This doesn't seem right. Maybe if we're assuming we can't saturate the 32-bit output? But if that's the case, we should eliminate the saturated addition entirely.
The fix is simple, but we ought to have some sort of test or check that the steps are sized appropriately.
This doesn't seem right. Maybe if we're assuming we can't saturate the 32-bit output? But if that's the case, we should eliminate the saturated addition entirely.
The fix is simple, but we ought to have some sort of test or check that the steps are sized appropriately.