Jump to content

  • Log In with Google      Sign In   
  • Create Account

To T&L or not to T&L


Old topic!
Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.

  • You cannot reply to this topic
7 replies to this topic

#1 marcus256   Members   -  Reputation: 122

Like
Likes
Like

Posted 28 August 2001 - 12:58 AM

There has been a lot of discussion about whether to use the driver-supplied T&L implementation (software or hardware) or to go through the complicated process of detecting if there is hardware T&L support, and if not, use your own "super optimized" T&L routines. Well, here is a little story on a related (?) topic: Once upon a time there was a wonderful computer called Amiga. It had cool graphics hardware with lots of cool acceleration features which allowed for cool 2D animations (in fact, a 7 MHz based Amiga was way faster than a 66 MHz PC on most 2D games). Enter Wolfenstein 3D and Doom. Oops! Bitplaned graphics was not good at all for software 3D rendering! For a couple of years the "elite" among programmers spent their valuable time writing "chunky to planar" converter routines, each one faster and more elaborate than the others. Of course, there was an operating system function that did just that, but it was super-slow (probably 20-40% slower than the fastest routines) = not acceptable. Enter the 24-bit graphic cards for the Amiga. Oops! 95% of the games and programs which used the custom converter routines did not even run on these new cards since they were so tightly coupled with the Amiga hardware (my 3D engine was one of those programs). The remaining 5% ran slower on those new graphic cards than they did before (the OS had to do an extra conversion pass back to "chunky" graphics that the graphic cards used). ALL of the programs would have run MUCH faster if they had used the operating system funcion, since it sent the graphics unmodified to the new graphic cards! The point is: If you are writing your own custom functions to do stuff that is already supported by standard functions, you had better know pretty darn well what you are doing, because some day there WILL be some super-cool hardware that you would never had dreamt about that would have made your program super-fast if only you had used that standard function instead of your own custom routine. Generally, writing custom routines will improve the overall speed of an application by 1-10%. What is that worth compared to the constant ongoing speed increase of CPUs, graphic cards and memories? There is probably a small percentage of programs out there that can gain a noticable amount of performance by using custom routines. How many programs would have gained performance by using the standard T&L functionality of OpenGL? There, now I have shared my view on the topic... Marcus

Sponsor:

#2 zedzeek   Members   -  Reputation: 528

Like
Likes
Like

Posted 28 August 2001 - 09:44 AM

>>The point is: If you are writing your own custom functions to do stuff that is already supported by standard functions, you had better know pretty darn well what you are doing, because some day there WILL be some super-cool hardware that you would never had dreamt about that would have made your program super-fast if only you had used that standard function instead of your own custom routine<<

exactlly this is what unreal/UT did ppl brought there geforces hoping to get a major speed boast yet it didnt happen, why not? cause the programmeurs of it decided to ignore the opengl matrix routines etc. whereas a game like q3a did use them + now u can run it at 200fps with a gefroce3 ( cant think of a reason why u would want to but ) though with UT youre stuck with 70fps

#3 Anonymous Poster_Anonymous Poster_*   Guests   -  Reputation:

Likes

Posted 28 August 2001 - 08:52 PM

You MUST use the OpenGL matrix function! It is true that, on some old card, you can get some FPS with your own custom routine but it is driver dependant. If you use your custom routine the acceleration will be only profitable to people who don''t have 3D card, because the Microsoft OpenGl software driver is very slower than others (not true for some old ATI card, where software mode is faster than hardware!!!).

#4 Pauly   Members   -  Reputation: 122

Like
Likes
Like

Posted 29 August 2001 - 02:13 AM

This is a bit of an age old problem really. As the developer you have to strike the right balance. In some situations you just can't use OGL's matrix mult functions.

It's all about getting at the data once OpenGL has calculated it. With these new 3d cards, the temptation is to make objects of massive complexity - loads of polygons. Using the OGL matrix routines is of course the prefered way to transform this around your 'world'.

However - you start coming into problems when you want your super complex space ship (or whatever) to interact with other objects in the world. Collision detection requires you have at least a bounding sphere in the same position as your space ship. Obviously, you need to use your own math routines to get this into position because if you used OpenGLs and the appropriate glGet you'd slow the rendering down.

But - bounding spheres really aren't any use to anyone. You need a more complex approximation of your space ship - again, this needs to transformed by your own routines so you can get at the data to perform collision detection etc. As the complexity of the models increases, so it does for your 'low detail' model. So you end up writing all these math routines anyway.

To avoid the player throwing the computer out the window when "It missed me by miles!" these 'low' details are actually quite high detail

I guess this isn't relavent to what you were saying - and it supports whats been said about QIII and UT.

Use OGL's functions for the display geometry, but don't think for a minute that'll prevent you from getting your hands dirty with your own routines...

Paul Groves
pauls opengl page

Edited by - Pauly on August 29, 2001 9:14:53 AM

#5 zedzeek   Members   -  Reputation: 528

Like
Likes
Like

Posted 29 August 2001 - 09:21 AM

true but collision models have usually a lot less polygons than the drawn model. eg i believe q3 uses 3 boxes to represent the collisions for a player instead of the 1000+ tris.


#6 Pauly   Members   -  Reputation: 122

Like
Likes
Like

Posted 29 August 2001 - 02:08 PM

QuakeIII''s a bad example though, with humanoid figures you could get away with very few bounding boxes like you said. You don''t need to go into proper point-in-poly intersection. I was talking about games you ''fly'' over stuff - like space ships games and things of that ilk.

Paul Groves
pauls opengl page

#7 zedzeek   Members   -  Reputation: 528

Like
Likes
Like

Posted 29 August 2001 - 02:48 PM

donno about that, humanoids are prolly one the hardest out there (as with a lot of cg we can model buildings trees cars etc that look pretty realistic + in some films look like reality yet when it comes to ppl it always looks like cg eg final fantasy )
with a human u have a head,body,arms,legs but each can move in a variaty of ways. a spacecraft you have the fusalarge + the wings. but they dont move with relation to each other(big difference) .
ideally a person on the ground will have one leg higher than the other ( u will need to model using bones ) or else what''ll happen one of the legs will go into the ground which looks bad (quake, tombraider etc unfortunatly also this is what my game does )

#8 Anonymous Poster_Anonymous Poster_*   Guests   -  Reputation:

Likes

Posted 05 September 2001 - 08:21 PM

quote:
Original post by Pauly
This is a bit of an age old problem really. As the developer you have to strike the right balance. In some situations you just can''t use OGL''s matrix mult functions.

...

Obviously, you need to use your own math routines to get this into position because if you used OpenGLs and the appropriate glGet you''d slow the rendering down.




Yes, you''re right. In many situations you need to do some transformations "on your own" (collision detection is perhaps the most important example). What I meant in my original post was that you should not replace standard routines with your own routines just for the sake of speed . I think we all agree that glGet is almost always a "don''t do" thing, since it would only make sense on totally un-pipelined and un-optimised hardware.

However, you can still use OpenGL matrix operations for the screen projection, which is still a significant part of the transformation operations. You can also use OpenGL lighting in most situations without interfering too much with your internal geometry representation and calculations (unless you need to do very special scene lighting).

If people did, we would probably see better optimised hardware drivers (and hardware). From what I''ve understood, the Voodoo 5 drivers were never very good at T&L, which is probably because the driver developers did not expect programmers to use T&L.

/Marcus




Old topic!
Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.



PARTNERS