[lldb][swift] Increase number of Rows inspected when unwinding async functions (#10698)

We were overly conservatives in our initial estimations: if callee saved
registers are used, they are usually spilled first, increasing the
number of rows in the prologue.

This also revealed a bug in computing the offset of a vector iterator:

```
-  int row_idx = fp_locs.end() - it;
+  int row_idx = it - fp_locs.begin();
```

Testing this is unfortunately difficult: a way of forcing register
spilling is required. For now, this commit tweaks one of the existing
tests in a way that seems to reliably spill more registers, but there
are no guarantees this will continue to be the case. Other testing
mechanisms need to be created.

(cherry picked from commit 3307906)

Author: felipepiovezan

lldb/source/Plugins/LanguageRuntime/Swift/SwiftLanguageRuntime.cpp

@@ -2588,9 +2588,13 @@ GetFrameSetupInfo(UnwindPlan &unwind_plan, RegisterContext &regctx) {
   if (!fp_unwind_regdomain)
     return fp_unwind_regdomain.takeError();
 
-  // Look at the first few (4) rows of the plan and store FP's location.
-  const int upper_bound = std::min(4, unwind_plan.GetRowCount());
-  llvm::SmallVector<AbstractRegisterLocation, 4> fp_locs;
+  // Look at the first few (12) rows of the plan and store FP's location.
+  // This number is based on AAPCS, with 10 callee-saved GPRs and 8 floating
+  // point registers. When STP instructions are used, the plan would have one
+  // initial row, nine rows of saving callee-saved registers, and two standard
+  // prologue rows (fp+lr and sp).
+  const int upper_bound = std::min(12, unwind_plan.GetRowCount());
+  llvm::SmallVector<AbstractRegisterLocation, 12> fp_locs;
   for (int row_idx = 0; row_idx < upper_bound; row_idx++) {
     auto *row = unwind_plan.GetRowAtIndex(row_idx);
     AbstractRegisterLocation regloc;
@@ -2617,7 +2621,7 @@ GetFrameSetupInfo(UnwindPlan &unwind_plan, RegisterContext &regctx) {
   // Use subsequent row, if available.
   // Pointer auth may introduce more instructions, but they don't affect the
   // unwinder rows / store to the stack.
-  int row_idx = fp_locs.end() - it;
+  int row_idx = it - fp_locs.begin();
   int next_row_idx = row_idx + 1;
 
   // If subsequent row is invalid, approximate through current row.

lldb/test/API/lang/swift/async/stepping/step_over_lots_of_tasks/TestSwiftAsyncSteppingManyTasks.py

@@ -55,7 +55,7 @@ def test_step_over_top_level_fibonacci(self):
         fib_bp.SetEnabled(False)
 
         fib_first_line = thread_in_fib.frames[0].GetLineEntry().GetLine()
-        num_lines_fib = 5
+        num_lines_fib = 7
         for line_offset in range(1, num_lines_fib):
             thread_in_fib.StepOver()
             self.assertEqual(process.GetSelectedThread(), thread_in_fib)

lldb/test/API/lang/swift/async/stepping/step_over_lots_of_tasks/main.swift

@@ -12,13 +12,17 @@
 
 func fib(n: Int) async -> Int {
   if (n == 0) {
+    print("done!")
     return 1
   }
   if (n == 1) {
+    print("done!")
     return 1
   }
   async let n1_task = fib(n: n - 1)  // Breakpoint fib
   async let n2_task = fib(n: n - 2)
+  print(n)
+  if (n == 6) { do { try await Task.sleep(for: .seconds(1)) } catch {} }
   let n1 = await n1_task
   let n2 = await n2_task
   return n1 + n2