/**
* @file nbody.c
- * @author Sam Moore (20503628) 2012
- * @purpose N-Body simulator - Definition of simulation functions; single threaded version
+ * @purpose Implentation of multi-threaded N-Body simulator using pthreads
+ * @author Sam Moore (20503628) - 2012
*/
-#include <stdlib.h>
-#include <stdio.h>
-#include <math.h>
-#include <time.h>
-#include <string.h>
-#include <stdbool.h>
-#include <stdlib.h>
-#include <unistd.h>
-
-#include "nbody.h"
-
-//Stuff to do with worker threads
-// The worker threads (if used) are spawned by the computation thread, not the main thread
-// The main thread (which handles initialisation, graphics, user input, cleanup and exit) does not need to know about these.
-#if NUM_THREADS > 1
+#include "nbody.h" // Declarations
+#include "../single-thread/nbody.c" // Include all functions from the single threaded version
+#include <assert.h>
+#include "graphics.h" // For declaration of Graphics_Run only
+#include "barrier.c"
+
+// --- Variable declarations --- //
+
+pthread_t compute_thread; // The thread responsible for computations; it spawns worker threads (* terms and conditions apply)
- pthread_t worker_thread[NUM_THREADS]; //Worker threads responsible for Force and Position updates
- System sub_system[NUM_THREADS]; //Array of System's used to divide up the main "universe" System for worker threads
- pthread_mutex_t mutex_workers;
- pthread_cond_t workers_done_cv;
- unsigned worker_count;
-#endif
+pthread_t * worker_thread = NULL; //Array of worker threads responsible for Force and Position updates
+System * sub_system = NULL; //Array of Systems used to divide up the main "universe" System for worker threads
+System * nested_sub_system = NULL; //Array of Systems for division of "universe" for the nested worker threads
-/**
- * Prints the body on screen
- */
-void Body_Print(Body * a)
-{
- printf("Body %p M=%f X=%f Y=%f Z=%f Fx=%f Fy=%f Fz=%f Vx=%f Vy=%f Vz=%f\n",
- (void*)a, a->mass, a->x[0], a->x[1], a->x[2], a->F[0], a->F[1], a->F[2], a->v[0], a->v[1], a->v[2]);
-}
+pthread_mutex_t mutex_runstate; // Mutex around the runstate
-/**
- * Computing forces
- */
-void Body_Force(Body * a, System * s)
-{
- double distance;
- double con;
- double gd;
- for (unsigned i = 0; i < DIMENSIONS; ++i)
- a->F[i] = 0;
- for (unsigned index = 0; index < s->N; ++index)
- {
- Body * b = s->body+index;
- if (b == a)
- continue;
-
- distance = 0.0;
- for (unsigned i = 0; i < DIMENSIONS; ++i)
- distance += square(b->x[i] - a->x[i]);
- distance = sqrt(distance);
- con = G * a->mass * b->mass / square(distance);
- gd = con / distance;
- for (unsigned i = 0; i < DIMENSIONS; ++i)
- a->F[i] += gd * (b->x[i] - a->x[i]);
- }
-}
-/**
- * Compute velocities
- */
-void Body_Velocity(Body * a)
-{
- for (unsigned i = 0; i < DIMENSIONS; ++i)
- a->v[i] += a->F[i] / a->mass * DELTA_T;
-}
+Barrier force_barrier; // I laughed at this variable name. A bit sad really.
+Barrier position_barrier;
-/**
- * Compute positions
- */
-void Body_Position(Body * a)
-{
- for (unsigned i = 0; i < DIMENSIONS; ++i)
- a->x[i] += a->v[i] * DELTA_T;
-}
-/**
- * Compute forces on each object in System s1 due to all bodies in System s2
- */
-void System_Forces(System * s1, System * s2)
-{
- //printf("Compute forces for %p - %d bodies...\n", (void*)s1, s1->N);
- for (unsigned i = 0; i < s1->N; ++i)
- {
- Body_Force(s1->body+i, s2);
- Body_Velocity(s1->body+i);
- }
- //printf("Compute positions for %p - Done!\n", (void*)s1);
-}
+Barrier graphics_barrier;
+
+pthread_mutex_t mutex_threads_running;
+int threads_running = 0;
/**
- * Compute positions of all objects in System
+ * @function Compute_Thread
+ * @purpose Thread - Continuously computes steps for a system of bodies. Seperate from graphics functions.
