|
|
|
date: Fri, 13 Jun 2008 00:35:00 -0700,
group: microsoft.public.win32.programmer.directx.graphics.shaders
back
How to use SV_Target[n] semantics correctly?
Hi,
First of all, I'd post the same question at XNA Creators club. I post the
almost-the-same context here because no one answers me there. Sorry if I make
you read this long post again and again.
Recently, I'm working on a project requiring MRT functional. I tried to
implement my algorithm with MRT controlled at primitive level in a geometry
shader. The program works fine.
However, for comparing performace, I want to implement a version with MRT
controlled at pixel level in a pixel shader. In the old days, we set multiple
render targets, and set COLOR[n] semantics in the output of a pixel shaders.
With DX10, I guess I should do the task by setting a render target array, and
set SV_Target[n] semantics:
So here is the code where I create a render target array in size of 3.
//create render target array
D3D10_VIEWPORT vp;
UINT n = 1;
mpDev->RSGetViewports(&n,&vp);
D3D10_TEXTURE2D_DESC td;
td.ArraySize = 3;
td.BindFlags = D3D10_BIND_RENDER_TARGET | D3D10_BIND_SHADER_RESOURCE;
td.CPUAccessFlags = 0;
td.Format = DXGI_FORMAT_R32G32B32A32_FLOAT;
td.Height = vp.Height;
td.Width = vp.Width;
td.MipLevels = 1;
td.MiscFlags = 0;
td.SampleDesc.Count = 1;
td.SampleDesc.Quality = 0;
td.Usage = D3D10_USAGE_DEFAULT;
VALI_RN(mpDev->CreateTexture2D(&td,NULL,&mpMidRenderTarget),E_FAIL);
D3D10_RENDER_TARGET_VIEW_DESC rtv;
rtv.Format = td.Format;
rtv.ViewDimension = D3D10_RTV_DIMENSION_TEXTURE2DARRAY;
rtv.Texture2DArray.ArraySize = 3;
rtv.Texture2DArray.FirstArraySlice = 0;
rtv.Texture2DArray.MipSlice = 0;
mpDev->CreateRenderTargetView(mpMidRenderTarget,&rtv,&mpMidRenderTargetView);
//create mid shader resource view
memset(&srv_desc,0,sizeof(srv_desc));
srv_desc.Format = td.Format;
srv_desc.ViewDimension = D3D10_SRV_DIMENSION_TEXTURE2DARRAY;
srv_desc.Texture2DArray.ArraySize = ATTR_COUNT;
srv_desc.Texture2DArray.FirstArraySlice = 0;
srv_desc.Texture2DArray.MipLevels = td.MipLevels;
srv_desc.Texture2DArray.MostDetailedMip = 0;
mpDev->CreateShaderResourceView(mpMidRenderTarget,&srv_desc,&mpsrvMid);
In fact, this code works fine in my geometry-shader version of code. I
render what I want to each render target without pain!
However, with pixel shader I just can't render anything to the render
targets besides the first one.
The following is a simplified version of what I wrote, there must be
something wrong...
struct PSOut{
float4 color0 : SV_Target0;
float4 color1 : SV_Target1;
float4 color2 : SV_Target2;
};
PSOut PS(PSIn input){
PSOut output = (PSOut)output;
output.color0 = input.color0;
output.color1 = input.color1;
outpur.color2 = input.color2;
return output;
}
Simple enough, I just want to pass different attributes to different render
targets, and I failed.
Any help is welcome!
Thanks!!!
date: Fri, 13 Jun 2008 00:35:00 -0700
author: Southp
Re: How to use SV_Target[n] semantics correctly?
[Posted answer in creators.xna.com too]
The problem is what to use MRT (instead on render target array and deciding
where to render in GS) you have to use the same pattern as in DX9, i.e. set
multiple rendertarget views in call to ID3D10Device::OMSetRenderTargets().
