NetNews Usenet Archive 1992 #30/NN_1992_30.iso/spool/comp/graphics/opengl/222

Internet Message Format | 1992-12-21 | 3.1 KB

Path: sparky!uunet!zaphod.mps.ohio-state.edu!usc!elroy.jpl.nasa.gov!ames!sgi!fido!cashew.asd.sgi.com!kurt
From: [e-mail] (Kurt Akeley)
Newsgroups: comp.graphics.opengl
Subject: Re: conformance question
Date: 18 Dec 1992 17:15:57 GMT
Organization: Silicon Graphics, Inc.
Lines: 45
Distribution: world
Message-ID: <[e-mail]>
References: <[e-mail]> <[e-mail]> <[e-mail]> <[e-mail]>
NNTP-Posting-Host: cashew.asd.sgi.com

--

In article <[e-mail]>, [e-mail] (Rich Thomson) writes:

|> You're missing my point; many applications just aren't useful to their
|> users unless the rendering is fast.  In fact, many new applications
|> have become possible simply by the advances in rendering speeds of
|> modern graphics systems.  For those applications, it isn't a matter of
|> detecting a visual with the given functionality, but it is a matter of
|> finding the visual with the given performance.  OpenGL guarantees that
|> a visual will exist which will have all the functionality, but it
|> doesn't provide any way for an application to determine which visuals
|> are implemented through hardware, software, or a combination of both.
|> At least, if it does provide such a mechanism I haven't found it in
|> the man pages or the spec.  I haven't read them exhaustively yet,
|> though.

You're correct that OpenGL has no direct interface to report the performance
of a given visual.  We haven't been able to come up with a language that
is suitable for this purpose.  For example, the distinction of software vs.
hardware is obviously flawed, as many respectably fast graphics systems do
much of their work in software on the "host" processor.  Even if agreement
for categorizing performance is achieved, what performance should be
reported?  A renderer might run "fast" with either zbuffering or stenciling
enabled, but not both.  OpenGL modes are almost entirely orthogonal, so
the enumerated list of all mode combinations is near infinite.

However, OpenGL also doesn't have a solids capping command, or even a
stroked font command, yet it is very capable of capping solids and rendering
stroked fonts using simple combinations of its low-level commands.  Likewise,
applications that care can use the available OpenGL command set and the
operating system clock to measure the performances of various visuals.
The OpenGL command glGetString(GL_RENDERER) returns the unambiguous name of
the renderer, so that an application can measure visual/renderer performance
just once and catalog the results, rather than testing all visuals each time
it is invoked.  (Taking care to distinguish between local and remote rendering,
and in the case of remote rendering, at the risk of variability due to
network loading.)

I believe that direct measurement is the only practical way to determine
the performance of an application's specific command sequences.  I may be
wrong, however.  The OpenGL ARB will remain open to suggestions for improved
performance queries.

-- Kurt Akeley        [e-mail]  [phone]  M/S 7U-550