Skip to content

Add some more functions and improve broadcasting support #69

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.

Sign up for GitHub

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

Merged
merged 4 commits into from
Jun 8, 2024
Merged

Add some more functions and improve broadcasting support #69

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
05f57fa
Add some more functions and improve broadcasting support
christiangnrd Mar 13, 2024
145a4da
Improve code coverage
christiangnrd Mar 13, 2024
decd0ce
More coverage
christiangnrd Mar 13, 2024
f5c2d71
More coverage
christiangnrd Mar 13, 2024
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
2 changes: 1 addition & 1 deletion src/AppleAccelerate.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -92,9 +92,9 @@ function __init__()
end

if Sys.isapple()
include("Util.jl")
include("Array.jl")
include("DSP.jl")
include("Util.jl")
end

end # module
136 changes: 135 additions & 1 deletion src/Array.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -116,6 +116,9 @@ for (T, suff) in ((Float64, ""), (Float32, "f"))
Base.copy(bc::Base.Broadcast.Broadcasted{Style, Axes, typeof($f), Tuple{Array{$T, N}}}) where {Style, Axes, N} = ($f)(bc.args...)
Base.copyto!(dest::Array{$T, N}, bc::Base.Broadcast.Broadcasted{Style, Axes, typeof($f), Tuple{Array{$T, N}}}) where {Style, Axes, N} = ($f!)(dest, bc.args...)
end
if T == Float32
@eval Base.broadcasted(::typeof($f), arg::Union{Array{F,N},Base.Broadcast.Broadcasted}) where {N,F<:Union{Float32,Float64}} = ($f)(maybecopy(arg))
end
end
for (f, fa) in (twoarg_funcs...,(:pow,:pow))
f! = Symbol("$(f)!")
Expand All @@ -124,14 +127,19 @@ for (T, suff) in ((Float64, ""), (Float32, "f"))
Base.copy(bc::Base.Broadcast.Broadcasted{Style, Axes, typeof($f), Tuple{Array{$T, N},Array{$T,N}}}) where {Style, Axes, N} = ($f)(bc.args...)
Base.copyto!(dest::Array{$T, N}, bc::Base.Broadcast.Broadcasted{Style, Axes, typeof($f), Tuple{Array{$T,N},Array{$T,N}}}) where {Style, Axes, N} = ($f!)(dest, bc.args...)
end
if T == Float32
@eval Base.broadcasted(::typeof($f), arg1::Union{Array{F, N},Base.Broadcast.Broadcasted}, arg2::Union{Array{F, N},Base.Broadcast.Broadcasted}) where {N,F<:Union{Float32,Float64}} = ($f)(maybecopy(arg1), maybecopy(arg2))
end
end
end

# Functions over single vectors that return scalars/tuples
for (T, suff) in ((Float32, ""), (Float64, "D"))

for (f, fa) in ((:maximum, :maxv), (:minimum, :minv), (:mean, :meanv),
(:meansqr, :measqv), (:meanmag, :meamgv), (:sum, :sve))
(:meanmag, :meamgv), (:meansqr, :measqv), (:meanssqr, :mvessq),
(:sum, :sve), (:summag, :svemg), (:sumsqr, :svesq),
(:sumssqr, :svs))
@eval begin
function ($f)(X::Vector{$T})
val = Ref{$T}(0.0)
Expand Down Expand Up @@ -192,7 +200,133 @@ for (T, suff) in ((Float32, ""), (Float64, "D"))
return result
end
end

@eval begin
# Broadcasting override such that f.(X) turns into f(X)
Base.copy(bc::Base.Broadcast.Broadcasted{Style, Axes, typeof($f), Tuple{Array{$T, N},Array{$T,N}}}) where {Style, Axes, N} = ($f)(bc.args...)
Base.copyto!(dest::Array{$T, N}, bc::Base.Broadcast.Broadcasted{Style, Axes, typeof($f), Tuple{Array{$T,N},Array{$T,N}}}) where {Style, Axes, N} = ($f!)(dest, bc.args...)
Base.broadcasted(::typeof($f), arg1::Union{Array{$T, N},Base.Broadcast.Broadcasted}, arg2::Union{Array{$T, N},Base.Broadcast.Broadcasted}) where {N} = ($f)(maybecopy(arg1), maybecopy(arg2))
end
end
end

# Element-wise operations over a vector and a scalar
for (T, suff) in ((Float32, ""), (Float64, "D"))

for (f, name) in ((:vsadd, "addition"), (:vsdiv, "division"), (:vsmul, "multiplication"))
f! = Symbol("$(f)!")

