10. Gem FAQUp one level
* : new question
+ : changed question
1.1) What is GEM?
1.2) What is Pd?
1.3) What platforms do GEM and Pd run on?
1.4.0) How do I install GEM and Pd on IRIX?
1.4.1) How do I install GEM and Pd on linux?
1.4.2) How do I install GEM and Pd on WinNT?
1.7) What is a good intro to OpenGL?
1.8) Are there any web sites for Pd or GEM?
1.9) What libraries does GEM use? (aka: Who does Mark want to thank?)
1.10) Are there any restrictions on GEM?
1.11) How do I use GEM in a performance?
2.1) How do I (???)
2.2) How do I make GEM run?
2.3) Why doesn't GEM run?
2.4) I've got it running. Now what?
2.5) On IRIX 5.3, why does GEM dump with an rld error?
2.6) Why can't I compile GEM on IRIX 5.3?
2.7) Why is GEM slow in general?
2.8) Why is GEM slow on IRIX?
2.9) Why is GEM slow on WinNT/Win95?
2.10) Why is GEM slow on Linux?
2.11) If I resize the window, everything looks strange.
2.12) Can GEM run on a 3Dfx Voodoo card?
2.13) Will GEM support hardware transform and lighting (T&L) ?
2.14) I get an error "GEM needs Truecolor visual support".
4.1) My image doesn't appear. What is going on?
4.2) My image looks strange. What is going on?
4.3) Why does GEM say that it can't handle a gray image?
4.4) What image formats can GEM handle?
4.5) What movie formats can GEM handle?
4.6) Why is pix_draw so slow?
7.1) Why doesn't <object> exist on <platform>?
7.2) Why doesn't gemtablet work?
7.3) I don't want GEM to take over my tablet. How do I stop it?
7.4) Why doesn't gemmouse work in IRIX/Linux?
7.5) Why doesn't gemorb work?
7.6) What is wrong with pix_video in WinNT?
1.1) What is GEM?
GEM is the Graphics Environment for Multimedia. It was originally written by Mark Danks to generate real-time computer graphics, especially for audio-visual compositions. It originally ran under FTS/Max (which is why you might see some papers reference it), but all new development is under Pd.
You can get GEM at http://gem.iem.at/
GEM was sponsored by a grant from Intel (http://www.intel.com)
GEM is now maintained by IOhannes m zmölnig.
the core-development team consists of
- chris clepper
- gÃ¼nter geiger
- daniel heckenberg
- james tittle
- IOhannes m zmÃ¶lnig
Pd is a real-time environment for audio and MIDI. It was written by Miller Puckette, who created FTS/Max when he was at IRCAM. Basically, Pd can be seen as the next generation of real-time visual programming languages. GEM runs inside of the Pd environment.
You can get Pd at http://www.crca.ucsd.edu/~msp/software.html
Pd is sponsored by a grant from Intel (http://www.intel.com)
GEM and Pd run on Windows (95, 98, ME, NT 4.0, 2000, XP), linux and macOS-X (>10.2). SGI-Irix (> 6.2) used to be supported but i don't have any prove that it still works). Günter Geiger has done an initial port of GEM and Pd to Linux http://gige.epy.co.at).
See the readme for installing Pd.
GEM should be at
If you run GEM.INSTALL.sh, then all of the example files and documention should be put in the correct locations.
See the readme for installing Pd.
GEM should be at
chdir to <gem>/src/Gnu and build Gem following the instructions in the README.build (./configure; make)
If you then make install, then all of the example files and documention should be put in the correct locations.
if you are using debian, Gem should be available via apt
if you are using an rpm-based distribution, check out the builds at planetCCRMA
See the readme for installing Pd.
unzip GEM so that it is at
If you run GEM.INSTALL.bat, then all of the example files and documentation should be put in the correct locations.
there is also an installer for windows.
See the readme for installing Pd.
there is also an installer for macOS.
The best book is the OpenGL Programming Manual by Mason and Woo. This is also called the "Red Book". If you search the web, there are many sites on OpenGL. A good starting point is http://www.opengl.org. Also, Mark Kilgard (who used to work for SGI) has a wonderful site with lots of links (http://reality.sgi.com/mjk) Also, Normal Lin has written another great book on 3D-graphics under linux
Except for the ones noted above, there is the Japanese
installation page at
There is a Pd mailing list. Subscription info is on IEM's site http://www.iem.at/mailinglists/pd-list
One of pd's unofficial home-pages is at http://pd.iem.at hosted by the Institute of Electronic Music and Acoustics, Graz, Austria
An interesting place might also be Günter Geiger's size http://gige.epy.co.at/
there are lot's of other cool pages (search the net...)
