• Advertisement
Sign in to follow this  

DX11 Puzzling Fill Rate

This topic is 2140 days old which is more than the 365 day threshold we allow for new replies. Please post a new topic.

If you intended to correct an error in the post then please contact us.

Recommended Posts

I've written a deferred renderer for DX11 and am optimizing its performance. While benchmarking fill rate, I ran into something puzzling. When I stub out one of the shaders used to render the scene (pre-lighting) so that it essentially has a null VS and its PS simply writes out constants to the GBuffers, I see a pretty significant increase in performance when rendering the scene. That seemed reasonable.

So then I started thinking about doing a depth prepass, but before going through the work of implementing it, I decided to first try a simple test to see if it had potential. For the test, I simply cleared the depth buffer to 0 instead of 1. The idea is that every pixel in the scene would then be rejected before the PS and I would see pretty much the same significant increase in performance as with the null shader above. However, I saw absolutely no speed increase at all.

Does this imply that my PS is being executed even if it's occluded by closer depth values? How is that possible?

Here's the PS and the functions it calls (see PSMain for the pixel shader):

[code]
StructuredBuffer<SpecBuf_Params_s> SpecBuf_Params;
SpecBuf_Params_s GetParams( in uint uInstIndex, in uint uSubMtlIndex ) {
return SpecBuf_Params[ Spec_GetParamsIndex( uInstIndex, uSubMtlIndex ) ];
}

float4 Spec_Motif( SpecMotif_s Motif ) {
float4 vTableColor = g_avMotifColor[ Motif.m_uMotifIndex ];
float4 vBiasedColor = (vTableColor * Motif.m_fScale + Motif.m_fOffset) * Motif.m_vBaseColor;
float3 vFinalColor = (Motif.m_uFlags & 1) ? Motif.m_vBaseColor.rgb : vBiasedColor.rgb;
float fFinalAlpha = (Motif.m_uFlags & 2) ? Motif.m_vBaseColor.a : vBiasedColor.a;
return float4( vFinalColor, fFinalAlpha );
}

uint Spec_AlphaToCoverage( float fUnitAlpha ) {
uint uCoverage;
#if SPEC_MSAA_COUNT == 2
if( fUnitAlpha < (1.0f / 3.0f) ) {
uCoverage = 0;
} else if( fUnitAlpha < (2.0f / 3.0f) ) {
uCoverage = 1;
} else {
uCoverage = 3;
}
#elif SPEC_MSAA_COUNT == 4
if( fUnitAlpha < (1.0f / 5.0f) ) {
uCoverage = 0;
} else if( fUnitAlpha < (2.0f / 5.0f) ) {
uCoverage = 1;
} else if( fUnitAlpha < (3.0f / 5.0f) ) {
uCoverage = 3;
} else if( fUnitAlpha < (4.0f / 5.0f) ) {
uCoverage = 7;
} else {
uCoverage = 15;
}
#else
uCoverage = 0xffffffff;
#endif
return uCoverage;
}

SpecRawGBuffer_s Spec_PackGBuffer( in SpecGBufferSource_s Source ) {
SpecRawGBuffer_s RawGBuffer;
// Compute flags field for Tex2...
uint uEdgePixel = any( frac( Source.m_vCentroidPosXY_SS ) - 0.5f );
uint uFlags = (Source.m_uNoAO << 5) | ((uEdgePixel & 1) << 4) | max( min( uint( Source.m_fSpecUnitSharpness * 15.0f ), 15 ), 1 );
// Store values in packed GBuffer...
RawGBuffer.m_vTex0 = float4( Source.m_vDiffuseColor, 0 );
RawGBuffer.m_vTex1 = float4( Source.m_vEmissiveColor, Source.m_fSpecUnitIntensity );
RawGBuffer.m_vuTex2 = uint4( (255.0f/2.0f) + (255.0f/2.0f)*Source.m_vUnitNorm_WS, uFlags );
return RawGBuffer;
}