@eval begin
@doc """
`$($f!)(result::Vector{$($T)}, X::Vector{$($T)}, c::$($T))`

Implements vector-scalar **$($name)** over **Vector{$($T)}** and $($T) and overwrites
the result vector with computed value. *Returns:* **Vector{$($T)}** `result`
""" ->
function ($f!)(result::Vector{$T}, X::Vector{$T}, c::$T)
ccall(($(string("vDSP_", f, suff), libacc)), Cvoid,
(Ptr{$T}, Int64, Ptr{$T}, Ptr{$T}, Int64, UInt64),
X, 1, Ref(c), result, 1, length(result))
return result
end
end

@eval begin
@doc """
`$($f)(X::Vector{$($T)}, c::$($T))`

Implements vector-scalar **$($name)** over **Vector{$($T)}** and $($T). Allocates
memory to store result. *Returns:* **Vector{$($T)}**
""" ->
function ($f)(X::Vector{$T}, c::$T)
result = similar(X)
($f!)(result, X, c)
return result
end
end
end
f = :vssub
f! = Symbol("$(f)!")

@eval begin
@doc """
`$($f!)(result::Vector{$($T)}, X::Vector{$($T)}, c::$($T))`

Implements vector-scalar **subtraction** over **Vector{$($T)}** and $($T) and overwrites
the result vector with computed value. *Returns:* **Vector{$($T)}** `result`
""" ->
function ($f!)(result::Vector{$T}, X::Vector{$T}, c::$T)
ccall(($(string("vDSP_vsadd", suff), libacc)), Cvoid,
(Ptr{$T}, Int64, Ptr{$T}, Ptr{$T}, Int64, UInt64),
X, 1, Ref(-c), result, 1, length(result))
return result
end
end

@eval begin
@doc """
`$($f)(X::Vector{$($T)}, c::$($T))`

Implements vector-scalar **subtraction** over **Vector{$($T)}** and $($T). Allocates
memory to store result. *Returns:* **Vector{$($T)}**
""" ->
function ($f)(X::Vector{$T}, c::$T)
result = similar(X)
($f!)(result, X, c)
return result
end
end

f = :svsub
f! = Symbol("$(f)!")

@eval begin
@doc """
`$($f!)(result::Vector{$($T)}, X::Vector{$($T)}, c::$($T))`

Implements vector-scalar **subtraction** over $($T) and **Vector{$($T)}** and overwrites
the result vector with computed value. *Returns:* **Vector{$($T)}** `result`
""" ->
function ($f!)(result::Vector{$T}, X::Vector{$T}, c::$T)
ccall(($(string("vDSP_vsadd", suff), libacc)), Cvoid,
(Ptr{$T}, Int64, Ptr{$T}, Ptr{$T}, Int64, UInt64),
-X, 1, Ref(c), result, 1, length(result))
return result
end
end

@eval begin
@doc """
`$($f)(X::Vector{$($T), c::$($T)})`

Implements vector-scalar **subtraction** over $($T) and **Vector{$($T)}**. Allocates
memory to store result. *Returns:* **Vector{$($T)}**
""" ->
function ($f)(X::Vector{$T}, c::$T)
result = similar(X)
($f!)(result, X, c)
return result
end
end

for f in (:vsadd, :vssub, :vsdiv, :vsmul)
f! = Symbol("$(f)!")
@eval begin
# Broadcasting override such that f.(X) turns into f(X)
Base.copy(bc::Base.Broadcast.Broadcasted{Style, Axes, typeof($f), Tuple{Array{$T, N},$T}}) where {Style, Axes, N} = ($f)(bc.args...)
Base.copyto!(dest::Array{$T, N}, bc::Base.Broadcast.Broadcasted{Style, Axes, typeof($f), Tuple{Array{$T,N},$T}}) where {Style, Axes, N} = ($f!)(dest, bc.args...)
Base.broadcasted(::typeof($f), arg1::Union{Array{$T, N},Base.Broadcast.Broadcasted}, arg2::$T) where {N} = ($f)(maybecopy(arg1), arg2)
end
end

f = :svsub
f! = Symbol("$(f)!")

