VideoCore: Refactor syncing.

7 files changed, 53 insertions, 12 deletions

diff --git a/src/core/core.cpp b/src/core/core.cpp
index 121092868..fa059a394 100644
--- a/src/core/core.cpp
+++ b/src/core/core.cpp
@@ -51,6 +51,7 @@
 #include "core/telemetry_session.h"
 #include "core/tools/freezer.h"
 #include "network/network.h"
+#include "video_core/host1x/host1x.h"
 #include "video_core/renderer_base.h"
 #include "video_core/video_core.h"
 
@@ -215,6 +216,7 @@ struct System::Impl {
 
         telemetry_session = std::make_unique<Core::TelemetrySession>();
 
+        host1x_core = std::make_unique<Tegra::Host1x::Host1x>();
         gpu_core = VideoCore::CreateGPU(emu_window, system);
         if (!gpu_core) {
             return SystemResultStatus::ErrorVideoCore;
@@ -373,6 +375,7 @@ struct System::Impl {
         app_loader.reset();
         audio_core.reset();
         gpu_core.reset();
+        host1x_core.reset();
         perf_stats.reset();
         kernel.Shutdown();
         memory.Reset();
@@ -450,6 +453,7 @@ struct System::Impl {
     /// AppLoader used to load the current executing application
     std::unique_ptr<Loader::AppLoader> app_loader;
     std::unique_ptr<Tegra::GPU> gpu_core;
+    std::unique_ptr<Tegra::Host1x::Host1x> host1x_core;
     std::unique_ptr<Hardware::InterruptManager> interrupt_manager;
     std::unique_ptr<Core::DeviceMemory> device_memory;
     std::unique_ptr<AudioCore::AudioCore> audio_core;
@@ -668,6 +672,14 @@ const Tegra::GPU& System::GPU() const {
     return *impl->gpu_core;
 }
 
+Tegra::Host1x::Host1x& System::Host1x() {
+    return *impl->host1x_core;
+}
+
+const Tegra::Host1x::Host1x& System::Host1x() const {
+    return *impl->host1x_core;
+}
+
 Core::Hardware::InterruptManager& System::InterruptManager() {
     return *impl->interrupt_manager;
 }
diff --git a/src/core/core.h b/src/core/core.h
index 0ce3b1d60..e4168a921 100644
--- a/src/core/core.h
+++ b/src/core/core.h
@@ -74,6 +74,9 @@ class TimeManager;
 namespace Tegra {
 class DebugContext;
 class GPU;
+namespace Host1x {
+class Host1x;
+} // namespace Host1x
 } // namespace Tegra
 
 namespace VideoCore {
@@ -260,6 +263,12 @@ public:
     /// Gets an immutable reference to the GPU interface.
     [[nodiscard]] const Tegra::GPU& GPU() const;
 
+    /// Gets a mutable reference to the Host1x interface
+    [[nodiscard]] Tegra::Host1x::Host1x& Host1x();
+
+    /// Gets an immutable reference to the Host1x interface.
+    [[nodiscard]] const Tegra::Host1x::Host1x& Host1x() const;
+
     /// Gets a mutable reference to the renderer.
     [[nodiscard]] VideoCore::RendererBase& Renderer();
 
diff --git a/src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp b/src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp
index b1c0e9eb2..e6a976714 100644
--- a/src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp
+++ b/src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp
@@ -50,7 +50,7 @@ void nvdisp_disp0::flip(u32 buffer_handle, u32 offset, android::PixelFormat form
                                                stride, format, transform, crop_rect};
 
     system.GetPerfStats().EndSystemFrame();
-    system.GPU().SwapBuffers(&framebuffer);
+    system.GPU().RequestSwapBuffers(&framebuffer, nullptr, 0);
     system.SpeedLimiter().DoSpeedLimiting(system.CoreTiming().GetGlobalTimeUs());
     system.GetPerfStats().BeginSystemFrame();
 }
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_ctrl.cpp b/src/core/hle/service/nvdrv/devices/nvhost_ctrl.cpp
index 54074af75..ffe42d423 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_ctrl.cpp
+++ b/src/core/hle/service/nvdrv/devices/nvhost_ctrl.cpp
@@ -18,6 +18,7 @@
 #include "core/hle/service/nvdrv/core/syncpoint_manager.h"
 #include "core/hle/service/nvdrv/devices/nvhost_ctrl.h"
 #include "video_core/gpu.h"
+#include "video_core/host1x/host1x.h"
 
 namespace Service::Nvidia::Devices {
 
@@ -129,7 +130,7 @@ NvResult nvhost_ctrl::IocCtrlEventWait(const std::vector<u8>& input, std::vector
         return NvResult::Success;
     }
 