+ * Spawns worker threads to divide up computation.
+ * @param arg - Can be cast to the System for which computations are performed
+ *
+ * NOTE:
+ * This will always be (void*)(&universe) where universe is the global System variable that has every Body in it.
+ * But I don't like global variables. And since the argument had to be passed, I thought I might as well use it.
+ * That way, when I change "universe" to "solar_system", I only have to change the argument where this is called, not all through it.
+ * Find and replace? Who uses that!?
*/
-void System_Positions(System * s)
+
+
+void * Compute_Thread(void * arg)
{
- //printf("Compute positions for %p - %d bodies...\n", (void*)s, s->N);
- for (unsigned i = 0; i < s->N; ++i)
- Body_Position(s->body+i);
- //printf("Compute positions for %p - Done!\n", (void*)s);
-}
+
+ System * s = (System*)(arg); //cast argument to a System*
+ // If no number of threads provided, use the default value, unless someone (me) changed that to a stupid value
+ if (options.num_threads <= 0)
+ options.num_threads = (DEFAULT_WORKING_THREADS > 1) ? DEFAULT_WORKING_THREADS : 1; // Fear the ternary operator!
-/*
- * This function reads an input file. You can change it if you choose a
- * different file format
- */
-#define LINE_SIZE BUFSIZ
-void System_Init(System * s, char *fileName)
-{
- char line[LINE_SIZE];
- char * token;
- FILE * file;
+ if (options.nested_threads <= 0)
+ options.nested_threads = 1;
- file = fopen(fileName, "rt");
- s->N = atoi(fgets(line, LINE_SIZE, file));
- s->body = (Body*) calloc((size_t)s->N, sizeof(Body));
+ // Do a sanity check; there is no point spawning more threads than bodies.
+ if (options.num_threads > s->N)
+ {
+ fprintf(stderr,
+ "Warning: Using %u threads instead of %u specified, because there are only %u bodies to simulate!\n",
+ s->N, options.num_threads, s->N);
+ options.num_threads = s->N;
+ }
- //printf("----------------------Initial field-------------------------------\n");
+ // Initialise the barriers ("shields up!")
+ Barrier_Init(&force_barrier, options.num_threads);
+ Barrier_Init(&position_barrier, options.num_threads);
+ Barrier_Init(&graphics_barrier, 1);
- for (unsigned i = 0; i < s->N; ++i)
+
+
+ if (options.num_threads > 1) // If we require additional worker threads...
{
- if (fgets(line, LINE_SIZE, file) != NULL)
+ // Allocate worker threads and sub systems
+
+ sub_system = Split_System(&universe, options.num_threads);
+
+ if (options.nested_threads > 1)
+ nested_sub_system = Split_System(&universe, options.nested_threads);
+
+ #ifdef PERSISTENT_THREADS // Code for the smart way of doing it (spawn threads once, keep them running)
+ worker_thread = Allocate_Threads(options.num_threads-1);
+ // Spawn a bunch of worker threads, and let them all do their thing.
+ // Note the "-1". Because this thread can do work too!
+ for (unsigned i = 0; i < options.num_threads-1; ++i)
{
- Body * a = s->body+i;
- token = strtok(line, ",; ");
- a->mass = atof(token);
-
- for (unsigned j = 0; j < DIMENSIONS; ++j)
- {
- token = strtok(NULL, ",; ");
- a->x[j] = atof(token);
- }
- for (unsigned j = 0; j < DIMENSIONS; ++j)
+ if (pthread_create(worker_thread+i, NULL, Worker_Thread, (void*)(sub_system+i)) != 0)
{
- token = strtok(NULL, ",; ");
- a->v[j] = atof(token);
+ perror("In compute thread, couldn't create worker thread");
+ QuitProgram(true);
+ pthread_exit(NULL);
}
- //Body_Print(a);
}
+ Worker_Thread((void*)(sub_system+options.num_threads-1)); // This thread becomes a worker thread
+ #else
+ worker_thread = Allocate_Threads(options.num_threads);
+ #endif //PERSISTENT_THREADS
+
}
+ #ifdef PERSISTENT_THREADS
+ else // We only require one worker thread...
