/*
* InitGlobalMemory ()
*
* Args : none.
*
* Returns : nothing.
*
* Side Effects : Allocates all the global storage for G_Memory.
*
*/
void
InitGlobalMemory ()
{
int i;
G_Memory = (g_mem *) G_MALLOC(sizeof(g_mem));
G_Memory->i_array = (int *) G_MALLOC(Number_Of_Processors * sizeof(int));
G_Memory->d_array = (double *) G_MALLOC(Number_Of_Processors
* sizeof(double));
if (G_Memory == NULL) {
printf("Ran out of global memory in InitGlobalMemory\n");
exit(-1);
}
G_Memory->count = 0;
G_Memory->id = 0;
LOCKINIT(G_Memory->io_lock);
LOCKINIT(G_Memory->mal_lock);
LOCKINIT(G_Memory->single_lock);
LOCKINIT(G_Memory->count_lock);
ALOCKINIT(G_Memory->lock_array, MAX_LOCKS);
BARINIT(G_Memory->synch);
BARINIT(G_Memory->done_barrier);
G_Memory->max_x = -MAX_REAL;
G_Memory->min_x = MAX_REAL;
G_Memory->max_y = -MAX_REAL;
G_Memory->min_y = MAX_REAL;
}
int main(int argc, char *argv[])
{
errval_t err;
if (argc != 2) {
printf("Usage %s: <Num additional threads>\n", argv[0]);
exit(-1);
}
//printf("main running on %d\n", disp_get_core_id());
int cores = strtol(argv[1], NULL, 10) + 1;
NPROC = cores -1;
BARINIT(barrier, NPROC);
uint64_t before = rdtsc();
times[0] = before;
trace_event(TRACE_SUBSYS_BENCH, TRACE_EVENT_BENCH_PCBENCH, 1);
for (int i = 1; i < cores; i++) {
err = domain_new_dispatcher(i + disp_get_core_id(),
domain_spanned_callback,
(void*)(uintptr_t)i);
if (err_is_fail(err)) {
USER_PANIC_ERR(err, "domain_new_dispatcher failed");
}
}
while (ndispatchers < cores) {
thread_yield();
}
uint64_t finish = rdtsc();
trace_event(TRACE_SUBSYS_BENCH, TRACE_EVENT_BENCH_PCBENCH, 0);
//sys_print("\nDone\n", 6);
printf("spantest: Done in %"PRIu64" cycles\n", finish-before);
//trace_dump();
for(int i = 1; i < cores; i++) {
err = domain_thread_create_on(i, remote, NULL);
assert(err_is_ok(err));
}
messages_handler_loop();
return 0;
}
void init_global(long process_id)
{
/* Clear BSP root pointer */
global->index = 1; /* ****** */
global->bsp_root = 0 ;
LOCKINIT(global->index_lock);
LOCKINIT(global->bsp_tree_lock);
/* Initialize radiosity statistics variables */
LOCKINIT(global->avg_radiosity_lock);
global->converged = 0 ;
global->prev_total_energy.r = 0.0 ;
global->prev_total_energy.g = 0.0 ;
global->prev_total_energy.b = 0.0 ;
global->total_energy.r = 1.0 ;
global->total_energy.g = 1.0 ;
global->total_energy.b = 1.0 ;
global->total_patch_area = 1.0 ;
global->iteration_count = -1 ; /* init_ray_task() increments to 0 */
/* Initialize the cost sum */
LOCKINIT(global->cost_sum_lock);
global->cost_sum = 0 ;
global->cost_estimate_sum = 0 ;
/* Initialize the barrier */
BARINIT(global->barrier, n_processors);
LOCKINIT(global->pbar_lock);
global->pbar_count = 0 ;
/* Initialize task counter */
global->task_counter = 0 ;
LOCKINIT(global->task_counter_lock);
/* Initialize task queue */
init_taskq(process_id) ;
/* Initialize Patch, Element, Interaction free lists */
init_patchlist(process_id) ;
init_elemlist(process_id) ;
init_interactionlist(process_id) ;
init_elemvertex(process_id) ;
init_edge(process_id) ;
/* Initialize statistical info */
init_stat_info(process_id) ;
}
int main(int argc, char *argv[])
{
long i;
long j;
long xextra;
long xportion;
long yextra;
long yportion;
long lower;
double procsqrt;
long k;
long logtest;
long my_num;
unsigned long computeend;
double min_total;
double max_total;
double avg_total;
double min_multi;
double max_multi;
double avg_multi;
double min_frac;
double max_frac;
double avg_frac;
extern char *optarg;
long ch;
unsigned long start;
CLOCK(start)
while ((ch = getopt(argc, argv, "n:p:e:r:t:soh")) != -1) {
switch(ch) {
case 'n': im = atoi(optarg);
if (im > IMAX) {
printerr("Max grid size exceeded\n");
exit(-1);
}
if (log_2(im-2) == -1) {
printerr("Grid must be ((power of 2)+2) in each dimension\n");
exit(-1);
}
break;
case 'p': nprocs = atoi(optarg);
if (nprocs < 1) {
printerr("P must be >= 1\n");
exit(-1);
}
if (log_2(nprocs) == -1) {
printerr("P must be a power of 2\n");
exit(-1);
}
break;
case 'e': tolerance = atof(optarg); break;
case 'r': res = atof(optarg); break;
case 't': dtau = atof(optarg); break;
case 's': do_stats = !do_stats; break;
case 'o': do_output = !do_output; break;
case 'h': printf("Usage: OCEAN <options>\n\n");
printf("options:\n");
printf(" -nN : Simulate NxN ocean. N must be (power of 2)+2.\n");
printf(" -pP : P = number of processors. P must be power of 2.\n");
printf(" -eE : E = error tolerance for iterative relaxation.\n");
printf(" -rR : R = distance between grid points in meters.\n");
printf(" -tT : T = timestep in seconds.\n");
printf(" -s : Print timing statistics.\n");
printf(" -o : Print out relaxation residual values.\n");
printf(" -h : Print out command line options.\n\n");
printf("Default: OCEAN -n%1d -p%1d -e%1g -r%1g -t%1g\n",
DEFAULT_N,DEFAULT_P,DEFAULT_E,DEFAULT_R,DEFAULT_T);
exit(0);
break;
}
}
MAIN_INITENV(,60000000)
logtest = im-2;
numlev = 1;
while (logtest != 1) {
if (logtest%2 != 0) {
printerr("Cannot determine number of multigrid levels\n");
exit(-1);
}
logtest = logtest / 2;
numlev++;