All copyrights and license info can be found in
Thanks to Sam Leffner for libTiff, the TIFF image loader.
Thanks to Masayuki Matsumoto for fstimage for OpenGL, the SGI
Thanks to the Independent JPEG Group for libjpeg, the JPEG image loader.
Thanks to Mark Kilgard at al. (and SGI) for glut, the openGL Utility Toolkit
Thanks to Stephane Rehel for GLTT, the OpenGL TrueType render.
Thanks to David Turner, Robert Wilhelm, and Werner Lemberg for
Freetype, a TrueType font rendering engine.
Thanks to the MPEG Software Simulation Group, for libmpeg, the
Thanks to Heroine for quicktime4linux a quickime Decoder and libmpeg3, another MPEG-2 Encoder/Decoder
Thanks to LCS/Telegraphics for Wintab, the Windows tablet library.
Thanks to David McAllister for the Particle System library.
Thanks to John Stone for the Space Orb library, libOrb
GEM is under the Gnu Public License. This basically means that it will always be free software.Check out http://www.gnu.org for more information and read the full license in GnuGPL.LICENSE in the GEM release.
This is a constant problem, because there is no consistent way to display video on any platform. Also, you usually do not want to send the entire screen, but only the GEM window. It is also useful to be able to edit/control the Pd patch window while the patch is actually running.
On SGIs, the best way is to get a video out option. On the SGI O2, Impact, and Onyx (Mark has used all of these), there is a simple connector or breakout box to do video. If you run the video out program, then you will get a rectangle on your screen which shows what is being sent out the video connector. Make your GEM window a little larger than 640x480 and center it in the rectangle. You can now project this with a standard video projector.
On PCs it is a bit harder. Several modern video-cards have the possibility to output several screens (either 2 (or more) VGA-screens or 1 VGA-screen and 1 TV (Composite or S-HVS) or a combination with DFTs) If you have a Canopus Voodoo2 card it has a video and s-video output on it. As described in question 2.12, you can get a Voodoo to work with GEM. If any one else has a better solution, please let me know. The nVidia Riva TNTs require that you output the full screen, so this is not a very good option. You can use a video scan convertor. Some of them only display a part of the scene, which is exactly what you want.
With modern multi-headed cards it is more simple:
Configure your card to display the desktop spread over your multiple screens
(e.g.: from left-to-right).
On windows and macOS you can do this via the display-properties dialog.
On linux you will have to edit your /etc/X11/XF86Config-4 file either by hand or
(if your system supports it) via an appropriate editor (yes, nowadays there are some).
Now create your gem-window on the second screen:
it should have the same dimensions as the 2nd screen (e.g: "[dimen 800 600(").
to place it at the second screen use the offset (e.g: if your primary sreen
(the one you want for patch-editing) has the dimension 1024x768 use "[offset 1024 0(",
which will create the gem-window 1024 pixels right of the upper-left corner
of the total screen (and 0 pixels below it),
which is exactly the upper-left corner of the 2nd screen.
You most probably want to turn off the borders with "[border 0(".
Note: some grafix-card have openGL-hardware-acceleration only on the 1st screen (so you should create the gem-window on the 1st screen and move your patches to the 2nd screen)
If you are using an XServer for displaying (under linux) you can also use another computer for rendering. You can specify the place where the gem-window should be created with something like "create <render.host>:0.0"
If you are doing audio with graphics, the only solution to prevent clicking (question 5.1) is to run 2 computers and have them communicate with netsend/netreceive. We are working on making Pd/GEM multi-processor friendly, so if you have a multi-processor system, you can run everything on one machine eventually.
2.1)How do I (???)
Many of the general usage questions are probably answered in the manual or release notes. The pd mailing list is also a good place to find answers as well.
GEM is not an executable. It requires Pd to work and is loaded in at run time. For example, I have an alias on the SGI which does
/usr/people/mdanks/pd/bin/pd -lib /usr/people/mdanks/pd/gem/Gem
and on WinNT
\pdDir\pd\bin\pd -lib /pdDir/pd/gem/Gem
on UNIX-systems you will probably want to use a .pdrc file, where you can put the command-line arguments for pd that you "always" need.
If you don't see a startup message from GEM, then something went wrong.
Most people use use the command shell to start Pd. It is not very difficult to configure Pd to run from double-clicking on the icon.
Notice that the -lib flag always requires Unix styles slashes. This is the case even on Windows.
