Remove undesired promotions with Float64 literals

Author: devmotion

src/deviation.jl

@@ -46,12 +46,14 @@ end
 Compute the squared L2 distance between two arrays: ``\\sum_{i=1}^n |a_i - b_i|^2``.
 Efficient equivalent of `sum(abs2, a - b)`.
 """
-function sqL2dist(a::AbstractArray{T}, b::AbstractArray{T}) where T<:Number
+function sqL2dist(a::AbstractArray{<:Number}, b::AbstractArray{<:Number})
     n = length(a)
     length(b) == n || throw(DimensionMismatch("Input dimension mismatch"))
-    r = 0.0
-    for i in eachindex(a, b)
-        r += abs2(a[i] - b[i])
+    if iszero(n)
+        r = zero(abs2(zero(eltype(a)) - zero(eltype(b))))
+    else
+        broadcasted = Broadcast.broadcasted((ai, bi) -> abs2(ai - bi), vec(a), vec(b))
+        r = sum(Broadcast.instantiate(broadcasted))
     end
     return r
 end
@@ -64,7 +66,7 @@ end
 Compute the L2 distance between two arrays: ``\\sqrt{\\sum_{i=1}^n |a_i - b_i|^2}``.
 Efficient equivalent of `sqrt(sum(abs2, a - b))`.
 """
-L2dist(a::AbstractArray{T}, b::AbstractArray{T}) where {T<:Number} = sqrt(sqL2dist(a, b))
+L2dist(a::AbstractArray{<:Number}, b::AbstractArray{<:Number}) = sqrt(sqL2dist(a, b))
 
 
 # L1 distance
@@ -74,12 +76,14 @@ L2dist(a::AbstractArray{T}, b::AbstractArray{T}) where {T<:Number} = sqrt(sqL2di
 Compute the L1 distance between two arrays: ``\\sum_{i=1}^n |a_i - b_i|``.
 Efficient equivalent of `sum(abs, a - b)`.
 """
-function L1dist(a::AbstractArray{T}, b::AbstractArray{T}) where T<:Number
+function L1dist(a::AbstractArray{<:Number}, b::AbstractArray{<:Number})
     n = length(a)
     length(b) == n || throw(DimensionMismatch("Input dimension mismatch"))
-    r = 0.0
-    for i in eachindex(a, b)
-        r += abs(a[i] - b[i])
+    if iszero(n)
+        r = zero(abs(zero(eltype(a)) - zero(eltype(b))))
+    else
+        broadcasted = Broadcast.broadcasted((ai, bi) -> abs(ai - bi), vec(a), vec(b))
+        r = sum(Broadcast.instantiate(broadcasted))
     end
     return r
 end
@@ -93,15 +97,14 @@ Compute the L∞ distance, also called the Chebyshev distance, between
 two arrays: ``\\max_{1≤i≤n} |a_i - b_i|``.
 Efficient equivalent of `maxabs(a - b)`.
 """
-function Linfdist(a::AbstractArray{T}, b::AbstractArray{T}) where T<:Number
+function Linfdist(a::AbstractArray{<:Number}, b::AbstractArray{<:Number})
     n = length(a)
     length(b) == n || throw(DimensionMismatch("Input dimension mismatch"))
-    r = 0.0
-    for i in eachindex(a, b)
-        v = abs(a[i] - b[i])
-        if r < v
-            r = v
-        end
+    if iszero(n)
+        r = zero(abs(zero(eltype(a)) - zero(eltype(b))))
+    else
+        broadcasted = Broadcast.broadcasted((ai, bi) -> abs(ai - bi), vec(a), vec(b))
+        r = maximum(Broadcast.instantiate(broadcasted))
     end
     return r
 end
@@ -115,19 +118,20 @@ Compute the generalized Kullback-Leibler divergence between two arrays:
 ``\\sum_{i=1}^n (a_i \\log(a_i/b_i) - a_i + b_i)``.
 Efficient equivalent of `sum(a*log(a/b)-a+b)`.
 """
-function gkldiv(a::AbstractArray{T}, b::AbstractArray{T}) where T<:AbstractFloat
+function gkldiv(a::AbstractArray{<:Real}, b::AbstractArray{<:Real})
     n = length(a)
-    r = 0.0
-    for i in eachindex(a, b)
-        ai = a[i]
-        bi = b[i]
-        if ai > 0
-            r += (ai * log(ai / bi) - ai + bi)
-        else
-            r += bi
+    length(b) == n || throw(DimensionMismatch("Input dimension mismatch"))
+    if iszero(n)
+        za = zero(eltype(a))
+        zb = zero(eltype(b))
+        r = zero(xlogy(za, za / zb) + (zb - za))
+    else
+        broadcasted = Broadcast.broadcasted(vec(a), vec(b)) do ai, bi
+            return xlogy(ai, ai / bi) + (bi - ai)
         end
+        return sum(Broadcast.instantiate(broadcasted))
     end
-    return r::Float64
+    return r
 end
 
