Mads

How far from the goal are you now, and have you identified where the time is spent? If you e.g. just manually launch the LabVIEW application and flag when it does its first action, how much of the minute is spent already and how much time does the rest of its tasks use? You say you want the program to be launched by Windows as quickly as possible too, how much time does it take for Windows to launch it now compared to any other simple application/service you can set up?

When I first saw the compount add I though that must be use because it is quicker than multiplying by 3, but in the tests on my machine it is the slower. With the replace logic though it does not matter anymore. A digression: Branch prediction is an interesting phenomenon when dealing with optimizations. In this particular case it did not come into play, but in other cases there might be a benefit in making consecutive operations identical 🙂 : https://stackoverflow.com/questions/289405/effects-of-branch-prediction-on-performance

You beat us well @ShaunR 🙂 Normally I would use replace in situations like this too, to get the advantage of the preallocated memory. I improved some of the OpenG array functions that exact way (which is now included in Hooovah's version), but lately I've gotten so used to the performance and simplicity of conditional indexing that I did not grab for it here 🤦‍♂️.

A bit quicker it seems is to use conditional indexing, like this: On my machine I got the following results (turned off debugging to avoid it interfering with test and removed the display of input as I would occationally get memory full errors with it): Original: 492 ms ensegre original: 95 ms ensegre with multiply by 3 instead of 3 adds: 85 ms Conditional indexing: 72 ms There might be even better ways..The efficiency of conditional indexing is often hard to beat though. demo_rev1.vi

On a related note GDevCon 4 just ended in Glasgow with a record number of attendees....👍 (Maybe that is partly because the alternatives are not as good/many anymore, but let me be optimistic for a while...) If Emerson does not want to kill or starve LabVIEW, maybe there is a nice future beyond 10 years as well. Retirement age is steadily getting pushed upwards though so personally it would be nice if things went on for another 25 years, or even better; allowed our kids to make a career working with G everywhere in 2050😀

We use the nuvola icon set for a lot of things.

Found it, here:

The link only points to Jordan's profile, but he might have been involved back then yes...

I had noted a simple recipe earlier from here somewhere for manually installing lvzlib.so (OpenGZip 4.2beta) on Linux RT 2020 and newer now that the "Custom Install" option no longer exists / works in NI Max, but I seem to have lost it and cannot find it again. Do anyone know where that discussion was, or has the recipe?

When it comes to "Remove Duplicates" the performance of OpenG is a bit of a curious case (unless I am just overlooking something obvious). The original OpenG function builds the output arrays one element at a time, but still (in my LabVIEW 2022 at least) outperforms the alternatives I make where I reuse the input array. The latter use less memory it seems, as expected, but it is as if LabVIEW has become smarter when there are repeated build operations (but not for the functions we have gone through previously...(yes debugging is disabled..) 🤨 The attached version is a simplified one of the original suggested revision. As mentioned it does seem to use less memory than the original OpenG, but is still slower 😞 Remove duplicates.zip

Not sorting the array we are processing, sorting the array of elements to delete. That is done just to facilitate the filtering algorithm, it does not affect the array that is to be processed. (Unless the sort has somehow gotten misplaced somewhere in the code that has been bounced around here...)

?

Performance mainly becomes an issue when the task at hand is heavy. I do not think saving microseconds on a small array/filter list is worth losing milliseconds or seconds when the array gets large. So that would be the rationale for simply ignoring the "quick on small arrays, slow on large" and choosing the one that is the fastest in the worst cases. However - there is one case where this can become an argument *for* keeping the other algorithms too; namely when you have to repeatedly run through small array/filtering jobs....🤨 Life is never easy😬 The "Revised version" is the best middle ground; quicker than all on most small and medium arrays, and still 2x times quicker than OpenG on worst case scenarios. The Rev2 one is 10 times quicker then...but in truly worst case scenarios you typically end up writing customized algorithms instead of using generic libraries anyway...(I would then often skip the deleted array indexes output to get things even quicker e.g.).

An inversion error snuck in, if you look at the code and invert the inversion its right 🙂 Attached here... Filter 1D test - fixed.zip

The old read me file says that only 8 functions were changed so I took a look at one of them now: Filter 1D array Improvement when filtering every 5th element in a 100k array: 2.4x Improvement when filtering every 100th element: 2.2x Improvement when filtering every 1000th element: 1.8x OK, but not that impressive... However, looking at the algorithm I saw that there are repeated 1D Array searches for the whole array of elements to filter, that is costly. So I threw in a new alternative where that search is based on the sorted search instead. That makes a big difference when the number of items to filter is big, for the test cases above the results on my computer were Improvement when filtering every 5th element in a 100k array: 227x Improvement when filtering every 100th element: 22x Improvement when filtering every 1000th element: 3x Now the last algorithm changed two things, it uses both conditional indexing and sort/search sorted array. The sorting bit could just as well be added to the original revision...I guess the reason why OpenG still does not use conditional indexing is to be backwards compatible (very far now though...), and then the in-built sorted search would not be available either. That could be replaced by a binary search function that was available back then too though. Filter 1D test.zip