Do some microptimization

Author: roflmaostc

Project.toml

@@ -9,10 +9,10 @@ FFTW = "7a1cc6ca-52ef-59f5-83cd-3a7055c09341"
 Tullio = "bc48ee85-29a4-5162-ae0b-a64e1601d4bc"
 
 [compat]
-julia = "1"
 ColorTypes = "0.11"
 FFTW = "1.4"
 Tullio = "0.3"
+julia = "1"
 
 [extras]
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"

examples/benchmarks/bin.jl

@@ -1,17 +1,34 @@
 using BenchmarkTools, MicroscopyTools
 
+function f(x)
+    @info "Random data with size $(size(x))"
+    print("Tullio: ")
+    @btime bin2($x)
+    print("CartesianIndices: ")
+    @btime bin($x, (2,2))
+end
+
 function main()
+    @info "Binning by (2,2)"
+    x = randn(Float32, (256, 512))
+    f(x)
+
+    x = randn(Float32, (8192, 7113))
+    f(x)
+
+    x = randn((97, 32))
+    f(x)
 
     x = randn((3097, 113, 32))
-    @warn "random data with size $(size(x))"
-    @btime bin2($x)
-    @btime bin($x, (2,2))
+    f(x)
+    
+    x = randn((23, 31, 8))
+    f(x)
 
+    x = randn(Float32, (3097, 4096, 32))
+    f(x)
 
-    @warn "random data with size $(size(x))"
-    x = randn((3097, 1113))
-    @btime bin2($x)
-    @btime bin($x, (2,2))
+    return 
 end
 
 main()

src/binning.jl

@@ -50,22 +50,36 @@ function bin(arr::AbstractArray{T, N}, binning::NTuple{M, Int}) where {T, N, M}
     # largest size which is possible according to binning
     out = similar(arr, size(arr) .÷ binning)
 
+    out_indices = CartesianIndices(out) 
+    z = zero(T)
+    neighbours = CartesianIndices(binning)
     # iterate over the indices of the out put array
-    for inds in CartesianIndices(out)
+    #@inbounds and @simd make it faster
+    @inbounds @simd for inds in out_indices
         # now we start to sum within binning
-        sum = zero(T)
+        s = z
         # sum all neighbours
-        for neighbours in CartesianIndices(binning)
-            pos = binning .* Tuple(inds) .- Tuple(neighbours) .+ 1
-            sum += arr[pos...]
+        base_inds = binning .* Tuple(inds) .+ 1
+        for neighbour in neighbours 
+            pos = base_inds .- Tuple(neighbour) 
+            s += arr[pos...]
         end
         # assign the sum
-        out[inds] = sum
+        out[inds] = s
     end
     return out
 end
 
 
+# was slightly slower with function barrier
+# Function barrier.
+# """
+# function _bin_loop!(out::AbstractArray{T, N}, 
+#                     arr::AbstractArray{T, N}, 
+#                     binning::NTuple{N, Int}) where {T, N}
+## end
+
+
 """
     bin2(arr)