Update multidimensional.jl

Use `_to_subscript_indices` and `_to_linear_index` rather than `divrem` to pick the outer view range

Author: N5N3

base/multidimensional.jl

@@ -612,23 +612,24 @@ module IteratorsMD
         return I, (I, n+1)
     end
 
+    @inline function _splitlinear(piter::CartesianIndices, oiter::CartesianIndices, i::Int)
+        ci = Base._to_subscript_indices(piter, i)
+        ci[1], Base._to_linear_index(oiter, tail(ci)...)
+    end
     @inline function simd_outer_range(iter::CartesianPartition)
         # In general, the CartesianPartition might start and stop in the middle of the outer
         # dimensions, thus the outer range itself is a CartesianPartition.
         piter = iter.parent.parent
-        ci = CartesianIndices(tail(piter.indices))
-        size1 = size(piter, 1)
-        vindʷ = only(iter.indices) .- firstindex(piter) .+ 1
-        d₁, r₁ = divrem(first(vindʷ), size1, RoundUp)
-        d₂, r₂ = divrem(last(vindʷ) , size1, RoundUp)
-        vindᵒ = (d₁:d₂) .+ firstindex(ci) .- 1
-        outer = @inbounds view(ci, vindᵒ)
+        ax1, oiter = split(piter, Val(1))
+        vindʷ = only(iter.indices)
+        fl, vl = _splitlinear(piter, oiter, first(vindʷ))
+        fr, vr = _splitlinear(piter, oiter, last(vindʷ))
+        outer = @inbounds view(oiter, vl:vr)
         # Use Generator to make inner loop branchless
         @inline function genouter(i::Int, I::CartesianIndex)
-            skip = i == 1 ? r₁ + size1 - 1 : 0
-            #TODO == d₂ - d₁ + 1 and == length(outer) is slower for 4d case
-            len = i == length(vindᵒ) ? r₂ + size1 : size1
-            skip, len - skip, I
+            l = i == 1 ? fl : firstindex(ax1)
+            r = i == length(outer) ? fr : lastindex(ax1)
+            l - firstindex(ax1), r - l + 1, I
         end
         (genouter(i, I) for (i, I) in Iterators.enumerate(outer))
     end