#include <bitfield/bitfield.h>
#include <bitfield/8byte.h>
-uint64_t eightbyte_encode_float(float value, uint8_t bit_offset, uint8_t bit_size,
- float factor, float offset) {
+static uint64_t float_to_fixed_point(const float value, const float factor,
+ const float offset) {
float raw = (value - offset) / factor;
if(raw > 0) {
// round up to avoid losing precision when we cast to an int
// rounding?
raw += 0.5;
}
+ return (uint64_t)raw;
+}
+
+uint64_t eightbyte_encode_float(float value, uint8_t bit_offset, uint8_t bit_size,
+ float factor, float offset) {
uint64_t result = 0;
- if(!eightbyte_set_bitfield(&result, (uint64_t)raw, bit_offset, bit_size)) {
+ if(!eightbyte_set_bitfield(float_to_fixed_point(value, factor, offset),
+ bit_offset, bit_size, &result)) {
// debug("%f will not fit in a %d bit field", value, bit_size);
}
return result;
return eightbyte_encode_float(value, bit_offset, bit_size, 1.0, 0);
}
-bool bitfield_encode_float(float value, uint8_t bit_offset,
- uint8_t bit_size, float factor, float offset, uint8_t destination[]) {
- // TODO
- return 0;
+bool bitfield_encode_float(const float value, const uint8_t bit_offset,
+ const uint8_t bit_size, const float factor, const float offset,
+ uint8_t destination[], const uint8_t destination_length) {
+ if(!set_bitfield(float_to_fixed_point(value, factor, offset), bit_offset,
+ bit_size, destination, destination_length)) {
+ // debug("%f will not fit in a %d bit field", value, bit_size);
+ return false;
+ }
+ return true;
+}
+
+bool bitfield_encode_bool(const bool value, const uint8_t bit_offset,
+ const uint8_t bit_size, uint8_t destination[],
+ const uint16_t destination_length) {
+ return bitfield_encode_float(value, bit_offset, bit_size, 1.0, 0,
+ destination, destination_length);
}