14 replies to this topic

[Chris_F]

To add to the ever growing ways in which OpenGL simply confuses the crap out of me, I have noticed when inspecting GL_EXTENSIONS and wglGetExtensionsStringARB that neither GL_ARB_robustness or WGL_ARB_create_context_robustness appear.

Doesn't ARB mean that they are part of the core api? I didn't think these sorts of things were optional.

Edit: Same goes for GL_ARB_debug_output and GL_ARB_cl_event. I'm starting to wonder what the phrase "OpenGL 4.2" support is even supposed to mean.

[AMD HD 5850 / Windows 8 / 3.3/4.2 Core Profile Forward Compat]

Edited by Chris_F, 27 September 2012 - 12:19 PM.

[Brother Bob]

The ARB tag means that it's an ARB approved extension. Core functions have no extension tag at all. If it's not in the list, then is simply means it's not supported by the driver. I quickly checked around for extension support, and indeed it looks like AMD drivers doesn't provide that extension.

[Chris_F]

The ARB tag means that it's an ARB approved extension. Core functions have no extension tag at all. If it's not in the list, then is simply means it's not supported by the driver. I quickly checked around for extension support, and indeed it looks like AMD drivers doesn't provide that extension.

What do you mean?

e.g.

ARB_geometry_shader4

Notice the ARB? ARB_geometry_shader4 is supposedly a core part of OGL 3.2.

Edited by Chris_F, 27 September 2012 - 12:39 PM.

[Brother Bob]

It is an extension, but it is an extension that has been promoted to the core API after it has been widely adopted and accepted. When it was promoted from extension to core function, it is entirely possible that functions change in the way they work, or that function names change (not counting dropping the ARB suffix), or that functions are removed entirely.

Lets say that there is an extension GL_ARB_foo which has a function named glFooARB. The extension is widely adopted and promoted to the core in version X of OpenGL. When promoted to the core, the extension suffix is dropped, and the function name becomes glFoo. At this point, a driver that doesn't provide support for OpenGL version X can choose to support GL_ARB_foo and must subsequently provide the function glFooARB. Once the driver implements OpenGL version X or later, it must now provide the function named glFoo since that is what's required for version X.

The driver can, if it chooses, provide an implementation for version X but not announce support for GL_ARB_foo. There is nothing wrong here, because GL_ARB_foo itself is not a part of the core API, but the function it provided was.

A driver is only required to provide functions for the version it claims to support and for the extensions listed in the extension string. If the extension string does not include GL_ARB_foo, then glFooARB is not provided. If version X is provided but GL_ARB_foo is not listed, then glFoo is available but not glFooARB, even though they may be the exact same functions.

If you look in the specification for ARB_geometry_shader4, you will see that all new functions has an ARB-suffix. This denotes that the function belongs to an ARB extension. If you look up the core API specification you will see that the corresponding functions does not have the ARB suffix. They are different functions as far as the specification goes.

Edited by Brother Bob, 27 September 2012 - 01:30 PM.

[Chris_F]

So if I understand correctly, the way it works is that anyone can create an extension to OpenGL, which would then have the suffix EXT. If the Architecture Review Board likes a particular extension, they can promote it to a ARB suffix which is their way of showing support for it, in hopes most likely that more vendors will implement it. Then if all goes well, it (or something very similar) could be incorporated into the Core spec where it looses all suffixes.

[Brother Bob]

In essence, yes, that's how it goes. Although the EXT-tag is an intermediate step between an individually proposed extension and a full ARB approved extension. Extensions typically start at a vendor-level with tags like NV, ATI, SGI, INTEL and AMD. The EXT-tag is for extensions that are supported by multiple vendors, and the ARB is the last step when it has more or less full vendor support. Extensions can also start as collaborations between vendors also and start directly at the EXT or even the ARB tag.