You may also want to use the -nosound flag. For instance, my PC has problems using audio (it leaks memory), so I just turn off the audio part of Pd. However, other people can't get GEM to work if the -nosound is used (on Win95). You can also try the -dac or -adc flags (for digital-analog-conversion only and analog-digital-conversion only).
Try out the manual. It will step you through
You will also want to look at the example files. Assuming that everything is installed correctly, you can get to the examples by going to the Help menu in Pd and selecting examples. A bunch of the patches should start with gem<something>. The best one is gem/01.basic/01.redSquare.pd It puts a red square up on the screen and allows you to rotate it. gemImage.pd shows how to load in a TIFF file. gem/03.lighting/04.moveSpheres.pd moves two spheres around the screen. Try the other ones.
Most of the GEM objects have test patches which give some information about the various controls for the object.
GEM only works under IRIX 6.2+. The rld error
is probably something about not having glBindTextureEXT (or something).
OpenGL 1.0 has some extensions to speed up texture mapping (which are an
integral part of OpenGL 1.1). However, these don't exist on IRIX
5.3. If you recompile GEM (see the next question), things should
I don't have access to an IRIX machine, so don't expect any builds from me. Upgrading to IRIX 6.2+ is worth it.
There was probably an error saying that the compiler couldn't find the file "dmedia/vl_vino.h" in pix_videoSGI.cpp. IRIX 6.2+ adds new functionality to the media libraries which makes life much easier. You cannot compile pix_video or pix_indycam as is under 5.3. You can remove them from the Pix/Makefile and from the linker part of the global Makefile. You will also need to recompile the Td and Tiff libraries.
There shouldn't be any problems doing this. I haven't tried any of this, so if it works for someone, please let me know.
Examine what you are doing. If you are constantly
changing textures, then this is probably your problem. If you have
models with a million triangles, then this is probably the problem.
Compare what you are doing with realistic specs on your system. Some
systems slow down when they have to draw very large polygons (slow fill
You can also turn on profiling to see how long it takes to render a frame. Send a profile message to the gemwin object. The number that is printed is the number of milliseconds one frame takes to render. 50 milliseconds is 20 frames per second. 'profile 2' is good if you want to see how long the image processing is taking.
profile 0 - turn off profiling
profile 1 - turn on profiling
profile 2 - turn on profiling and don't cache pixes
If you are having major slowdowns, then please let me know. I have gotten very good performance on most machines (Indy, O2, Impact, Onyx2).
You probably don't have hardware acceleration.
You can use software rendering, but it basically useless except for extremely
basic patches. You can get a good graphics accelerator for really
cheap these days. I recommend a card based on nVidia's chipsets,
such as the TNT2 or GeForce, but there are other companies such as 3dfx
and Matrox. Make sure that you are running the latest drivers for
your card. The basic drivers that come with the cards are usually
Also, PCs don't deal with lots of texture maps very well (they are bus limited, at least until AGP), so if you are trying to use lots of constantly changing texture maps (especially with [pix_multiimage], [pix_video] or [pix_film]), that will cause problems.
It is because you have to use Mesa, which might be
running iin software. Mesa (http://www.mesa.org)
is an awesome package by Brian Paul (firstname.lastname@example.org) which "emulates"
OpenGL. Basically, it is a fully compliant OpenGL package, but it
isn't officially sanctioned by the OpenGL ARB, such, it is doesn't have
the OpenGL name. There is an acceleration package for the many graphics
card, but I don't know anything about it.
nVidia is being very supportive of Linux: their TNT2 and GeForce cards work under Linux with hardware-acceleration of openGL. (but the drivers are proprietary)
radeon cards should also be supported very well under linux (even with open-source drivers)
GEM doesn't trap resize events in IRIX or Linux (this is not a problem in WinNT). This means that OpenGL doesn't have the correct information to render properly. If you want to resize the window, send a 'dimen x y' message to gemwin before you create the window.
I (this is: Mark Danks) have a Voodoo2 card, which runs fine under WinNT. I use the OpenGL beta driver from 3Dfx at work all the time without any problems and, except that the Voodoo takes over the full screen, it seems to work fine. You will need to download the OpenGL Beta driver from 3Dfx's web site at http://www.3dfx.com and put the OpenGL32.dll into the same directory as pd.exe (NOT gem.dll). Debugging patches is much easier if you have two monitors, one for the 3-D card and one for the 2-D card.
IMPORTANT: You MUST set the environment variable
GEM_SINGLE_CONTEXT = 1
to make the Voodoo card work. It will make a window 640x480 (which
is the correct size for TV video out on my Canopus V2 card). On WinNT,
right click "My Computer" and go to "Properties". On the "Environment"
tab, you need to add the variable "GEM_SINGLE_CONTEXT" with a value of
Resizing the GEM window with a Voodoo card is not a great idea. The Voodoo card can only display certain window sizes and will clip the graphics.
