Flat Sequence Causes Buffer Allocation Shift and Impacts Benchmark Results in LabVIEW

[Andrey_Dmitriev]

Dear colleagues,

I observed some strange behavior during a simple benchmarking experiment.

The code is trivial: two U16 arrays are added (no need to worry about overflow, this is just dummy calc):

On my Xeon W5-2445, the execution takes around 125–130 ms for both runs, which is expected. Nothing unusual so far.

However, when I added a Flat Sequence around the constant (without any fundamental change in dataflow), the result changed — the first test now takes almost twice as long:

It seems like buffer allocation was moved into the benchmarked section. Let’s check, indeed:

Interestingly, if I apply "Copy Buffer" only to the upper array, the timing remains the same:

But if I apply it only to the second array, the doubled execution time returns:

Tested on LabVIEW 2025 Q1 (64-bit) 25.1.2f2 and LabVIEW 2018 SP1 (32-bit) 18.0.1f4.

Yes, the buffer allocation dot is here:

But why was it "asymmetrically" moved to this location just because a "meaningless" sequence was added? It looks like some hidden compiler's "optimizations" are at play...

[LVNinja]

After runnin example 1 

After runnin example 2

It looks like by adding the  sequence is reducing the data usage, would that explain the speed ? it runs faster but consumes more memory. (Don't aske me why...)

[Martin_Henz]

Hi Andrey,

I have seen some strange effects when using flat sequences during speed tests in the past. I am very uncertain in this regard. I suspect that there are situations where the first frame could start execution before all the data is available at the sequence.

However, the effect you describe is new to me.

I recommend using the stacked sequence structure for time measurements. I haven't seen such effects there so far. If you replace the flat sequence with a stacked sequence, then this effect will also disappear.

[Yamaeda]

That's an interesting find indeed!