It typically goes both ways, but normally ARB approves common extensions and much less often approves them to promote them. And yes, anyone is free to write and submit an extension, but vendors are equally free to not implement it, and the ARB is also free to not allocate for example unique identifiers for any constants you define.

But note that, as was mentioning in the foo extension example, that each promotion, be it to another tag or to the core API, technically creates a completely new set of functions and symbols. It is not correct to check the extension string for GL_ARB_foo to verify that foo is supported and then proceed to use glFoo(), since the two are from different sets; one from the extension and one from the core, and the presence of one does not imply the presence of the other.

That is why you may not have seen the robustness extension even though version 4.3 was reported; version 4.3 does not imply the presence of the extension, but it does guarantee the presence of the core-promoted functions.

[Chris_F]

OK, I find that I am still having trouble understanding this.

For example, take this excerpt from Wikipedia about OpenGL 4.2:

Support for shaders with atomic counters and load/store/atomic read-modify-write operations to a single level of a texture.

Does this mean that OpenGL 4.2 has guaranteed support for atomic counters, or does it just mean that starting with OpenGL 4.2 GL_ARB_shader_atomic_counters might be available on supporting hardware.

I'm also trying to figure out whether GL_ARB_robustness was made core in version 4.3. I have no interest in thing which only might be available.

Edited by Chris_F, 29 September 2012 - 12:23 PM.

[mhagain]

Fair warning that this may confuse you even more.

In the past extensions worked much like Brother Bob described. In recent GL_VERSIONs many ARB extensions have worked slightly different; a new GL_VERSION is released accompanied by a slew of extensions which are not part of it but which serve the purpose of enabling exposure of as much functionality from that version on older hardware as possible. This is effectively the opposite of the previous mechanism; rather than gathering together a bunch of previously existing extensions and making a new GL_VERSION including them in core, these newer versions instead define a new spec and spawn new extensions out of it for use by older hardware.

So to answer your question: if a piece of functionality is included in GL4.3 then you're guaranteed availability of it on all GL4.3 hardware. You don't need to check the extension string; once you see 4.3 in the GL_VERSION string you can just load any required entry points (without the -ARB suffix) and start using it.

If the hardware is not GL4.3 capable then it may still be able to support the functionality, but as an extension. So query the extension string, see if GL_ARB_whatever is present, load the entry points, run any further checks you might need (these will be documented in the extension documentation) and use it.

Now to go down the rabbit hole.

In the past entry points and GLenums for all -ARB extensions had an ARB suffix on them; you can still find old VBO tutorials (for example) that use glGenBuffersARB, glBindBufferARB, GL_ARRAY_BUFFER_ARB, and so on. Once these extensions moved into core you would use the versions without the suffix which were normally (but not always - GL_ARB_shader_objects is a case in point) otherwise absolutely identical. If an extension is in core then drivers don't have to export the extension string any more, but as a general rule they always will in order to provide support for existing programs that may have used the extension version of a piece of functionality. So while that can be relied on, strictly speaking it shouldn't be.

These newer extensions don't always have an -ARB suffix present. See issue number 6 in the documentation for GL_ARB_vertex_array_object for the reasoning behind this: http://www.opengl.or...rray_object.txt.

If by now your head is spinning a little then I don't really blame you; it can be quite bewildering on first encounter for sure, so the best approach might be to dive in and start writing some code. That may be an easier way for you to get your head around the way these things interact and interoperate.

Edited by mhagain, 29 September 2012 - 04:06 PM.

[Aks9]

Everything said by Brother Bob and Mhagain is correst, but maybe it sounds a little bit complicated (although it is not).
Let me try to make it clear...

1. Not anyone, but only hardware vendors can make extensions. It is logical since extensions are part of the drivers, and only driver writers know what their hardware supports.

2. If only one vendor supports some functionality, then appropriate prefix/suffix is used (NV, AMD,...) Sorry, Brother Bob, for repeating this.

3. If at least two vendors support some functionality, then EXT prefix/suffix is used (prefix for constants, suffix for functions).

