gnmdriver, amdgpu: added gpu idle IRQ; submission lock logic improved

src/core/libraries/gnmdriver/gnmdriver.cpp

@@ -3,6 +3,7 @@
 
 #include "common/assert.h"
 #include "common/config.h"
+#include "common/debug.h"
 #include "common/logging/log.h"
 #include "common/path_util.h"
 #include "common/slot_vector.h"
@@ -264,6 +265,7 @@ static_assert(CtxInitSequence400.size() == 0x61);
 
 // In case if `submitDone` is issued we need to block submissions until GPU idle
 static u32 submission_lock{};
+std::condition_variable cv_lock{};
 static std::mutex m_submission{};
 static u64 frames_submitted{};      // frame counter
 static bool send_init_packet{true}; // initialize HW state before first game's submit in a frame
@@ -277,6 +279,18 @@ struct AscQueueInfo {
 static Common::SlotVector<AscQueueInfo> asc_queues{};
 static constexpr VAddr tessellation_factors_ring_addr = 0xFF0000000ULL;
 
+static void ResetSubmissionLock(Platform::InterruptId irq) {
+    std::unique_lock lock{m_submission};
+    submission_lock = 0;
+    cv_lock.notify_all();
+}
+
+static void WaitGpuIdle() {
+    HLE_TRACE;
+    std::unique_lock lock{m_submission};
+    cv_lock.wait(lock, [] { return submission_lock == 0; });
+}
+
 static void DumpCommandList(std::span<const u32> cmd_list, const std::string& postfix) {
     using namespace Common::FS;
     const auto dump_dir = GetUserPath(PathType::PM4Dir);
@@ -465,14 +479,9 @@ void PS4_SYSV_ABI sceGnmDingDong(u32 gnm_vqid, u32 next_offs_dw) {
         return;
     }
 
-    std::unique_lock lock{m_submission};
+    WaitGpuIdle();
-    if (submission_lock != 0) {
-        liverpool->WaitGpuIdle();
 
-        // Suspend logic goes here
+    /* Suspend logic goes here */
-
-        submission_lock = 0;
-    }
 
     auto vqid = gnm_vqid - 1;
     auto& asc_queue = asc_queues[{vqid}];
@@ -1930,13 +1939,9 @@ s32 PS4_SYSV_ABI sceGnmSubmitCommandBuffers(u32 count, const u32* dcb_gpu_addrs[
         }
     }
 
-    if (submission_lock != 0) {
+    WaitGpuIdle();
-        liverpool->WaitGpuIdle();
 
-        // Suspend logic goes here
+    /* Suspend logic goes here */
-
-        submission_lock = 0;
-    }
 
     if (send_init_packet) {
         if (sdk_version <= 0x1ffffffu) {
@@ -1990,7 +1995,6 @@ int PS4_SYSV_ABI sceGnmSubmitDone() {
     if (!liverpool->IsGpuIdle()) {
         submission_lock = true;
     }
-    liverpool->NotifySubmitDone();
     send_init_packet = true;
     ++frames_submitted;
     return ORBIS_OK;
@@ -2471,6 +2475,9 @@ void RegisterlibSceGnmDriver(Core::Loader::SymbolsResolver* sym) {
         sdk_version = 0;
     }
 
+    Platform::IrqC::Instance()->Register(Platform::InterruptId::GpuIdle, ResetSubmissionLock,
+                                         nullptr);
+
     LIB_FUNCTION("b0xyllnVY-I", "libSceGnmDriver", 1, "libSceGnmDriver", 1, 1, sceGnmAddEqEvent);
     LIB_FUNCTION("b08AgtPlHPg", "libSceGnmDriver", 1, "libSceGnmDriver", 1, 1,
                  sceGnmAreSubmitsAllowed);

src/core/platform.h

@@ -26,6 +26,7 @@ enum class InterruptId : u32 {
     Compute6RelMem = 6u,
     GfxEop = 7u,
     GfxFlip = 8u,
+    GpuIdle = 9u,
 };
 
 using IrqHandler = std::function<void(InterruptId)>;

src/video_core/amdgpu/liverpool.cpp

@@ -66,21 +66,10 @@ void Liverpool::Process(std::stop_token stoken) {
             }
         }
 
-        if (submit_done) {
+        Platform::IrqC::Instance()->Signal(Platform::InterruptId::GpuIdle);
-            std::scoped_lock lk{submit_mutex};
-            submit_cv.notify_all();
-            submit_done = false;
-        }
     }
 }
 
-void Liverpool::WaitGpuIdle() {
-    RENDERER_TRACE;
-
-    std::unique_lock lk{submit_mutex};
-    submit_cv.wait(lk, [this] { return num_submits == 0; });
-}
-
 Liverpool::Task Liverpool::ProcessCeUpdate(std::span<const u32> ccb) {
     TracyFiberEnter(ccb_task_name);
 

src/video_core/amdgpu/liverpool.h

@@ -937,18 +937,10 @@ public:
     void SubmitGfx(std::span<const u32> dcb, std::span<const u32> ccb);
     void SubmitAsc(u32 vqid, std::span<const u32> acb);
 
-    void WaitGpuIdle();
-
     bool IsGpuIdle() const {
         return num_submits == 0;
     }
 
-    void NotifySubmitDone() {
-        std::scoped_lock lk{submit_mutex};
-        submit_done = true;
-        submit_cv.notify_all();
-    }
-
     void BindRasterizer(Vulkan::Rasterizer* rasterizer_) {
         rasterizer = rasterizer_;
     }
@@ -1017,7 +1009,6 @@ private:
     u32 num_submits{};
     std::mutex submit_mutex;
     std::condition_variable_any submit_cv;
-    std::atomic<bool> submit_done{};
 };
 
 static_assert(GFX6_3D_REG_INDEX(ps_program) == 0x2C08);