-    auto& gpu = system.GPU();
+    auto& host1x_syncpoint_manager = system.Host1x().GetSyncpointManager();
     const u32 target_value = params.fence.value;
 
     auto lock = NvEventsLock();
@@ -149,7 +150,7 @@ NvResult nvhost_ctrl::IocCtrlEventWait(const std::vector<u8>& input, std::vector
         if (events[slot].fails > 2) {
             {
                 auto lk = system.StallProcesses();
-                gpu.WaitFence(fence_id, target_value);
+                host1x_syncpoint_manager.WaitHost(fence_id, target_value);
                 system.UnstallProcesses();
             }
             params.value.raw = target_value;
@@ -198,7 +199,15 @@ NvResult nvhost_ctrl::IocCtrlEventWait(const std::vector<u8>& input, std::vector
     }
     params.value.raw |= slot;
 
-    gpu.RegisterSyncptInterrupt(fence_id, target_value);
+    event.wait_handle =
+        host1x_syncpoint_manager.RegisterHostAction(fence_id, target_value, [this, slot]() {
+            auto& event = events[slot];
+            if (event.status.exchange(EventState::Signalling, std::memory_order_acq_rel) ==
+                EventState::Waiting) {
+                event.kevent->GetWritableEvent().Signal();
+            }
+            event.status.store(EventState::Signalled, std::memory_order_release);
+        });
     return NvResult::Timeout;
 }
 
@@ -288,8 +297,10 @@ NvResult nvhost_ctrl::IocCtrlClearEventWait(const std::vector<u8>& input, std::v
     auto& event = events[event_id];
     if (event.status.exchange(EventState::Cancelling, std::memory_order_acq_rel) ==
         EventState::Waiting) {
-        system.GPU().CancelSyncptInterrupt(event.assigned_syncpt, event.assigned_value);
+        auto& host1x_syncpoint_manager = system.Host1x().GetSyncpointManager();
+        host1x_syncpoint_manager.DeregisterHostAction(event.assigned_syncpt, event.wait_handle);
         syncpoint_manager.RefreshSyncpoint(event.assigned_syncpt);
+        event.wait_handle = {};
     }
     event.fails++;
     event.status.store(EventState::Cancelled, std::memory_order_release);
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_ctrl.h b/src/core/hle/service/nvdrv/devices/nvhost_ctrl.h
index d56aea405..136a1e925 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_ctrl.h
+++ b/src/core/hle/service/nvdrv/devices/nvhost_ctrl.h
@@ -11,6 +11,7 @@
 #include "common/common_types.h"
 #include "core/hle/service/nvdrv/devices/nvdevice.h"
 #include "core/hle/service/nvdrv/nvdrv.h"
+#include "video_core/host1x/syncpoint_manager.h"
 
 namespace Service::Nvidia::NvCore {
 class Container;
@@ -78,6 +79,9 @@ private:
         // Tells if an NVEvent is registered or not
         bool registered{};
 
+        // Used for waiting on a syncpoint & canceling it.
+        Tegra::Host1x::SyncpointManager::ActionHandle wait_handle{};
+
         bool IsBeingUsed() {
             const auto current_status = status.load(std::memory_order_acquire);
             return current_status == EventState::Waiting ||
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp b/src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp
index 38d45cb79..db3e266ad 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp
+++ b/src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp
@@ -210,10 +210,10 @@ NvResult nvhost_gpu::AllocateObjectContext(const std::vector<u8>& input, std::ve
 