4. If at least two vendors support some functionality, and the committee (ARB) decides it is something that should be a part of the core, the prefix/suffix changes to ARB.

5. Almost all ARB extensions become core functionality in the next revision of OpenGL. In the recent years, ARB extensions are published at the same time with the new release of the specification of the core, so ARB overnight becomes core. Core names lose ARB prefix/suffix.

6. If the driver supports particular ARB extension it also supports core equivalent (please correct me if I'm wrong; maybe there were some functions differ from the core version in the list of parameters), and vice versa. I saw a lot of advices on the forums claiming that ARB version of the function should not be used if the core version exists. Well, if the core version exists, both ARB and core functions point to the same entry. The function pointers are identical. Do you think driver writers would implement several version of the same function? Be aware, OpenGL is a mammoth with decades of adding new functions. Also, it doesn't matter how the function is named. You can use whatever name you want as long as you are accessing the right pointer. void glVertexAttrib1f (GLuint index, GLfloat x) has the same functionality as void Marry (GLuint index, GLfloat x) if they point to the same entry, have the same return-value and use the same parameters.

7. The extensions names returned by glGetStringi(GL_EXTENSIONS, i) should report the functionality. If there is no adequate ARB string, the functionality is probably not supported even if should be included into the core of the version the driver supports. You can be a 100% sure only after retrieving the function pointer and try to execute it properly. Considering OPs question, I can bet that mentioned functionality (part of GL_ARB_robustness and WGL_ARB_create_context_robustness) is not supported in any form.

8. Claiming that appropriate version of OpenGL is supported is not a guarantee that it is true. I won't start a debate about it, but there is a plenty of examples.

9. Starting from OpenGL 4.0, the addition of new functionality does not follow new hardware capabilities. For example, 82% of OpenGL 4.3 functionality is supported by the old hardware (SM4). The same is true for OpenGL 4.1 and 4.2 (70+%). So, even if a driver claims it supports OpenGL 3.3, it actually could support more than 70% of all versions above (up to 4.3). The exception is only OpenGL 4.0, because it requires SM5 hardware exclusively.

Sorry for repeating your words, Brother Bob and Mhagain, but I tried to summarize all you sad. I hope it is clearer now.

[mhagain]

Regarding your number 6, the example of GL_ARB_shader_objects, which was quite different in some key areas to the core 2.0 implementation of GLSL, should be sufficient warning that ARB and core are not necessarily always identical. Having said that, I don't think we'll be seeing that happen again.

[Brother Bob]

Everything said by Brother Bob and Mhagain is correst, but maybe it sounds a little bit complicated (although it is not).
Let me try to make it clear...

1. Not anyone, but only hardware vendors can make extensions. It is logical since extensions are part of the drivers, and only driver writers know what their hardware supports.

There are some non-hardware vendor-written extensions in the registry already which indicates that anyone can, in fact, write extensions. It doesn't take be a hardware manufacturer to suggest something. A lot of useful support functions can be suggested that no hardware manufacturer have thought of yet. For example, GL_MESA_window_pos was first suggested by Mesa, and was then fairly quickly adopted by the bigger vendors and promoted to an ARB extension, and then finally promoted into the core API.

The authors of gDEBugger also has an extension in the registry.

2. If only one vendor supports some functionality, then appropriate prefix/suffix is used (NV, AMD,...) Sorry, Brother Bob, for repeating this.

3. If at least two vendors support some functionality, then EXT prefix/suffix is used (prefix for constants, suffix for functions).

4. If at least two vendors support some functionality, and the committee (ARB) decides it is something that should be a part of the core, the prefix/suffix changes to ARB.

The numbers are not necessarily that clear and well defined. There are extensions with vendor-tags that are supported by multiple vendors, and even extensions that have not been further promoted at all so there's no EXT or ARB equivalent. And funnily enough, I did find some ARB extensions with only one supported hardware vendor. Although, that database did look a bit incomplete or outdated so I would take that information with a grain of salt. In any case, there are no definite rules like that.

