Hi Wojciech,
Thank you for the very extensive explanation and tests.
It seems it is either a driver problem or I solved the problem
accidentally. Tests on my office PC worked with setting up multisampled
textures bound to a non-multisampled FBO. I'm able to "resolve" them
manually in the shader via texelFetch.
So if I get it correctly there is no implict blit if FBO is 0/0
multisamples and bound texture is multisampled.
Btw: Can you point me to the line where the implicit blitting is done,
so I can debug it?
cheers
Sebastian
Hi,
Ok. I used verb copy to avoid verb resolve, not being sure how
advanced you are... But yes that 'copy' is a resolve operation using
glBlitFramebuffer internally.
I have not tried multisampled textures as multirender targets so I
guess your case is bit more complex and I guess its the reason why
your results vary. My observations described earlier, come from
experiment with single color and single depth multisampled textures
attached as COLOR and DEPTH attachments. Attaching them with samples /
color samples = 0 was creating single FBO and rendered directly to
these textures. Attaching with samples / color samples = 8 was
internally creating render and resolve FBOs. First render FBO was
rendering to Renderbuffer with 8 samples. And attachment multisample
textures were bound to second resolve FBO. So final glBlitFarmebuffer
was 'copying' contents from Renderbuffer to my multisampled textures.
With first case (samples/color samples = 0) I was still able to use
the textures as input to some other shaders by declaring them as
sampler2DMS uniforms. Below is my shader I used to 'resolve' the
textures myself in input phase.
#version 150
#extension GL_ARB_texture_multisample : enable
uniform sampler2DMS colorTex;
uniform sampler2DMS depthTex;
const int NUM_SAMPLES = 8;
const float NUM_SAMPLES_INV = 1. / float( NUM_SAMPLES );
void main( void )
{
ivec2 itc = ivec2( gl_FragCoord.xy );
float depth = 0.;
vec4 color = vec4(0.);
for ( int i=0; i<NUM_SAMPLES; i++ ) {
depth += texelFetch( depthTex, itc, i ).x;
color += texelFetch( colorTex, itc, i );
}
gl_FragDepth = depth * NUM_SAMPLES_INV;
gl_FragColor = color * NUM_SAMPLES_INV;
}
Cheers,
Wojtek
2013/11/12 Sebastian Messerschmidt <[email protected]
<mailto:[email protected]>>
Am 11.11.2013 22:44, schrieb Wojciech Lewandowski:
Hi,
I guess answer to main question would be go ahead and add this.
Okay, I just wanted to make sure they are not left out intentionally.
But I just want to write about something else here. If you attach
Texture2DMultisample to Camera you would want to leave the number
of samples and color samples at 0. Thats because these params are
used to set up rendering to multisampled Renderbuffer objects and
in post step that Renderbuffer is copied to attached texture. If
you set Texture2DMultisample and additionaly set numbers of
samples in attach it will do rendering to Renderbuffer and then
will copy the result to your Texture2DMultisample attachment. I
guess its not what you are after.
That is strange. In this case I need some explanation. I want to
render several multisampled color attachments and resolve them in
later passes. In this case I create a multisampled color texture
and attach it to the camera. In consecutive passes I rebind those
textures as input and resolve them there. If I leave the
colorsamples and sample count at the FBO I get a
FRAMEBUFFER_ATTACHMENT_INCOMPLETE error for the first pass.
Setting them solved the problem and I got a correct rendering.
So what is the correct thing to do here?
You mentioned a copy. I always thought binding a texture to a FBO
and reusing it later in another FBO will not copy it, so I'm quite
puzzled. Am I'm doing something fundamentally wrong here?
cheers
Sebastian
Cheers,
Wojtek Lewandowski.
2013/11/11 Sebastian Messerschmidt
<[email protected]
<mailto:[email protected]>>
Hi,
maybe this seems like s stupid question but why does
osg::Texture2DMultisample does not have a getNumSamples()
function?
This might come in handy if, like in my case, a texture is
passed around for MRT binding (camera->attach), to determine
the number of samples for the attach function.
Would it be okay to add the get function?
cheers
Sebastian
_______________________________________________
osg-users mailing list
[email protected]
<mailto:[email protected]>
http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
_______________________________________________
osg-users mailing list
[email protected]
<mailto:[email protected]>
http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
_______________________________________________
osg-users mailing list
[email protected]
<mailto:[email protected]>
http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
_______________________________________________
osg-users mailing list
[email protected]
http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
_______________________________________________
osg-users mailing list
[email protected]
http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