+ {
+ // So just do all computations in this thread
+ while (!ExitCondition())
+ {
+
+ System_Forces(s, s);
+ // If required, wait for graphics to finish drawing
+ if (options.draw_graphics && options.pedantic_graphics)
+ Barrier_Wait(&graphics_barrier);
+ System_Positions(s);
+ StepFunction(s);
- //printf("--------------------------------------------------------------\n");
-
- fclose(file);
-}
-
-/**
- * Cleans up the universe by freeing all memory
- */
-void Universe_Cleanup()
-{
- free(universe.body);
-
-}
+ if (options.draw_graphics && options.pedantic_graphics)
+ Barrier_ForceExit(&position_barrier); //Make the graphics continue
+ }
+ QuitProgram(false);
+ pthread_exit(NULL);
+ }
-/**
- * Thread - Continuously computes steps for a system of bodies
- */
-void * Compute_Thread(void * arg)
-{
+ #else // Code for the stupid way of doing it (respawn threads each step)
+ // (ie: The way I immediately implemented and didn't realise was stupid until someone told me)
- System * s = (System*)(arg); //cast argument to a System*
- // First set up the "sub_system" array
- #if NUM_THREADS > 1
- unsigned bodies_per_system = (s->N) / NUM_THREADS;
- unsigned remainder = (s->N) % NUM_THREADS;
- for (unsigned i = 0; i < NUM_THREADS; ++i)
- {
- sub_system[i].body = (s->body)+(i*bodies_per_system);
- sub_system[i].N = bodies_per_system;
- if (i == NUM_THREADS - 1)
- sub_system[i].N += remainder;
- }
- pthread_attr_t attr; //thread attribute for the workers
- pthread_attr_init(&attr);
- pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
+
- #endif
- while (true)
+ // Run until we can't run anymore
+ while (!ExitCondition())
{
- if (runstate != RUN) pthread_exit(NULL); //Check whether the main thread wants to exit
-
-
- #if NUM_THREADS <= 1
- //No worker threads; do everything in this thread
+ if (options.num_threads <= 1) // If there is only 1 worker thread...
+ {
+ // Just do everything in this thread
System_Forces(s, s);
+ // If required, wait for graphics to finish drawing
+ if (options.draw_graphics && options.pedantic_graphics)
+ Barrier_Join(&graphics_barrier);
+
System_Positions(s);
- #else
+ StepFunction(s);
-
+ if (options.draw_graphics && options.pedantic_graphics)
+ Barrier_Join(&position_barrier); //Make the graphics continue
+ continue;
+ }
- worker_count = NUM_THREADS; //All threads working
- //Compute forces
- for (unsigned i = 0; i < NUM_THREADS; ++i)
+ //Compute forces by spawning threads, each thread gets a sub system
+ for (unsigned i = 0; i < options.num_threads; ++i)
{
- if (pthread_create(worker_thread+i, &attr, Force_Thread, (void*)(sub_system+i)) != 0)
+ if (pthread_create(worker_thread+i, NULL, Force_Thread, (void*)(sub_system+i)) != 0)
{
perror("In compute thread, couldn't create worker thread (force)");
QuitProgram(true);
pthread_exit(NULL);
}
+
}
+ for (unsigned i = 0; i < options.num_threads; ++i)
+ pthread_join(worker_thread[i], NULL);
-
- //Wait for forces to be computed
- pthread_mutex_lock(&mutex_workers);
- while (worker_count > 0)
- pthread_cond_wait(&workers_done_cv, &mutex_workers);
- pthread_mutex_unlock(&mutex_workers);
- worker_count = NUM_THREADS; //All threads working
+ // If required, wait for graphics to finish drawing
+ if (options.draw_graphics && options.pedantic_graphics)
+ Barrier_Wait(&graphics_barrier);
- //Compute positions
- for (unsigned i = 0; i < NUM_THREADS; ++i)
+
+ Barrier_Enter(&position_barrier);
+
+ //Compute positions by spawning a bunch of threads to do it
+ for (unsigned i = 0; i < options.num_threads; ++i)
{
