11 replies to this topic

[goowik]

Hi,

I've started learning OpenGL a few days back and love it
It's been a few months to almost a year I've been creating games in 2D, time to change!

As of now I always created my primitive objects with an immediate rendering coding
For example:

glBegin(GL_TRIANGLES);
glColor3f(1, 0, 0);
glVertex2f(-0.5f, -0.5f);
glColor3f(0, 1, 0);
glVertex2f(0.5f, -0.5f);
glColor3f(0, 0, 1);
glVertex2f(0.5f, 0.5f);
glEnd();

After reading some articles on performances I found out there are several more rendering "patterns". There is:

• Display lists
• Vertex arrays
• Vertex buffer object
• and Immediate

Does it come to personal preferences when choosing one of them? Or is there a major difference?

Also let's say you have a cube. How can you cull 3 of the invisible sides and make it change each time you move around it.
Is this coding different per rendering pattern?

Kind regards
Peter

Edited by goowik, 03 July 2012 - 08:32 AM.

[larspensjo]

You are using legacy OpenGL. It is fine for testing small simple things, but not good enough for "real" applications. And it is usually not feasible to start with legacy, and then upgrade, because it is done differently.

Have a look at Learning Modern 3D Graphics Programming, and you will quickly learn the modern way of doing it.

Unfortunately, most of the tutorials "out there" are done in the old way.

[mark ds]

Read this post.

[goowik]

@lardpensjo:
Just so I understand better: By Legacy OpenGL you mean the way of writing code? Not the library I'm using?
Thanks a bunch for the link! Guess I have some work to do testing out

@mark ds:
okay, I guess I forgot to mention that I'm using LWJGL (The Lightweight Java Game Library). But I guess the OpenGL code isn't much different from c++.

[mhagain]

Specific differences between the 4 items you mentioned.

Immediate mode is fine for rapid prototyping and experimental work. It can be fine for learning, provided you recognise and accept that it's a suboptimal path that can lead you into bad habits, and that you plan to move away from it as quickly as possible. It's suboptimal for a number of reasons - more calls into the driver means significantly higher function call overhead, and modern drivers are likely to emulate it via a dynamic VBO behind the scenes - not a good idea if the data is not dynamic. It also involves transfer of a lot of data from system memory to the GPU, which again is not good for data that is not dynamic in nature. It's been deprecated from recent GL versions so you need to be aware that you're learning something of very limited future utility.

Display lists provide a mechanism for caching GL commands and data on the driver for subsequent future use. The driver may cache them in GPU memory or in system memory (you have no control over this) and there are a set of highly complex rules for what state gets cached and in what circumstances it happens. In the right cases and with the right hardware it can be the fastest method available, but not all drivers or hardware are equal. Once created you can't modify a display list - you can only execute it or destroy it. If state or data needs to change you're SoL. Display lists are also deprecated so clean interaction with current and future GL functionality may not be guaranteed in all cases - approach with extreme caution in other words.

Vertex arrays allow you to specify a vertex layout and data in system memory in a handful of calls, and transfer it to the GPU very quickly in a single call. This retains the immediate mode overhead of needing to transfer data that may not change every time, but removes the function call overhead. For older code or code that needs to be compatible with older drivers (and note that we're talking much older here - in the order of 10 years or so) this should be the preferred option. These are also deprecated, but - since VBOs are built on top of vertex arrays - are more likely to work well with modern OpenGL (e.g. instancing works perfectly fine with vertex arrays).

VBOs are the modern OpenGL way of doing everything. They're built on top of vertex arrays so they share the advantage of needing very few calls to specify and draw geometry, and have the added advantage that the data may be stored in memory that is more optimal for the driver. This suits static data perfectly, but does mean that you need to be slightly more careful about how you operate with VBOs if you're using dynamic data (although do note that using shaders as well means that much data which formerly needed to be dynamic does not necessarily need to be so any more). As a general rule VBOs should be fastest of all on the widest range of hardware, but do note that it's incredibly easy to write code using VBOs that runs slower than anything else if you're sloppy or careless.

With all that in mind my recommendation is to use vertex arrays for learning, but keep an eye on moving to VBOs. The reason why is that vertex arrays can share a lot of the same coding style as VBOs so you're primed for an easier move, but avoid traps for the unwary that can cause pipeline stalls throughout your program. Vertex arrays have also been available since OpenGL 1.1 so you can be very confident that driver support is ubiquitous. The same applies to VBOs, of course (although it's 1.5 rather than 1.1). So vertex arrays first and - when you get comfortable with them and build up more understanding of how things work in general - switch to VBOs.

[goowik]

@mhagain:
Thanks for the detailed explanation. But after searching OpenGL's wiki I found the Vertex Buffer Object page.


Unfortunatly the following is written:

Legacy Note: Versions of OpenGL prior to 3.0 allowed the use of client-side data for vertex rendering, but GL 3.0 deprecated this. Core GL 3.1 and greater forbid client-side vertex data, so VBOs are the only means for rendering.

And they state offcourse that it is recommended that we do not use any of these functionality in our programs.

