mpipe_workqueue.h File Reference

Shared P4WQ pool for element-level work parallelism. More...

#include <zephyr/sys/p4wq.h>

Go to the source code of this file.

Variables
struct k_p4wq *const	mpipe_p4wq
	Shared mpipe P4WQ pool instance.

Detailed Description

Shared P4WQ pool for element-level work parallelism.

Provides a shared P4WQ (Pooled Parallel Preemptible Priority-based Work Queue) pool that elements can use to offload and parallelize work items (e.g., tile-based AI inference, DCT transforms, etc.).

Elements use the native Zephyr P4WQ API directly (k_p4wq_submit, k_p4wq_wait) with the shared mpipe_p4wq instance, without wrapping the API.

Example usage in an element with parallel tile processing:

#include <zephyr/mpipe/mpipe_workqueue.h>
 
static void tile_handler(struct k_p4wq_work *work) {
    // process one tile...
}
 
static bool my_chain_fn(struct mpipe_pad *pad, struct net_buf *in,
                       struct net_buf **out) {
    struct k_p4wq_work tile_work[4];
 
    for (int i = 0; i < 4; i++) {
        tile_work[i].handler = tile_handler;
        tile_work[i].priority = 7;
        tile_work[i].deadline = 0;
        tile_work[i].sync = true;
        k_sem_init(&tile_work[i].done_sem, 0, 1);
        k_p4wq_submit(mpipe_p4wq, &tile_work[i]);
    }
    for (int i = 0; i < 4; i++) {
        k_p4wq_wait(&tile_work[i], K_FOREVER);
    }
    // assemble output...
    *out = result;
    return true;
}