 
@@ -137,8 +141,7 @@ end
 
 Return the mean absolute deviation between two arrays: `mean(abs, a - b)`.
 """
-meanad(a::AbstractArray{T}, b::AbstractArray{T}) where {T<:Number} =
-    L1dist(a, b) / length(a)
+meanad(a::AbstractArray{<:Number}, b::AbstractArray{<:Number}) = L1dist(a, b) / length(a)
 
 
 # MaxAD: maximum absolute deviation
@@ -147,7 +150,7 @@ meanad(a::AbstractArray{T}, b::AbstractArray{T}) where {T<:Number} =
 
 Return the maximum absolute deviation between two arrays: `maxabs(a - b)`.
 """
-maxad(a::AbstractArray{T}, b::AbstractArray{T}) where {T<:Number} = Linfdist(a, b)
+maxad(a::AbstractArray{<:Number}, b::AbstractArray{<:Number}) = Linfdist(a, b)
 
 
 # MSD: mean squared deviation
@@ -156,8 +159,7 @@ maxad(a::AbstractArray{T}, b::AbstractArray{T}) where {T<:Number} = Linfdist(a,
 
 Return the mean squared deviation between two arrays: `mean(abs2, a - b)`.
 """
-msd(a::AbstractArray{T}, b::AbstractArray{T}) where {T<:Number} =
-    sqL2dist(a, b) / length(a)
+msd(a::AbstractArray{<:Number}, b::AbstractArray{<:Number}) = sqL2dist(a, b) / length(a)
 
 
 # RMSD: root mean squared deviation
@@ -168,13 +170,14 @@ Return the root mean squared deviation between two optionally
 normalized arrays. The root mean squared deviation is computed
 as `sqrt(msd(a, b))`.
 """
-function rmsd(a::AbstractArray{T}, b::AbstractArray{T}; normalize::Bool=false) where T<:Number
+function rmsd(a::AbstractArray{<:Number}, b::AbstractArray{<:Number}; normalize::Bool=false)
     v = sqrt(msd(a, b))
     if normalize
         amin, amax = extrema(a)
-        v /= (amax - amin)
+        return v / (amax - amin)
+    else
+        return v
     end
-    return v
 end
 
 
@@ -186,6 +189,6 @@ Compute the peak signal-to-noise ratio between two arrays `a` and `b`.
 `maxv` is the maximum possible value either array can take. The PSNR
 is computed as `10 * log10(maxv^2 / msd(a, b))`.
 """
-function psnr(a::AbstractArray{T}, b::AbstractArray{T}, maxv::Real) where T<:Real
-    20. * log10(maxv) - 10. * log10(msd(a, b))
+function psnr(a::AbstractArray{<:Real}, b::AbstractArray{<:Real}, maxv::Real)
+    20 * log10(maxv) - 10 * log10(msd(a, b))
 end