The allocation sequence should be:
  // Allocate the instance:
  // If TLAB is enabled:
  // Try to allocate in the TLAB.
  // If fails, go to the slow path.
  // Else If inline contiguous allocations are enabled:
  // Try to allocate in eden.
  // If fails due to heap end, go to slow path.
  //
  // If TLAB is enabled OR inline contiguous is enabled:
  // Initialize the allocation.
  // Exit.
  //
  // Go to slow path.

But by mistake, when doing this for x86, I seem to have made the the "If inline contiguous" become "always try inline contiguous".