Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Make op(Array, Scalar) use broadcast #19692

Merged
merged 1 commit into from
Dec 28, 2016
Merged

Make op(Array, Scalar) use broadcast #19692

merged 1 commit into from
Dec 28, 2016

Conversation

andreasnoack
Copy link
Member

@andreasnoack andreasnoack commented Dec 23, 2016

These are broadcast operations and could therefore just use the broadcast functionality. @martinholters I had to tighten a single type test that you added but I believe that this can just be considered an improvement.

@stevengj
Copy link
Member

The original thinking was that, for example, we shouldn't widen Array{Float32} even when multiplying it by a Float64. But in cases where this matters, one probably wants to use the (new) in-place operations anyway, so this seems like a reasonable change to me.

Needs a NEWS item under breaking changes, though.

@martinholters
Copy link
Member

martinholters commented Dec 23, 2016

I had to tighten a single type test that you added but I believe that this can just be considered an improvement.

I had the same change in the branch mentioned in #19669, so 👍.

@kshyatt kshyatt added the broadcast Applying a function over a collection label Dec 24, 2016
@kshyatt
Copy link
Contributor

kshyatt commented Dec 27, 2016

Can we merge this?

@andreasnoack
Copy link
Member Author

Still needs an updated NEWS.md. Will do that now.

@andreasnoack
Copy link
Member Author

Done.

@tkelman
Copy link
Contributor

tkelman commented Dec 27, 2016

@nanosoldier runbenchmarks(ALL, vs = ":master")

@KristofferC
Copy link
Member

So code like f(x::Vector) = x[1] < 0.5 ? x[1] * x : 0.1 * x will now be type unstable for Vector{Float32}?

@andreasnoack
Copy link
Member Author

Yes. Unless we change the behavior of broadcast promotion. I don't really think it makes sense to have different promotion behavior between these and broadcast. The main point here is to make the two similar but I also think it makes more sense to follow the scalar-scalar promotion rules. The function is type unstable in the same way f(x::Number) = x < 0.5 ? x*x : 0.1*x is type unstable for Float32.

@nanosoldier
Copy link
Collaborator

Your benchmark job has completed - possible performance regressions were detected. A full report can be found here. cc @jrevels

@andreasnoack
Copy link
Member Author

Couldn't reproduce the regression locally so I assume it is noise.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
broadcast Applying a function over a collection
Projects
None yet
Development

Successfully merging this pull request may close these issues.

7 participants