@eval begin
# Broadcasting override such that f.(X) turns into f(X)
Base.copy(bc::Base.Broadcast.Broadcasted{Style, Axes, typeof($f), Tuple{Array{$T, N}, $T}}) where {Style, Axes, N} = ($f)(bc.args...)
Base.copyto!(dest::Array{$T, N}, bc::Base.Broadcast.Broadcasted{Style, Axes, typeof($f), Tuple{Array{$T,N}, $T}}) where {Style, Axes, N} = ($f!)(dest, bc.args...)
Base.broadcasted(::typeof($f), arg1::Union{Array{$T, N},Base.Broadcast.Broadcasted}, arg2::$T) where {N} = ($f)(maybecopy(arg1), arg2)
end
end

22 changes: 21 additions & 1 deletion src/Util.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,6 +2,13 @@

tupletypelength(a)=length(a.parameters)

@inline maybecopy(x::T) where {T <: Base.Broadcast.Broadcasted} = copy(x)
@inline maybecopy(x::T) where {T <: Array} = x

const OPS = Dict{Symbol,Tuple{Symbol, Symbol, Symbol}}(:+ => (:vadd, :vsadd, :vsadd),
:- => (:vsub, :vssub, :svsub),
:* => (:vadd, :vsmul, :vsmul),
:/ => (:vadd, :vsdiv, :vsdiv),)

macro replaceBase(fs...)
b = Expr(:block)
Expand All @@ -28,16 +35,29 @@ macro replaceBase(fs...)
e = quote
(Base.$f)(X::Array{T}) where {T <: Union{Float64,Float32}} = ($fa)(X)
(Base.$f)(X::Union{Float64,Float32}) = ($fa)([X])[1]
Base.broadcasted(::typeof(Base.$f), arg::Union{Array{F,N},Base.Broadcast.Broadcasted}) where {N,F<:Union{Float32,Float64}} = ($fa)(maybecopy(arg))
end
arg_consumed = true
end
if fa in (:copysign,:atan,:pow,:rem,:div_float, :vadd, :vsub, :vmul)
if fa in (:copysign,:atan,:pow,:rem)
e = quote
(Base.$f)(X::Array{T},Y::Array{T}) where {T <: Union{Float32,Float64}} = ($fa)(X,Y)
(Base.$f)(X::T,Y::T) where {T <: Union{Float32,Float64}} = ($fa)([X],[Y])[1]
Base.broadcasted(::typeof(Base.$f), arg1::Union{Array{F, N},Base.Broadcast.Broadcasted}, arg2::Union{Array{F, N},Base.Broadcast.Broadcasted}) where {N,F<:Union{Float32,Float64}} = ($fa)(maybecopy(arg1), maybecopy(arg2))
end
arg_consumed = true
end
if f in (:+,:-,:*,:/)
e = quote
(Base.$f)(X::Array{T},Y::Array{T}) where {T <: Union{Float32,Float64}} = ($(OPS[f][1]))(X,Y)
(Base.$f)(X::T,Y::T) where {T <: Union{Float32,Float64}} = ($(OPS[f][1]))([X],[Y])[1]
Base.broadcasted(::typeof(Base.$f), arg1::Union{Array{F, N},Base.Broadcast.Broadcasted}, arg2::Union{Array{F, N},Base.Broadcast.Broadcasted}) where {N,F<:Union{Float32,Float64}} = ($(OPS[f][1]))(maybecopy(arg1), maybecopy(arg2))

Base.broadcasted(::typeof(Base.$f), arg1::Union{Array{T, N},Base.Broadcast.Broadcasted}, arg2::T) where {N, T <: Union{Float32,Float64}} = ($(OPS[f][2]))(maybecopy(arg1), arg2)
Base.broadcasted(::typeof(Base.$f), arg1::T, arg2::Union{Array{T, N},Base.Broadcast.Broadcasted}) where {N, T <: Union{Float32,Float64}} = ($(OPS[f][3]))(maybecopy(arg2), arg1)
end
arg_consumed = true
end
if !arg_consumed
error("Function $f not defined by AppleAccelerate.jl")
end
Expand Down
73 changes: 69 additions & 4 deletions test/runtests.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,14 +10,62 @@ end
Random.seed!(7)
N = 1_000

@testset "AppleAccelerate.jl" begin
for T in (Float32, Float64)
@testset "Element-wise Operators::$T" begin
X::Vector{T} = randn(N)
Y::Vector{T} = randn(N)
Z::Vector{T} = similar(X)
# Vector-vector
@test (X .+ Y) ≈ AppleAccelerate.vadd(X, Y)
@test (X .- Y) ≈ AppleAccelerate.vsub(X, Y)
@test (X .* Y) ≈ AppleAccelerate.vmul(X, Y)
@test (X ./ Y) ≈ AppleAccelerate.vdiv(X, Y)

