[d3d11] Flush more aggressively on stalling Event queries

Increases GPU utilization in Quake Champions.

src/d3d11/d3d11_context_imm.cpp

@@ -80,17 +80,33 @@ namespace dxvk {
     SynchronizeCsThread();
 
     // Get query status directly from the query object
-    HRESULT hr = static_cast<D3D11Query*>(pAsync)->GetData(pData, GetDataFlags);
+    auto query = static_cast<D3D11Query*>(pAsync);
+    HRESULT hr = query->GetData(pData, GetDataFlags);
     
     // If we're likely going to spin on the asynchronous object,
     // flush the context so that we're keeping the GPU busy
-    if (hr == S_FALSE)
+    if (hr == S_FALSE) {
+      query->NotifyStall();
       FlushImplicit(FALSE);
+    }
     
     return hr;
   }
   
   
+  void STDMETHODCALLTYPE D3D11ImmediateContext::End(ID3D11Asynchronous* pAsync) {
+    D3D11DeviceContext::End(pAsync);
+
+    auto query = static_cast<D3D11Query*>(pAsync);
+    if (unlikely(query && query->IsEvent())) {
+      query->NotifyEnd();
+      query->IsStalling()
+        ? Flush()
+        : FlushImplicit(TRUE);
+    }
+  }
+
+
   void STDMETHODCALLTYPE D3D11ImmediateContext::Flush() {
     m_parent->FlushInitContext();
     

src/d3d11/d3d11_context_imm.h

@@ -32,6 +32,8 @@ namespace dxvk {
             UINT                              DataSize,
             UINT                              GetDataFlags);
     
+    void STDMETHODCALLTYPE End(ID3D11Asynchronous *pAsync);
+    
     void STDMETHODCALLTYPE Flush();
     
     void STDMETHODCALLTYPE ExecuteCommandList(

src/d3d11/d3d11_query.h

@@ -46,6 +46,23 @@ namespace dxvk {
             UINT                              GetDataFlags);
     
     DxvkBufferSlice GetPredicate(DxvkContext* ctx);
+
+    bool IsEvent() const {
+      return m_desc.Query == D3D11_QUERY_EVENT;
+    }
+
+    bool IsStalling() const {
+      return m_stallFlag;
+    }
+
+    void NotifyEnd() {
+      m_stallMask <<= 1;
+    }
+
+    void NotifyStall() {
+      m_stallMask |= 1;
+      m_stallFlag |= bit::popcnt(m_stallMask) >= 16;
+    }
     
     D3D10Query* GetD3D10Iface() {
       return &m_d3d10;
@@ -66,6 +83,9 @@ namespace dxvk {
 
     D3D10Query m_d3d10;
 
+    uint32_t m_stallMask = 0;
+    bool     m_stallFlag = false;
+
     UINT64 GetTimestampQueryFrequency() const;
     
   };