NVIDIA releases OpenGL 4.2 drivers

[Groovounet]

You're right, the spec requires it. The driver never complained before...

It was a known an famous NVIDIA drivers bug. It's good to see it fixed despite that VAOs are nothing else than annoying most of the time.

A workaround and quick fix for you application is to create and bind a VAO at the beginning, (right after the context creation) you can then forget about it.

No the ultimate fixed but it get you application running again with no other change.

[Groovounet]

A little bit of fun, the following image show Fermi rasterizer pattern using the atomic counter:

[Chris Lux]

I have the following problem with immutable textures: I create a mipmapped 2d texture using glTexStorage2D, requesting the full amount of mip map levels. After this is upload the data of the single mip maps using glTexSubImage. The problem is the resulting texture objects does not have any mip maps. When i exchange the glTexStorage call by calls to glTexImag2D for all mip map levels the texture is correctly initialized with all mip maps filled.

NOT working:

glTexStorage2D(object_target(),
               init_mip_levels,
               util::gl_internal_format(in_desc._format),
               in_desc._size.x, in_desc._size.y);
// no error reported
for (unsigned i = 0; i < init_mip_levels; ++i) {
    math::vec2ui lev_size      = util::mip_level_dimensions(in_desc._size, i);
    const void*  init_lev_data = in_initial_mip_level_data[i];
    glTexSubImage2D(object_target(),
                    i,
                    0, 0,
                    lev_size.x, lev_size.y,
                    gl_base_format,
                    gl_base_type,
                    init_lev_data);
}
// still no errors reported

working:

for (unsigned i = 0; i < init_mip_levels; ++i) {
    math::vec2ui lev_size      = util::mip_level_dimensions(in_desc._size, i);
    glTexImage2D(object_target(),
                 i,
                 util::gl_internal_format(in_desc._format),
                 lev_size.x, lev_size.y,
                 0,
                 gl_base_format,
                 gl_base_type,
                 0);
}
 
// no error reported
for (unsigned i = 0; i < init_mip_levels; ++i) {
    math::vec2ui lev_size      = util::mip_level_dimensions(in_desc._size, i);
    const void*  init_lev_data = in_initial_mip_level_data[i];
    glTexSubImage2D(object_target(),
                    i,
                    0, 0,
                    lev_size.x, lev_size.y,
                    gl_base_format,
                    gl_base_type,
                    init_lev_data);
}
// still no errors reported

[mfort]

I have the following problem with immutable ...

Well, it works in my code. I've implemented it today.

The only road block was the incompatible internal format. It requires sized format. Such as GL_RGBA8 instead of GL_RGBA. The documentation states that correctly. (My fault).

I am going to test the performance changes ....

[Chris Lux]

Could you check that you really see trilinear or anisotropic filtering on your tests? I always see the base level as if I had only a single level texture in every test I did using sampler objects and plain texture parameters.

[Groovounet]

Works well for me like what follows but I haven't use the sampler object with it yet:

	gli::texture2D Image = gli::load(TEXTURE_DIFFUSE);
 
	glPixelStorei(GL_UNPACK_ALIGNMENT, 1);
 
	glGenTextures(1, &TextureName);
	glActiveTexture(GL_TEXTURE0);
	glBindTexture(GL_TEXTURE_2D, TextureName);
	glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_SWIZZLE_R, GL_RED);
	glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_SWIZZLE_G, GL_GREEN);
	glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_SWIZZLE_B, GL_BLUE);
	glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_SWIZZLE_A, GL_ALPHA);
	glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_BASE_LEVEL, 0);
	glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAX_LEVEL, 1000);
	glTexStorage2D(GL_TEXTURE_2D, GLint(Image.levels()), GL_RGBA8, GLsizei(Image[0].dimensions().x), GLsizei(Image[0].dimensions().y));
 
	for(std::size_t Level = 0; Level < Image.levels(); ++Level)
	{
		glTexSubImage2D(
			GL_TEXTURE_2D, 
			GLint(Level), 
			0, 0, 
			GLsizei(Image[Level].dimensions().x), 
			GLsizei(Image[Level].dimensions().y), 
			GL_BGRA, GL_UNSIGNED_BYTE, 
			Image[Level].data());
	}
 
	glPixelStorei(GL_UNPACK_ALIGNMENT, 4);

[Chris Lux]

Ok i triple checked using sampler objects and no sampler objects with essentially the code by Groovounet. I still do not get trilinear filtering (checked with colored mip levels), i can clearly see massive texture aliasing and only see the date from level 0. When switching to the glTexImage loop of my original post everything works as expected.

I am on Windows 7 x64 using the 280.28 driver. The context is a 4.2 core profile context (also checked with compatibility).

[Alfonse Reinheart]

It requires sized format.

It does? Oh thank God. It's about time they shoved those unsized formats out the door.

[Chris Lux]

Ok, Groovounet my friend , you have the exact same problem... you did not test what i was describing:

1. your dds image did not contain mipmaps!
2. you didn't even try to enable trilinear filtering!

you can find attached a modified sample and sample image which clearly shows the problem i described.

the fun part is this:

#if 1
    glTexStorage2D(GL_TEXTURE_2D, GLint(Image.levels()), GL_RGBA8, GLsizei(Image[0].dimensions().x), GLsizei(Image[0].dimensions().y));
#else
    for(std::size_t Level = 0; Level < Image.levels(); ++Level)
    {
        glTexImage2D(
            GL_TEXTURE_2D, 
            GLint(Level), 
            GL_RGBA8, 
            GLsizei(Image[Level].dimensions().x), 
            GLsizei(Image[Level].dimensions().y),
            0, 
            GL_BGRA, GL_UNSIGNED_BYTE, 
            0);
    }
#endif

[Groovounet]

LOL, i made a couple of tests and I was about to write that I had the same problem but you beat me that it!