}
if (numlev > MAX_LEVELS) {
printerr("Max grid levels exceeded for multigrid\n");
exit(-1);
}
jm = im;
printf("\n");
printf("Ocean simulation with W-cycle multigrid solver\n");
printf(" Processors : %1ld\n",nprocs);
printf(" Grid size : %1ld x %1ld\n",im,jm);
printf(" Grid resolution (meters) : %0.2f\n",res);
printf(" Time between relaxations (seconds) : %0.0f\n",dtau);
printf(" Error tolerance : %0.7g\n",tolerance);
printf("\n");
gp = (struct Global_Private *) G_MALLOC((nprocs+1)*sizeof(struct Global_Private));
for (i=0;i<nprocs;i++) {
gp[i].multi_time = 0;
gp[i].total_time = 0;
}
global = (struct global_struct *) G_MALLOC(sizeof(struct global_struct));
fields = (struct fields_struct *) G_MALLOC(sizeof(struct fields_struct));
fields2 = (struct fields2_struct *) G_MALLOC(sizeof(struct fields2_struct));
wrk1 = (struct wrk1_struct *) G_MALLOC(sizeof(struct wrk1_struct));
wrk3 = (struct wrk3_struct *) G_MALLOC(sizeof(struct wrk3_struct));
wrk2 = (struct wrk2_struct *) G_MALLOC(sizeof(struct wrk2_struct));
wrk4 = (struct wrk4_struct *) G_MALLOC(sizeof(struct wrk4_struct));
wrk6 = (struct wrk6_struct *) G_MALLOC(sizeof(struct wrk6_struct));
wrk5 = (struct wrk5_struct *) G_MALLOC(sizeof(struct wrk5_struct));
frcng = (struct frcng_struct *) G_MALLOC(sizeof(struct frcng_struct));
iter = (struct iter_struct *) G_MALLOC(sizeof(struct iter_struct));
guess = (struct guess_struct *) G_MALLOC(sizeof(struct guess_struct));
multi = (struct multi_struct *) G_MALLOC(sizeof(struct multi_struct));
locks = (struct locks_struct *) G_MALLOC(sizeof(struct locks_struct));
bars = (struct bars_struct *) G_MALLOC(sizeof(struct bars_struct));
LOCKINIT(locks->idlock)
LOCKINIT(locks->psiailock)
LOCKINIT(locks->psibilock)
LOCKINIT(locks->donelock)
LOCKINIT(locks->error_lock)
LOCKINIT(locks->bar_lock)
BARINIT(bars->iteration)
BARINIT(bars->gsudn)
BARINIT(bars->p_setup)
BARINIT(bars->p_redph)
BARINIT(bars->p_soln)
BARINIT(bars->p_subph)
BARINIT(bars->sl_prini)
BARINIT(bars->sl_psini)
BARINIT(bars->sl_onetime)
BARINIT(bars->sl_phase_1)
BARINIT(bars->sl_phase_2)
BARINIT(bars->sl_phase_3)
BARINIT(bars->sl_phase_4)
BARINIT(bars->sl_phase_5)
BARINIT(bars->sl_phase_6)
BARINIT(bars->sl_phase_7)
BARINIT(bars->sl_phase_8)
BARINIT(bars->sl_phase_9)
BARINIT(bars->sl_phase_10)
BARINIT(bars->error_barrier)
imx[numlev-1] = im;
jmx[numlev-1] = jm;
lev_res[numlev-1] = res;
lev_tol[numlev-1] = tolerance;
multi->err_multi = 0.0;
multi->numspin = 0;
for (i=0;i<nprocs;i++) {
multi->spinflag[i] = 0;
}
for (i=numlev-2;i>=0;i--) {
imx[i] = ((imx[i+1] - 2) / 2) + 2;
jmx[i] = ((jmx[i+1] - 2) / 2) + 2;
lev_res[i] = lev_res[i+1] * 2;
}
xprocs = 0;
yprocs = 0;
procsqrt = sqrt((double) nprocs);
j = (long) procsqrt;
while ((xprocs == 0) && (j > 0)) {
k = nprocs / j;
if (k * j == nprocs) {
if (k > j) {
xprocs = j;
yprocs = k;
} else {
xprocs = k;
yprocs = j;
}
}
j--;
}
if (xprocs == 0) {
printerr("Could not find factors for subblocking\n");
exit(-1);
}
/* Determine starting coord and number of points to process in */
/* each direction */
for (i=0;i<numlev;i++) {
xportion = (jmx[i] - 2) / xprocs;
xextra = (jmx[i] - 2) % xprocs;
for (j=0;j<xprocs;j++) {
if (xextra == 0) {
for (k=0;k<yprocs;k++) {
gp[k*xprocs+j].rel_start_x[i] = j * xportion + 1;
gp[k*xprocs+j].rel_num_x[i] = xportion;
}
} else {
if (j + 1 > xextra) {
for (k=0;k<yprocs;k++) {
lower = xextra * (xportion + 1);
gp[k*xprocs+j].rel_start_x[i] = lower + (j - xextra) * xportion + 1;
gp[k*xprocs+j].rel_num_x[i] = xportion;
}
} else {
for (k=0;k<yprocs;k++) {
gp[k*xprocs+j].rel_start_x[i] = j * (xportion + 1) + 1;
gp[k*xprocs+j].rel_num_x[i] = xportion + 1;
}
}
}
}
yportion = (imx[i] - 2) / yprocs;
yextra = (imx[i] - 2) % yprocs;
for (j=0;j<yprocs;j++) {
if (yextra == 0) {
for (k=0;k<xprocs;k++) {
gp[j*xprocs+k].rel_start_y[i] = j * yportion + 1;
gp[j*xprocs+k].rel_num_y[i] = yportion;
}
} else {
if (j + 1 > yextra) {
for (k=0;k<xprocs;k++) {
lower = yextra * (yportion + 1);
gp[j*xprocs+k].rel_start_y[i] = lower + (j - yextra) * yportion + 1;
gp[j*xprocs+k].rel_num_y[i] = yportion;
}
} else {
for (k=0;k<xprocs;k++) {
gp[j*xprocs+k].rel_start_y[i] = j * (yportion + 1) + 1;
gp[j*xprocs+k].rel_num_y[i] = yportion + 1;
}
}
}
}
}
i_int_coeff[0] = 0.0;
j_int_coeff[0] = 0.0;
for (i=0;i<numlev;i++) {
i_int_coeff[i] = 1.0/(imx[i]-1);
j_int_coeff[i] = 1.0/(jmx[i]-1);
}
for (my_num=0;my_num<nprocs;my_num++) {
for (i=0;i<numlev;i++) {
gp[my_num].rlist[i] = gp[my_num].rel_start_y[i];
gp[my_num].rljst[i] = gp[my_num].rel_start_x[i];
gp[my_num].rlien[i] = gp[my_num].rlist[i] + gp[my_num].rel_num_y[i] - 1;
gp[my_num].rljen[i] = gp[my_num].rljst[i] + gp[my_num].rel_num_x[i] - 1;
gp[my_num].iist[i] = gp[my_num].rel_start_y[i];
gp[my_num].ijst[i] = gp[my_num].rel_start_x[i];
gp[my_num].iien[i] = gp[my_num].iist[i] + gp[my_num].rel_num_y[i] - 1;
gp[my_num].ijen[i] = gp[my_num].ijst[i] + gp[my_num].rel_num_x[i] - 1;
gp[my_num].pist[i] = gp[my_num].rel_start_y[i];
gp[my_num].pjst[i] = gp[my_num].rel_start_x[i];
gp[my_num].pien[i] = gp[my_num].pist[i] + gp[my_num].rel_num_y[i] - 1;
