Posted by SprinterSB:
Fri. Oct 5, 2012 - 03:15 PM
Hi, I am interested in ASF's code metrics for fixed-point square root computation for unsigned 0.16 Q-format numbers, i.e. no integral bits and 16 fractional bits.

- Maximum absolute error (over all inputs)
- Maximum execution time (over all inputs)
- Code size

A granularity of "have / don't have MUL instruction" is enough.

Is there such data available?

The function is mentioned in "fixed_t.h File Reference" as

fixed_t fixed_sqrt (fixed_t f, int Q)

and I am interested for the case Q=16.

AVR-Libc's float sqrtf takes a maximum execution time of ~500 ticks if MUL is available (MUL should not matter for sqrtf implementation) so I'd guess fixed point math is considerably faster? AVR-Libc computer 23 fractional digits which is ~22 Ticks per digit.

avrfreaks does not support Opera. Profile inactive.

## Tags: