From 8340e615d46615894b44b4ffce5dc2dd86cbad40 Mon Sep 17 00:00:00 2001
From: Vito Caputo <vcaputo@gnugeneration.com>
Date: Tue, 13 Dec 2016 07:57:56 -0800
Subject: ray: introduce a rudimentary ray tracer

My first ray tracer, it only has spheres, planes, and point light sources.

No texture mapping, no soft shadows, no global illumination.

This is all very basic right now, the camera movement is simple and boring, but
sufficient for further development and optimization.

I made some effort to support multiple CPUs, it should detect the number of
CPUs in the system and use enough pthreads to keep them busy.

Jacco Bikker's tutorial on flipcode was the original impetus to do this, and
definitely served as a guide early on.
---
 modules/ray/ray_threads.c | 111 ++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 111 insertions(+)
 create mode 100644 modules/ray/ray_threads.c

(limited to 'modules/ray/ray_threads.c')

diff --git a/modules/ray/ray_threads.c b/modules/ray/ray_threads.c
new file mode 100644
index 0000000..2369687
--- /dev/null
+++ b/modules/ray/ray_threads.c
@@ -0,0 +1,111 @@
+#include <pthread.h>
+#include <stdlib.h>
+
+#include "fb.h"
+
+#include "ray_scene.h"
+#include "ray_threads.h"
+
+#define BUSY_WAIT_NUM	1000000000	/* How much to spin before sleeping in pthread_cond_wait() */
+
+/* for now assuming x86 */
+#define cpu_relax() \
+	__asm__ __volatile__ ( "pause\n" : : : "memory")
+
+/* This is a very simple/naive implementation, there's certainly room for improvement.
+ *
+ * Without the BUSY_WAIT_NUM spinning this approach seems to leave a fairly
+ * substantial proportion of CPU idle while waiting for the render thread to
+ * complete on my core 2 duo.
+ *
+ * It's probably just latency in getting the render thread woken when the work
+ * is submitted, and since the fragments are split equally the main thread gets
+ * a head start and has to wait when it finishes first.  The spinning is just
+ * an attempt to avoid going to sleep while the render threads finish, there
+ * still needs to be improvement in how the work is submitted.
+ *
+ * I haven't spent much time on optimizing the raytracer yet.
+ */
+
+static void * ray_thread_func(void *_thread)
+{
+	ray_thread_t	*thread = _thread;
+
+	for (;;) {
+		pthread_mutex_lock(&thread->mutex);
+		while (thread->fragment == NULL)
+			pthread_cond_wait(&thread->cond, &thread->mutex);
+
+		ray_scene_render_fragment(thread->scene, thread->camera, thread->fragment);
+		thread->fragment = NULL;
+		pthread_mutex_unlock(&thread->mutex);
+		pthread_cond_signal(&thread->cond);
+	}
+
+	return NULL;
+}
+
+
+void ray_thread_fragment_submit(ray_thread_t *thread, ray_scene_t *scene, ray_camera_t *camera, fb_fragment_t *fragment)
+{
+	pthread_mutex_lock(&thread->mutex);
+	while (thread->fragment != NULL)	/* XXX: never true due to ray_thread_wait_idle() */
+		pthread_cond_wait(&thread->cond, &thread->mutex);
+
+	thread->fragment = fragment;
+	thread->scene = scene;
+	thread->camera = camera;
+
+	pthread_mutex_unlock(&thread->mutex);
+	pthread_cond_signal(&thread->cond);
+}
+
+
+void ray_thread_wait_idle(ray_thread_t *thread)
+{
+	unsigned	n;
+
+	/* Spin before going to sleep, the other thread should not take substantially longer. */
+	for (n = 0; thread->fragment != NULL && n < BUSY_WAIT_NUM; n++)
+		cpu_relax();
+
+	pthread_mutex_lock(&thread->mutex);
+	while (thread->fragment != NULL)
+		pthread_cond_wait(&thread->cond, &thread->mutex);
+	pthread_mutex_unlock(&thread->mutex);
+}
+
+
+ray_threads_t * ray_threads_create(unsigned num)
+{
+	ray_threads_t	*threads;
+	unsigned	i;
+
+	threads = malloc(sizeof(ray_threads_t) + sizeof(ray_thread_t) * num);
+	if (!threads)
+		return NULL;
+
+	for (i = 0; i < num; i++) {
+		pthread_mutex_init(&threads->threads[i].mutex, NULL);
+		pthread_cond_init(&threads->threads[i].cond, NULL);
+		threads->threads[i].fragment = NULL;
+		pthread_create(&threads->threads[i].thread, NULL, ray_thread_func, &threads->threads[i]);
+	}
+	threads->n_threads = num;
+
+	return threads;
+}
+
+
+void ray_threads_destroy(ray_threads_t *threads)
+{
+	unsigned	i;
+
+	for (i = 0; i < threads->n_threads; i++)
+		pthread_cancel(threads->threads[i].thread);
+
+	for (i = 0; i < threads->n_threads; i++)
+		pthread_join(threads->threads[i].thread, NULL);
+
+	free(threads);	
+}
-- 
cgit v1.2.3