Hello, everyone. OpenGL question here.
I am trying to use DEPTH32F_STENCIL8 format with the FLOAT_32_UNSIGNED_INT_24_8_REV type. My use case is the following. I am doing deferred rendering. Depth is generated in the geometry pass and I reconstruct position from depth during lighting passes. Before I do a lighting pass and use depth for getting the position, I stencil my light volume. When I sample depth in the lighting pass, I don't do writes to the depth_stencil texture (neither depth values, nor stencil values).
What I get is shown in depth32F_stencil8.jpg attachment.
Using FLOAT32F I lose stencilling, but everything looks fine, see depth32F.jpg attachment.
Same issue when using DEPTH24_STENCIL8 format with UNSIGNED_INT_24_8 type.
I am aware of TexParameter GL_DEPTH_STENCIL_TEXTURE_MODE, which should be set (and is by default set) to DEPTH_COMPONENT to indicate which value I want to get when I sample. However this parameter is only available since GL 4.3, whereas I am using GL 3.3.
Configuration: AMD HD5145, Win7 x64, 13.1 Legacy Driver.
Question is, should I expect reading depth from packed depth_stencil texture to work under GL 3.3 or is this a bug? I don't have access to Nvidia hardware, so I can't tell if it behaves the same.
I've seen pretty much the same questions here and the OpenGL forum. It was suggested by a person from AMD that a fix was supposed to hit the driver quite some time ago. Is AMD aware this is still an issue? This is kind of a big deal.
I am willing to conjure up a code sample for verification and testing purposes, if someone from AMD shows up.
I have similar issues with the 24bit EXT_packed_depth_stencil format across multiple drivers, if i switch to the straight 32bit depth with no stencil it works fine. http://devgurus.amd.com/message/1286318#1286318
Im also finding the problems when reconstructing position from depth, all of these problems seem to be AMD related. When I run it on my nvidia 670 I dont get the problem, so there seems to be a problem with packed depth stencil formats.
The 24_8 format was promoted to core 3.0.