When to use alpha premultiplied images?

I am using LVGL 9.2 and observed that the multiple ARGB case is not performing good. On narrowing down further I found that HW premultiplication is not supported in my case so it takes software path and aplha premultiplication in lv_image_decoder_post_process() is consuming lot of time.
If I use premultiplied ARGB image then this step is skipped and I get good FPS but the rotated ARGB case is degrading in performance.
If I use premultiplied image in benchmark demo:
Multiple ARGB images : 10 FPS → 17 FPS (+7 FPS)
Multiple rotated ARGB images : 10 FPS → 2 FPS (-8 FPS)

So I had 2 questions:

  1. Is it ok to always use premultipled ARGB images or are there any side effects of that
  2. What could be the reason behind decrease in FPS in rotated ARGB test if I use premultiplied image?
    Thanks