For the tech heads in the audience...I create an OpenGL context at startup and never actually display its associated window. This means that GEM objects can create display lists, call OpenGL commands, etc. in their constructors, even if no window is actually being displayed. However, with the Voodoo card, there can only be one OpenGL context. So, instead of creating one context and just holding onto it in the background, I create the normal GEM window and associate the OpenGL context with it...and the user can never destroy or close that window.
Absolutely! Unlike some other APIs, OpenGL will automatically use hardware accelerated transform and lighting if the card has it. GEM gets great performance from cards like nVidia's GeForce.
This error means that your X display is running with paletted colors, which is the result of limited color depth. If you start the X display with
startx -- -bpp 16
or some higher number, then it should work fine. 32-bit color is the best.
3.1)Why does everything seem dim?<
You probably turned on lighting but don't have any lights in the world. Either add a light with world_light or light or turn lighting off by sending a message 'lighting 0' to the gemwin. You can also send a reset message to gemwin to set it back to the startup state (which doesn't have any lighting).
See question 3.1.
If you are using view in your patch to change the viewpoint, you may not be pointing in the correct direction. You also might have translated everything outside of the current viewport.
Also, if you have been using single buffering ('buffer 1' message to gemwin), then you might still be in that mode. Either send a 'buffer 2' message or a 'reset' message to gemwin. Then, destroy and create your window.
4.1) My image doesn't appear. What is going on?
Normally images have to be texture-mapped onto Geos. You have to use [pix_texture] to map the current image onto a Geo. "Current" means that any pix-manipulation that is done after texturing will not be displayed.
Any Geo has a color (which is initially set to white). If you have set the color to black, your Geo (including the image) might be very dark. If you are using alpha-blending, make sure that the Geo is not invisible.
Normally images that want to be texture mapped with openGL should have dimensions that are a power of 2 in both height and width. Now [pix_texture] will make this totally transparent to you (so normally you don't have to care about the size of the image). However with non-power-of-2 images pix_coordinate might not behave as expected, because these images need absolute texture-coordinates rather than normalized ones (as are used with power-of-2 images): so if the texture-coordinates are set to "(0,0) (1,0) (1,1) (0,1)" you might see only the first pixel of the image (which might be black).
Also, make sure that GEM can find your image (ie, that the path name is correct).
GEM supports gray8, YUV, and RGBA images. If
it sees that the number of bits per channel and the number of channels
is something that it should be able to handle, it tries to load the raw
data. If you have compressed or stored the pixel data in some "strange"
format, then GEM will probably not read the information correctly.
Also, if it is an RGBA image, then make sure that the alpha channel is something useful (this only matters if you are using the alpha channel, like in the alpha object or pix_mask).
This error message occurs whenever a pix object receives a gray8 image and the implementor hasn't provided a way to deal with that format of image. (Implementors often only provide functions for GEM's native color-format RGBA. Any other color-format (like BGR) will try to call the function for gray8 images, which might not be supported.) If you do not want to change the image format with some extern image-programm (like Photoshop or the Gimp) you might want to try pix_rgba or harass whoever made the object to add the functionality.
GEM can read in TIFF, JPEG, and SGI images. These can be in any color format. Gray scale images are loaded in as gray scale (ie, one byte per pixel). Everything else is loaded in or converted to an RGBA image (ie, four bytes per pixel). If there is an alpha channel, then it will be respected. Otherwise, the alpha channel will be set to fully opaque (alpha == 255).
GEM can write TIFF and JPEG images. TIFF-images will be full RGBA-images, wheras JPEG-files only support (compressed) RGB.
The movie formats GEM can handle (still) depend on the platform you are using.
On Windoze you can read all AVI-files you have codecs for
On linux the readable formats depend on the libraries you had installed when you compiled GEM. Currently there is (optional) support for AVI, quicktime (*.MOV) and MPEG (*.MPG) files. Not all quicktime-formats are supported. This is unfortunate but is due to linux restrictions. I highly recommend that you install the mpeg3-library from Heroine because it is much more stable than mpeg1 (which comes with many linux-distributions). If you have compiled in support for libavifile, you will be able to open Micro$oft-AVI-files. If you have installed the proper codecs (libavifile supports a mechanism for loading codecs from windows-DLLs) you should be able to open almost any format. If you have serious problems, mail them to me. (Be ready to upload the movie-file that won't work)
pix_draw is almost never hardware accelerated on PCs graphics accelerator. This means that it runs extremely slowly. Always use pix_texture, even if you are just displaying an image.
