Add dot(x,A,y) (#683)

* add some dot(x,A,y) methods

* Bump version to 3.2.0

Co-authored-by: Fredrik Ekre <[email protected]>
Co-authored-by: Martin Holters <[email protected]>

Project.toml

@@ -1,6 +1,6 @@
 name = "Compat"
 uuid = "34da2185-b29b-5c13-b0c7-acf172513d20"
-version = "3.1.0"
+version = "3.2.0"
 
 [deps]
 Base64 = "2a0f44e3-6c83-55bd-87e4-b1978d98bd5f"

README.md

@@ -44,6 +44,8 @@ Please check the list below for the specific syntax you need.
 
 ## Supported features
 
+* `dot` now has a 3-argument method `dot(x, A, y)` without storing the intermediate result `A*y` ([#32739]). (since Compat 3.2.0)
+
 * `pkgdir(m)` returns the root directory of the package that imported module `m` ([#33128]). (since Compat 3.2.0)
 
 * `filter` can now act on a `Tuple` [#32968]. (since Compat 3.1.0)
@@ -104,6 +106,7 @@ Note that you should specify the correct minimum version for `Compat` in the
 [#29674]: https://github.com/JuliaLang/julia/issues/29674
 [#29749]: https://github.com/JuliaLang/julia/issues/29749
 [#32628]: https://github.com/JuliaLang/julia/issues/32628
+[#32739]: https://github.com/JuliaLang/julia/pull/32739
 [#33129]: https://github.com/JuliaLang/julia/issues/33129
 [#33568]: https://github.com/JuliaLang/julia/pull/33568
 [#33128]: https://github.com/JuliaLang/julia/pull/33128

src/Compat.jl

@@ -1,5 +1,8 @@
 module Compat
 
+import LinearAlgebra
+using LinearAlgebra: Adjoint, Diagonal, Transpose, UniformScaling, RealHermSymComplexHerm
+
 include("compatmacro.jl")
 
 # https://github.com/JuliaLang/julia/pull/29679
@@ -88,6 +91,87 @@ if VERSION < v"1.3.0-alpha.8"
     Base.mod(i::Integer, r::AbstractUnitRange{<:Integer}) = mod(i-first(r), length(r)) + first(r)
 end
 
