 deletions

diff --git a/src/core/hle/service/nvdrv/devices/nvhost_nvdec.cpp b/src/core/hle/service/nvdrv/devices/nvhost_nvdec.cpp
index d8735491c..36970f828 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_nvdec.cpp
+++ b/src/core/hle/service/nvdrv/devices/nvhost_nvdec.cpp
@@ -11,8 +11,9 @@
 
 namespace Service::Nvidia::Devices {
 
-nvhost_nvdec::nvhost_nvdec(Core::System& system, std::shared_ptr<nvmap> nvmap_dev)
-    : nvhost_nvdec_common(system, std::move(nvmap_dev)) {}
+nvhost_nvdec::nvhost_nvdec(Core::System& system, std::shared_ptr<nvmap> nvmap_dev,
+                           SyncpointManager& syncpoint_manager)
+    : nvhost_nvdec_common(system, std::move(nvmap_dev), syncpoint_manager) {}
 nvhost_nvdec::~nvhost_nvdec() = default;
 
 NvResult nvhost_nvdec::Ioctl1(Ioctl command, const std::vector<u8>& input,
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_nvdec.h b/src/core/hle/service/nvdrv/devices/nvhost_nvdec.h
index 79b8b6de1..77ef53cdd 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_nvdec.h
+++ b/src/core/hle/service/nvdrv/devices/nvhost_nvdec.h
@@ -11,7 +11,8 @@ namespace Service::Nvidia::Devices {
 
 class nvhost_nvdec final : public nvhost_nvdec_common {
 public:
-    explicit nvhost_nvdec(Core::System& system, std::shared_ptr<nvmap> nvmap_dev);
+    explicit nvhost_nvdec(Core::System& system, std::shared_ptr<nvmap> nvmap_dev,
+                          SyncpointManager& syncpoint_manager);
     ~nvhost_nvdec() override;
 
     NvResult Ioctl1(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) override;
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.cpp b/src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.cpp
index b49cecb42..4898dc27a 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.cpp
+++ b/src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.cpp
@@ -11,6 +11,7 @@
 #include "core/core.h"
 #include "core/hle/service/nvdrv/devices/nvhost_nvdec_common.h"
 #include "core/hle/service/nvdrv/devices/nvmap.h"
+#include "core/hle/service/nvdrv/syncpoint_manager.h"
 #include "core/memory.h"
 #include "video_core/memory_manager.h"
 #include "video_core/renderer_base.h"
@@ -36,8 +37,9 @@ std::size_t WriteVectors(std::vector<u8>& dst, const std::vector<T>& src, std::s
 }
 } // Anonymous namespace
 
-nvhost_nvdec_common::nvhost_nvdec_common(Core::System& system, std::shared_ptr<nvmap> nvmap_dev)
-    : nvdevice(system), nvmap_dev(std::move(nvmap_dev)) {}
+nvhost_nvdec_common::nvhost_nvdec_common(Core::System& system, std::shared_ptr<nvmap> nvmap_dev,
+                                         SyncpointManager& syncpoint_manager)
+    : nvdevice(system), nvmap_dev(std::move(nvmap_dev)), syncpoint_manager(syncpoint_manager) {}
 nvhost_nvdec_common::~nvhost_nvdec_common() = default;
 
 NvResult nvhost_nvdec_common::SetNVMAPfd(const std::vector<u8>& input) {
@@ -71,10 +73,15 @@ NvResult nvhost_nvdec_common::Submit(const std::vector<u8>& input, std::vector<u
     offset = SpliceVectors(input, wait_checks, params.syncpoint_count, offset);
     offset = SpliceVectors(input, fences, params.fence_count, offset);
 
-    // TODO(ameerj): For async gpu, utilize fences for syncpoint 'max' increment
-
     auto& gpu = system.GPU();
-
+    if (gpu.UseNvdec()) {
+        for (std::size_t i = 0; i < syncpt_increments.size(); i++) {
+            const SyncptIncr& syncpt_incr = syncpt_increments[i];
+            fences[i].id = syncpt_incr.id;
+            fences[i].value =
+                syncpoint_manager.IncreaseSyncpoint(syncpt_incr.id, syncpt_incr.increments);
+        }
+    }
     for (const auto& cmd_buffer : command_buffers) {
         auto object = nvmap_dev->GetObject(cmd_buffer.memory_id);
         ASSERT_OR_EXECUTE(object, return NvResult::InvalidState;);
@@ -89,7 +96,13 @@ NvResult nvhost_nvdec_common::Submit(const std::vector<u8>& input, std::vector<u
                                       cmdlist.size() * sizeof(u32));
         gpu.PushCommandBuffer(cmdlist);
     }
+    if (gpu.UseNvdec()) {
 
