# Special case \`lmul!(0, ::AbstractArray)\`?
**URL:** https://discourse.julialang.org/t/special-case-lmul-0-abstractarray/93557
**Category:** Internals & Design
**Tags:** array, linearalgebra
**Created:** [January 26, 2023, 7:07am UTC](https://discourse.julialang.org/t/special-case-lmul-0-abstractarray/93557 "2023-01-26T07:07:26Z")
**Posts on this page:** 4
**Page:** 1
### Author:  [@jishnub](https://discourse.julialang.org/u/jishnub)
#### Post date: [January 26, 2023, 7:07am UTC](https://discourse.julialang.org/t/special-case-lmul-0-abstractarray/93557/1 "2023-01-26T07:07:26Z")
```julia
julia> using LinearAlgebra
julia> function lmul2!(s::Number, X::AbstractArray)
iszero(s) && return fill!(X, s)
@simd for I in eachindex(X)
@inbounds X[I] = s*X[I]
end
X
end
lmul2! (generic function with 1 method)
julia> A = rand(1000, 1000);
julia> @btime (A -> lmul!(0, A))($A);
290.313 μs (0 allocations: 0 bytes)
julia> @btime (A -> lmul2!(0, A))($A);
222.413 μs (0 allocations: 0 bytes)
```
Perhaps the compiler should transform the loop to a `fill!`. It does seem to handle `1` specially, transforming the loop to a no-op.
```julia
julia> @btime (A -> lmul!(1, A))($A);
2.883 ns (0 allocations: 0 bytes)
```
If this is complicated, adding an `iszero` check may help with performance.
---
### Author:  [@mbauman](https://discourse.julialang.org/u/mbauman)
#### Post date: [January 26, 2023, 2:25pm UTC](https://discourse.julialang.org/t/special-case-lmul-0-abstractarray/93557/2 "2023-01-26T14:25:47Z")
The compiler is doing the best it can in the face of `NaN`s and `Inf`s, I think. `1*[2.3 NaN Inf]` is indeed `[2.3 NaN Inf]`, but `0*[2.3 NaN Inf]` isn’t `[0.0 0.0 0.0]`.
---
### Author:  [@mikmoore](https://discourse.julialang.org/u/mikmoore)
#### Post date: [January 26, 2023, 4:41pm UTC](https://discourse.julialang.org/t/special-case-lmul-0-abstractarray/93557/3 "2023-01-26T16:41:37Z")
One can debate the merits, but there is precedent for strong zeros in linear algebra functions:
```julia-repl
using LinearAlgebra
julia> A = fill(1.0,3,3); B = fill(NaN,3,3); C = fill(NaN,3,3);
julia> mul!(C,A,B,0.0,0.0)
3×3 Matrix{Float64}:
0.0 0.0 0.0
0.0 0.0 0.0
0.0 0.0 0.0
```
Even if we don’t like numeric zeros acting as strong zeros, we could certainly adopt `false` since it already acts that way.
---
### Author:  [@Dan](https://discourse.julialang.org/u/Dan)
#### Post date: [January 26, 2023, 5:00pm UTC](https://discourse.julialang.org/t/special-case-lmul-0-abstractarray/93557/4 "2023-01-26T17:00:35Z")
> [@jishnub](#):
>
> `iszero(s) && return fill!(X, s)`
Regardless of other discussion,
```julia
iszero(s) && return fill!(X, s*one(eltype(X)))
```
might be appropriate (on the fence, the original looks much cleaner)