Efficient simulation method for video processing FPGA?

[thiagu_comp]

I am trying to simulate a video processing FPGA. It processes HD resolution videos (1920x1080). Its very hard to simulate by providing this video input, as it takes about 4 days to send one frame of data (i.e., (1920+100)x(1080+45)=2272500pixels, 30bits each). Due to this limitation, its hardly possible to simulate. But I wonder how other HD video processing chips are simulated. Can someone suggest some efficient simulation / validation methods?

 

[supercat]

I am trying to simulate a video processing FPGA.... Can someone suggest some efficient simulation / validation methods?

I think the key to having your simulations run in a reasonable length of time is going to be to figure out all the different cases that really need testing and force the system state as appropriate to test them.

If, for example, your video controller were simply clocking data from a memory bitmap, the behavior of the scan lines between the third and third-to-last lines would probably not be very interesting, except possibly for those where counters hit interesting values (e.g. counting from 0111111111 to 1000000000). It may thus be helpful to skip the boring scan lines in simulation.

Of course, sometimes that approach isn't as easy as it sounds. Especially if your system uses pipelining, it may be hard to force the system state to 'jump ahead' over the 'boring' parts. Further, it's not always possible to predict which state transitions are going to be boring and which ones might have lurking bugs. Nonetheless, simulating a few "key" scan lines under various conditions and ensuring everything is working as expected will probably be more helpful than trying to simulate whole frames.

 

[kvingle]

In a case like this I will use hardware acceleration instead of using pure simulation...
these are the tools that let you download some of your hardware in actual FPGA and run it.


Another thing is as you are dealing with FPGAs, why not get the first prototype ready and evaluate on it directly....?

 

[thiagu_comp]

Thanks for the replies.
To simulate only significant places and to jump ahead is being evaluated already. But it seems to be a difficult approach.
A prototype is already available. We had already got to see the video on the screen. But we can not validate all parts of the code. So we want to use simulation to ensure everything is fine and to see some intricate cases. I am not sure about hardware acceleration, though I have heard of it. I searched in edaboard and there are no satisfying answers, though there are some posts regarding acceleration, prototyping and emulation. Can you please help me understand this, so that we can analyse the feasibility to use it. Thanks in advance.

 

[kvingle]

Sure!
read
**broken link removed**

As you hardware is ready a cheaper alternative of identify pro is also good..


**broken link removed**

 

[zape]

You can also think about using some tools integrated in Matlab, as Xilinx System Generator, that allows you to get your hardware in the loop.

Working in this way your can use Matlab for data injection, your HW for running the algorithm and once again Matlab for collecting data output and performing the validation. There is a JTAG connection between Matlab and your evaluation HW which allows this stuff.