- if (pthread_create(worker_thread+i, &attr, Position_Thread, (void*)(sub_system+i)) != 0)
+ if (pthread_create(worker_thread+i, NULL, Position_Thread, (void*)(sub_system+i)) != 0)
{
perror("In compute thread, couldn't create worker thread (position)");
QuitProgram(true);
}
}
- //Wait for positions to be computed
- pthread_mutex_lock(&mutex_workers);
- while (worker_count > 0)
- pthread_cond_wait(&workers_done_cv, &mutex_workers);
- pthread_mutex_unlock(&mutex_workers);
+
+ for (unsigned i = 0; i < options.num_threads; ++i)
+ pthread_join(worker_thread[i], NULL);
+
+ StepFunction(s); // Execute single threaded stuff
- #endif
}
+ QuitProgram(false);
+ #endif //PERSISTENT_THREADS
+ return NULL;
}
-void * Force_Thread(void * s)
+/**
+ * @function BeforeDraw
+ * @purpose Called in graphics thread before the draw loop
+ * When --pedantic-graphics enabled, will wait for position computations to finish before drawing
+ * Otherwise does nothing
+ *
+ * This originally seemed like a good place to put the code now in StepFunction(), since only one thread runs this
+ * But then I realised that the graphics might be disabled,
+ * and there was no point having a thread that only existed to call that code.
+ *
+ * So I changed it to the horrible solution that I currently have.
+ */
+void BeforeDraw()
+{
+
+ //printf("BEFORE DRAW\n");
+ if (!options.pedantic_graphics)
+ return;
+
+ //printf("Graphics thread waits on position barrier\n");
+ Barrier_Wait(&position_barrier);
+ //printf("\tGraphics thread wakes up\n");
+ Barrier_Enter(&graphics_barrier);
+}
+
+/**
+ * @function AfterDraw
+ * @purpose Called in graphics thread after the draw loop
+ * When --pedantic-graphics is supplied, will signal computation threads that drawing is finished
+ * So that positions can be safely altered
+ * Otherwise does nothing
+ */
+void AfterDraw()
{
+ //universe.steps += 1;
+ if (!options.pedantic_graphics)
+ return;
+ Barrier_Join(&graphics_barrier);
- System_Forces((System*)s, &universe);
- pthread_mutex_lock(&mutex_workers);
- worker_count -= 1;
- if (worker_count == 0)
+}
+
+#ifdef PERSISTENT_THREADS
+
+/**
+ * @function Worker_Thread
+ * @purpose Thread - A self contained worker thread to compute a particular sub system of bodies
+ *
+ * This is the "smart" way to do it, because threads are only created once, and compute both force and position.
+ * The greatest difficulty with pthreads is getting a *single* thread from the team to execute certain code
+ * (ie: The stuff in StepFunction()).
+ * With the "continuously respawning threads of stupidity" approach,
+ * because there is one "master" thread (not necessarilly the main thread... don't get confused now)
+ * to keep respawning the workers, the single threaded code can just be executed in the master thread.
+ *
+ * With this approach, I have created a hacky solution so that the *last* thread to leave the position barrier gets to call StepFunction.
+ *
+ */
+void * Worker_Thread(void * arg)
+{
+ System * s = (System*)(arg); // This is mainly to save typing the RHS a lot of times
+
+ // Each thread runs until the whole program is supposed to end
+ while (!ExitCondition())
{
- pthread_cond_signal(&workers_done_cv);
+
+
+ System_Forces(s, &universe); // Each thread computes the forces for its share of bodies
+
+ // Do not confuse with "Barrier_Wait".
+ // Barrier_Wait does not affect the barrier; it just waits for it
+ // Barrier_Join actively updates the state of the barrier, and wakes up sleeping threads if required.
+
+ Barrier_Join(&force_barrier); // All threads must reach here before moving on.