# Vector-vector non-allocating
AppleAccelerate.vadd!(Z, X, Y)
@test (X .+ Y) ≈ Z
AppleAccelerate.vsub!(Z, X, Y)
@test (X .- Y) ≈ Z
AppleAccelerate.vmul!(Z, X, Y)
@test (X .* Y) ≈ Z
AppleAccelerate.vdiv!(Z, X, Y)
@test (X ./ Y) ≈ Z

# Vector-vector broadcasting
@test (X .+ Y) ≈ AppleAccelerate.vadd.(X, Y)
@test (X .- Y) ≈ AppleAccelerate.vsub.(X, Y)
@test (X .* Y) ≈ AppleAccelerate.vmul.(X, Y)
@test (X ./ Y) ≈ AppleAccelerate.vdiv.(X, Y)

#Vector-scalar
c::T = randn()
@test (X .+ c) ≈ AppleAccelerate.vsadd.(X, c)
@test (X .- c) ≈ AppleAccelerate.vssub.(X, c)
@test (c .- X) ≈ AppleAccelerate.svsub.(X, c)
@test (X .* c) ≈ AppleAccelerate.vsmul.(X, c)
@test (X ./ c) ≈ AppleAccelerate.vsdiv.(X, c)

#Vector-scalar non-allocating
AppleAccelerate.vsadd!(Y, X, c)
@test (X .+ c) ≈ Y
AppleAccelerate.vssub!(Y, X, c)
@test (X .- c) ≈ Y
AppleAccelerate.svsub!(Y, X, c)
@test (c .- X) ≈ Y
AppleAccelerate.vsmul!(Y, X, c)
@test (X .* c) ≈ Y
AppleAccelerate.vsdiv!(Y, X, c)
@test (X ./ c) ≈ Y

#Vector-scalar broadcasting
@test (X .+ c) ≈ AppleAccelerate.vsadd.(X, c)
@test (X .- c) ≈ AppleAccelerate.vssub.(X, c)
@test (c .- X) ≈ AppleAccelerate.svsub.(X, c)
@test (X .* c) ≈ AppleAccelerate.vsmul.(X, c)
@test (X ./ c) ≈ AppleAccelerate.vsdiv.(X, c)

@test (X .+ Y .+ Y) ≈ AppleAccelerate.vadd.(X, Y .+ Y)
end
end

Expand Down Expand Up @@ -198,12 +246,28 @@ for T in (Float32, Float64)
@test fa(X)[2] ≈ fb(X)[2]
end

@testset "Testing meanmag::$T" begin
@test AppleAccelerate.meanmag(X) ≈ mean(abs, X)
end

@testset "Testing meansqr::$T" begin
@test AppleAccelerate.meansqr(X) ≈ mean(X .*X)
@test AppleAccelerate.meansqr(X) ≈ mean(X .* X)
end

@testset "Testing meanmag::$T" begin
@test AppleAccelerate.meanmag(X) ≈ mean(abs.(X))
@testset "Testing meanssqr::$T" begin
@test AppleAccelerate.meanssqr(X) ≈ mean(X .* abs.(X))
end

@testset "Testing summag::$T" begin
@test AppleAccelerate.summag(X) ≈ sum(abs, X)
end

@testset "Testing sumsqr::$T" begin
@test AppleAccelerate.sumsqr(X) ≈ sum(abs2, X)
end

@testset "Testing sumssqr::$T" begin
@test AppleAccelerate.sumssqr(X) ≈ sum(X .* abs.(X))
end

end
Expand Down Expand Up @@ -270,6 +334,7 @@ Y::Array{T} = abs.(randn(N))
@test X ./ Y == AppleAccelerate.div_float(X, Y)
end
=#
end

if AppleAccelerate.get_macos_version() < v"13.3"
@info("AppleAccelerate.jl needs macOS >= 13.3 for BLAS forwarding. Not testing forwarding capabilities.")
Expand Down Expand Up @@ -338,6 +403,6 @@ end
end

run(`$(Base.julia_cmd()) --project=$(Base.active_project()) $(dir)/runtests.jl LinearAlgebra/blas LinearAlgebra/lapack`)
end;
end;
end