issue work in nworkers()-sized batches

src/dualtvdd.jl

@@ -143,7 +143,8 @@ function step!(ctx::DualTVDDState)
     cids = alg.parallel ? [eachindex(ctx.subax)] :
         chessboard_coloring(size(ctx.subax))
     for (color, ids) in enumerate(cids)
-        for i in ids
+        for pids in Iterators.partition(ids, length(alg.workers))
+            for i in pids
                 sax = ctx.subax[i]
 
                 # update remaining old contribution
@@ -166,7 +167,7 @@ function step!(ctx::DualTVDDState)
                 # start computation
                 put!(ctx.cons[i], sg)
             end
-        for i in ids
+            for i in pids
                 sax = ctx.subax[i]
 
                 sp = take!(ctx.cons[i])
@@ -176,6 +177,7 @@ function step!(ctx::DualTVDDState)
                     (1 .- σ) .* theta.(Ref(ax), Ref(sax), Ref(overlap), CartesianIndices(sax)) .* view(ctx.p, sax...) .+ σ .* sp
             end
         end
+    end
 
     ctx.p .= p_don