SpecRawGBuffer_s PSMain( VS_Out Input, out uint uCoverage : SV_Coverage ) : SV_TARGET {
SpecBuf_Params_s Params = GetParams( Input.uInstanceID, Input.uSubMtlIndex );
uint uFlags = Params.m_uFlags;
uint uFlag_VtxRGB_Tint = uFlags & FLAG_VTX_RGB_TINT;
uint uFlag_VtxRGB_Emis = uFlags & FLAG_VTX_RGB_EMIS;
uint uFlag_VtxA_Tint = uFlags & FLAG_VTX_A_TINT;
uint uFlag_VtxA_Emis = uFlags & FLAG_VTX_A_EMIS;
uint uFlag_VtxA_Opac = uFlags & FLAG_VTX_A_OPAC;
uint uFlag_VtxA_Glos = uFlags & FLAG_VTX_A_GLOS;
uint uFlag_BaseA_Emis = uFlags & FLAG_BASE_A_EMIS;
uint uFlag_BaseA_Opac = uFlags & FLAG_BASE_A_OPAC;
uint uFlag_BaseA_Glos = uFlags & FLAG_BASE_A_GLOS;

float4 vMotifTintOpac = Spec_Motif( Params.m_MotifTintOpac );
float4 vMotifEmisGlos = Spec_Motif( Params.m_MotifEmisGlos );
float3 vVtxTint = (uFlag_VtxRGB_Tint ? Input.vColorVtx.rgb : 1) * (uFlag_VtxA_Tint ? Input.vColorVtx.a : 1);
float3 vVtxEmis = Params.m_fAddEmis + (uFlag_VtxRGB_Emis ? Input.vColorVtx.rgb : 0) + (uFlag_VtxA_Emis ? Input.vColorVtx.a : 0);
float fVtxOpac = (uFlag_VtxA_Opac ? Input.vColorVtx.a : 1);
float fVtxGlos = (uFlag_VtxA_Glos ? Input.vColorVtx.a : 0);
float4 vVtxTintOpac = float4( vVtxTint, fVtxOpac ) * vMotifTintOpac;
float4 vVtxEmisGlos = float4( vVtxEmis, fVtxGlos );

// Compute normal...
float3 vUnitNorm_WS = normalize( Input.vNormal_WS );

// Compute base color...
float3 vTC_BaseRGB = float3( Input.vTC_Base.xy, Params.m_uTexSliceIndexBaseRGB );
float3 vTC_BaseA = float3( Params.m_fTexCoordScale_BaseA * Input.vTC_Base.zw, Params.m_uTexSliceIndexBaseA );
float3 vTexColorBaseRGB = TexBase.Sample( SamplerBase, vTC_BaseRGB ).rgb;
float fTexColorBaseA = TexBase.Sample( SamplerBase, vTC_BaseA ).a;
float fDetailMult = lerp( 1, 2 * fTexColorBaseA, Params.m_Switch_fBaseA_Detl );
vTexColorBaseRGB *= fDetailMult;

// Compute reflection color...
float4 vMotifCube = Spec_Motif( Params.m_MotifCube );
float3 vUnitVtxToCam_WS = normalize( Spec_GetCamPos() - Input.vPos_WS );
float3 vUnitReflect_WS = reflect( -vUnitVtxToCam_WS, vUnitNorm_WS );
float3 vReflectionColor = vMotifCube.rgb * TexCube.Sample( SamplerCube, float4( vUnitReflect_WS, Params.m_uTexSliceIndexCube ) ).rgb;

// Compute final values...
float fFinalOpac = Input.fUnitFadeAlpha * vVtxTintOpac.a * (uFlag_BaseA_Opac ? fTexColorBaseA : 1);
float fFinalGlos = vMotifEmisGlos.a * (vVtxEmisGlos.a + (uFlag_BaseA_Glos ? fTexColorBaseA : 0));
float3 vFinalDiff = saturate( vVtxTintOpac.rgb * vTexColorBaseRGB ) + fFinalGlos * vReflectionColor;
float3 vFinalEmis = vMotifEmisGlos.rgb * (vVtxEmisGlos.rgb + (uFlag_BaseA_Emis ? fTexColorBaseA : 0));
uCoverage = Spec_AlphaToCoverage( fFinalOpac );

// Store everything into our gbuffers...
SpecGBufferSource_s GBufSource = Spec_GetDefaultGBufSource();
GBufSource.m_vDiffuseColor = vFinalDiff;
GBufSource.m_vEmissiveColor = vFinalDiff * vFinalEmis;
GBufSource.m_vCentroidPosXY_SS = Input.vPos_HS.xy;
GBufSource.m_vUnitNorm_WS = vUnitNorm_WS;
GBufSource.m_fSpecUnitSharpness = Params.m_fSpecUnitSharpness;
GBufSource.m_fSpecUnitIntensity = fFinalGlos;
return Spec_PackGBuffer( GBufSource );
}
[/code]

Edit: After some more testing, it's really looking like early Z rejection just isn't working. Though, I'm not sure why yet. Do NVIDIA and ATI provide docs that describe the conditions which must be met to keep early Z rejection enabled?

