Brian Beach
Note: This document uses the O2 VL DMbuffer API instead of the cross-platform VL DMbuffer API (see What Are the SGI Video-Related Libraries? for more info). It refers to IRIX 6.3 man pages and documentation.
If you've read through the overview of the use of DMbuffers with video and graphics, you already know most of what's needed to use video as a texture.
The basic steps are as follows:
There are a number of code fragments on this page, all of which came from the example program "composite". You'll find the source for it in: /usr/share/src/dmedia/video/vl/OpenGL/composite.c.
The image layout must be either VL_LAYOUT_GRAPHICS or VL_LAYOUT_MIPMAP. Use VL_LAYOUT_GRAPHICS if your rendering does not use mipmapping; the non-mipmapped images will be smaller and save memory.
The packing must be either VL_PACKING_ARGB_1555 or VL_PACKING_ARGB_8. For mipmapped textures, the only choice is VL_PACKING_ARGB_1555, because of the increased I/O bandwidth required for the larger mipmapped images.
The tricky part is the size of the image. To use them as a texture, both the width and the height must be a power of 2. Standard video is not the right size. The video input path on O2 support arbitrary scaling down of the input image. To use an NTSC input (640x480) as a texture, you'll have to scale it down to 512x256. Here is the complete list of settings for the memory node, as set by the composite example:
int width = 512;
int height = 256;
setIntControl ( path, drain, VL_PACKING, VL_PACKING_ARGB_1555 );
setIntControl ( path, drain, VL_CAP_TYPE, VL_CAPTURE_INTERLEAVED );
setIntControl ( path, drain, VL_FORMAT, VL_FORMAT_RGB );
setXYControl ( path, drain, VL_MVP_ZOOMSIZE, width, height );
setXYControl ( path, drain, VL_SIZE, width, height );
setFractControl( path, drain, VL_RATE, 30, 1 );
setIntControl ( path, drain, VL_LAYOUT, VL_LAYOUT_MIPMAP );
A frame buffer configuration (fbconfig) must be chosen that matches the VL packing. For VL_PACKING_ARGB_1555, you must choose an fb config where the red, green, and blue components are exactly 5 bits deep, and the alpha is exactly 1 bit deep. Choosing the usual 8888 config will not work. Here is the code used by composite to get the right config:
GLXFBConfigSGIX getFBConfig(
void
)
{
int i;
/*
* Find a frame buffer configuration that suits our needs: it must
* work with both the pbuffers and the window, and must have 5551 pixels,
* because that is the only format that can be used to capture
* mipmapped video.
*/
GLXFBConfigSGIX config = NULL;
int params[] = {
GLX_RENDER_TYPE_SGIX, GLX_RGBA_BIT_SGIX,
GLX_DRAWABLE_TYPE_SGIX, GLX_PBUFFER_BIT_SGIX,
GLX_ALPHA_SIZE, 1,
GLX_RED_SIZE, 5,
GLX_GREEN_SIZE, 5,
GLX_BLUE_SIZE, 5,
GLX_X_RENDERABLE_SGIX, False,
(int)None,
};
int configCount;
GLXFBConfigSGIX* configs =
glXChooseFBConfigSGIX(
theDisplay,
DefaultScreen(theDisplay),
params,
&configCount
);
CHECK( configs != NULL || configCount != 0,
"Could not get FBConfig" );
for ( i = 0; i < configCount; i++ )
{
int value;
glXGetFBConfigAttribSGIX(theDisplay, configs[i], GLX_ALPHA_SIZE, &value);
if ( value != 1 ) continue;
glXGetFBConfigAttribSGIX(theDisplay, configs[i], GLX_RED_SIZE, &value);
if ( value != 5 ) continue;
glXGetFBConfigAttribSGIX(theDisplay, configs[i], GLX_GREEN_SIZE, &value);
if ( value != 5 ) continue;
glXGetFBConfigAttribSGIX(theDisplay, configs[i], GLX_BLUE_SIZE, &value);
if ( value != 5 ) continue;
break;
}
if ( i != configCount )
{
config = configs[i];
}
CHECK( config != NULL, "No 5551 FBconfigs were found" );
XFree( configs );
return config;
}
Once the fb config is chosen, you just have to create a dm pbuffer that is exactly the same width and height as the texture:
GLXFBConfigSGIX fbconfig = getFBConfig();
int attrib[] = {
GLX_PRESERVED_CONTENTS_SGIX, GL_TRUE,
GLX_DIGITAL_MEDIA_PBUFFER_SGIX, GL_TRUE,
None};
pbuffer = glXCreateGLXPbufferSGIX(
theDisplay, fbconfig, 512, 256, attrib);
This needs to be done after creating the video path (so that the VL's pool constraints can be queried), but before starting the path (because it needs a buffer pool to put the input frames into).
