AVR Core¶

lbuild module: modm:platform:core

Implements the interfaces defined in the :architecture modules. Adds the -mmcu={target} compile flag and the F_CPU={core frequency} define.

Fuses¶

You can define the fuse values in your source code via the FUSES struct ELF section mechanism, they are automatically used by avrdude. For fuse values see the AVR Fuse Calculator.

#include <avr/io.h>
FUSES =
{
    LFUSE_DEFAULT, // .low
    HFUSE_DEFAULT, // .high
    EFUSE_DEFAULT, // .extended
};

Blocking Delay¶

The delay functions as defined by modm:architecture:delay and called with a constant value are implemented directly via GCC's __builtin_avr_delay_cycles(ns / (1e9 / F_CPU)). Constant delays are therefore fairly accurate.

For delays with a dynamic time, the following limitations apply for nanoseconds:

The input is limited to 16-bit nanoseconds, so ~65 microseconds.
The minimum delay is between 5-10 cycles, so even at the fastest clock rate of 32MHz a delay below ~250ns is not possible. At 8MHz delays less than 1us are not possible.
For longer delays, the 3-cycle _delay_loop_1(loops) is used and the number of loops should be calculated by dividing the input value by ns_per_loop = 3e9/F_CPU. However, division is prohibitively slow and thus very coarsly approximated by shifting the input: ns >> ceil(log_2(ns_per_loop)). This is only accurate for 3e9/2^7 = 23.4MHz, 11.7MHz, 5.8MHz, etc. For all other clock speeds the error is bound by a factor of 2, so the delay is at most twice as long as expected.

For micro- and milliseconds delays with dynamic time:

Microseconds delay is implemented fairly accurately in 1us steps with a maximum time delay of 65ms for clocks larger than 6MHz, or .
Millisecond delay is implemented fairly accurately in 1ms steps on 32-bits of input time.

This module is only available for avr.

Options¶

f_cpu¶

CPU clock frequency

Default: None
Inputs: [1 ... 32000000]

Dependencies¶

Limited availability: Check with 'lbuild discover' if this module is available for your target!