Clang - generate code for loop rerolling pass

Beginners
1

I was looking at one of the tests for the LLVM’s -loop-reroll pass - llvm-project/basic.ll at main · llvm/llvm-project · GitHub

As in the test, the pass is able to take something like this -

; Function Attrs: nounwind sspstrong uwtable
define dso_local void @bar(ptr nocapture noundef readnone %0) local_unnamed_addr #0 {
  br label %3

2:                                                ; preds = %3
  ret void

3:                                                ; preds = %1, %3
  %4 = phi i32 [ 0, %1 ], [ %10, %3 ]
  %5 = tail call i32 @foo(i32 noundef %4) #2
  %6 = add nuw nsw i32 %4, 1
  %7 = tail call i32 @foo(i32 noundef %6) #2
  %8 = add nuw nsw i32 %4, 2
  %9 = tail call i32 @foo(i32 noundef %8) #2
  %10 = add nuw nsw i32 %4, 3
  %11 = icmp eq i32 %10, 500
  br i1 %11, label %3, label %2, !llvm.loop !5
}

declare i32 @foo(i32 noundef) local_unnamed_addr #1

and perform a loop rerolling optimization on it.

But the given C code in the test that is expected to generate this LLVM IR -

int foo(int a);
 void bar(int *x) {
   for (int i = 0; i < 500; i += 3) {
     foo(i);
     foo(i+1);
     foo(i+2);
   }
 }

instead generates (with clang) something akin to -

; Function Attrs: nounwind sspstrong uwtable
define dso_local void @bar(ptr nocapture noundef readnone %0) local_unnamed_addr #0 {
  br label %3

2:                                                ; preds = %3
  ret void

3:                                                ; preds = %1, %3
  %4 = phi i32 [ 0, %1 ], [ %10, %3 ]
  %5 = tail call i32 @foo(i32 noundef %4) #2
  %6 = add nuw nsw i32 %4, 1
  %7 = tail call i32 @foo(i32 noundef %6) #2
  %8 = add nuw nsw i32 %4, 2
  %9 = tail call i32 @foo(i32 noundef %8) #2
  %10 = add nuw nsw i32 %4, 3
  %11 = icmp ult i32 %4, 497
  br i1 %11, label %3, label %2, !llvm.loop !5
}

declare i32 @foo(i32 noundef) local_unnamed_addr #1

which the loop rerolling pass is unable to work on (it looks like the pass is only able to work on icmp eq llvm-project/LoopRerollPass.cpp at main · llvm/llvm-project · GitHub so this might be an expected result).

Note the difference:
In the test, we have - %11 = icmp eq i32 %10, 500.
Clang generates - %11 = icmp ult i32 %4, 497.

Is there a way for Clang to generate the former, so that the loop rerolling pass can work on the generated LLVM IR?