Be sure to query the constraints of both the video input path, using vlDMPoolGetParams, and the pbuffer, using dmBufferGetGLPoolParams. Then create the pool using dmBufferPoolCreate.
After the pool is created it must be bound to the video path using vlDMPoolRegister.
Don't forget this step. It's the usual reason that my VL programs don't do anything.
The code looks like this:
VLEvent event;
DMbuffer buffer;
int vlstatus;
vlstatus = vlEventRecv( theVideoServer, path, &event );
while ( vlstatus == -1 && vlGetErrno() == VLAgain )
{
sginap( 1 ); // XXX select() is better in real programs
vlstatus = vlEventRecv( theVideoServer, path, &event );
}
CHECK( vlstatus == 0,
"Could not get video input event" );
CHECK( event.reason == VLTransferComplete,
"Unexpected video event received" );
vlstatus = vlEventToDMBuffer( &event, &buffer );
CHECK( vlstatus == 0, "Video event does not have a buffer" );
This make the contents of the DMbuffer be the contents of the color buffer in the pbuffer:
Bool ok;
ok = glXAssociateDMPbufferSGIX( theDisplay, pbuffer, format, dmbuffer );
CHECK( ok, "Could not associate path2 DMbuffer with pbuffer" );
The format is a DMparams list that contains with width, height, layout, and packing of the image being passed in.
Before coping the texture into place, two things must be prepared.
First, there must already be a texture of the same size in the current context. It can be any image, it just has to be the right size, including all of the mipmap levels:
int w = MY_TEX_WIDTH;
int h = MY_TEX_HEIGHT;
int level = 0;
void* black = calloc( w * h, 4 );
level = 0;
while ( 1 )
{
glTexImage2D(
GL_TEXTURE_2D,
level,
GL_RGB5_A1_EXT, /* number of components */
w, h,
0, /* border */
GL_RGBA,
GL_UNSIGNED_BYTE,
black
);
printf( "%d: %d x %d\n", level, w, h );
if ( w == 1 && h == 1 )
{
break;
}
if ( w > 1 ) w /= 2;
if ( h > 1 ) h /= 2;
level++;
}
free( black );
Second, the current readable and drawable must be set to copy from the image to be used as texture. Usually this can be done once at startup when setting the drawable to be a window (or wherever the rendering is to happen):
Bool ok;
ok = glXMakeCurrentReadSGI( theDisplay, drawable, pbuffer, context );
CHECK( ok, "glXMakeCurrent failed" );
Under the right circumstances, described in intro.html, glCopyTexSubImage2DEXT will do a copy-by-reference. The image in the pbuffer will be used directly as a texture, without having to copy the pixels. Also, for video input, the mipmaps are already present and they are also used directly, with no need to create them in software. Before loading a mipmapped texture, the mipmap generation extension must be turned on. (O2 doesn't support this extension, but requires that it be turned on in this case.)
glTexParameteri(GL_TEXTURE_2D, GL_GENERATE_MIPMAP_SGIS, GL_TRUE);
glCopyTexSubImage2DEXT(
GL_TEXTURE_2D,
0, /* level */
0, 0, /* x, y offset into target */
0, 0, /* lower-left corner of copied rect */
512, 256 /* size of copied rect */
);
That's all there is to it! Only a couple of hundred lines of code to do a very simple thing. (Sarcasm intended.)
Now that the texture is loaded, all of the normal GL rendering calls can use the texture.