Bound Checks slowing code down?

[matzhaugen]

Hey I was wondering, do the bound checks for all assignment operations slow down the code? For example if I did

use ndarray::Array2;
let p: usize = 100;
let matrix1 = Array2::<f64>::zeros((p, p));
let mut matrix2 = Array2::<f64>::ones((p,  p));

matrix2 = &matrix1;

would I do p*p bound checks?

If so how can one avoid this without doing unsafe rust?

[bluss]

Since it's a question, I moved it over here to the Q&A section. It helps keeps the issue list focused.

Your code doesn't compile :) Can you fix that and mark which line you are talking about? Variable assignments/value moves in Rust don't run any library defined code at all, so that doesn't bounds check.

[matzhaugen]

Hey @mulimoen! Unfortunately that slowed the code down.
I made a little repo to illustrate my example https://github.com/matzhaugen/ndarray-bench . If you have a mac and g++ installed it should be easy to run.

[mulimoen]

I ser you use cargo run to run the binary. This will run a non-optimized binary. Either run target/release/foo or use cargo run --release

[matzhaugen]

That increased the speed by 500x. My bad, thanks for clarifying :)

[matzhaugen]

Why isn't this zip contraption the default in the assignment operations? It seems like it should be the norm. Otherwise, one has to write this zip expression for all assignments if you want the code to be fast, no?

[bluss]

There is no evidence to suggest that the zip thing is required to make the unary negation operation fast.

Also, again, you can't "overload" assignment in Rust, and assignment is not where the interesting operations happen here. For example the unary minus operation is "interesting".