What is the most efficient way to replace array subset in this case on LabVIEW FPGA?

[UMASO]

Hello, this is a kind of quiz which asks the most efficient way to do the following thing on FPGA. 

The VI attached on this post is the way I applied to what I would like to accomplish.  

What I would like to accomplish is the following.  

1. There is an U32 array with 64 elements
2. Specify an array index 
3. Replace all values before the index with the value specified by the array index.  

Let's say, I have an array of [5, 6, 7, 8, 9].  If I specify 3 for array index, that points to 8 in that array.  Then, the pointed value replaces all values before the index, so, resulted array is [8,8,8,8,9].  If I specify 1 for array index, 6 is pointed by the index, and the resulted array will be [6, 6, 7, 8, 9].

What is the most efficient way to accomplish this on LabVIEW FPGA?  I tried various array functions such as Delete from Array, Replace Array Subset, Array Subset, etc, but Undefined Array Size error pops up at intermediate file generation.  

If any smart people could share any idea, it would be really appreciated.  

Thanks much in advance.

[JensG69]

I would do something like this:

For added efficiency and speed, replace the while loop with a SCTL.

Regards, Jens

[nathand]

Auto-indexed for loop. If the index is less than or equal to the target index, use the new value; otherwise, use the existing value. You can even put this for loop inside a single-cycle timed loop (although note that there was and perhaps still is a bug where Select doesn't work inside a for loop inside a single-cycle timed loop; if so, use a case structure instead, see https://knowledge.ni.com/KnowledgeArticleDetails?id=kA00Z0000019OLtSAM&l=en-US).

It would be an interesting experiment to look at the relative FPGA resource use of doing this inside versus outside a single-cycle timed loop as the array size grows. For a large array, putting it inside a single-cycle loop will cause a lot of copies of the comparator (which isn't that big, and if you know the value will fit in a U8 then I would cast to that to save space) but will of course execute in a single cycle. For a small array I doubt there will be much difference between inside and outside an SCTL.

[UMASO]

Hello JensG69,  

thanks for taking time to provide the idea.

 I apologize that I did not clearly describe the requirement.  I need to do  the process within one clock cycle of SCTL.  So, I cannot use any iterations.  

[UMASO]

Hello nathand ,  

thanks for taking time to provide the idea.

 I apologize that I did not clearly describe the requirement.  I need to do  the process within one clock cycle of SCTL.  So, I cannot use any iterations.  

[nathand]

As I mentioned in my post, you can put the for loop inside a single cycle timed loop. All iterations will execute in parallel.

If that's not clear, I'll try to post an example later when I'm in front of a computer.

[UMASO]

Hi nathand, 

Thanks much for the advice.  Please do not take any more time at this moment, as I will try your idea right today.  I will get back once I find anything.  Thanks!

[nathand · Accepted solution]

I'd just coded this snippet when I saw your reply so I might as well post it. I don't have the FPGA toolkit installed but this is the right idea, just with the wrong appearance for the timed loop.

[UMASO]

Hello nathand, 

Thanks much again for teaching me the trick!  

I tested it and it worked!  I will follow your method in my project and it was a great learning.  Thanks a lot!

Just FYI, below is the result of my test.  

Test conditions.

• LabVIEW 2017 SP1
• PXIe-6592R
• not removed implicit enable
• compiler setting == default 

Below is your method and has many advantages compared to my original code. 

• much much simpler than my original code.
• while Registers are used in almost the same amount, LUTs are saved about 2000 in Kintex7.  

Below is my original code, 

• its may reach up to 290MHz.