Share this post


Link to post
Share on other sites
Advertisement
Nvidia has some guidelines in [url="http://developer.download.nvidia.com/GPU_Programming_Guide/GPU_Programming_Guide_G80.pdf"]this doc[/url], but they might be a but out of date depending on which hardware you're working with. Alpha to coverage or outputting SV_Coverage can definitely mess with Z cull, so you might want to try disabling that to see if it makes a difference.

Also, if you want to see whether your pixel shader is actually running you can use the D3D11_QUERY_DATA_PIPELINE_STATISTICS to get the number of pixel shader invocations.

Share this post


Link to post
Share on other sites
Thanks for the info MJP. I was indeed not meeting some of those requirements. However, even after fixing things up the query reports no change in the number of PS invocations.

- PS no longer outputs SV_Coverage.
- Using ClearDepthStencilView() to clear the depth buffer.
- PS doesn't write depth.
- The direction of the depth test is <= while both writing and comparing the depth buffer and doesn't change in between.
- Depth buffer is a Texture2DMS (no array).
- The PS uses the XY components of the SV_Position semantic, but not the z component.
- The PS doesn't use clip, texkil, or discard.
- Alpha to coverage is disabled
- SampleMask is always 0xFFFFFFFF.

The depth buffer is, however, DXGI_FORMAT_D32_FLOAT, but the NVIDIA doc doesn't list that as a reason early Z would be disabled. I don't have a stencil buffer. I'm am using 2x MSAA render targets. I can see in Pix that the depth buffer has been written to with the pre-pass.

When writing the depth, I bind a read/write DSV to the pipeline and use this state:

D3D11_BLEND_DESC:
AlphaToCoverageEnable = 0
IndependentblenEnable = 0
BlendEnable = 0
SrcBlend = ONE
DestBlend = ZERO
BlendOp = ADD
SrcBlendAlpha = ONE
DestBlendAlpha = ZERO
BlendOpAlpha = ADD
RenderTargetWriteMask = 0

D3D11_DEPTH_STENCIL_DESC:
DepthEnable = 1
DepthWriteMask = ALL
DepthFunc = LESS_EQUAL
(all stencil members are 0)

D3D11_RASTERIZER_DESC:
FillMode = SOLID
CullMode = BACK
FrontCounterClockwise = 1
DepthBias = 0
DepthBiasClamp = 0
SlopeScaledDepthBias = 0
DepthClipEnable = 1
ScissorEnable = 0
MultisampleEnable = 0
AntialiasedLineEnable = 0


When rendering the scene, I bind a read-only DSV to the pipeline and use this state:

D3D11_BLEND_DESC:
AlphaToCoverageEnable = 0
IndependentblenEnable = 0
BlendEnable = 0
SrcBlend = ONE
DestBlend = ZERO
BlendOp = ADD
SrcBlendAlpha = ONE
DestBlendAlpha = ZERO
BlendOpAlpha = ADD
RenderTargetWriteMask = 15

D3D11_DEPTH_STENCIL_DESC:
DepthEnable = 1
DepthWriteMask = ZERO
DepthFunc = LESS_EQUAL
(all stencil members are 0)

D3D11_RASTERIZER_DESC:
FillMode = SOLID
CullMode = BACK
FrontCounterClockwise = 1
DepthBias = 0
DepthBiasClamp = 0
SlopeScaledDepthBias = 0
DepthClipEnable = 1
ScissorEnable = 0
MultisampleEnable = 0
AntialiasedLineEnable = 0

Here's the depth prepass shader:

[code]
// Constant buffer with cam info:
cbuffer SpecBuf_Camera {
row_major float4x4 g_ProjCamMtx;
};

// Vertex in:
struct VS_In {
float3 vPos_WS : POSITION;
};

// Vertex out:
struct VS_Out {
float4 vPos_HS : SV_POSITION;
};

// Vertex shader:
VS_Out VS( VS_In Input ) {
VS_Out Output;
Output.vPos_HS = mul( g_ProjCamMtx, float4( Input.vPos_WS, 1 ) );
return Output;
}

// Pixel shader:
float4 PS( VS_Out Input ) : SV_TARGET {
return 0;
}

// Technique:
technique11 Terrain {
pass P1 {
SetVertexShader( CompileShader( vs_5_0, VS() ) );
SetPixelShader( CompileShader( ps_5_0, PS() ) );
}
}
[/code]


What else could I be doing to turn off early Z?

