added thread pool, started fleshing out the render job system

[erebus2020] / liberebus / src / erebus.c

#include <stdio.h>
#include <pthread.h>
#include "erebus.h"
#include "tpool.h"

struct render_job {
        struct erb_rend *erb;
        int x, y, width, height;
        struct render_job *next;
};

struct erb_rend {
        int fb_width, fb_height, fb_npix;
        float *fb_pixels;
        int *fb_nsamples;
};

static void proc_render_job(void *cls);
static struct render_job *alloc_job(void);
static void free_job(struct render_job *job);

static struct thread_pool *tpool;


struct erb_rend *erb_create(void)
{
        struct erb_rend *erb;

        if(!tpool) {
                if(!(tpool = tpool_create(0))) {
                        fprintf(stderr, "erb_create: fatal error, failed to create thread pool!\n");
                        return 0;
                }
        }

        if(!(erb = calloc(1, sizeof *erb))) {
                return 0;
        }
        return erb;
}

void erb_destroy(struct erb_rend *erb)
{
        if(!erb) return;
        free(erb->fb_pixels);
        free(erb);
}

int erb_allocframe(struct erb_rend *erb, int width, int height)
{
        float *newfb;
        int *newns;
        int npix;

        if(width == erb->fb_width && height == erb->fb_height) {
                return 0;
        }

        npix = width * height;

        if(!(newfb = malloc(npix * 3 * sizeof *erb->fb_pixels))) {
                goto err;
        }
        if(!(newns = malloc(npix * sizeof *erb->fb_nsamples))) {
                free(newfb);
                goto err;
        }

        free(erb->fb_pixels);
        free(erb->fb_nsamples);

        erb->fb_pixels = newfb;
        erb->fb_nsamples = newns;
        erb->fb_width = width;
        erb->fb_height = height;
        erb->fb_npix = npix;
        return 0;

err:
        fprintf(stderr, "erb_allocframe: failed to allocate %dx%d framebuffer\n", width, height);
        return -1;
}

float *erb_getframe(struct erb_rend *erb)
{
        return erb->fb_pixels;
}

void erb_begin(struct erb_rend *erb)
{
        memset(erb->fb_pixels, 0, erb->fb_npix * 3 * sizeof *erb->fb_pixels);
        memset(erb->fb_nsamples, 0, erb->fb_npix * sizeof *erb->fb_nsamples);
}

float *erb_end(struct erb_rend *erb)
{
        int i, n;
        float s;
        float *pptr = erb->fb_pixels;
        int *nptr = erb->fb_nsamples;

        for(i=0; i<erb->fb_npix; i++) {
                n = *nptr;
                if(n > 1) {
                        s = 1.0f / n;
                        *pptr++ *= s;
                        *pptr++ *= s;
                        *pptr++ *= s;
                        *nptr++ = 1;
                } else {
                        pptr += 3;
                        nptr++;
                }
        }

        return erb->fb_pixels;
}

void erb_queue_frame(struct erb_rend *erb)
{
        erb_queue_block(erb, 0, 0, erb->fb_width, erb->fb_height);
}

void erb_queue_block(struct erb_rend *erb, int x, int y, int width, int height)
{
        struct render_job *job;

        if(!(job = alloc_job())) {
                fprintf(stderr, "erb_queue_block: failed to allocate rendering job\n");
                return;
        }

        job->erb = erb;
        job->x = x;
        job->y = y;
        job->width = width;
        job->height = height;

        tpool_enqueue(tpool, job, proc_render_job, 0);
}

void erb_wait(struct erb_rend *erb)
{
        /* XXX should we have a per-renderer instance thread pool, to wait only for our own jobs? */
        tpool_wait(tpool);
}

void erb_primary_ray(struct erb_rend *erb, struct erb_ray *ray, int sample)
{
        /* TODO */
}

void erb_sample_ray(struct erb_rend *erb, struct erb_ray *ray, float *col)
{
        /* TODO */
        col[0] = fmod(col[0] + 1.0f, 1.0f);
        col[1] = col[2] = fmod(col[1] + 0.33f, 1.0f);
}

static void proc_render_job(void *cls)
{
        int i, j, fboffs;
        struct erb_rend *erb;
        struct render_job *job = cls;
        float *fbptr;
        int *nsptr;
        struct erb_ray ray;

        erb = job->erb;
        fboffs = job->y * erb->fb_width + job->x;
        fbptr = erb->fb_pixels + fboffs * 3;
        nsptr = erb->fb_nsamples + fboffs;

        for(i=0; i<job->height; i++) {
                for(j=0; j<job->width; j++) {
                        erb_primary_ray(erb, &ray, *nsptr++);
                        erb_sample_ray(erb, &ray, fbptr);
                        fbptr += 3;
                }
        }
        free_job(job);
}


#define MAX_JOB_POOL    128
static struct render_job *job_pool;
static int num_free_jobs;
static pthread_mutex_t job_pool_lock = PTHREAD_MUTEX_INITIALIZER;

static struct render_job *alloc_job(void)
{
        struct render_job *job;

        pthread_mutex_lock(&job_pool_lock);
        if(!job_pool) {
                pthread_mutex_unlock(&job_pool_lock);
                return malloc(sizeof *job);
        }

        job = job_pool;
        job_pool = job->next;
        num_free_jobs--;
        pthread_mutex_unlock(&job_pool_lock);
        return job;
}

static void free_job(struct render_job *job)
{
        pthread_mutex_lock(&job_pool_lock);
        if(num_free_jobs >= MAX_JOB_POOL) {
                pthread_mutex_unlock(&job_pool_lock);
                free(job);
                return;
        }

        job->next = job_pool;
        job_pool = job;
        num_free_jobs++;
        pthread_mutex_unlock(&job_pool_lock);
}