So I have a two questions about this:

• What is the meaning of Legacy? as 'lardpensjo' stated.
• What do they mean by client-side data? Isn't everything client sided?

[Brother Bob]

So I have a two questions about this:

• What is the meaning of Legacy? as 'lardpensjo' stated.
• What do they mean by client-side data? Isn't everything client sided?

Legacy refers to the old OpenGL API up to version 2.1. Much of OpenGL 2.1 was removed from the API when 3.0 was introduced (there was a combatibility mode for some versions, but that compatiblity mode has been removed as well in 3.3), incluiding things such as display lists and client-side vertex arrays. Legacy refers to the now-deprecated API of version 2.1 and earlier.

Client-side and server-side in this context refers to the application (the client) and the OpenGL implementation/driver (the server). These are typically on the same computer (you run the application on the same computer you have your graphics card on), but that does not have to be the case. Especially on unix-platforms and their windowing systems, you can basically have the application run on one computer and have the display on another computer.

Client-side data in this context means the data is stored in memory managed by the application. For example, when you allocate the memory with malloc or new, or store the data in an std::vector). Server-side data means that the data is stored in memory managed by OpenGL. For example, texture data stored in a texture object with glTexImage or vertex arrays stored in a VBO are both stored in memory managed by OpenGL. They don't have to be physically stored on the graphics card, only that the memory is managed by OpenGL.

Server-side data is the only way to store any data in modern OpenGL. You may have to load your vertex arrays into your own memory before uploading them to your buffer objects of course, but you cannot use the vexrtex arrays to draw something until it has been uploaded into VBOs. Server-side is mandatory for everything involving user data at the moment, not just vertex arrays.

[CodedVentures]

If you have the choice, consider the API that is geared for the future of OpenGL. On mobile devices that would be OpenGL ES 2.0. Even if OpenGL 4.2 still supports the legacy api, you should try to avoid it if you can...
That being said, using immediate mode is still ok for quick testing and learning. But for the long run, and production code, avoid legacy API.
This page has some info about OpenGL Core profile: http://www.opengl.org/wiki/Core_And_Compatibility_in_Contexts

Look for glCullFace to find out how you can cull the invisible faces of your cube. glCullFace will always work no matter the rendering API you use.

[goowik]

@Brother Bob:
Okay now I get it! So most of the code is backwards compatible till OpenGL 3.0.
Just out of curiosity, client/server side is this applicable to most graphic libraries (such as directX)?

@CodedVentures:
K so that clears most of it. To recap you simply have 2 "packages" (if I may call it this way):

• The Core package
• The Compatibility (which was introduced in 3.2)

Will the core always be the same? As all deprecated methods were removed?


EDIT:
I also was looking for a decent book to buy/rent and came across the "OpenGL Superbible"
But on Amazon it does seem to have a low score (3.5/5) and the reviews are very various.

Edited by goowik, 04 July 2012 - 12:59 AM.

[Brother Bob]

@Brother Bob:
Okay now I get it! So most of the code is backwards compatible till OpenGL 3.0.

That is correct. The compatibility mode of the modern API lets you extent the backward compatibility of the legacy API to version 3.1 even.

The modern API is mostly backward compatible as well, although there are some changes that makes it not. For example, vertex array objects (VAO) are not required in version 3, but are in version 4. That means that you cannot run a version 3 compatible program on a version 4 context. But that change is trivial; just use VAO in version 3 as well and you will have no problems with version 4.

Just out of curiosity, client/server side is this applicable to most graphic libraries (such as directX)?

I have no idea whether Direct3D makes the difference explicit in some way or not. I would, however, guess that pretty much the same requirements on memory management applies to both Direct3D and OpenGL: manual memory management is not possible, you have to let the API handle your resources.

@CodedVentures:
K so that clears most of it. To recap you simply have 2 "packages" (if I may call it this way):

• The Core package
• The Compatibility (which was introduced in 3.2)

Will the core always be the same? As all deprecated methods were removed?

The core changes, but (with few exceptions, see for example my comment above on VAO in version 3 and 4) new stuff are only added. What works on an earlier version typically works on a later version as well. Only exception of course is the major change from the legacy API to the modern API.

The compatibility mode was introduced in 3.0 though. From version 3.2 and onwards, compatibility mode is not available. Thus, it is only available for the modern API for versions 3.0 and 3.1.

[larspensjo]

I also was looking for a decent book to buy/rent and came across the "OpenGL Superbible"
But on Amazon it does seem to have a low score (3.5/5) and the reviews are very various.

I have that book, and I think the weakness is that it is that the examples are based on a home made support library. The disadvantage of that is for beginners, that do want simple examples where you can see the whole solution on the same page. The advantage is for the intermediate or advanced reader, that no longer cares about the basics but want to learn and understand advanced concepts.

If you have understood the basics, then the book is excellent.

Actually, a member of the forum, japro, recently created basic minimal examples, perfect for beginners to start with. See https://github.com/progschj/OpenGL-Examples.

[goowik]

Thanks, will look further into it!