gp[my_num].pjen[i] = gp[my_num].pjst[i] + gp[my_num].rel_num_x[i] - 1;
if (gp[my_num].pist[i] == 1) {
gp[my_num].pist[i] = 0;
}
if (gp[my_num].pjst[i] == 1) {
gp[my_num].pjst[i] = 0;
}
if (gp[my_num].pien[i] == imx[i] - 2) {
gp[my_num].pien[i] = imx[i]-1;
}
if (gp[my_num].pjen[i] == jmx[i] - 2) {
gp[my_num].pjen[i] = jmx[i]-1;
}
if (gp[my_num].rlist[i] % 2 == 0) {
gp[my_num].eist[i] = gp[my_num].rlist[i];
gp[my_num].oist[i] = gp[my_num].rlist[i] + 1;
} else {
gp[my_num].eist[i] = gp[my_num].rlist[i] + 1;
gp[my_num].oist[i] = gp[my_num].rlist[i];
}
if (gp[my_num].rljst[i] % 2 == 0) {
gp[my_num].ejst[i] = gp[my_num].rljst[i];
gp[my_num].ojst[i] = gp[my_num].rljst[i] + 1;
} else {
gp[my_num].ejst[i] = gp[my_num].rljst[i] + 1;
gp[my_num].ojst[i] = gp[my_num].rljst[i];
}
if (gp[my_num].rlien[i] == imx[i]-2) {
gp[my_num].rlien[i] = gp[my_num].rlien[i] - 1;
if (gp[my_num].rlien[i] % 2 == 0) {
gp[my_num].ojest[i] = gp[my_num].ojst[i];
gp[my_num].ejest[i] = gp[my_num].ejst[i];
} else {
gp[my_num].ojest[i] = gp[my_num].ejst[i];
gp[my_num].ejest[i] = gp[my_num].ojst[i];
}
}
if (gp[my_num].rljen[i] == jmx[i]-2) {
gp[my_num].rljen[i] = gp[my_num].rljen[i] - 1;
if (gp[my_num].rljen[i] % 2 == 0) {
gp[my_num].oiest[i] = gp[my_num].oist[i];
gp[my_num].eiest[i] = gp[my_num].eist[i];
} else {
gp[my_num].oiest[i] = gp[my_num].eist[i];
gp[my_num].eiest[i] = gp[my_num].oist[i];
}
}
}
}
/* initialize constants and variables
id is a global shared variable that has fetch-and-add operations
performed on it by processes to obtain their pids. */
global->id = 0;
global->psibi = 0.0;
pi = atan(1.0);
pi = 4.*pi;
factjacob = -1./(12.*res*res);
factlap = 1./(res*res);
eig2 = -h*f0*f0/(h1*h3*gpr);
jmm1 = jm-1 ;
ysca = ((double) jmm1)*res ;
for (i=0;i<im;i++) {
for (j=0;j<jm;j++) {
guess->oldga[i][j] = 0.0;
guess->oldgb[i][j] = 0.0;
}
}
if (do_output) {
printf(" MULTIGRID OUTPUTS\n");
}
CREATE(slave, nprocs);
WAIT_FOR_END(nprocs);
CLOCK(computeend)
printf("\n");
printf(" PROCESS STATISTICS\n");
printf(" Total Multigrid Multigrid\n");
printf(" Proc Time Time Fraction\n");
printf(" 0 %15.0f %15.0f %10.3f\n", gp[0].total_time,gp[0].multi_time, gp[0].multi_time/gp[0].total_time);
if (do_stats) {
min_total = max_total = avg_total = gp[0].total_time;
min_multi = max_multi = avg_multi = gp[0].multi_time;
min_frac = max_frac = avg_frac = gp[0].multi_time/gp[0].total_time;
for (i=1;i<nprocs;i++) {
if (gp[i].total_time > max_total) {
max_total = gp[i].total_time;
}
if (gp[i].total_time < min_total) {
min_total = gp[i].total_time;
}
if (gp[i].multi_time > max_multi) {
max_multi = gp[i].multi_time;
}
if (gp[i].multi_time < min_multi) {
min_multi = gp[i].multi_time;
}
if (gp[i].multi_time/gp[i].total_time > max_frac) {
max_frac = gp[i].multi_time/gp[i].total_time;
}
if (gp[i].multi_time/gp[i].total_time < min_frac) {
min_frac = gp[i].multi_time/gp[i].total_time;
}
avg_total += gp[i].total_time;
avg_multi += gp[i].multi_time;
avg_frac += gp[i].multi_time/gp[i].total_time;
}
avg_total = avg_total / nprocs;
avg_multi = avg_multi / nprocs;
avg_frac = avg_frac / nprocs;
for (i=1;i<nprocs;i++) {
printf(" %3ld %15.0f %15.0f %10.3f\n", i, gp[i].total_time, gp[i].multi_time, gp[i].multi_time/gp[i].total_time);
}
printf(" Avg %15.0f %15.0f %10.3f\n", avg_total,avg_multi,avg_frac);
printf(" Min %15.0f %15.0f %10.3f\n", min_total,min_multi,min_frac);
printf(" Max %15.0f %15.0f %10.3f\n", max_total,max_multi,max_frac);
}
printf("\n");
global->starttime = start;
printf(" TIMING INFORMATION\n");
printf("Start time : %16lu\n", global->starttime);
printf("Initialization finish time : %16lu\n", global->trackstart);
printf("Overall finish time : %16lu\n", computeend);
printf("Total time with initialization : %16lu\n", computeend-global->starttime);
printf("Total time without initialization : %16lu\n", computeend-global->trackstart);
printf(" (excludes first timestep)\n");
printf("\n");
MAIN_END
}
int main(int argc, CHAR *argv[])
{
INT i;
UINT begin;
UINT end;
UINT lapsed;
MATRIX vtrans, Vinv; /* View transformation and inverse. */
/*
* First, process command line arguments.
*/
i = 1;
while ((i < argc) && (argv[i][0] == '-')) {
switch (argv[i][1]) {
case '?':
case 'h':
case 'H':
Usage();
exit(1);
case 'a':
case 'A':
AntiAlias = TRUE;
if (argv[i][2] != '\0') {
NumSubRays = atoi(&argv[i][2]);
} else {
NumSubRays = atoi(&argv[++i][0]);
}
break;
case 'm':
if (argv[i][2] != '\0') {
MaxGlobMem = atoi(&argv[i][2]);
} else {
MaxGlobMem = atoi(&argv[++i][0]);
}
break;
case 'p':
if (argv[i][2] != '\0') {
nprocs = atoi(&argv[i][2]);
} else {
nprocs = atoi(&argv[++i][0]);
}
break;
case 's':
case 'S':
dostats = TRUE;
break;
default:
fprintf(stderr, "%s: Invalid option \'%c\'.\n", ProgName, argv[i][0]);
exit(1);
}
i++;
}
if (i == argc) {
Usage();
exit(1);
}
/*
* Make sure nprocs is within valid range.
*/
if (nprocs < 1 || nprocs > MAX_PROCS)
{
fprintf(stderr, "%s: Valid range for #processors is [1, %d].\n", ProgName, MAX_PROCS);
exit(1);
}
/*
* Print command line parameters.