WORKING WITH PD
5.1) Why do I get clicks in the audio?
If you are getting a constant stream of clicks in
your audio, then it is probably because you are trying to do graphics and
audio in the same process. Rendering a graphics frame usually takes
longer than the size of the audio buffer, which is why you get clicks (the
clicks are usually at 20Hz...the typical frame rate).
One way around this is to use two computers, one for graphics and one for audio. If you have enough processing power (or dual processors), then you can run two versions of Pd, one for graphics and one for audio. Just use netsend and netreceive to have the two versions of Pd talk to each other.
One simple way to get raw audio values right now is to use snapshot~. Just set up a metro which bangs snapshot~ and use the floating point value. If you want "musical" information, then use objects such as env~. You might also have a look at the pix_sig2pix~ which interprets audio-data as pixels and its counterpart pix_pix2sig~
This means that GEM can't locate the file. If you use an absolute path (with / for instance), then GEM will look there. Otherwise, GEM will look in the directory of where the patch is. Then pd/GEM will search the paths you specified at startup with the -path flag.
Check the following:
1) Does the file exist?
2) Did you make a typo in the filename?
3) Is the file in the search-path ?
One of the biggest performance hits is having UI
elements in your patch which have to be updated. The biggest performance
hog is the number box. While the number box is great for debugging,
make sure that they are all gone from your "release" patch. If you
run a performance meter, you will see that whenever Tcl/Tk has to update
the user interface, it sucks the entire processor. Another examples
of this is when you move a lot of objects at once, everything jerks and
slides across the screen. There are probably ways to improve this...
Another problem is doing unneccessary calculations. When you are throwing lots of numbers around, especially packing/unpacking, doing vector math, etc., they add up. If the calculations are going unused (for instance, that part of the patch is turned off), then do not trigger the math objects. Use spigot or gate and block the events early. This is especially important with objects that send a lot of numbers, like ~ objects or line/tripleLine.
WRITING NEW GEM OBJECTS
6.1) How do I write a new GEM object?
For the time being, you have to look at the code. It is fairly well documented and straight forward (if you know C++ and OOP). Start with an object which is similar to what you want and derive a new class. The biggest issue right now is how to load in GEM as a DSO/DLL. For SGIs, you will need to setenv LD_LIBRARY_PATH. On NT, you will need to have your path include the directory with GEM.
GemMan (and by association, gemwin) disables alpha testing, alpha blending, culling, and lighting. Lighting defaults to two sided, with GL_COLOR_MATERIAL enabled. The viewport is set to
float xDivy = (float)m_width / (float)m_height;
glFrustum(-xDivy, xDivy, -1.0, 1.0, 1.0, 20.0);
gluLookAt(0.0, 0.0, 4.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0);
glViewport(0, 0, m_width, m_height);
which gives a range of about -4 to 4 in X and Y at the origin. This is a small range, but changing it now would break a lot of patches.
The specific functions to look at are:
7.1) Why doesn't <object> exist on <platform>?
Usually, this is because I don't have the resources to get the object running on that platform. If an object that you want doesn't exist on your platform, then ask for it! However, if it is tied to hardware, then it is much less likely that I will be able to do anything about it (unless someone donates the hardware to me...)
gemtablet only works on WinNT. I don't
have drivers for IRIX or Linux (also, see question 7.4)
If GEM can find the tablet, then it will print a message at window creation time. If you don't see a message, then GEM doesn't think that you have a tablet.
The tablet is mapped to the size of the GEM graphics window.
Set the environment variable
GEM_NO_TABLET = 1
Basically, I don't have physical access to an SGI machine. This makes it hard to do some of the OS specific work. It should be straightforward to do the event handling, so if someone gets it working, I would love to include it (and give you credit). All you have to do is call the correct event functions from GemEvent.h and everything should just start to work (ie, gemmouse doesn't have any OS specific code in it).
You need to make sure that your SpaceOrb is hooked
up correctly. I am using a library which isn't supported by SpaceTec so
there can be problems, although I have not had any.
<RANT> When will companies wake up and actually provide drivers and support for their products under WinNT? </RANT>
I haven't completely figured out how to get access to the video stream in WinNT. I'm using Video for Windows with a Connectix QuickCam, as well as an Intel Video Capture Card, and it seems to assume that you are only writing to a file or previewing into a window. Windows tries to take over the system and doesn't really provide any stable hooks (unlike IRIX). If anyone knows how to deal with this, please let me know.