ElPeriodic 1.5

Phase-locked userland scheduling library

libElPeriodic

Library to run frequent periodic tasks.

Principle of Operation

The libElPeriodic is designed to simplify writing control loops that are expected to run at constant "tick" intervals with smallest possible overhead and little or no support from the underlying run time environment.

The library is optimized to align active periods of the control loop to the set frequency (and optionally phase as well) by applying phase locked loop design with a proportional phase detector and a low-pass filter as an error amplifier.

Basic Usage

Sample usage pattern is demonstrated below. The code block denoted by the square brackets will be executing 125.5 times a second, untul the value returned by the is_runnable() routine is non-zero. Provided of course that the "logic" does not take more than 0.01 second to run on average and that OS scheduler plays the ball.

#include <assert.h>
#include <time.h>
#include <elperiodic.h>

extern int is_runnable(void);

void
event_loop(void)
{
    double frequency = 125.5; /* Hz */
    void *elp;

    elp = prdic_init(frequency, 0.0);
    assert(elp != NULL);

    while (is_runnable()) {
//      [----------------------];
//      [Insert your logic here];
//      [----------------------];
        prdic_procrastinate(elp);
    }
    prdic_free(elp);
}

Dispatching Calls from Worker Threads

The library also supports simple FIFO queue of function calls that have to be dispatched by the library asynchronously in the main thread during the "procrastination" time intervals. This allows I/O, timer and other type of events to enter into the processing loop in a thread-safe manner.

This can be accomplished by enabling the functionality using the prdic_CFT_enable() API and then scheduling necessary calls via the prdic_call_from_thread() in a worker thread(s).

#include <assert.h>
#include <time.h>
#include <elperiodic.h>
#include <pthread.h>
#include <signal.h>

extern int do_something(void);

struct prd_ctx {
    /* This member is initialized in the main thread */
    void *elp;
    /* This member only accessible by the main thread */
    int is_runnable;
    /* This member is filled in by the worker thread before exit */
    int result;
};

static void shutdown(struct prd_ctx *ctxp) {ctxp->is_runnable = 0;}

static void
worker_thread(void *ap)
{
    struct prd_ctx *ctxp = (struct prd_ctx *)ap;

    ctxp->result = do_something();
    prdic_call_from_thread(ctxp->elp, (void (*)(void *))shutdown, ctxp);
}

int
event_loop(void)
{
    struct prd_ctx ctx = {.is_runnable = 1};
    double freq = 125.5; /* Hz */
    pthread_t wthr;

    ctx.elp = prdic_init(freq, 0.0);
    assert(ctx.elp != NULL);
    assert(prdic_CFT_enable(ctx.elp, SIGUSR1) == 0);

    assert(pthread_create(&wthr, NULL, (void *(*)(void *))worker_thread, &ctx) == 0);

    while (ctx.is_runnable) {
//      [----------------------];
//      [Insert your logic here];
//      [----------------------];
        prdic_procrastinate(ctx.elp);
    }
    pthread_join(wthr, NULL);
    prdic_free(ctx.elp);

    return ctx.result;
}

Story

It came about having to write the same code over and over again in multiple real-time projects, ranging from game Digger, RTP relay server RTPProxy. It has also been recently utilized to replace a heavy-weight (and at the time not portable to Python 3) "Twisted" framework in the Python Sippy B2BUA project.