*/
printf("\n");
printf("Number of processors: \t%ld\n", nprocs);
printf("Global shared memory size:\t%ld MB\n", MaxGlobMem);
printf("Samples per pixel: \t%ld\n", NumSubRays);
printf("\n");
/*
* Initialize the shared memory environment and request the total
* amount of amount of shared memory we might need. This
* includes memory for the database, grid, and framebuffer.
*/
MaxGlobMem <<= 20; /* Convert MB to bytes. */
MAIN_INITENV(,MaxGlobMem + 512*1024)
THREAD_INIT_FREE();
gm = (GMEM *)G_MALLOC(sizeof(GMEM));
/*
* Perform shared environment initializations.
*/
gm->nprocs = nprocs;
gm->pid = 0;
gm->rid = 1;
BARINIT(gm->start, nprocs)
LOCKINIT(gm->pidlock)
LOCKINIT(gm->ridlock)
LOCKINIT(gm->memlock)
ALOCKINIT(gm->wplock, nprocs)
/* POSSIBLE ENHANCEMENT: Here is where one might distribute the
raystruct data structure across physically distributed memories as
desired. */
if (!GlobalHeapInit(MaxGlobMem))
{
fprintf(stderr, "%s: Cannot initialize global heap.\n", ProgName);
exit(1);
}
/*
* Initialize HUG parameters, read environment and geometry files.
*/
Huniform_defaults();
ReadEnvFile(/* *argv*/argv[i]);
ReadGeoFile(GeoFileName);
OpenFrameBuffer();
/*
* Compute view transform and its inverse.
*/
CreateViewMatrix();
MatrixCopy(vtrans, View.vtrans);
MatrixInverse(Vinv, vtrans);
MatrixCopy(View.vtransInv, Vinv);
/*
* Print out what we have so far.
*/
printf("Number of primitive objects: \t%ld\n", prim_obj_cnt);
printf("Number of primitive elements:\t%ld\n", prim_elem_cnt);
/*
* Preprocess database into hierarchical uniform grid.
*/
if (TraversalType == TT_HUG)
BuildHierarchy_Uniform();
/*
* Now create slave processes.
*/
CLOCK(begin)
CREATE(StartRayTrace, gm->nprocs);
WAIT_FOR_END(gm->nprocs);
CLOCK(end)
/*
* We are finished. Clean up, print statistics and run time.
*/
CloseFrameBuffer(PicFileName);
PrintStatistics();
lapsed = (end - begin) & 0x7FFFFFFF;
printf("TIMING STATISTICS MEASURED BY MAIN PROCESS:\n");
printf(" Overall start time %20lu\n", begin);
printf(" Overall end time %20lu\n", end);
printf(" Total time with initialization %20lu\n", lapsed);
printf(" Total time without initialization %20lu\n", end - gm->par_start_time);
if (dostats) {
unsigned totalproctime, maxproctime, minproctime;
printf("\n\n\nPER-PROCESS STATISTICS:\n");
printf("%20s%20s\n","Proc","Time");
printf("%20s%20s\n\n","","Tracing Rays");
for (i = 0; i < gm->nprocs; i++)
printf("%20ld%20ld\n",i,gm->partime[i]);
totalproctime = gm->partime[0];
minproctime = gm->partime[0];
maxproctime = gm->partime[0];
for (i = 1; i < gm->nprocs; i++) {
totalproctime += gm->partime[i];
if (gm->partime[i] > maxproctime)
maxproctime = gm->partime[i];
if (gm->partime[i] < minproctime)
minproctime = gm->partime[i];
}
printf("\n\n%20s%20d\n","Max = ",maxproctime);
printf("%20s%20d\n","Min = ",minproctime);
printf("%20s%20d\n","Avg = ",(int) (((double) totalproctime) / ((double) (1.0 * gm->nprocs))));
}
MAIN_END
}
int main(int argc, char *argv[])
{
long i;
long c;
extern char *optarg;
long m1;
long factor;
long pages;
unsigned long start;
CLOCK(start);
while ((c = getopt(argc, argv, "p:m:n:l:stoh")) != -1) {
switch(c) {
case 'p': P = atoi(optarg);
if (P < 1) {
printerr("P must be >= 1\n");
exit(-1);
}
if (log_2(P) == -1) {
printerr("P must be a power of 2\n");
exit(-1);
}
break;
case 'm': M = atoi(optarg);
m1 = M/2;
if (2*m1 != M) {
printerr("M must be even\n");
exit(-1);
}
break;
case 'n': num_cache_lines = atoi(optarg);
orig_num_lines = num_cache_lines;
if (num_cache_lines < 1) {
printerr("Number of cache lines must be >= 1\n");
exit(-1);
}
break;
case 'l': log2_line_size = atoi(optarg);
if (log2_line_size < 0) {
printerr("Log base 2 of cache line length in bytes must be >= 0\n");
exit(-1);
}
break;
case 's': dostats = !dostats;
break;
case 't': test_result = !test_result;
break;
case 'o': doprint = !doprint;
break;
case 'h': printf("Usage: FFT <options>\n\n");
printf("options:\n");
printf(" -mM : M = even integer; 2**M total complex data points transformed.\n");
printf(" -pP : P = number of processors; Must be a power of 2.\n");
printf(" -nN : N = number of cache lines.\n");
printf(" -lL : L = Log base 2 of cache line length in bytes.\n");
printf(" -s : Print individual processor timing statistics.\n");
printf(" -t : Perform FFT and inverse FFT. Test output by comparing the\n");
printf(" integral of the original data to the integral of the data that\n");
printf(" results from performing the FFT and inverse FFT.\n");
printf(" -o : Print out complex data points.\n");
printf(" -h : Print out command line options.\n\n");
printf("Default: FFT -m%1d -p%1d -n%1d -l%1d\n",
DEFAULT_M,DEFAULT_P,NUM_CACHE_LINES,LOG2_LINE_SIZE);
exit(0);
break;
}
}
MAIN_INITENV(,80000000);
N = 1<<M;
rootN = 1<<(M/2);
rowsperproc = rootN/P;
if (rowsperproc == 0) {
printerr("Matrix not large enough. 2**(M/2) must be >= P\n");
exit(-1);
}
line_size = 1 << log2_line_size;
if (line_size < 2*sizeof(double)) {
printf("WARNING: Each element is a complex double (%ld bytes)\n",2*sizeof(double));
printf(" => Less than one element per cache line\n");
printf(" Computing transpose blocking factor\n");
factor = (2*sizeof(double)) / line_size;
num_cache_lines = orig_num_lines / factor;
}
if (line_size <= 2*sizeof(double)) {
pad_length = 1;
} else {
pad_length = line_size / (2*sizeof(double));
}
if (rowsperproc * rootN * 2 * sizeof(double) >= PAGE_SIZE) {
pages = (2 * pad_length * sizeof(double) * rowsperproc) / PAGE_SIZE;
if (pages * PAGE_SIZE != 2 * pad_length * sizeof(double) * rowsperproc) {
pages ++;
}
pad_length = (pages * PAGE_SIZE) / (2 * sizeof(double) * rowsperproc);
} else {
pad_length = (PAGE_SIZE - (rowsperproc * rootN * 2 * sizeof(double))) /
(2 * sizeof(double) * rowsperproc);
if (pad_length * (2 * sizeof(double) * rowsperproc) !=
(PAGE_SIZE - (rowsperproc * rootN * 2 * sizeof(double)))) {
printerr("Padding algorithm unsuccessful\n");
exit(-1);
}
}
Global = (struct GlobalMemory *) G_MALLOC(sizeof(struct GlobalMemory));
x = (double *) G_MALLOC(2*(N+rootN*pad_length)*sizeof(double)+PAGE_SIZE);
trans = (double *) G_MALLOC(2*(N+rootN*pad_length)*sizeof(double)+PAGE_SIZE);
umain = (double *) G_MALLOC(2*rootN*sizeof(double));
umain2 = (double *) G_MALLOC(2*(N+rootN*pad_length)*sizeof(double)+PAGE_SIZE);
Global->transtimes = (long *) G_MALLOC(P*sizeof(long));
Global->totaltimes = (long *) G_MALLOC(P*sizeof(long));
if (Global == NULL) {
printerr("Could not malloc memory for Global\n");
exit(-1);
} else if (x == NULL) {
printerr("Could not malloc memory for x\n");
exit(-1);
} else if (trans == NULL) {
printerr("Could not malloc memory for trans\n");
exit(-1);
} else if (umain == NULL) {
printerr("Could not malloc memory for umain\n");
exit(-1);
} else if (umain2 == NULL) {
printerr("Could not malloc memory for umain2\n");
exit(-1);
}
x = (double *) (((unsigned long) x) + PAGE_SIZE - ((unsigned long) x) % PAGE_SIZE);
trans = (double *) (((unsigned long) trans) + PAGE_SIZE - ((unsigned long) trans) % PAGE_SIZE);
umain2 = (double *) (((unsigned long) umain2) + PAGE_SIZE - ((unsigned long) umain2) % PAGE_SIZE);
/* In order to optimize data distribution, the data structures x, trans,
and umain2 have been aligned so that each begins on a page boundary.
