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

Some linear algebra, via reinterpret #437

Draft
wants to merge 6 commits into
base: master
Choose a base branch
from
Draft
Show file tree
Hide file tree
Changes from 2 commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions src/Unitful.jl
Original file line number Diff line number Diff line change
Expand Up @@ -69,5 +69,6 @@ include("logarithm.jl")
include("complex.jl")
include("pkgdefaults.jl")
include("dates.jl")
include("linearalgebra.jl")

end
42 changes: 42 additions & 0 deletions src/linearalgebra.jl
Original file line number Diff line number Diff line change
@@ -0,0 +1,42 @@
using LinearAlgebra

# This function is re-defined during testing, to check we hit the fast path:
linearalgebra_count() = nothing

function LinearAlgebra.mul!(C::StridedVecOrMat{<:AbstractQuantity{T}},
A::StridedMatrix{<:AbstractQuantity{T}},
B::StridedVecOrMat{<:AbstractQuantity{T}},
alpha::Bool, beta::Bool) where {T<:Base.HWNumber}
Copy link
Contributor Author

@mcabbott mcabbott Apr 13, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The case α, β::Bool is what A * B produces. Allowing other pure numbers ought to be fine. Allowing these to have units I haven't looked into.

I have restricted to A, B, C having the same eltype, as I think this is what BLAS handles. Julia sometimes copies a matrix to promote e.g. Float64 * Float32, but I don't recall at what stage.

5-arg mul! doesn't exist on 1.0, but that's OK, this will just never be called, so you'll get the slow but correct fallback.

# This is exactly how A * B creates C = similar(B, T, ...)
eltype(C) == Base.promote_op(LinearAlgebra.matprod, eltype(A), eltype(B)) || error("bad eltypes")
C0 = ustrip(C)
A0 = ustrip(A)
B0 = ustrip(B)
mul!(C0, A0, B0)
linearalgebra_count()
return C
end

function LinearAlgebra.mul!(C::StridedVecOrMat{<:AbstractQuantity{T}},
A::LinearAlgebra.AdjOrTransAbsMat{<:AbstractQuantity{T}, <:StridedMatrix},
B::StridedVecOrMat{<:AbstractQuantity{T}},
alpha::Bool, beta::Bool) where {T<:Base.HWNumber}

eltype(C) == Base.promote_op(LinearAlgebra.matprod, eltype(A), eltype(B)) || error("bad eltypes")
C0 = ustrip(C)
A0 = A isa Adjoint ? adjoint(ustrip(parent(A))) : transpose(ustrip(parent(A)))
B0 = ustrip(B)
mul!(C0, A0, B0)
linearalgebra_count()
return C
end

function LinearAlgebra.dot(A::StridedArray{<:AbstractQuantity{T}},
B::StridedArray{<:AbstractQuantity{T}}) where {T<:Base.HWNumber}
A0 = ustrip(A)
B0 = ustrip(B)
C0 = dot(A0, B0)
linearalgebra_count()
C = C0 * oneunit(eltype(A)) * oneunit(eltype(B)) # surely there is an official way
return C
end
2 changes: 1 addition & 1 deletion src/utils.jl
Original file line number Diff line number Diff line change
Expand Up @@ -75,7 +75,7 @@ julia> a[1] = 3u"m"; b
2
```
"""
@inline ustrip(A::Array{Q}) where {Q <: Quantity} = reinterpret(numtype(Q), A)
@inline ustrip(A::StridedArray{Q}) where {Q <: Quantity} = reinterpret(numtype(Q), A)

@deprecate(ustrip(A::AbstractArray{T}) where {T<:Number}, ustrip.(A))

Expand Down
44 changes: 44 additions & 0 deletions test/runtests.jl
Original file line number Diff line number Diff line change
Expand Up @@ -41,6 +41,16 @@ using Dates:

const colon = Base.:(:)

ambig = sort(detect_ambiguities(Unitful), by = a -> [string(a[1].name), string(a[2].module)])
if length(ambig) > 0
println(stdout, "detect_ambiguities(Unitful) found $(length(ambig)) issues:")
Copy link
Contributor Author

@mcabbott mcabbott Apr 13, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I was concerned about creating ambiguities, so made it print them before starting.

In the end, overloading mul! not * seems not to create problems. But there are (for me) 30 other ambiguities this detects (and prints out).

Now moved to #439 instead.

for i in 1:length(ambig)
println(stdout, "[",i, "]:")
println(stdout, " ", ambig[i][1])
println(stdout, " ", ambig[i][2])
end
println(stdout)
end
@testset "Construction" begin
@test isa(NoUnits, FreeUnits)
@test typeof(𝐋) === Unitful.Dimensions{(Unitful.Dimension{:Length}(1),)}
Expand Down Expand Up @@ -84,6 +94,40 @@ const colon = Base.:(:)
@test ConstructionBase.constructorof(typeof(1.0m))(2) === 2m
end

@testset "LinearAlgebra functions" begin
CNT = Ref(0)
Unitful.linearalgebra_count() = (CNT[] += 1; nothing)
@testset "> Matrix multiplication: *" begin
M = rand(3,3) .* u"m"
M_ = view(M,:,1:3)
v = rand(3) .* u"V"
v_ = view(v, 1:3)

CNT[] = 0

@test unit(first(M * M)) == u"m*m"
@test M * M == M_ * M == M * M_ == M_ * M_

@test unit(first(M * v)) == u"m*V"
@test M * v == M_ * v == M * v_ == M_ * v_

@test CNT[] == 10

@test unit(first(v' * M)) == u"m*V"
@test v' * M == v_' * M == v_' * M == v_' * M_

@test CNT[] == 15

@test unit(v' * v) == u"V*V"
@test v' * v == v_' * v == v_' * v == v_' * v_

@test CNT[] == 20
end
@testset "> Matrix multiplication: mul!" begin

end
end

@testset "Types" begin
@test Base.complex(Quantity{Float64,NoDims,NoUnits}) ==
Quantity{Complex{Float64},NoDims,NoUnits}
Expand Down