Running Sum with array functions only

[DaIn]

Hi,

i'm currently struggling to find an array function to generate a running sum array.

My current implementation is working so far. But with bigger arrays (>100000 elements) the throughput is not sufficient.

Running sum means:

Each array element is the sum of the previous elements + the value of the according Input-array-element.

INPUT Array:

[100, 10, 95, 95, 5,5,100]

OUTPUT Array:

[100,110,205,300,305,310,410]

Does anybody know an array/matrix function to speed up this operation? 

If possible without any loops.

[GerdW]

Hi Daniel,

adding up 100k array elements takes ~1ms on my PC:

Using HighResolutionRelativeSeconds it just reports ~400µs for this routine.

With 1M elements it reports ~4ms, so this scales linearly with number of elements…

How much do you want to speed this up?

[DaIn]

As fast as possible. 

Down to a few us would be a big advantage.

But in my opinion this is only possible by using array functions.

So the question is: 

   Exists an array or matrix function which is doing this operation?

[altenbach]

---

DaIn wrote:

But in my opinion this is only possible by using array functions.

---

Based on what evidence?

The autoindexing FOR loop is extremely efficient. What makes you think otherwise?

[DaIn]

Using array functions instead of for/while loops was in the past in the most cases a good tool to speed up such operations.

This operation is implemented in the SW part of a high-throughput FPGA application.

Maybe using an auto-indexing FOR loop is the best and fastest solution.

But I want to be sure that I'm using the most efficient implementation.

[GerdW]

Hi Daln,

ohh wait: we talk about LabVIEW on FPGA now? Why didn't you mention it in the first place?

- LV-FPGA can use arrays, BUT usually insists on fixed-length arrays…

- LV-FPGA needs a lot of space to handle big arrays, so you need some streamlining of smaller chunks anyway…

[DaIn]

No no....

This function is implemented in the SOFTWARE PART of an FPGA application.

So the FPGA is generating the data. This data is streamed to the Host-vi which is running on a PC.

And this "Running sum" operation is implemented in one of the Sub-vis on the PC.

No FPGA solution needed! 

[altenbach]

---

DaIn wrote:

Down to a few us would be a big advantage.

---

Do the math!

To add 100000 elements in a microsecond, an single addition would have to take 10ps.

Even if the CPU could do an addition in a single clock cycle, you still would need a 100GHz processor. Not realistic!

[DaIn]

From that point of view: You are correct.

Like stated above: 

  I only wanted to be sure, that there is no faster way to do this operation in Labview.

  This is 100% fine for me if it is the fastest way.

[Yamaeda]

For a running sum, especially on a large array you'll only need to keep the previous sum and add the new number, no need to recalculate the previous values that'll be constant.

/Y