This ensures that the amount of padding calculated by the program is
such that each processor's partition ends on a page boundary, thus
ensuring that all data from these structures that are needed by a
processor can be allocated to its local memory */
/* POSSIBLE ENHANCEMENT: Here is where one might distribute the x,
trans, and umain2 data structures across physically distributed
memories as desired.
One way to place data is as follows:
double *base;
long i;
i = ((N/P)+(rootN/P)*pad_length)*2;
base = &(x[0]);
for (j=0;j<P;j++) {
Place all addresses x such that (base <= x < base+i) on node j
base += i;
}
The trans and umain2 data structures can be placed in a similar manner.
*/
printf("\n");
printf("FFT with Blocking Transpose\n");
printf(" %ld Complex Doubles\n",N);
printf(" %ld Processors\n",P);
if (num_cache_lines != orig_num_lines) {
printf(" %ld Cache lines\n",orig_num_lines);
printf(" %ld Cache lines for blocking transpose\n",num_cache_lines);
} else {
printf(" %ld Cache lines\n",num_cache_lines);
}
printf(" %d Byte line size\n",(1 << log2_line_size));
printf(" %d Bytes per page\n",PAGE_SIZE);
printf("\n");
BARINIT(Global->start, P);
LOCKINIT(Global->idlock);
Global->id = 0;
InitX(x); /* place random values in x */
if (test_result) {
ck1 = CheckSum(x);
}
if (doprint) {
printf("Original data values:\n");
PrintArray(N, x);
}
InitU(N,umain); /* initialize u arrays*/
InitU2(N,umain2,rootN);
/* fire off P processes */
CREATE(SlaveStart, P);
WAIT_FOR_END(P);
if (doprint) {
if (test_result) {
printf("Data values after inverse FFT:\n");
} else {
printf("Data values after FFT:\n");
}
PrintArray(N, x);
}
transtime = Global->transtimes[0];
printf("\n");
printf(" PROCESS STATISTICS\n");
printf(" Computation Transpose Transpose\n");
printf(" Proc Time Time Fraction\n");
printf(" 0 %10ld %10ld %8.5f\n",
Global->totaltimes[0],Global->transtimes[0],
((double)Global->transtimes[0])/Global->totaltimes[0]);
if (dostats) {
transtime2 = Global->transtimes[0];
avgtranstime = Global->transtimes[0];
avgcomptime = Global->totaltimes[0];
maxtotal = Global->totaltimes[0];
mintotal = Global->totaltimes[0];
maxfrac = ((double)Global->transtimes[0])/Global->totaltimes[0];
minfrac = ((double)Global->transtimes[0])/Global->totaltimes[0];
avgfractime = ((double)Global->transtimes[0])/Global->totaltimes[0];
for (i=1;i<P;i++) {
if (Global->transtimes[i] > transtime) {
transtime = Global->transtimes[i];
}
if (Global->transtimes[i] < transtime2) {
transtime2 = Global->transtimes[i];
}
if (Global->totaltimes[i] > maxtotal) {
maxtotal = Global->totaltimes[i];
}
if (Global->totaltimes[i] < mintotal) {
mintotal = Global->totaltimes[i];
}
if (((double)Global->transtimes[i])/Global->totaltimes[i] > maxfrac) {
maxfrac = ((double)Global->transtimes[i])/Global->totaltimes[i];
}
if (((double)Global->transtimes[i])/Global->totaltimes[i] < minfrac) {
minfrac = ((double)Global->transtimes[i])/Global->totaltimes[i];
}
printf(" %3ld %10ld %10ld %8.5f\n",
i,Global->totaltimes[i],Global->transtimes[i],
((double)Global->transtimes[i])/Global->totaltimes[i]);
avgtranstime += Global->transtimes[i];
avgcomptime += Global->totaltimes[i];
avgfractime += ((double)Global->transtimes[i])/Global->totaltimes[i];
}
printf(" Avg %10.0f %10.0f %8.5f\n",
((double) avgcomptime)/P,((double) avgtranstime)/P,avgfractime/P);
printf(" Max %10ld %10ld %8.5f\n",
maxtotal,transtime,maxfrac);
printf(" Min %10ld %10ld %8.5f\n",
mintotal,transtime2,minfrac);
}
Global->starttime = start;
printf("\n");
printf(" TIMING INFORMATION\n");
printf("Start time : %16lu\n",
Global->starttime);
printf("Initialization finish time : %16lu\n",
Global->initdonetime);
printf("Overall finish time : %16lu\n",
Global->finishtime);
printf("Total time with initialization : %16lu\n",
Global->finishtime-Global->starttime);
printf("Total time without initialization : %16lu\n",
Global->finishtime-Global->initdonetime);
printf("Overall transpose time : %16ld\n",
transtime);
printf("Overall transpose fraction : %16.5f\n",
((double) transtime)/(Global->finishtime-Global->initdonetime));
printf("\n");
if (test_result) {
ck3 = CheckSum(x);
printf(" INVERSE FFT TEST RESULTS\n");
printf("Checksum difference is %.3f (%.3f, %.3f)\n",
ck1-ck3, ck1, ck3);
if (fabs(ck1-ck3) < 0.001) {
printf("TEST PASSED\n");
} else {
printf("TEST FAILED\n");
}
}
MAIN_END;
}
int
main(
int argc, /* arg count */
char * argv[] /* arg vector */
){
static char * context = "main(life)";
bool2D world; /* world to evolve */
int nr, nc; /* matrix size */
int iters; /* number of iterations */
char * infn = NULL; /* input file name */
char * outfn = NULL; /* output file name */
int argd = 1; /* argument index */
void * args[5];
/* arguments */
#if NUMA
MAIN_INITENV(,32000000)
BARINIT(GlobalBar);
#endif
while (argd < argc){
CHECK(argv[argd][0] == '-',
fail(context, "bad argument", "index", "%d", argd, NULL));
switch(argv[argd][1]){
case 'L' :
iters = arg_int(context, argc, argv, argd+1, argv[argd]);
argd += 2;