+        fences[0].value = syncpoint_manager.IncreaseSyncpoint(fences[0].id, 1);
+
+        Tegra::ChCommandHeaderList cmdlist{{(4 << 28) | fences[0].id}};
+        gpu.PushCommandBuffer(cmdlist);
+    }
     std::memcpy(output.data(), &params, sizeof(IoctlSubmit));
     // Some games expect command_buffers to be written back
     offset = sizeof(IoctlSubmit);
@@ -98,6 +111,7 @@ NvResult nvhost_nvdec_common::Submit(const std::vector<u8>& input, std::vector<u
     offset = WriteVectors(output, reloc_shifts, offset);
     offset = WriteVectors(output, syncpt_increments, offset);
     offset = WriteVectors(output, wait_checks, offset);
+    offset = WriteVectors(output, fences, offset);
 
     return NvResult::Success;
 }
@@ -107,9 +121,10 @@ NvResult nvhost_nvdec_common::GetSyncpoint(const std::vector<u8>& input, std::ve
     std::memcpy(&params, input.data(), sizeof(IoctlGetSyncpoint));
     LOG_DEBUG(Service_NVDRV, "called GetSyncpoint, id={}", params.param);
 
-    // We found that implementing this causes deadlocks with async gpu, along with degraded
-    // performance. TODO: RE the nvdec async implementation
-    params.value = 0;
+    if (device_syncpoints[params.param] == 0 && system.GPU().UseNvdec()) {
+        device_syncpoints[params.param] = syncpoint_manager.AllocateSyncpoint();
+    }
+    params.value = device_syncpoints[params.param];
     std::memcpy(output.data(), &params, sizeof(IoctlGetSyncpoint));
 
     return NvResult::Success;
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.h b/src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.h
index d9f95ba58..4c9d4ba41 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.h
+++ b/src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.h
@@ -10,12 +10,16 @@
 #include "common/swap.h"
 #include "core/hle/service/nvdrv/devices/nvdevice.h"
 
-namespace Service::Nvidia::Devices {
+namespace Service::Nvidia {
+class SyncpointManager;
+
+namespace Devices {
 class nvmap;
 
 class nvhost_nvdec_common : public nvdevice {
 public:
-    explicit nvhost_nvdec_common(Core::System& system, std::shared_ptr<nvmap> nvmap_dev);
+    explicit nvhost_nvdec_common(Core::System& system, std::shared_ptr<nvmap> nvmap_dev,
+                                 SyncpointManager& syncpoint_manager);
     ~nvhost_nvdec_common() override;
 
 protected:
@@ -157,8 +161,10 @@ protected:
     s32_le nvmap_fd{};
     u32_le submit_timeout{};
     std::shared_ptr<nvmap> nvmap_dev;
-
+    SyncpointManager& syncpoint_manager;
+    std::array<u32, MaxSyncPoints> device_syncpoints{};
     // This is expected to be ordered, therefore we must use a map, not unordered_map
     std::map<GPUVAddr, BufferMap> buffer_mappings;
 };
-}; // namespace Service::Nvidia::Devices
+}; // namespace Devices
+} // namespace Service::Nvidia
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_vic.cpp b/src/core/hle/service/nvdrv/devices/nvhost_vic.cpp
index 805fe86ae..72499654c 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_vic.cpp
+++ b/src/core/hle/service/nvdrv/devices/nvhost_vic.cpp
@@ -10,8 +10,9 @@
 #include "video_core/renderer_base.h"
 
 namespace Service::Nvidia::Devices {
-nvhost_vic::nvhost_vic(Core::System& system, std::shared_ptr<nvmap> nvmap_dev)
-    : nvhost_nvdec_common(system, std::move(nvmap_dev)) {}
+nvhost_vic::nvhost_vic(Core::System& system, std::shared_ptr<nvmap> nvmap_dev,
+                       SyncpointManager& syncpoint_manager)
+    : nvhost_nvdec_common(system, std::move(nvmap_dev), syncpoint_manager) {}
 
 nvhost_vic::~nvhost_vic() = default;
 