+ if (ExitCondition()) return NULL;
+
+
+ //fprintf(stderr,"Thread %p - force barrier finished\n", arg);
+ //printf("Computed ALL forces\n");
+
+
+ // If required, wait for the graphics to finish drawing stuff
+ if (options.draw_graphics && options.pedantic_graphics)
+ {
+ //printf("Worker %p waits on graphics barrier\n", arg);
+ Barrier_Wait(&graphics_barrier);
+ //printf("\tWorker %p wakes up after graphics barrier\n", arg);
+ if (ExitCondition()) return NULL;
+ }
+
+
+
+ Barrier_Enter(&position_barrier);
+ System_Positions(s); // Each thread updates the positions for its share of bodies
+
+
+ // Barrier_JoinCall behaves in the same way as Barrier_Join, except the *last* thread
+ // (ie: the one that wakes up the others) also calls the function with arguments given.
+ Barrier_JoinCall(&position_barrier, StepFunction, (void*)(&universe));
+ if (ExitCondition()) return NULL;
+ //Barrier_Join(&position_barrier);
+
+ // All threads have computed positions, and *one* thread calls StepFunction()
+
}
- pthread_mutex_unlock(&mutex_workers);
+ QuitProgram(false); // Set the run state of the program
return NULL;
}
-void * Position_Thread(void * s)
+#endif //PERSISTENT_THREADS
+
+/**
+ * @function Force_Thread
+ * @purpose Thread - Calculates the forces on objects in a System
+ * @param s - Can be cast to the System*
+ */
+void * Force_Thread(void * s)
{
+ //System_ForcePair * pair = (System_ForcePair*)s;
+
- System_Positions((System*)s);
- pthread_mutex_lock(&mutex_workers);
- worker_count -= 1;
- if (worker_count == 0)
- {
- pthread_cond_signal(&workers_done_cv);
- }
- pthread_mutex_unlock(&mutex_workers);
+ System_Forces(s, &universe); //Simple wrapper
+ //printf("Force_Thread waits\n");
+
+ return NULL;
+}
+
+
+/**
+ * @function Position_Thread
+ * @purpose Thread - Calculates the positions of objects in a System
+ * @param s - Can be cast to the System*
+ */
+void * Position_Thread(void * s)
+{
+
+ System_Positions((System*)s); // Simple wrapper
+ Barrier_Join(&position_barrier); // This needed so that graphics will wait
+
return NULL;
}
/**
- * Child threads can call this to signal the main thread to quit the program
- * The main thread also uses this to tell child threads that the program is quitting
- * Note that the function doesn't call exit(), that is done by the main thread
+ * @function QuitProgram
+ * @purpose This function can be called in any thread to signal all threads to exit
+ * Repeated calls to this function have no effect
+ *
+ * All threads periodically call ExitCondition(), which will return true if the program should exit.
+ * One (not the only way) to return true is if this function has been called.
+ * Threads will call this function if they detect ExitCondition() is true. Only the first call has any effect.
*/
-void QuitProgram(bool error)
+inline void QuitProgram(bool error)
{
+ //If already quitting, don't do anything
+ if (runstate == QUIT || runstate == QUIT_ERROR)
+ return;
+
- pthread_mutex_lock(&mutex_runstate);
- if (error)
- runstate = QUIT_ERROR;
+
+ // set the runstate (checked in ExitCondition())
+
+ pthread_mutex_lock(&mutex_runstate); // aquire mutex
+ if (error)
+ runstate = QUIT_ERROR; // Program is exiting due to an error
else
- runstate = QUIT;
- pthread_mutex_unlock(&mutex_runstate);
+ runstate = QUIT; // Program is exiting naturally
+ pthread_mutex_unlock(&mutex_runstate); //release mutex
+
+
+}
+
+/**
+ * @function Thread_Cleanup
+ * @purpose Will be called in the *main* thread when exit() is called
+ * Ensures working threads will exit, and waits for them to finish.
+ * Then waits for them to finish.
+ * Also frees memory associated with the worker threads.
+ */
+void Thread_Cleanup(void)
+{
+
+
+ // Threads recheck the exit condition whenever they leave a barrier.
+ // These calls will stop any threads waiting forever in a barrier for threads that exited before getting to the barrier.