Share this post


Link to post
Share on other sites
Matt, thanks to your book for pointing out the [earlydepthstencil] attribute for pixel shaders! I tried this and the query now reports over a million invocations of the PS have been eliminated (without a depth pre-pass - just drawing the scene from front to back as much as possible). Though, I didn't notice much of a gain in terms of performance. Doing a full depth prepass increases the draw call count substantially which negatively impacts frame rate.

If anyone's interested, the book I'm referring to is called Practical Rendering & Computation with Direct3D 11, and it's proven to be a valuable resource to me during my adventure through DX11.

Share this post


Link to post
Share on other sites
Sign in to follow this  

  • Advertisement
  • Advertisement
  • Popular Now

  • Advertisement
  • Similar Content

    • By evelyn4you
      hi,
      until now i use typical vertexshader approach for skinning with a Constantbuffer containing the transform matrix for the bones and an the vertexbuffer containing bone index and bone weight.
      Now i have implemented realtime environment  probe cubemaping so i have to render my scene from many point of views and the time for skinning takes too long because it is recalculated for every side of the cubemap.
      For Info i am working on Win7 an therefore use one Shadermodel 5.0 not 5.x that have more options, or is there a way to use 5.x in Win 7
      My Graphic Card is Directx 12 compatible NVidia GTX 960
      the member turanszkij has posted a good for me understandable compute shader. ( for Info: in his engine he uses an optimized version of it )
      https://turanszkij.wordpress.com/2017/09/09/skinning-in-compute-shader/
      Now my questions
       is it possible to feed the compute shader with my orignial vertexbuffer or do i have to copy it in several ByteAdressBuffers as implemented in the following code ?
        the same question is about the constant buffer of the matrixes
       my more urgent question is how do i feed my normal pipeline with the result of the compute Shader which are 2 RWByteAddressBuffers that contain position an normal
      for example i could use 2 vertexbuffer bindings
      1 containing only the uv coordinates
      2.containing position and normal
      How do i copy from the RWByteAddressBuffers to the vertexbuffer ?
       
      (Code from turanszkij )
      Here is my shader implementation for skinning a mesh in a compute shader:
      1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 struct Bone { float4x4 pose; }; StructuredBuffer<Bone> boneBuffer;   ByteAddressBuffer vertexBuffer_POS; // T-Pose pos ByteAddressBuffer vertexBuffer_NOR; // T-Pose normal ByteAddressBuffer vertexBuffer_WEI; // bone weights ByteAddressBuffer vertexBuffer_BON; // bone indices   RWByteAddressBuffer streamoutBuffer_POS; // skinned pos RWByteAddressBuffer streamoutBuffer_NOR; // skinned normal RWByteAddressBuffer streamoutBuffer_PRE; // previous frame skinned pos   inline void Skinning(inout float4 pos, inout float4 nor, in float4 inBon, in float4 inWei) {  float4 p = 0, pp = 0;  float3 n = 0;  float4x4 m;  float3x3 m3;  float weisum = 0;   // force loop to reduce register pressure  // though this way we can not interleave TEX - ALU operations  [loop]  for (uint i = 0; ((i &lt; 4) &amp;&amp; (weisum&lt;1.0f)); ++i)  {  m = boneBuffer[(uint)inBon].pose;  m3 = (float3x3)m;   p += mul(float4(pos.xyz, 1), m)*inWei;  n += mul(nor.xyz, m3)*inWei;   weisum += inWei;  }   bool w = any(inWei);  pos.xyz = w ? p.xyz : pos.xyz;  nor.xyz = w ? n : nor.xyz; }   [numthreads(1024, 1, 1)] void main( uint3 DTid : SV_DispatchThreadID ) {  const uint fetchAddress = DTid.x * 16; // stride is 16 bytes for each vertex buffer now...   uint4 pos_u = vertexBuffer_POS.Load4(fetchAddress);  uint4 nor_u = vertexBuffer_NOR.Load4(fetchAddress);  uint4 wei_u = vertexBuffer_WEI.Load4(fetchAddress);  uint4 bon_u = vertexBuffer_BON.Load4(fetchAddress);   float4 pos = asfloat(pos_u);  float4 nor = asfloat(nor_u);  float4 wei = asfloat(wei_u);  float4 bon = asfloat(bon_u);   Skinning(pos, nor, bon, wei);   pos_u = asuint(pos);  nor_u = asuint(nor);   // copy prev frame current pos to current frame prev pos streamoutBuffer_PRE.Store4(fetchAddress, streamoutBuffer_POS.Load4(fetchAddress)); // write out skinned props:  streamoutBuffer_POS.Store4(fetchAddress, pos_u);  streamoutBuffer_NOR.Store4(fetchAddress, nor_u); }  
    • By mister345
      Hi, can someone please explain why this is giving an assertion EyePosition!=0 exception?
       