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_vic.h b/src/core/hle/service/nvdrv/devices/nvhost_vic.h
index b2e11f4d4..f401c61fa 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_vic.h
+++ b/src/core/hle/service/nvdrv/devices/nvhost_vic.h
@@ -7,11 +7,11 @@
 #include "core/hle/service/nvdrv/devices/nvhost_nvdec_common.h"
 
 namespace Service::Nvidia::Devices {
-class nvmap;
 
 class nvhost_vic final : public nvhost_nvdec_common {
 public:
-    explicit nvhost_vic(Core::System& system, std::shared_ptr<nvmap> nvmap_dev);
+    explicit nvhost_vic(Core::System& system, std::shared_ptr<nvmap> nvmap_dev,
+                        SyncpointManager& syncpoint_manager);
     ~nvhost_vic();
 
     NvResult Ioctl1(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) override;
diff --git a/src/core/hle/service/nvdrv/nvdrv.cpp b/src/core/hle/service/nvdrv/nvdrv.cpp
index e03195afe..620c18728 100644
--- a/src/core/hle/service/nvdrv/nvdrv.cpp
+++ b/src/core/hle/service/nvdrv/nvdrv.cpp
@@ -55,9 +55,11 @@ Module::Module(Core::System& system) : syncpoint_manager{system.GPU()} {
     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, syncpoint_manager);
-    devices["/dev/nvhost-nvdec"] = std::make_shared<Devices::nvhost_nvdec>(system, nvmap_dev);
+    devices["/dev/nvhost-nvdec"] =
+        std::make_shared<Devices::nvhost_nvdec>(system, nvmap_dev, syncpoint_manager);
     devices["/dev/nvhost-nvjpg"] = std::make_shared<Devices::nvhost_nvjpg>(system);
-    devices["/dev/nvhost-vic"] = std::make_shared<Devices::nvhost_vic>(system, nvmap_dev);
+    devices["/dev/nvhost-vic"] =
+        std::make_shared<Devices::nvhost_vic>(system, nvmap_dev, syncpoint_manager);
 }
 
 Module::~Module() = default;
diff --git a/src/video_core/cdma_pusher.cpp b/src/video_core/cdma_pusher.cpp
index e3e7432f7..94679d5d1 100644
--- a/src/video_core/cdma_pusher.cpp
+++ b/src/video_core/cdma_pusher.cpp
@@ -33,8 +33,7 @@ CDmaPusher::CDmaPusher(GPU& gpu_)
     : gpu{gpu_}, nvdec_processor(std::make_shared<Nvdec>(gpu)),
       vic_processor(std::make_unique<Vic>(gpu, nvdec_processor)),
       host1x_processor(std::make_unique<Host1x>(gpu)),
-      nvdec_sync(std::make_unique<SyncptIncrManager>(gpu)),
-      vic_sync(std::make_unique<SyncptIncrManager>(gpu)) {}
+      sync_manager(std::make_unique<SyncptIncrManager>(gpu)) {}
 
 CDmaPusher::~CDmaPusher() = default;
 
@@ -110,10 +109,10 @@ void CDmaPusher::ExecuteCommand(u32 state_offset, u32 data) {
             const auto syncpoint_id = static_cast<u32>(data & 0xFF);
             const auto cond = static_cast<u32>((data >> 8) & 0xFF);
             if (cond == 0) {
-                nvdec_sync->Increment(syncpoint_id);
+                sync_manager->Increment(syncpoint_id);
             } else {
-                nvdec_sync->IncrementWhenDone(static_cast<u32>(current_class), syncpoint_id);
-                nvdec_sync->SignalDone(syncpoint_id);
+                sync_manager->SignalDone(
+                    sync_manager->IncrementWhenDone(static_cast<u32>(current_class), syncpoint_id));
             }
             break;
         }
@@ -135,10 +134,10 @@ void CDmaPusher::ExecuteCommand(u32 state_offset, u32 data) {
             const auto syncpoint_id = static_cast<u32>(data & 0xFF);
             const auto cond = static_cast<u32>((data >> 8) & 0xFF);
             if (cond == 0) {
-                vic_sync->Increment(syncpoint_id);
+                sync_manager->Increment(syncpoint_id);
             } else {
-                vic_sync->IncrementWhenDone(static_cast<u32>(current_class), syncpoint_id);
-                vic_sync->SignalDone(syncpoint_id);
+                sync_manager->SignalDone(
+                    sync_manager->IncrementWhenDone(static_cast<u32>(current_class), syncpoint_id));
             }
             break;
         }
diff --git a/src/video_core/cdma_pusher.h b/src/video_core/cdma_pusher.h
index 0db1cd646..8ca70b6dd 100644
--- a/src/video_core/cdma_pusher.h
+++ b/src/video_core/cdma_pusher.h
@@ -116,12 +116,10 @@ private:
     void ThiStateWrite(ThiRegisters& state, u32 state_offset, const std::vector<u32>& arguments);
 
