Hello, Is there a nice way to use vectors in arrays without spending too much time on allocations? I have some parallel for loops and if I set 3 scalar variables to certain values, it seems to perform better than if I set one array with 3 elements (and it works in CUDA kernels). I don’t really kno…

Is there a reason you can’t use a multidimensional array?

@Mason you mean within my for loop have one variable per vertex per element? Because that would take up way too much memory and most of those variables are temporary. Thanks!

[image] Ribeiro: vertices = points[panels[k]]; for v in 1:panelType vnext = v%numVertices+1; e[v] = vertices[vnext]-vertices[v]; E[v] = norm(e[v]); end You can do that with a vector of static arrays. In that code e[v] does not seem necessáry as an array (coukd be only a temporary vector…

I don’t think your question is phrased very clearly and I’m having some trouble understanding. I am suggesting that instead of writing code like v = [[1, 2, 3], [4, 5, 6]] and accessing elements like v[i][j] you should instead write m = [1 4 2 5 3 6] and access elements as m[i, j] T…

If you show us an actually runnable snippet of code that demonstrates a simplified version of your problem, we might be able to offer more specific advice. [image] Please read: make it easier to help you Meta Discussion Welcome to the Julia Discourse! We are enthusiastic …

I might be wrong, but maybe this note I have may be helpful: Immutable variables · JuliaNotes.jl

@Mason I started with a regular matrix and when I switched to a Vector of StaticArrays my code became 10x faster. @lmiq Thanks for the link. I learned some stuff =]. I thought this was a common enough problem that I wouldn’t need to make a full example, but here’s one: N=1000; vertexList = Vector…

[image] Ribeiro: So temp1 is waaaay better You mean temp2 is better right? Note that N=length(vertexList), you don’t need to pass it as a parameter. In temp1 you should passa v and vNext as parameters, otherwise they are global variables. I think you can do something on the line you want. …

Sorry, I just edited that mistake. I wrote the sample code real quick, so some stuff is odd. But the changes you mentioned don’t make a difference. Thanks! I’m messing around a bit and this helps a lot: edges = @SVector [vertices[v%elementLength+1]-vertices[v] for v in 1:elementLength]; Becaus…

[image] Ribeiro: [vertices[v%elementLength+1] Probably this is allocating something. Can you make it a static vector as well? Just a curiosity: I have never seen the % notation before. Is that from some package? Where did you find it? Small things: while you may not see a difference in one…

Local arrays in for loops

New to Julia

Mason February 28, 2021, 11:11pm 6

If you show us an actually runnable snippet of code that demonstrates a simplified version of your problem, we might be able to offer more specific advice.

5 Likes

Topic		Replies	Views
Avoiding allocations of small but non-trivial arrays (work array alternative?) Performance question	38	4828	November 17, 2022
Help with allocations (variable length least squares) New to Julia memory-allocation	10	1261	March 25, 2021
Common allocation mistakes Performance memory-allocation	47	7719	August 21, 2023
Memory allocations when returning vectors General Usage array , memory-allocation	15	1571	June 6, 2018
Creating a Static Array with a loop General Usage	10	1066	April 12, 2022

Local arrays in for loops

Related topics