      _lightBufferVS->viewMatrix = DirectX::XMMatrixLookAtLH(XMLoadFloat3(&_lightBufferVS->position), XMLoadFloat3(&_lookAt), XMLoadFloat3(&up));
      It looks like DirectX doesnt want the 2nd parameter to be a zero vector in the assertion, but I passed in a zero vector with this exact same code in another program and it ran just fine. (Here is the version of the code that worked - note XMLoadFloat3(&m_lookAt) parameter value is (0,0,0) at runtime - I debugged it - but it throws no exceptions.
          m_viewMatrix = DirectX::XMMatrixLookAtLH(XMLoadFloat3(&m_position), XMLoadFloat3(&m_lookAt), XMLoadFloat3(&up)); Here is the repo for the broken code (See LightClass) https://github.com/mister51213/DirectX11Engine/blob/master/DirectX11Engine/LightClass.cpp
      and here is the repo with the alternative version of the code that is working with a value of (0,0,0) for the second parameter.
      https://github.com/mister51213/DX11Port_SoftShadows/blob/master/Engine/lightclass.cpp
    • By mister345
      Hi, can somebody please tell me in clear simple steps how to debug and step through an hlsl shader file?
      I already did Debug > Start Graphics Debugging > then captured some frames from Visual Studio and
      double clicked on the frame to open it, but no idea where to go from there.
       
      I've been searching for hours and there's no information on this, not even on the Microsoft Website!
      They say "open the  Graphics Pixel History window" but there is no such window!
      Then they say, in the "Pipeline Stages choose Start Debugging"  but the Start Debugging option is nowhere to be found in the whole interface.
      Also, how do I even open the hlsl file that I want to set a break point in from inside the Graphics Debugger?
       
      All I want to do is set a break point in a specific hlsl file, step thru it, and see the data, but this is so unbelievably complicated
      and Microsoft's instructions are horrible! Somebody please, please help.
       
       
       

    • By mister345
      I finally ported Rastertek's tutorial # 42 on soft shadows and blur shading. This tutorial has a ton of really useful effects and there's no working version anywhere online.
      Unfortunately it just draws a black screen. Not sure what's causing it. I'm guessing the camera or ortho matrix transforms are wrong, light directions, or maybe texture resources not being properly initialized.  I didnt change any of the variables though, only upgraded all types and functions DirectX3DVector3 to XMFLOAT3, and used DirectXTK for texture loading. If anyone is willing to take a look at what might be causing the black screen, maybe something pops out to you, let me know, thanks.
      https://github.com/mister51213/DX11Port_SoftShadows
       
      Also, for reference, here's tutorial #40 which has normal shadows but no blur, which I also ported, and it works perfectly.
      https://github.com/mister51213/DX11Port_ShadowMapping
       
    • By xhcao
      Is Direct3D 11 an api function like glMemoryBarrier in OpenGL? For example, if binds a texture to compute shader, compute shader writes some values to texture, then dispatchCompute, after that, read texture content to CPU side. I know, In OpenGL, we could call glMemoryBarrier before reading to assure that texture all content has been updated by compute shader.
      How to handle incoherent memory access in Direct3D 11? Thank you.
  • Advertisement