Re: [CAD_CAM_EDM_DRO] Imbedded computers

Codesuidae wrote:

>Stephen Wille Padnos wrote:
>
>
>>The one thing that is different from most desktops is embedded video.
>>Not just "video on the motherboard", but "video using shared memory
>>access". This can be a problem for interrupt latency, and the CPU can't
>>necessarily even tell that anything bad is happening - it gets
>>effectively stopped while the video subsystem is using main memory
>>
>>
>A question regarding systems that may exibit this problem. AGP uses
>main memory to expand texture memory (to reduce the need to use more
>expensive video memory and the requirement to move large textures across
>the PCI bus into video memory). I believe all AGP video cards use
>onboard video memory for mosts tasks, and the amount of main memory used
>for textures is configurable. Does disabling AGP use of main memory for
>textures correct the problem? Since motion control software is not
>particularly noted for its intensive use of 3D texture mapping :),
>disabling use of main memory for textures shouldn't present any problems.
>
>Dave K
>
>

Isn't it usually just the opposite - ie, the textures are transferred to
the card once (possibly even compressed), then used many times in rendering?

The trouble with the shared memory systems is that it's main memory on
*both* sides of the AGP "bus". Part of main memory is "hidden" from the
CPU by the chipset, but it's still accessed on the same bus.

Unfortunately I don't have any (personal) data on this, so I may do some
testing one of these days. I have a system that has an onboard video
controller, though it's not integrated into the chipset. Using
Linux/RTAI, I see latencies of around 5-6 microseconds, which isn't
bad. That's using the RTAI test programs though, so I'm not sure that
they can measure what really happens on the parallel port pins. I'll
figure out some way to check, and put the scope on it one of these days.

- Steve