Unnecessary GC root preceding arrayset

[yuyichao]

This is the problem I mentioned in #15402 (comment) which I realized later is actually a different issue. On the commit right before #15717 (comment), the test case is,

julia> function f(a, b)
           @inbounds a[] = b
           nothing
       end

With IR

julia> @code_llvm f(Any[], 1)

define void @julia_f_23824(%jl_value_t*, i64) #0 {
top:
  %2 = alloca [3 x %jl_value_t*], align 8
  %.sub = getelementptr inbounds [3 x %jl_value_t*], [3 x %jl_value_t*]* %2, i64 0, i64 0
  %3 = getelementptr [3 x %jl_value_t*], [3 x %jl_value_t*]* %2, i64 0, i64 2
  store %jl_value_t* null, %jl_value_t** %3, align 8
  %4 = bitcast [3 x %jl_value_t*]* %2 to i64*
  store i64 2, i64* %4, align 8
  %5 = getelementptr [3 x %jl_value_t*], [3 x %jl_value_t*]* %2, i64 0, i64 1
  %6 = load i64, i64* bitcast (%jl_value_t*** @jl_tls_states to i64*), align 8
  %7 = bitcast %jl_value_t** %5 to i64*
  store i64 %6, i64* %7, align 8
  store %jl_value_t** %.sub, %jl_value_t*** @jl_tls_states, align 8
  %8 = getelementptr inbounds %jl_value_t, %jl_value_t* %0, i64 2
  %9 = bitcast %jl_value_t* %8 to i16*
  %10 = load i16, i16* %9, align 2
  %11 = and i16 %10, 3
  %12 = icmp eq i16 %11, 3
  br i1 %12, label %array_owned, label %merge_own

array_owned:                                      ; preds = %top
  %13 = getelementptr inbounds %jl_value_t, %jl_value_t* %0, i64 5, i32 0
  %14 = load %jl_value_t*, %jl_value_t** %13, align 8
  br label %merge_own

merge_own:                                        ; preds = %array_owned, %top
  %15 = phi %jl_value_t* [ %0, %top ], [ %14, %array_owned ]
  %16 = bitcast %jl_value_t* %0 to %jl_value_t***
  %17 = load %jl_value_t**, %jl_value_t*** %16, align 8
  %18 = call %jl_value_t* @jl_box_int64(i64 signext %1)
  store %jl_value_t* %18, %jl_value_t** %3, align 8 ;; <<------ boxed int rooted here unnecessarily
  %19 = getelementptr inbounds %jl_value_t, %jl_value_t* %15, i64 -1, i32 0
  %20 = bitcast %jl_value_t** %19 to i64*
  %21 = load i64, i64* %20, align 8
  %22 = and i64 %21, 1
  %23 = icmp eq i64 %22, 0
  br i1 %23, label %cont, label %wb_may_trigger

wb_may_trigger:                                   ; preds = %merge_own
  %24 = getelementptr inbounds %jl_value_t, %jl_value_t* %18, i64 -1, i32 0
  %25 = bitcast %jl_value_t** %24 to i64*
  %26 = load i64, i64* %25, align 8
  %27 = and i64 %26, 1
  %28 = icmp eq i64 %27, 0
  br i1 %28, label %wb_trigger, label %cont

wb_trigger:                                       ; preds = %wb_may_trigger
  call void @jl_gc_queue_root(%jl_value_t* %15)
  br label %cont

cont:                                             ; preds = %merge_own, %wb_trigger, %wb_may_trigger
  store %jl_value_t* %18, %jl_value_t** %17, align 8
  %29 = load i64, i64* %7, align 8
  store i64 %29, i64* bitcast (%jl_value_t*** @jl_tls_states to i64*), align 8
  ret void
}

Which roots the boxed integer before assigning it to the array. setfield! doesn't have this problem (on this commit).