8. Claiming that appropriate version of OpenGL is supported is not a guarantee that it is true. I won't start a debate about it, but there is a plenty of examples.

In practice I agree that the support may not be what it's supposed to be, but it is a violation with the licensing of the OpenGL brand and any claim the vendor makes about supporting a version. I would call that a driver bug rather than an issue with relying on the version information. It is a problem with the driver that we software developers end up having to deal with.

[mhagain]

In practice I agree that the support may not be what it's supposed to be, but it is a violation with the licensing of the OpenGL brand and any claim the vendor makes about supporting a version. I would call that a driver bug rather than an issue with relying on the version information. It is a problem with the driver that we software developers end up having to deal with.

Worth noting that in some cases there is leeway in the spec for an implementation to report an extension but not actually support it. I'm primarily thinking of GL_ARB_occlusion_query here, where implementations are allowed to report 0 query counter bits. That's why it's important to read documentation, be aware of these cases, and run the appropriate glGet calls before assuming that something is available.

Also worth noting that the GL spec itself makes no guarantee that a particular piece of functionality will actually be hardware accelerated. It has happened in the past that vendors have provided a software implementation of some features in order to claim a higher GL_VERSION than what they actually did support in hardware, and that attempting to use it could drop your entire pipeline down to software emulation. Fortunately this doesn't happen so much any more, but there are still trouble spots around accumulation buffers and drawing to the front buffer where some lower-end implementations (*cough*Intel*cough*) can give you problems.

A general rule I use is this: if it's not available in D3D then trying to use it in GL can be an unwise move on certain implementations, even if the GL spec indicates that it should be available.

[Aks9]

Regarding your number 6, the example of GL_ARB_shader_objects, which was quite different in some key areas to the core 2.0 implementation of GLSL, should be sufficient warning that ARB and core are not necessarily always identical. Having said that, I don't think we'll be seeing that happen again.

Thanks for the example! I know it is not always the case, but an exception only confirms the rule. ;)

There are some non-hardware vendor-written extensions in the registry already which indicates that anyone can, in fact, write extensions.

I agree that extensions can be proposed from "the outer world", but GL functionality is implemented in drivers, so actual development is done on the hardware vendors' side. OK, I as to restrictive. My mistake.

In practice I agree that the support may not be what it's supposed to be, but it is a violation with the licensing of the OpenGL brand and any claim the vendor makes about supporting a version. I would call that a driver bug rather than an issue with relying on the version information. It is a problem with the driver that we software developers end up having to deal with.

As mhagain already said, what's the purpose of GL_ARB_occlusion_query if it actually doesn't function. Another case (Intel again) is problem with multisampling. Intel does not support multisampling in pre-HD2000 GPUs, although it claims GL 2.1 is supported. Well, my humble opinion is that it is brainless to advertise GL 1.2 (because of missing functionality) when there is a full shader support. Extension strings should list all supported functionality regardless of the advertised GL version.

[Aks9]

I have to apologize because my statement number 6 is not correct.

6. If the driver supports particular ARB extension it also supports core equivalent and vice versa. ... Well, if the core version exists, both ARB and core functions point to the same entry.

I carried out some experiments, and they proved that core, ARB and EXT versions of the same function may point to different entries.

[mhagain]

I carried out some experiments, and they proved that core, ARB and EXT versions of the same function may point to different entries.

There may be cases where these are just stubs that call the real function internally in the driver, and that is common to all. The real determining factor would be whether the spec wording is the same or not.

Even more confusingly however, there are also cases where core, ARB and EXT GLenums are identical, and if these are taken as params to a core function, then the core function should be expected to behave the same (how could it possibly not?) for all. This is old OpenGL, but I'm thinking glTexEnv with the combine extensions here.