 static std::vector<Tegra::CommandHeader> BuildWaitCommandList(NvFence fence) {
     return {
-        Tegra::BuildCommandHeader(Tegra::BufferMethods::FenceValue, 1,
+        Tegra::BuildCommandHeader(Tegra::BufferMethods::SyncpointPayload, 1,
                                   Tegra::SubmissionMode::Increasing),
         {fence.value},
-        Tegra::BuildCommandHeader(Tegra::BufferMethods::FenceAction, 1,
+        Tegra::BuildCommandHeader(Tegra::BufferMethods::SyncpointOperation, 1,
                                   Tegra::SubmissionMode::Increasing),
         BuildFenceAction(Tegra::Engines::Puller::FenceOperation::Acquire, fence.id),
     };
@@ -222,12 +222,12 @@ static std::vector<Tegra::CommandHeader> BuildWaitCommandList(NvFence fence) {
 static std::vector<Tegra::CommandHeader> BuildIncrementCommandList(NvFence fence,
                                                                    u32 add_increment) {
     std::vector<Tegra::CommandHeader> result{
-        Tegra::BuildCommandHeader(Tegra::BufferMethods::FenceValue, 1,
+        Tegra::BuildCommandHeader(Tegra::BufferMethods::SyncpointPayload, 1,
                                   Tegra::SubmissionMode::Increasing),
         {}};
 
     for (u32 count = 0; count < add_increment; ++count) {
-        result.emplace_back(Tegra::BuildCommandHeader(Tegra::BufferMethods::FenceAction, 1,
+        result.emplace_back(Tegra::BuildCommandHeader(Tegra::BufferMethods::SyncpointOperation, 1,
                                                       Tegra::SubmissionMode::Increasing));
         result.emplace_back(
             BuildFenceAction(Tegra::Engines::Puller::FenceOperation::Increment, fence.id));
@@ -239,7 +239,7 @@ static std::vector<Tegra::CommandHeader> BuildIncrementCommandList(NvFence fence
 static std::vector<Tegra::CommandHeader> BuildIncrementWithWfiCommandList(NvFence fence,
                                                                           u32 add_increment) {
     std::vector<Tegra::CommandHeader> result{
-        Tegra::BuildCommandHeader(Tegra::BufferMethods::WaitForInterrupt, 1,
+        Tegra::BuildCommandHeader(Tegra::BufferMethods::WaitForIdle, 1,
                                   Tegra::SubmissionMode::Increasing),
         {}};
     const std::vector<Tegra::CommandHeader> increment{
diff --git a/src/core/hle/service/nvflinger/nvflinger.cpp b/src/core/hle/service/nvflinger/nvflinger.cpp
index 8c3013f83..aa112021d 100644
--- a/src/core/hle/service/nvflinger/nvflinger.cpp
+++ b/src/core/hle/service/nvflinger/nvflinger.cpp
@@ -24,6 +24,8 @@
 #include "core/hle/service/vi/layer/vi_layer.h"
 #include "core/hle/service/vi/vi_results.h"
 #include "video_core/gpu.h"
+#include "video_core/host1x/host1x.h"
+#include "video_core/host1x/syncpoint_manager.h"
 
 namespace Service::NVFlinger {
 
@@ -267,12 +269,12 @@ void NVFlinger::Compose() {
             return; // We are likely shutting down
         }
 
-        auto& gpu = system.GPU();
+        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];
-            gpu.WaitFence(fence.id, fence.value);
+            syncpoint_manager.WaitGuest(fence.id, fence.value);
         }
         guard->lock();
 
@@ -284,6 +286,7 @@ void NVFlinger::Compose() {
         auto nvdisp = nvdrv->GetDevice<Nvidia::Devices::nvdisp_disp0>(disp_fd);
         ASSERT(nvdisp);
 
+        guard->unlock();
         Common::Rectangle<int> crop_rect{
             static_cast<int>(buffer.crop.Left()), static_cast<int>(buffer.crop.Top()),
             static_cast<int>(buffer.crop.Right()), static_cast<int>(buffer.crop.Bottom())};
@@ -292,6 +295,8 @@ void NVFlinger::Compose() {
                      igbp_buffer.Width(), igbp_buffer.Height(), igbp_buffer.Stride(),
                      static_cast<android::BufferTransformFlags>(buffer.transform), crop_rect);
 
+        guard->lock();
+
         swap_interval = buffer.swap_interval;
 
         auto fence = android::Fence::NoFence();