Programming model on FPGA

hi awa, I took a short glance at some papers about your topic.

The term low level programming (model) seems to be used with multiple meanings, thats why you are confused. (me too, to be honest)

One thing is programming on the assembly level, creating fast libraries for image processing algorithms. But then it is only refered as "low level programming" w/o "model".

The other is the implementation of the basic algorithms. As in arithmetics on FPGAs, where you can choose between parallel, serial and pipelined architectures, the main question in image processing is: Using a sequential or a parallel approach.

In both cases each approach has its advantages and disadvantages. The main disadvantage for FPGAs seems to be that (besides their ability to be reconfigurable, but that also needs some kind of control) they are not flexible enough when you want or need to change/adapt your algorithm to a specific problem. In Software, once the algorithm is written, you check how many Processors and Memory is available and look in the library for the best algorithm for your problem and (if applicable) share the data and (sub)programs dynamically over the processors.

In hardware you don't have this flexibility, or have to use a large controlling overhead, which minimizes the processing gain you normally have by using FPGAs instead of µPs.

In general an FPGA has the advantage of massive parralel computation power, but for immage processing, some algorithms have data depentant runtimes. So synchronizing results becomes an additional task.

There may be an approach to overcome these problem. One is FPGA-internal processors making it a hardware software codesign,(Hardware for algorithms, software for data handling) but then it may be no longer seen as low level...?

If you have to do image processing in a very specialized way (e.g. image based quality control in production flows) where the same kind of image has to be processed always in the same manner with one specialized piece of hardware the FPGA will surely beat any ordinary computer hardware.

So, you have to look which field you are really working in. Scientific image processing or industrial image processing?

Best regards Eilrt

awa schrieb:

Thanks a lot backhus. I really appreaciate your effort. The information is helpful. Thanks again.

Cheers!