[d3d11] Map() optimization removed, needs buffer renaming

The naive optimization to use staging buffers rather than actual mapping
turned out to be no more efficient than the previous approach. In order
to achieve good performance, buffer renaming must be implemented instead.

src/d3d11/d3d11_buffer.cpp

@@ -91,44 +91,41 @@ namespace dxvk {
       if (MapFlags & D3D11_MAP_FLAG_DO_NOT_WAIT)
         return DXGI_ERROR_WAS_STILL_DRAWING;
       
-      if (MapType == D3D11_MAP_WRITE_DISCARD) {
-        // Instead of synchronizing with the device, which is
-        // highly inefficient, return a host-local buffer to
-        // the application and upload its contents on unmap()
-        // TODO evaluate whether this improves performance
-        m_mapData = new DxvkDataBuffer(buffer->info().size);
-        
-        pMappedSubresource->pData      = m_mapData->data();
-        pMappedSubresource->RowPitch   = buffer->info().size;
-        pMappedSubresource->DepthPitch = buffer->info().size;
-        return S_OK;
-      } else {
-        // We have to wait for the device to complete
-        pContext->Flush();
-        pContext->Synchronize();
-        
-        pMappedSubresource->pData      = buffer->mapPtr(0);
-        pMappedSubresource->RowPitch   = buffer->info().size;
-        pMappedSubresource->DepthPitch = buffer->info().size;
-        return S_OK;
-      }
+      // TODO optimize this. In order to properly cover common use cases
+      // like frequent constant buffer updates, we must implement buffer
+      // renaming techniques. The current approach is inefficient as it
+      // leads to a lot of Flush() and Synchronize() calls.
+      // 
+      // Possible solution:
+      //  (1) Create buffers with a significantly larger size if they
+      //      can be mapped by the host for writing. If mapping the
+      //      buffer would stall on D3D11_MAP_WRITE_DISCARD, map the
+      //      next slice. on D3D11_MAP_WRITE_NO_OVERWRITE, return the
+      //      current slice. If the buffer is bound, update bindings.
+      //  (2) If no more slices are available, create a new buffer.
+      //      Limit the number of buffers to a small, fixed number.
+      //  (3) If no more buffers are available, flush and synchronize.
+      //  (4) When renaming the buffer internally, all active bindings
+      //      need to be updated internally as well.
+      // 
+      // In order to support deferred contexts, the immediate context
+      // must commit all changes to the initial buffer slice prior to
+      // executing a command list. When mapping on deferred contexts,
+      // the deferred context shall create local buffer objects.
+      pContext->Flush();
+      pContext->Synchronize();
+      
+      pMappedSubresource->pData      = buffer->mapPtr(0);
+      pMappedSubresource->RowPitch   = buffer->info().size;
+      pMappedSubresource->DepthPitch = buffer->info().size;
+      return S_OK;
     }
   }
   
   
   void D3D11Buffer::Unmap(
           D3D11DeviceContext*       pContext) {
-    if (m_mapData != nullptr) {
-      const Rc<DxvkContext> context
-        = pContext->GetDXVKContext();
-        
-      context->updateBuffer(
-        m_resource->GetDXVKBuffer(),
-        0, m_mapData->size(),
-        m_mapData->data());
-      
-      m_mapData = nullptr;
-    }
+    // Nothing to see here, folks
   }
   
   

src/d3d11/d3d11_buffer.h

@@ -55,8 +55,6 @@ namespace dxvk {
     Com<IDXGIBufferResourcePrivate> m_resource;
     D3D11_BUFFER_DESC               m_desc;
     
-    Rc<DxvkDataBuffer>              m_mapData;
-    
   };
   
 }