Fix sort overwriting values in target array (#823)

src/sorting.jl

@@ -225,7 +225,11 @@ elements spaced by `stride`. Good for sampling pivot values as well as short sor
             end
             sync_threads()
             if 1 <= buddy <= L && threadIdx().x <= L
-                if (threadIdx().x < buddy) != flex_lt(swap[threadIdx().x], buddy_val, false, lt, by)
+                is_left = threadIdx().x < buddy
+                # flex_lt needs to handle equivalence in opposite ways for the
+                # two threads in each swap pair. Otherwise, if there are two
+                # different values with the same by, one will overwrite the other
+                if is_left != flex_lt(swap[threadIdx().x], buddy_val, is_left, lt, by)
                     swap[threadIdx().x] = buddy_val
                 end
             end

test/sorting.jl

@@ -252,7 +252,15 @@ end
 
     # using a `by` argument
     @test check_sort(Float32, 100000; by=x->abs(x - 0.5))
-    @test check_sort(Float64, (4, 100000); by=x->8*x-round(8*x), dims=2)
+    @test check_sort!(Float32, (100000, 4); by=x->abs(x - 0.5), dims=1)
+    @test check_sort!(Float32, (4, 100000); by=x->abs(x - 0.5), dims=2)
+    @test check_sort!(Float64, 400000; by=x->8*x-round(8*x))
+    @test check_sort!(Float64, (100000, 4); by=x->8*x-round(8*x), dims=1)
+    @test check_sort!(Float64, (4, 100000); by=x->8*x-round(8*x), dims=2)
+    # target bubble sort by using sub-blocksize input:
+    @test check_sort!(Int, 200; by=x->x % 2)
+    @test check_sort!(Int, 200; by=x->x % 3)
+    @test check_sort!(Int, 200; by=x->x % 4)
 end
 
 end