     GPU& gpu;
-
-    std::shared_ptr<Nvdec> nvdec_processor;
-    std::unique_ptr<Vic> vic_processor;
-    std::unique_ptr<Host1x> host1x_processor;
-    std::unique_ptr<SyncptIncrManager> nvdec_sync;
-    std::unique_ptr<SyncptIncrManager> vic_sync;
+    std::shared_ptr<Tegra::Nvdec> nvdec_processor;
+    std::unique_ptr<Tegra::Vic> vic_processor;
+    std::unique_ptr<Tegra::Host1x> host1x_processor;
+    std::unique_ptr<SyncptIncrManager> sync_manager;
     ChClassId current_class{};
     ThiRegisters vic_thi_state{};
     ThiRegisters nvdec_thi_state{};
diff --git a/src/video_core/command_classes/host1x.cpp b/src/video_core/command_classes/host1x.cpp
index c4dd4881a..b12494528 100644
--- a/src/video_core/command_classes/host1x.cpp
+++ b/src/video_core/command_classes/host1x.cpp
@@ -10,22 +10,14 @@ Tegra::Host1x::Host1x(GPU& gpu_) : gpu(gpu_) {}
 
 Tegra::Host1x::~Host1x() = default;
 
-void Tegra::Host1x::StateWrite(u32 offset, u32 arguments) {
-    u8* const state_offset = reinterpret_cast<u8*>(&state) + offset * sizeof(u32);
-    std::memcpy(state_offset, &arguments, sizeof(u32));
-}
-
-void Tegra::Host1x::ProcessMethod(Method method, const std::vector<u32>& arguments) {
-    StateWrite(static_cast<u32>(method), arguments[0]);
+void Tegra::Host1x::ProcessMethod(Method method, u32 argument) {
     switch (method) {
-    case Method::WaitSyncpt:
-        Execute(arguments[0]);
-        break;
     case Method::LoadSyncptPayload32:
-        syncpoint_value = arguments[0];
+        syncpoint_value = argument;
         break;
+    case Method::WaitSyncpt:
     case Method::WaitSyncpt32:
-        Execute(arguments[0]);
+        Execute(argument);
         break;
     default:
         UNIMPLEMENTED_MSG("Host1x method 0x{:X}", static_cast<u32>(method));
@@ -34,6 +26,5 @@ void Tegra::Host1x::ProcessMethod(Method method, const std::vector<u32>& argumen
 }
 
 void Tegra::Host1x::Execute(u32 data) {
-    // This method waits on a valid syncpoint.
-    // TODO: Implement when proper Async is in place
+    gpu.WaitFence(data, syncpoint_value);
 }
diff --git a/src/video_core/command_classes/host1x.h b/src/video_core/command_classes/host1x.h
index 013eaa0c1..7e94799dd 100644
--- a/src/video_core/command_classes/host1x.h
+++ b/src/video_core/command_classes/host1x.h
@@ -14,64 +14,23 @@ class Nvdec;
 
 class Host1x {
 public:
-    struct Host1xClassRegisters {
-        u32 incr_syncpt{};
-        u32 incr_syncpt_ctrl{};
-        u32 incr_syncpt_error{};
-        INSERT_PADDING_WORDS(5);
-        u32 wait_syncpt{};
-        u32 wait_syncpt_base{};
-        u32 wait_syncpt_incr{};
-        u32 load_syncpt_base{};
-        u32 incr_syncpt_base{};
-        u32 clear{};
-        u32 wait{};
-        u32 wait_with_interrupt{};
-        u32 delay_use{};
-        u32 tick_count_high{};
-        u32 tick_count_low{};
-        u32 tick_ctrl{};
-        INSERT_PADDING_WORDS(23);
-        u32 ind_ctrl{};
-        u32 ind_off2{};
-        u32 ind_off{};
-        std::array<u32, 31> ind_data{};
-        INSERT_PADDING_WORDS(1);
-        u32 load_syncpoint_payload32{};
-        u32 stall_ctrl{};
-        u32 wait_syncpt32{};
-        u32 wait_syncpt_base32{};
-        u32 load_syncpt_base32{};
-        u32 incr_syncpt_base32{};
-        u32 stall_count_high{};
-        u32 stall_count_low{};
-        u32 xref_ctrl{};
-        u32 channel_xref_high{};
-        u32 channel_xref_low{};
-    };
-    static_assert(sizeof(Host1xClassRegisters) == 0x164, "Host1xClassRegisters is an invalid size");
-
     enum class Method : u32 {
-        WaitSyncpt = offsetof(Host1xClassRegisters, wait_syncpt) / 4,
-        LoadSyncptPayload32 = offsetof(Host1xClassRegisters, load_syncpoint_payload32) / 4,
-        WaitSyncpt32 = offsetof(Host1xClassRegisters, wait_syncpt32) / 4,
+        WaitSyncpt = 0x8,
+        LoadSyncptPayload32 = 0x4e,
+        WaitSyncpt32 = 0x50,
     };
 
     explicit Host1x(GPU& gpu);
     ~Host1x();
 
