The division routine supplied with the ARM libraries provides good overall performance. However, the amount of time required to perform a division depends on the input values. For example, a division that generates a four-bit quotient might require only 12 cycles while a 32-bit quotient might require 96 cycles. Depending on your target, some applications require a faster worst-case cycle count at the expense of lower average performance. For this reason, the ARM library provides two divide routines.
The real-time routine:
always executes in fewer than 45 cycles
is faster than the standard division routine for larger quotients
is slower than the standard division routine for typical quotients
returns the same results
does not require any change in the surrounding code.
Note
Real-time division is not available in the libraries for Cortex-M1 or Cortex-M0.
The Cortex-R4 and Cortex-M3 processors support hardware floating-point divide, so they do not require the library divide routines.
See also
PDF version