+# https://github.com/JuliaLang/julia/pull/32739
+# This omits special methods for more exotic matrix types, Triangular and worse.
+if VERSION < v"1.4.0-DEV.92" # 2425ae760fb5151c5c7dd0554e87c5fc9e24de73
+
+    # stdlib/LinearAlgebra/src/generic.jl
+    LinearAlgebra.dot(x, A, y) = LinearAlgebra.dot(x, A*y) # generic fallback
+
+    function LinearAlgebra.dot(x::AbstractVector, A::AbstractMatrix, y::AbstractVector)
+        (axes(x)..., axes(y)...) == axes(A) || throw(DimensionMismatch())
+        T = typeof(LinearAlgebra.dot(first(x), first(A), first(y)))
+        s = zero(T)
+        i₁ = first(eachindex(x))
+        x₁ = first(x)
+        @inbounds for j in eachindex(y)
+            yj = y[j]
+            if !iszero(yj)
+                temp = zero(adjoint(A[i₁,j]) * x₁)
+                @simd for i in eachindex(x)
+                    temp += adjoint(A[i,j]) * x[i]
+                end
+                s += LinearAlgebra.dot(temp, yj)
+            end
+        end
+        return s
+    end
+    LinearAlgebra.dot(x::AbstractVector, adjA::Adjoint, y::AbstractVector) =
+         adjoint(LinearAlgebra.dot(y, adjA.parent, x))
+    LinearAlgebra.dot(x::AbstractVector, transA::Transpose{<:Real}, y::AbstractVector) =
+        adjoint(LinearAlgebra.dot(y, transA.parent, x))
+
+    # stdlib/LinearAlgebra/src/diagonal.jl
+    function LinearAlgebra.dot(x::AbstractVector, D::Diagonal, y::AbstractVector)
+        mapreduce(t -> LinearAlgebra.dot(t[1], t[2], t[3]), +, zip(x, D.diag, y))
+    end
+
+    # stdlib/LinearAlgebra/src/symmetric.jl
+    function LinearAlgebra.dot(x::AbstractVector, A::RealHermSymComplexHerm, y::AbstractVector)
+        require_one_based_indexing(x, y)
+        (length(x) == length(y) == size(A, 1)) || throw(DimensionMismatch())
+        data = A.data
+        r = zero(eltype(x)) * zero(eltype(A)) * zero(eltype(y))
+        if A.uplo == 'U'
+            @inbounds for j = 1:length(y)
+                r += LinearAlgebra.dot(x[j], real(data[j,j]), y[j])
+                @simd for i = 1:j-1
+                    Aij = data[i,j]
+                    r += LinearAlgebra.dot(x[i], Aij, y[j]) +
+                        LinearAlgebra.dot(x[j], adjoint(Aij), y[i])
+                end
+            end
+        else # A.uplo == 'L'
+            @inbounds for j = 1:length(y)
+                r += LinearAlgebra.dot(x[j], real(data[j,j]), y[j])
+                @simd for i = j+1:length(y)
+                    Aij = data[i,j]
+                    r += LinearAlgebra.dot(x[i], Aij, y[j]) +
+                        LinearAlgebra.dot(x[j], adjoint(Aij), y[i])
+                end
+            end
+        end
+        return r
+    end
+
+    # stdlib/LinearAlgebra/src/uniformscaling.jl
+    LinearAlgebra.dot(x::AbstractVector, J::UniformScaling, y::AbstractVector) =
+        LinearAlgebra.dot(x, J.λ, y)
+    LinearAlgebra.dot(x::AbstractVector, a::Number, y::AbstractVector) =
+        sum(t -> LinearAlgebra.dot(t[1], a, t[2]), zip(x, y))
+    LinearAlgebra.dot(x::AbstractVector, a::Union{Real,Complex}, y::AbstractVector) =
+        a*LinearAlgebra.dot(x, y)
+end
+
+# https://github.com/JuliaLang/julia/pull/30630
+if VERSION < v"1.2.0-DEV.125" # 1da48c2e4028c1514ed45688be727efbef1db884
+    require_one_based_indexing(A...) = !Base.has_offset_axes(A...) || throw(ArgumentError(
+        "offset arrays are not supported but got an array with index other than 1"))
+# At present this is only used in Compat inside the above dot(x,A,y) functions, #32739
+elseif VERSION < v"1.4.0-DEV.92"
+    using Base: require_one_based_indexing
+end
+
 # https://github.com/JuliaLang/julia/pull/33568
 if VERSION < v"1.4.0-DEV.329"
     Base.:∘(f, g, h...) = ∘(f ∘ g, h...)