     /// Writes the method into the state, Invoke Execute() if encountered
-    void ProcessMethod(Method method, const std::vector<u32>& arguments);
+    void ProcessMethod(Method method, u32 argument);
 
 private:
     /// For Host1x, execute is waiting on a syncpoint previously written into the state
     void Execute(u32 data);
 
-    /// Write argument into the provided offset
-    void StateWrite(u32 offset, u32 arguments);
-
     u32 syncpoint_value{};
-    Host1xClassRegisters state{};
     GPU& gpu;
 };
 
diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h
index ad86c50b4..d1080300f 100644
--- a/src/video_core/texture_cache/texture_cache.h
+++ b/src/video_core/texture_cache/texture_cache.h
@@ -61,7 +61,7 @@ using VideoCore::Surface::SurfaceType;
 template <class P>
 class TextureCache {
     /// Address shift for caching images into a hash table
-    static constexpr u64 PAGE_SHIFT = 20;
+    static constexpr u64 PAGE_BITS = 20;
 
     /// Enables debugging features to the texture cache
     static constexpr bool ENABLE_VALIDATION = P::ENABLE_VALIDATION;
@@ -184,8 +184,8 @@ private:
     template <typename Func>
     static void ForEachPage(VAddr addr, size_t size, Func&& func) {
         static constexpr bool RETURNS_BOOL = std::is_same_v<std::invoke_result<Func, u64>, bool>;
-        const u64 page_end = (addr + size - 1) >> PAGE_SHIFT;
-        for (u64 page = addr >> PAGE_SHIFT; page <= page_end; ++page) {
+        const u64 page_end = (addr + size - 1) >> PAGE_BITS;
+        for (u64 page = addr >> PAGE_BITS; page <= page_end; ++page) {
             if constexpr (RETURNS_BOOL) {
                 if (func(page)) {
                     break;
@@ -708,7 +708,7 @@ void TextureCache<P>::InvalidateDepthBuffer() {
 template <class P>
 typename P::ImageView* TextureCache<P>::TryFindFramebufferImageView(VAddr cpu_addr) {
     // TODO: Properly implement this
-    const auto it = page_table.find(cpu_addr >> PAGE_SHIFT);
+    const auto it = page_table.find(cpu_addr >> PAGE_BITS);
     if (it == page_table.end()) {
         return nullptr;
     }
@@ -1170,13 +1170,13 @@ void TextureCache<P>::UnregisterImage(ImageId image_id) {
     ForEachPage(image.cpu_addr, image.guest_size_bytes, [this, image_id](u64 page) {
         const auto page_it = page_table.find(page);
         if (page_it == page_table.end()) {
-            UNREACHABLE_MSG("Unregistering unregistered page=0x{:x}", page << PAGE_SHIFT);
+            UNREACHABLE_MSG("Unregistering unregistered page=0x{:x}", page << PAGE_BITS);
             return;
         }
         std::vector<ImageId>& image_ids = page_it->second;
         const auto vector_it = std::ranges::find(image_ids, image_id);
         if (vector_it == image_ids.end()) {
-            UNREACHABLE_MSG("Unregistering unregistered image in page=0x{:x}", page << PAGE_SHIFT);
+            UNREACHABLE_MSG("Unregistering unregistered image in page=0x{:x}", page << PAGE_BITS);
             return;
         }
         image_ids.erase(vector_it);
diff --git a/src/video_core/vulkan_common/vulkan_library.cpp b/src/video_core/vulkan_common/vulkan_library.cpp
index 27c958221..557871d81 100644
--- a/src/video_core/vulkan_common/vulkan_library.cpp
+++ b/src/video_core/vulkan_common/vulkan_library.cpp
@@ -20,7 +20,7 @@ Common::DynamicLibrary OpenLibrary() {
         // Use the libvulkan.dylib from the application bundle.
         const std::string filename =
             Common::FS::GetBundleDirectory() + "/Contents/Frameworks/libvulkan.dylib";
-        library.Open(filename.c_str());
+        void(library.Open(filename.c_str()));
     }
 #else
     std::string filename = Common::DynamicLibrary::GetVersionedFilename("vulkan", 1);