XNA 4.0 beta: EffectPass not a simple wrapper!

　　So, the beta is out!

      XNA 4.0進行了大幅更新，大部分改動都不錯，不過一直有2點讓我非常不爽：1. render target必須和某個depth buffer綁定到一起；2. 沒有了SetxxShader和SetShaderConstant等底層函數。前者對deffered shading非常不友好，後者讓我寫的整個材質系統都不能用了。相比之下，後者更讓我擔心，因為CTP中的Effect和EffectPass僅僅是DX Effect系統之上的一個簡單wrapper眾所周知，DirectX中Effect的效能一直是被人詬病的地方。好在Shawn告訴我：

“The Windows EffectPass.Apply implementation in the CTP is not complete, and many optimizations not yet implemented. The CTP implementation is functionally correct, but not from a performance standpoint, so you should not worry too much about Windows performance in the CTP.”

 

     似乎Beta版已經解決了這個問題，4.0中的EffectPass不再是DX Effect上的一個簡單wrapper，xna team做了很大的努力減少狀態改變，以至於PIX分析xna 4.0的程式時有可能出錯，Shawn在blog中說：

"The problem occurs because PIX does not support some crazy tricks that GS4 uses to optimize state management" 。

 

      下面是同一段代碼在CTP和beta版下的對比：

//test code//
.....set parameter for first object
effectPass.Apply();
graphicsDevice.DrawIndexedPrimitives(PrimitiveType.TriangleList, 0, 0,vertices.Count, 0, primitiveCount);
//only change world matrix and draw the same object again
effect.World = Matrix.CreateTranslation(1, 0, 0);
effectPass.Apply();
graphicsDevice.DrawIndexedPrimitives(PrimitiveType.TriangleList, 0, 0,vertices.Count, 0, primitiveCount);

//Pix data for ctp:
ID3DXEffect::BeginPass(0)                          
ID3DXEffect::EndPass()                         
ID3DXEffect::End()                         
IDirect3DDevice9::BeginScene()                         
IDirect3DDevice9::DrawIndexedPrimitive(D3DPT_TRIANGLELIST, 0, 0, 24, 0, 12)//draw the 1 primitive                  
ID3DXEffect::SetMatrix(0xFF6B4D6B, 0x0039E6F0)                     
ID3DXEffect::SetMatrix(0xFF6B4E13, 0x0039E684)                     
ID3DXEffect::SetMatrix(0xFF6B4DBF, 0x0039E684)                     
ID3DXEffect::Begin(0x0039E7EC, 0x00000001)                 
ID3DXEffect::BeginPass(0)                          
 IDirect3DDevice9::SetVertexShader(0x06487820)                     //redundance!!!!!!
 IDirect3DDevice9::SetVertexShaderConstantF(0, 0x015851C8, 1)      //redundance!!!!!!!!      
 IDirect3DDevice9::SetVertexShaderConstantF(14, 0x015852A8, 1)     //redundance !!!!!!    
 IDirect3DDevice9::SetVertexShaderConstantF(15, 0x015852B8, 4)             
 IDirect3DDevice9::SetVertexShaderConstantF(19, 0x015852F8, 3)             
 IDirect3DDevice9::SetVertexShaderConstantF(23, 0x01585338, 3)             
 IDirect3DDevice9::SetPixelShader(0x06486AE8)   3914655838         //redundance!!!!!
 IDirect3DDevice9::SetPixelShaderConstantF(0, 0x01586D40, 15)      //redundance!!!!!!    
ID3DXEffect::EndPass()                         
ID3DXEffect::End()                     
IDirect3DDevice9::DrawIndexedPrimitive(D3DPT_TRIANGLELIST, 0, 0, 24, 0, 12)// draw the 2 primitive      

//Pix data for beta with d3dx function call
ID3DXEffect::BeginPass(0)                
IDirect3DDevice9::BeginScene()
IDirect3DDevice9::DrawIndexedPrimitive(D3DPT_TRIANGLELIST, 0, 0, 24, 0, 12)        
ID3DXEffect::SetMatrix(0xFF84502B, 0x0023E444)                    
ID3DXEffect::SetMatrix(0xFF8450D3, 0x0023E3D8)                    
ID3DXEffect::SetMatrix(0xFF84507F, 0x0023E3D8)                    
ID3DXEffect::CommitChanges()                    
  IDirect3DDevice9::SetVertexShaderConstantF(15, 0x00FB5668, 4)
  IDirect3DDevice9::SetVertexShaderConstantF(19, 0x00FB56A8, 3)
  IDirect3DDevice9::SetVertexShaderConstantF(23, 0x00FB56E8, 3)
IDirect3DDevice9::DrawIndexedPrimitive(D3DPT_TRIANGLELIST, 0, 0, 24, 0, 12)

//Pix data for beta without d3dx function:
IDirect3DDevice9::BeginScene()                
IDirect3DDevice9::DrawIndexedPrimitive(D3DPT_TRIANGLELIST, 0, 0, 24, 0, 12)
IDirect3DDevice9::SetVertexShaderConstantF(15, 0x007D5660, 4)
IDirect3DDevice9::SetVertexShaderConstantF(19, 0x007D56A0, 3)
IDirect3DDevice9::SetVertexShaderConstantF(23, 0x007D56E0, 3)
IDirect3DDevice9::DrawIndexedPrimitive(D3DPT_TRIANGLELIST, 0, 0, 24, 0, 12)                    

 

 

     可以看到beta版中，完全過濾了冗餘的狀態設定，測試代碼中的effect.World會更新world marix，worldViewProjMatrix以及WorldInverseTranspose，因此渲染第二個對象前出現3個參數更新是正確的。測試程式裡只渲染了2個圖形，因此，BeginScene在第一次DP之前出現也是正確的。