break;
#if GRAPHICS
case 'g' :
gfx_open(app_life, arg_gfxCtrl(context, argc, argv, argd+1, argv[argd]));
argd += 2;
break;
#endif
#if PARALLEL
case 'p' :
DataDist = arg_dataDist(context, argc, argv, argd+1, argv[argd]);
ParWidth = arg_int(context, argc, argv, argd+2, argv[argd]);
argd += 3;
break;
#endif
case 'i' :
infn = arg_str(context, argc, argv, argd+1, argv[argd]);
argd += 2;
break;
case 'o' :
outfn = arg_str(context, argc, argv, argd+1, argv[argd]);
argd += 2;
break;
case 'u' :
io_init(FALSE);
argd += 1;
break;
default :
fail(context, "unknown flag", "flag", "%s", argv[argd], NULL);
break;
}
}
/* setup */
sch_init(DataDist);
CHECK(0 < iters,
fail(context, "non-positive number of iterations",
"number of iterations", "%d", iters, NULL));
io_rdBool2D(context, infn, world, &nr, &nc);
/* run */
TP_any(args, 0, world);
TP_any(args, 1, nr);
TP_any(args, 2, nc);
TP_any(args, 3, iters);
thr_grp(life_thr, args);
/* takedown */
io_wrBool2D(context, outfn, world, nr, nc);
#if GRAPHICS
gfx_close();
#endif
#if IEEE
ieee_retrospective(stderr);
#endif
#if NUMA
BARFREE(GlobalBar);
MAIN_END;
#endif
return 0;
}
int main(int argc, char **argv)
{
/* default values for the control parameters of the driver */
/* are in parameters.h */
if ((argc == 2) && ((strncmp(argv[1],"-h",strlen("-h")) == 0) || (strncmp(argv[1],"-H",strlen("-H")) == 0))) {
printf("Usage: WATER-SPATIAL < infile, where the contents of infile can be\nobtained from the comments at the top of water.C and the first scanf \nin main() in water.C\n\n");
exit(0);
}
#else
int main(void)
{
#endif
/* POSSIBLE ENHANCEMENT: One might bind the first process to a processor
here, even before the other (child) processes are bound later in mdmain().
*/
six = stdout;
TEMP =298.0;
RHO =0.9980;
/* read input */
#ifndef SIM_SOCLIB
if (scanf("%lf%ld%ld%ld%ld%ld%ld%ld%ld%lf",&TSTEP, &NMOL, &NSTEP, &NORDER, &NSAVE, &NRST, &NPRINT, &NFMC,&NumProcs, &CUTOFF) != 10)
fprintf(stderr,"ERROR: Usage: water < infile, which must have 10 fields, see SPLASH documentation or comment at top of water.C\n");
#else
TSTEP = 1.5e-16;
NMOL = NMOLS;
NSTEP = 3;
NORDER = 6;
NSAVE = -1 ;
NRST = 3000 ;
NPRINT = 3 ;
NFMC = 0;
NumProcs = NB_P;
CUTOFF = 6.212752;
#endif
printf("Using %ld procs on %ld steps of %ld mols\n", NumProcs, NSTEP, NMOL);
printf("Other parameters:\n\tTSTEP = %8.2e\n\tNORDER = %ld\n\tNSAVE = %ld\n",TSTEP,NORDER,NSAVE);
printf("\tNRST = %ld\n\tNPRINT = %ld\n\tNFMC = %ld\n\tCUTOFF = %lf\n\n",NRST,NPRINT,NFMC,CUTOFF);
/* set up scaling factors and constants */
NORD1=NORDER+1;
CNSTNT(NORD1,TLC); /* sub. call to set up constants */
SYSCNS(); /* sub. call to initialize system constants */
printf("%ld boxes with %ld processors\n\n",
BOX_PER_SIDE * BOX_PER_SIDE * BOX_PER_SIDE, NumProcs);
if (NumProcs > (BOX_PER_SIDE * BOX_PER_SIDE * BOX_PER_SIDE)) {
fprintf(stderr,"ERROR: less boxes (%ld) than processors (%ld)\n",
BOX_PER_SIDE * BOX_PER_SIDE * BOX_PER_SIDE, NumProcs);
fflush(stderr);
exit(-1);
}
fprintf(six,"\nTEMPERATURE = %8.2f K\n",TEMP);
fprintf(six,"DENSITY = %8.5f G/C.C.\n",RHO);
fprintf(six,"NUMBER OF MOLECULES = %8ld\n",NMOL);
fprintf(six,"NUMBER OF PROCESSORS = %8ld\n",NumProcs);
fprintf(six,"TIME STEP = %8.2e SEC\n",TSTEP);
fprintf(six,"ORDER USED TO SOLVE F=MA = %8ld \n",NORDER);
fprintf(six,"NO. OF TIME STEPS = %8ld \n",NSTEP);
fprintf(six,"FREQUENCY OF DATA SAVING = %8ld \n",NSAVE);
fprintf(six,"FREQUENCY TO WRITE RST FILE= %8ld \n",NRST);
fflush(six);
{ /* do memory initializations */
long procnum, i, j, k, l;
struct list_of_boxes *temp_box;
long xprocs, yprocs, zprocs;
long x_inc, y_inc, z_inc;
long x_ct, y_ct, z_ct;
long x_left, y_left, z_left;
long x_first, y_first, z_first;
long x_last, y_last, z_last;
double proccbrt;
long gmem_size = sizeof(struct GlobalMemory);
MAIN_INITENV((NumProcs),40000000,); /* macro call to initialize
shared memory etc. */
/* Allocate space for main (BOX) data structure as well as
* synchronization variables
*/
start_end = (first_last_array **)
G_MALLOC(sizeof(first_last_array *) * NumProcs);
for (i=0; i < NumProcs; i++) {
start_end[i] = (first_last_array *)
G_MALLOC(sizeof(first_last_array));
}
/* Calculate start and finish box numbers for processors */
xprocs = 0;
yprocs = 0;
proccbrt = (double) pow((double) NumProcs, 1.0/3.0) + 0.00000000000001;
j = (long) proccbrt;
if (j<1) j = 1;
while ((xprocs == 0) && (j>0)) {
k = (long) sqrt((double) (NumProcs / j));
if (k<1) k=1;
while ((yprocs == 0) && (k>0)) {
l = NumProcs/(j*k);
if ((j*k*l) == NumProcs) {
xprocs = j;
yprocs = k;
zprocs = l;
} /* if */
k--;
} /* while yprocs && k */
j--;
} /* while xprocs && j */
printf("xprocs = %ld\typrocs = %ld\tzprocs = %ld\n",
xprocs, yprocs, zprocs);
fflush(stdout);
/* Fill in start_end array values */
procnum = 0;
x_inc = BOX_PER_SIDE/xprocs;
y_inc = BOX_PER_SIDE/yprocs;
z_inc = BOX_PER_SIDE/zprocs;
x_left = BOX_PER_SIDE - (xprocs*x_inc);
y_left = BOX_PER_SIDE - (yprocs*y_inc);
z_left = BOX_PER_SIDE - (zprocs*z_inc);