You can use your existing Texture2DArray, you just need to query for each
separate slice of array and set these slices in SRT() call as separate
surfaces.
"Southp" wrote in message
news:D437E95E-2013-4C19-87D8-679804C1A549@microsoft.com...
>
> Hi,
> First of all, I'd post the same question at XNA Creators club. I post the
> almost-the-same context here because no one answers me there. Sorry if I
> make
> you read this long post again and again.
>
> Recently, I'm working on a project requiring MRT functional. I tried to
> implement my algorithm with MRT controlled at primitive level in a
> geometry
> shader. The program works fine.
>
> However, for comparing performace, I want to implement a version with MRT
> controlled at pixel level in a pixel shader. In the old days, we set
> multiple
> render targets, and set COLOR[n] semantics in the output of a pixel
> shaders.
> With DX10, I guess I should do the task by setting a render target array,
> and
> set SV_Target[n] semantics:
>
> So here is the code where I create a render target array in size of 3.
> //create render target array
> D3D10_VIEWPORT vp;
> UINT n = 1;
> mpDev->RSGetViewports(&n,&vp);
> D3D10_TEXTURE2D_DESC td;
> td.ArraySize = 3;
> td.BindFlags = D3D10_BIND_RENDER_TARGET | D3D10_BIND_SHADER_RESOURCE;
> td.CPUAccessFlags = 0;
> td.Format = DXGI_FORMAT_R32G32B32A32_FLOAT;
> td.Height = vp.Height;
> td.Width = vp.Width;
> td.MipLevels = 1;
> td.MiscFlags = 0;
> td.SampleDesc.Count = 1;
> td.SampleDesc.Quality = 0;
> td.Usage = D3D10_USAGE_DEFAULT;
> VALI_RN(mpDev->CreateTexture2D(&td,NULL,&mpMidRenderTarget),E_FAIL);
>
> D3D10_RENDER_TARGET_VIEW_DESC rtv;
> rtv.Format = td.Format;
> rtv.ViewDimension = D3D10_RTV_DIMENSION_TEXTURE2DARRAY;
> rtv.Texture2DArray.ArraySize = 3;
> rtv.Texture2DArray.FirstArraySlice = 0;
> rtv.Texture2DArray.MipSlice = 0;
>
> mpDev->CreateRenderTargetView(mpMidRenderTarget,&rtv,&mpMidRenderTargetView);
>
> //create mid shader resource view
> memset(&srv_desc,0,sizeof(srv_desc));
> srv_desc.Format = td.Format;
> srv_desc.ViewDimension = D3D10_SRV_DIMENSION_TEXTURE2DARRAY;
> srv_desc.Texture2DArray.ArraySize = ATTR_COUNT;
>
>
> srv_desc.Texture2DArray.FirstArraySlice = 0;
> srv_desc.Texture2DArray.MipLevels = td.MipLevels;
> srv_desc.Texture2DArray.MostDetailedMip = 0;
>
> mpDev->CreateShaderResourceView(mpMidRenderTarget,&srv_desc,&mpsrvMid);
>
> In fact, this code works fine in my geometry-shader version of code. I
> render what I want to each render target without pain!
>
> However, with pixel shader I just can't render anything to the render
> targets besides the first one.
>
> The following is a simplified version of what I wrote, there must be
> something wrong...
>
>
>
>
> struct PSOut{
> float4 color0 : SV_Target0;
> float4 color1 : SV_Target1;
> float4 color2 : SV_Target2;
> };
>
> PSOut PS(PSIn input){
> PSOut output = (PSOut)output;
> output.color0 = input.color0;
> output.color1 = input.color1;
> outpur.color2 = input.color2;
> return output;
> }
>
>
> Simple enough, I just want to pass different attributes to different
> render
> targets, and I failed.
>
> Any help is welcome!
>
> Thanks!!!
>
>
date: Sun, 15 Jun 2008 15:35:55 +0400
author: Alexey Barkovoy
|
|
