GY-63_MS5611/libraries/runningAngle/examples/time-add/time-add.ino

108 lines
3.3 KiB
Arduino
Raw Normal View History

2021-01-29 06:31:58 -05:00
/*
* time-add.ino: Measure the average execution time of
* runningAngle::add().
*
* This test sketch feeds pseudo-random angles to runningAngle::add()
* in order to measure its average execution time in CPU cycles. The
* input angles are within 0..90 deg, which ensures there will be no
* wrapping. Wrapping would make the method slightly slower, but it is
* expected to be infrequent in typical use cases.
*
* This test is meant to run on AVR-based Arduinos only.
*/
2021-12-28 03:50:06 -05:00
2021-01-29 06:31:58 -05:00
#include <runningAngle.h>
2021-12-28 03:50:06 -05:00
2021-01-29 06:31:58 -05:00
// Uncomment the line below in order to do the computation in radians.
//#define USE_RADIANS
// We could test both angle units within the same code. Typical user
// code, however, is likely to use only a single unit, and this can
// enable some compiler optimizations. Choosing the unit at compile time
// gives the compiler the same optimization opportunities as in typical
// user code.
#ifdef USE_RADIANS
# define ANGLE_UNIT RADIANS
# define ANGLE_SYMBOL "rad"
# define ANGLE_MAX (PI/2)
#else
# define ANGLE_UNIT DEGREES
# define ANGLE_SYMBOL "deg"
# define ANGLE_MAX 90.0
#endif
2021-12-28 03:50:06 -05:00
2021-01-29 06:31:58 -05:00
// Overhead of the timing code, in CPU cycles. This was found by
// disassembling and counting cycles.
const uint16_t timing_overhead = 8;
// Iterations for averaging the execution time.
const int iterations = 100;
// Prevent unwanted optimizations. This is most useful for preventing
// constant folding when feeding constants to the filter. It can also
// prevent instruction reordering that could move parts of the
// computation of the input angle within the timed portion of the code.
static float unoptimize(float x)
{
volatile float y = x;
return y;
}
2021-12-28 03:50:06 -05:00
2021-01-29 06:31:58 -05:00
runningAngle heading(runningAngle::ANGLE_UNIT);
void setup() {
Serial.begin(9600);
// Set Timer 1 to count in normal mode at the full CPU frequency.
// The timer value, TCNT1, can then be used as a clock with
// single-cycle resolution.
TCCR1A = 0;
TCCR1B = _BV(CS10);
long total_time = 0;
for (int i = 0; i <= iterations; i++) {
float angle = (rand() + 0.5) / (RAND_MAX + 1.0) * ANGLE_MAX;
angle = unoptimize(angle);
// Timed part.
uint16_t start_time = TCNT1;
heading.add(angle);
uint16_t end_time = TCNT1;
// Do not use the time of the first execution of add(), as it
// goes through a non typical and much shorter execution path.
if (i != 0) {
// Add the execution time of this iteration. Note that timer
// rollover is not an issue as long as the timed code takes
// less that 65536 cycles, and the timing computations are
// performed in uint16_t, as this type rolls over in the
// same manner as the timer itself.
total_time += end_time - start_time - timing_overhead;
}
}
// Use the resulting average. Otherwise the whole averaging code
// could be optimized away.
Serial.print("Average angle: ");
Serial.print(heading.getAverage());
Serial.println(" " ANGLE_SYMBOL);
// Print the timing result.
Serial.print("Average time: ");
Serial.print((float) total_time / iterations);
Serial.println(" CPU cycles");
// This can be used to exit a simulation on simavr.
Serial.flush();
cli();
asm("sleep");
}
void loop(){}
2021-12-28 03:50:06 -05:00