JuliaLang / julia / #37757 / 1
77%
master: 77%

DEFAULT BRANCH: master
Ran
Files 361
Run time 9s
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst
coverage: 87.428% (+0.03%) from 87.395%
#37757.1

push

local

web-flow 
LinearAlgebra: copyto! between banded matrix types (#54041)

This specialized `copyto!` for combinations of banded structured matrix
types so that the copy may be O(N) instead of the fallback O(N^2)
implementation.

E.g.:
```julia
julia> T = Tridiagonal(zeros(999), zeros(1000), zeros(999));

julia> B = Bidiagonal(ones(1000), fill(2.0, 999), :U);

julia> @btime copyto!($T, $B);
  1.927 ms (0 allocations: 0 bytes) # master
  229.870 ns (0 allocations: 0 bytes) # PR
```

This also changes the `copyto!` implementation for mismatched matrix
sizes, bringing it closer to the docstring. So, the following works on
master:
```julia
julia> Ddest = Diagonal(zeros(4));

julia> Dsrc = Diagonal(ones(2));

julia> copyto!(Ddest, Dsrc)
4×4 Diagonal{Float64, Vector{Float64}}:
 1.0   ⋅    ⋅    ⋅ 
  ⋅   1.0   ⋅    ⋅ 
  ⋅    ⋅   0.0   ⋅ 
  ⋅    ⋅    ⋅   0.0
```
but this won't work anymore with this PR. This was inconsistent anyway,
as materializing the matrices produces a different result, which
shouldn't be the case:
```julia
julia> copyto!(Matrix(Ddest), Dsrc)
4×4 Matrix{Float64}:
 1.0  0.0  0.0  0.0
 0.0  1.0  0.0  0.0
 0.0  0.0  0.0  0.0
 1.0  0.0  0.0  0.0
```
After this PR, the way to carry out the copy would be
```julia
julia> copyto!(Ddest, CartesianIndices(Dsrc), Dsrc, CartesianIndices(Dsrc))
4×4 Diagonal{Float64, Vector{Float64}}:
 1.0   ⋅    ⋅    ⋅ 
  ⋅   1.0   ⋅    ⋅ 
  ⋅    ⋅   0.0   ⋅ 
  ⋅    ⋅    ⋅   0.0
```
This change fixes https://github.com/JuliaLang/julia/issues/46005.

Also fixes https://github.com/JuliaLang/julia/issues/53997
After this,
```julia
julia> @btime copyto!(C, B) setup=(n = 1_000; B = Bidiagonal(randn(n), randn(n-1), :L); C = Bidiagonal(randn(n), randn(n-1), :L));
  158.405 ns (0 allocations: 0 bytes)

julia> @btime copyto!(C, B) setup=(n = 10_000; B = Bidiagonal(randn(n), randn(n-1), :L); C = Bidiagonal(randn(n), randn(n-1), :L));
  4.706 μs (0 allocations: 0 bytes)

julia> @btim... (continued)

76360 of 87340 relevant lines covered (87.43%)

16050766.29 hits per line

Source Files on job #37757.1