Assigning array to a slice of another array

Hi, I’ve been trying to vectorize the following operation

for i = 1:N
    arr[i, :, :] .= arr[a, :, :]

over the indices ‘i’ as

@. arr[b:c, :, :] = arr[a, :, :]

(where ‘b’ = 1, ‘c’ = N)
such that all of the elements b:c are copied from index ‘a’. However, this throws an error:
DimensionMismatch(“array could not be broadcast to match destination”)

Makes sense, since I’m trying to assign, e.g. a 5x5 array (RHS) to a 2x5x5 array (LHS). What I want is for all the elements sliced on the LHS array to be the elements copied from arr[a, :, :]. Is there a way to do this?

arr[b:c, :, :] .= arr[a:a,:, :] should work.
Note arr[a:a,:,:] is a 1x5x5 array and not a 5x5 array like arr[a,:,:]. This is important to let broadcasting .= know which dimension to replicate on.

1 Like

As a side note, in case you’re doing this for performance reasons: Loops and general non-vectorized code in Julia have no inherent penalty to them, and can be as fast as (or faster than) vectorized code. So your for loop can stand as-is unless you have some other reason to change it.

1 Like

Another thing is that Julia Arrays are column major, unlike in numpy/C++/etc. So it would be (possibly much) faster to arrange your data so you can do

arr[:, :, i] .= arr[:, :, a]

In general, it’s a good idea to take this into account when designing your data layout.

1 Like

For better performance, you may consider using views:

arr[i, :, :] .=  view(arr, a, :, :)

Thanks for the tip, I’ll see if I can do so!

Thanks, I’ve added the @views macro to the relevant lines.