I cannot use the Fixed-point Tool when using dual-port RAMs in my design - strange bug

Hello Community

I have a Simulink design which requires quite a large word lenght (64bit signed fixed-point) to run without overflows.

Now I want to optimize my design by using the Fixed-point Tool Add-On in order to identify the required resolution of sub-blocks in my design.

However, I get a strange error when I try to collect the ranges.

The block "select" is inside the "Simple dual-port RAM" block and is locked, i.e. cannot be edited. The signal "Idx1", which is connected to the "rd_addr" port, is driven by a unsigned integer (unsigned fixed-point number with no fraction). By monitoring this signal I can confirm that no signal with the value "-16" enters the RAM block. Therefore, I do not know what causes this error. Unfortunately, the RAM block is locked and I cannot monitor the internal signals.

I attached a simple Simulink model which reproduces the problem.

How can I fix this?

Answer 1

[Fangjun Jiang]

Internal "Idx1" traces back to input port "rd_addr". The error is complaining that the "rd_addr" can't be negative. In your model, the "rd_addr" could be negative due to the subtraction of 16.

[Dominik Hiltbrunner]

But "rd_addr" is driven with a unsigned data type which wraps from 0 to 63, i.e. there never is a negative number.

[Fangjun Jiang]

For "Range Collection Mode", select "Derived ranges", not "Simulation ranges". It will give you the analysis result. It shows overflows.

Although you wrap it, the value between 0 to 63 subtracted by 16 will cause negative value. Maybe the analysis tool didn't consider the wrap.

It might be worth to report this to the Mathworks tech support.

[Dominik Hiltbrunner]

Hello, thanks for your answer.

I have been playing around with the fixed-point designer for a couple of days now. Seems like there is no way around this issue. So far I use a workaround where I replace my RAM and the counters by other blocks that yield the same functional behaviour.

I will write a ticket to the support by time.