printf("x_inc = %ld\t y_inc = %ld\t z_inc = %ld\n",x_inc,y_inc,z_inc);
printf("x_left = %ld\t y_left = %ld\t z_left = %ld\n",x_left,y_left,z_left);
fflush(stdout);
x_first = 0;
x_ct = x_left;
x_last = -1;
x_inc++;
for (i=0; i<xprocs; i++) {
y_ct = y_left;
if (x_ct == 0) x_inc--;
x_last += x_inc;
y_first = 0;
y_last = -1;
y_inc++;
for (j=0; j<yprocs; j++) {
z_ct = z_left;
if (y_ct == 0) y_inc--;
y_last += y_inc;
z_first = 0;
z_last = -1;
z_inc++;
for (k=0; k<zprocs; k++) {
if (z_ct == 0) z_inc--;
z_last += z_inc;
start_end[procnum]->box[XDIR][FIRST] = x_first;
start_end[procnum]->box[XDIR][LAST] =
min(x_last, BOX_PER_SIDE - 1);
start_end[procnum]->box[YDIR][FIRST] = y_first;
start_end[procnum]->box[YDIR][LAST] =
min(y_last, BOX_PER_SIDE - 1);
start_end[procnum]->box[ZDIR][FIRST] = z_first;
start_end[procnum]->box[ZDIR][LAST] =
min(z_last, BOX_PER_SIDE - 1);
z_first = z_last + 1;
z_ct--;
procnum++;
}
y_first = y_last + 1;
y_ct--;
}
x_first = x_last + 1;
x_ct--;
}
/* Allocate space for my_boxes array */
my_boxes = (box_list **) G_MALLOC(NumProcs * sizeof(box_list *));
/* Set all box ptrs to null */
for (i=0; i<NumProcs; i++) my_boxes[i] = NULL;
/* Set up links for all boxes for initial interf and intraf */
temp_box = my_boxes[0];
while (temp_box) {
temp_box = temp_box->next_box;
}
/* Allocate space for BOX array */
BOX = (box_type ***) G_MALLOC(BOX_PER_SIDE * sizeof(box_type **));
for (i=0; i < BOX_PER_SIDE; i++) {
BOX[i] = (box_type **) G_MALLOC( BOX_PER_SIDE * sizeof(box_type *));
for (j=0; j < BOX_PER_SIDE; j++) {
BOX[i][j] = (box_type *) G_MALLOC(BOX_PER_SIDE * sizeof(box_type));
for (k=0; k < BOX_PER_SIDE; k++) {
BOX[i][j][k].list = NULL;
LOCKINIT(BOX[i][j][k].boxlock);
}
}
} /* for i */
gl = (struct GlobalMemory *) G_MALLOC(gmem_size);
/* macro calls to initialize synch variables */
BARINIT(gl->start, NumProcs);
BARINIT(gl->InterfBar, NumProcs);
BARINIT(gl->PotengBar, NumProcs);
LOCKINIT(gl->IOLock);
LOCKINIT(gl->IndexLock);
LOCKINIT(gl->IntrafVirLock);
LOCKINIT(gl->InterfVirLock);
LOCKINIT(gl->KinetiSumLock);
LOCKINIT(gl->PotengSumLock);
}
fprintf(six,"SPHERICAL CUTOFF RADIUS = %8.4f ANGSTROM\n",CUTOFF);
fflush(six);
IRST=0;
/* call initialization routine */
INITIA();
gl->tracktime = 0;
gl->intratime = 0;
gl->intertime = 0;
/* initialize Index to 1 so that the first created child gets
id 1, not 0 */
gl->Index = 1;
if (NSAVE > 0) { /* not true for input decks provided */
fprintf(six,"COLLECTING X AND V DATA AT EVERY %4ld TIME STEPS \n",NSAVE);
}
/* spawn helper processes */
CLOCK(gl->computestart);
CREATE(WorkStart, NumProcs);
/* macro to make main process wait for all others to finish */
WAIT_FOR_END(NumProcs);
CLOCK(gl->computeend);
printf("COMPUTESTART (after initialization) = %lu\n",gl->computestart);
printf("COMPUTEEND = %lu\n",gl->computeend);
printf("COMPUTETIME (after initialization) = %lu\n",gl->computeend-gl->computestart);
printf("Measured Time (2nd timestep onward) = %lu\n",gl->tracktime);
printf("Intramolecular time only (2nd timestep onward) = %lu\n",gl->intratime);
printf("Intermolecular time only (2nd timestep onward) = %lu\n",gl->intertime);
printf("Other time (2nd timestep onward) = %lu\n",gl->tracktime - gl->intratime - gl->intertime);
printf("\nExited Happily with XTT = %g (note: XTT value is garbage if NPRINT > NSTEP)\n", XTT);
MAIN_END;
} /* main.c */
void Frame()
{
long starttime,stoptime,exectime,i;
Init_Options();
printf("*****Entering init_decomposition with num_nodes = %ld\n",num_nodes);
fflush(stdout);
Init_Decomposition();
printf("*****Exited init_decomposition with num_nodes = %ld\n",num_nodes);
fflush(stdout);
Global = (struct GlobalMemory *)NU_MALLOC(sizeof(struct GlobalMemory),0);
BARINIT(Global->SlaveBarrier, num_nodes);
BARINIT(Global->TimeBarrier, num_nodes);
LOCKINIT(Global->IndexLock);
LOCKINIT(Global->CountLock);
ALOCKINIT(Global->QLock,MAX_NUMPROC+1);
/* load dataset from file to each node */
#ifndef RENDER_ONLY
CLOCK(starttime);
Load_Map(filename);
CLOCK(stoptime);
mclock(stoptime,starttime,&exectime);
printf("wall clock execution time to load map: %lu ms\n", exectime);
#endif
CLOCK(starttime);
#ifndef RENDER_ONLY
Compute_Normal();
#ifdef PREPROCESS
Store_Normal(filename);
#endif
#else
Load_Normal(filename);
#endif
CLOCK(stoptime);
mclock(stoptime,starttime,&exectime);
printf("wall clock execution time to compute normal: %lu ms\n", exectime);
CLOCK(starttime);
#ifndef RENDER_ONLY
Compute_Opacity();
#ifdef PREPROCESS
Store_Opacity(filename);
#endif
#else
Load_Opacity(filename);
#endif
CLOCK(stoptime);
mclock(stoptime,starttime,&exectime);
printf("wall clock execution time to compute opacity: %lu ms\n", exectime);
Compute_Pre_View();
shd_length = LOOKUP_SIZE;
Allocate_Shading_Table(&shd_address,shd_length);
/* allocate space for image */
image_len[X] = frust_len;
image_len[Y] = frust_len;
image_length = image_len[X] * image_len[Y];
Allocate_Image(&image_address,image_length);
if (num_nodes == 1) {
block_xlen = image_len[X];
block_ylen = image_len[Y];