+ Barrier_ForceExit(&force_barrier);
+ Barrier_ForceExit(&position_barrier);
+
+
+ if (options.draw_graphics) // If the graphics are enabled...
+ {
+ // Then there is a computation thread, since graphics are done in the main thread
+ pthread_join(compute_thread, NULL);
+ }
+
+ #ifdef PERSISTENT_THREADS
+ for (unsigned i = 0; i < options.num_threads-1; ++i)
+ {
+ pthread_join(worker_thread[i], NULL);
+ }
+ #else
+ // All other worker threads (if they were spawned) are terminated in Compute_Thread
+ #endif //PERSISTENT_THREADS
+
+ // Scary memory management here.
+ if (worker_thread != NULL)
+ free(worker_thread);
+ if (sub_system != NULL)
+ free(sub_system);
+ worker_thread = NULL;
+ sub_system = NULL;
+
}
+/**
+ * @function Simulation_Run
+ * @purpose Initialise and start the simulation. Will be called in the main thread.
+ * Replaces the single-threaded macro that does nothing, and sets up the graphics and computation threads
+ * @param argc - Number of arguments - Passed to Graphics_Run if needed
+ * @param argv - Argument strings - Passed to Graphics_Run if needed
+ */
+void Simulation_Run(int argc, char ** argv)
+{
+ atexit(Thread_Cleanup);
+ if (options.draw_graphics) // The graphics are enabled
+ {
+ // I have chosen to do graphics in the main thread in this case.
+ // A *single* seperate thread is spawned here to do computations.
+ // This computation thread will spawn any additional worker threads required.
+ if (pthread_create(&compute_thread, NULL, Compute_Thread, (void*)&universe) != 0)
+ {
+ perror("Error creating compute thread");
+ exit(EXIT_FAILURE);
+ }
+
+ // This is run in the main thread
+ // It is effectively the graphics initialisation, followed by the glut loop
+ Graphics_Run(argc, argv);
+
+ // The main thread reaches here after leaving the glut loop when ExitCondition() returns true.
+
+ QuitProgram(false);
+
+ exit(EXIT_SUCCESS); // This is the main thread; use exit()
+
+ }
+ else //The graphics are disabled
+ {
+ // If graphics are disabled, there is no point spawning an extra thread.
+ // In this case, the *main* thread starts computations.
+ // Note that it will probably spawn additional worker threads (unless options.num_threads <= 1)
+ Compute_Thread((void*)(&universe));
+ QuitProgram(false);
+ exit(EXIT_SUCCESS);
+ }
+}
+
+
+
+
+
+/**
+ * @function Allocate_Threads
+ * @purpose Helper function to allocate an array of pthread_t objects
+ * Handles all the pointless, er, "important" error checking that should be done
+ * @param n - Number of threads in the array
+ *
+ * WARNING: Remember to free() the array!!!
+ */
+pthread_t * Allocate_Threads(unsigned n)
+{
+ pthread_t * result = (pthread_t*)(calloc(n, sizeof(pthread_t)));
+ if (result == NULL)
+ {
+ perror("Unable to allocate memory for threads");
+ QuitProgram(true);
+ pthread_exit(NULL);
+ }
+ return result;
+}
+
+/**
+ * @function StepFunction
+ * @purpose Helper to perform stuff in a single thread every step, after position computations are done
+ * The reason this has void* all over the place is so that I can pass the function pointer (to horrible dragons and fiendish demons).
+ * @param arg - Can be cast to System* for which steps are to be updated
+ * Will always be (void*)(&universe). But I have been brainwashed into the "global variables are baaaaad" philosophy.
+ * @returns arg
+ */
+void * StepFunction(void * arg)
+{
+ //fprintf(stderr, "StepFunction called\n");
+ System * s = (System*)(arg);
+ s->steps += 1; //Increment number of steps computed
+
+ if (options.verbosity != 0 && s->steps % options.verbosity == 0)
+ DisplayStatistics();
+
+
+ return arg;
+}
+
git.ucc.asn.au / matches / honours.git / blobdiff
UCC git Repository :: git.ucc.asn.au