test/runtests.jl

@@ -90,6 +90,101 @@ end
     @test_throws DivideError mod(3, 1:0)
 end
 
+using LinearAlgebra
+
+@testset "generalized dot #32739" begin
+    # stdlib/LinearAlgebra/test/generic.jl
+    for elty in (Int, Float32, Float64, BigFloat, Complex{Float32}, Complex{Float64}, Complex{BigFloat})
+        n = 10
+        if elty <: Int
+            A = rand(-n:n, n, n)
+            x = rand(-n:n, n)
+            y = rand(-n:n, n)
+        elseif elty <: Real
+            A = convert(Matrix{elty}, randn(n,n))
+            x = rand(elty, n)
+            y = rand(elty, n)
+        else
+            A = convert(Matrix{elty}, complex.(randn(n,n), randn(n,n)))
+            x = rand(elty, n)
+            y = rand(elty, n)
+        end
+        @test dot(x, A, y) ≈ dot(A'x, y) ≈ *(x', A, y) ≈ (x'A)*y
+        @test dot(x, A', y) ≈ dot(A*x, y) ≈ *(x', A', y) ≈ (x'A')*y
+        elty <: Real && @test dot(x, transpose(A), y) ≈ dot(x, transpose(A)*y) ≈ *(x', transpose(A), y) ≈ (x'*transpose(A))*y
+        B = reshape([A], 1, 1)
+        x = [x]
+        y = [y]
+        @test dot(x, B, y) ≈ dot(B'x, y)
+        @test dot(x, B', y) ≈ dot(B*x, y)
+        elty <: Real && @test dot(x, transpose(B), y) ≈ dot(x, transpose(B)*y)
+    end
+
+    # stdlib/LinearAlgebra/test/symmetric.jl
+    n = 10
+    areal = randn(n,n)/2
+    aimg  = randn(n,n)/2
+    @testset for eltya in (Float32, Float64, ComplexF32, ComplexF64, BigFloat, Int)
+        a = eltya == Int ? rand(1:7, n, n) : convert(Matrix{eltya}, eltya <: Complex ? complex.(areal, aimg) : areal)
+        asym = transpose(a) + a                 # symmetric indefinite
+        aherm = a' + a                 # Hermitian indefinite
+        apos  = a' * a                 # Hermitian positive definite
+        aposs = apos + transpose(apos)        # Symmetric positive definite
+        ε = εa = eps(abs(float(one(eltya))))
+        x = randn(n)
+        y = randn(n)
+        b = randn(n,n)/2
+        x = eltya == Int ? rand(1:7, n) : convert(Vector{eltya}, eltya <: Complex ? complex.(x, zeros(n)) : x)
+        y = eltya == Int ? rand(1:7, n) : convert(Vector{eltya}, eltya <: Complex ? complex.(y, zeros(n)) : y)
+        b = eltya == Int ? rand(1:7, n, n) : convert(Matrix{eltya}, eltya <: Complex ? complex.(b, zeros(n,n)) : b)
+
+        @testset "generalized dot product" begin
+            for uplo in (:U, :L)
+                @test dot(x, Hermitian(aherm, uplo), y) ≈ dot(x, Hermitian(aherm, uplo)*y) ≈ dot(x, Matrix(Hermitian(aherm, uplo)), y)
+                @test dot(x, Hermitian(aherm, uplo), x) ≈ dot(x, Hermitian(aherm, uplo)*x) ≈ dot(x, Matrix(Hermitian(aherm, uplo)), x)
+            end
+            if eltya <: Real
+                for uplo in (:U, :L)
+                    @test dot(x, Symmetric(aherm, uplo), y) ≈ dot(x, Symmetric(aherm, uplo)*y) ≈ dot(x, Matrix(Symmetric(aherm, uplo)), y)
+                    @test dot(x, Symmetric(aherm, uplo), x) ≈ dot(x, Symmetric(aherm, uplo)*x) ≈ dot(x, Matrix(Symmetric(aherm, uplo)), x)
+                end
+            end
+        end
+    end
+
+    # stdlib/LinearAlgebra/test/uniformscaling.jl
+    @testset "generalized dot" begin
+        x = rand(-10:10, 3)
+        y = rand(-10:10, 3)
+        λ = rand(-10:10)
+        J = UniformScaling(λ)
+        @test dot(x, J, y) == λ*dot(x, y)
+    end
+
+    # stdlib/LinearAlgebra/test/bidiag.jl
+    # The special method for this is not in Compat #683, so this tests the generic fallback
+    @testset "generalized dot" begin
+        for elty in (Float64, ComplexF64)
+            dv = randn(elty, 5)
+            ev = randn(elty, 4)
+            x = randn(elty, 5)
+            y = randn(elty, 5)
+            for uplo in (:U, :L)
+                B = Bidiagonal(dv, ev, uplo)
+                @test dot(x, B, y) ≈ dot(B'x, y) ≈ dot(x, Matrix(B), y)
+            end
+        end
+    end
+
+    # Diagonal -- no such test in Base.
+    @testset "diagonal" begin
+        x = rand(-10:10, 3) .+ im
+        y = rand(-10:10, 3) .+ im
+        d = Diagonal(rand(-10:10, 3) .+ im)
+        @test dot(x,d,y) == dot(x,collect(d),y) == dot(x, d*y)
+    end
+end
+
 # https://github.com/JuliaLang/julia/pull/33568
 @testset "function composition" begin
     @test ∘(x -> x-2, x -> x-3, x -> x+5)(7) == 7