num_blocks = 1;
num_xblocks = 1;
num_yblocks = 1;
image_block = image_address;
}
else {
num_xblocks = ROUNDUP((float)image_len[X]/(float)block_xlen);
num_yblocks = ROUNDUP((float)image_len[Y]/(float)block_ylen);
num_blocks = num_xblocks * num_yblocks;
Lallocate_Image(&image_block,block_xlen*block_ylen);
}
CLOCK(starttime);
#ifndef RENDER_ONLY
Compute_Octree();
#ifdef PREPROCESS
Store_Octree(filename);
#endif
#else
Load_Octree(filename);
#endif
CLOCK(stoptime);
mclock(stoptime,starttime,&exectime);
printf("wall clock execution time to compute octree: %lu ms\n", exectime);
#ifdef PREPROCESS
return;
#endif
if (adaptive) {
printf("1.\n");
for (i=0; i<NI; i++) {
mask_image_len[i] = image_len[i];
}
mask_image_length = image_length;
Allocate_MImage(&mask_image_address, mask_image_length);
if (num_nodes == 1)
mask_image_block = (PIXEL *)mask_image_address;
else
Lallocate_Image(&mask_image_block, block_xlen*block_ylen);
printf("2.\n");
}
#ifndef RENDER_ONLY
Deallocate_Map(&map_address);
#endif
Global->Index = NODE0;
printf("\nRendering...\n");
printf("node\tframe\ttime\titime\trays\thrays\tsamples trilirped\n");
CREATE(Render_Loop, num_nodes);
}
int main(int argc, char **argv)
{
/* default values for the control parameters of the driver */
/* are in parameters.h */
if ((argc == 2) &&((strncmp(argv[1],"-h",strlen("-h")) == 0) || (strncmp(argv[1],"-H",strlen("-H")) == 0))) {
printf("Usage: WATER-NSQUARED < infile, where the contents of infile can be\nobtained from the comments at the top of water.C and the first scanf \nin main() in water.C\n\n");
exit(0);
}
/* POSSIBLE ENHANCEMENT: Here's where one might bind the main process
(process 0) to a processor if one wanted to. Others can be bound in
the WorkStart routine.
*/
six = stdout; /* output file */
TEMP =298.0;
RHO =0.9980;
CUTOFF=0.0;
/* read input */
if (scanf("%lf%ld%ld%ld%ld%ld%ld%ld%ld%lf",&TSTEP, &NMOL, &NSTEP, &NORDER,
&NSAVE, &NRST, &NPRINT, &NFMC,&NumProcs, &CUTOFF) != 10)
fprintf(stderr,"ERROR: Usage: water < infile, which must have 10 fields, see SPLASH documentation or comment at top of water.C\n");
if (NMOL > MAXLCKS) {
fprintf(stderr, "Just so you know ... Lock array in global.H has size %ld < %ld (NMOL)\n code will still run correctly but there may be lock contention\n\n", MAXLCKS, NMOL);
}
printf("Using %ld procs on %ld steps of %ld mols\n", NumProcs, NSTEP, NMOL);
printf("Other parameters:\n\tTSTEP = %8.2e\n\tNORDER = %ld\n\tNSAVE = %ld\n",TSTEP,NORDER,NSAVE);
printf("\tNRST = %ld\n\tNPRINT = %ld\n\tNFMC = %ld\n\tCUTOFF = %lf\n\n",NRST,NPRINT,NFMC,CUTOFF);
/* SET UP SCALING FACTORS AND CONSTANTS */
NORD1=NORDER+1;
CNSTNT(NORD1,TLC); /* sub. call to set up constants */
{ /* Do memory initializations */
long pid;
long mol_size = sizeof(molecule_type) * NMOL;
long gmem_size = sizeof(struct GlobalMemory);
/* POSSIBLE ENHANCEMENT: One might bind the first process to
a processor here, even before the other (child) processes are
bound later in mdmain().
*/
MAIN_INITENV(,70000000,); /* macro call to initialize
shared memory etc. */
THREAD_INIT_FREE();
/* allocate space for main (VAR) data structure as well as
synchronization variables */
/* POSSIBLE ENHANCEMENT: One might want to allocate a process's
portion of the VAR array and what it points to in its local
memory */
VAR = (molecule_type *) G_MALLOC(mol_size);
gl = (struct GlobalMemory *) G_MALLOC(gmem_size);
/* POSSIBLE ENHANCEMENT: One might want to allocate process i's
PFORCES[i] array in its local memory */
PFORCES = (double ****) G_MALLOC(NumProcs * sizeof (double ***));
{ long i,j,k;
for (i = 0; i < NumProcs; i++) {
PFORCES[i] = (double ***) G_MALLOC(NMOL * sizeof (double **));
for (j = 0; j < NMOL; j++) {
PFORCES[i][j] = (double **) G_MALLOC(NDIR * sizeof (double *));
for (k = 0; k < NDIR; k++) {
PFORCES[i][j][k] = (double *) G_MALLOC(NATOM * sizeof (double));
}
}
}
}
/* macro calls to initialize synch varibles */
BARINIT(gl->start, NumProcs);
BARINIT(gl->InterfBar, NumProcs);
BARINIT(gl->PotengBar, NumProcs);
LOCKINIT(gl->IOLock);
LOCKINIT(gl->IndexLock);
LOCKINIT(gl->IntrafVirLock);
LOCKINIT(gl->InterfVirLock);
LOCKINIT(gl->FXLock);
LOCKINIT(gl->FYLock);
LOCKINIT(gl->FZLock);
if (NMOL < MAXLCKS) {
ALOCKINIT(gl->MolLock, NMOL);
}
else {
ALOCKINIT(gl->MolLock, MAXLCKS);
}
LOCKINIT(gl->KinetiSumLock);
LOCKINIT(gl->PotengSumLock);
/* set up control for static scheduling */
MolsPerProc = NMOL/NumProcs;
StartMol[0] = 0;
for (pid = 1; pid < NumProcs; pid += 1) {
StartMol[pid] = StartMol[pid-1] + MolsPerProc;
}
StartMol[NumProcs] = NMOL;
}
SYSCNS(); /* sub. call to initialize system constants */
fprintf(six,"\nTEMPERATURE = %8.2f K\n",TEMP);
fprintf(six,"DENSITY = %8.5f G/C.C.\n",RHO);
fprintf(six,"NUMBER OF MOLECULES = %8ld\n",NMOL);
fprintf(six,"NUMBER OF PROCESSORS = %8ld\n",NumProcs);
fprintf(six,"TIME STEP = %8.2e SEC\n",TSTEP);
fprintf(six,"ORDER USED TO SOLVE F=MA = %8ld \n",NORDER);
fprintf(six,"NO. OF TIME STEPS = %8ld \n",NSTEP);
fprintf(six,"FREQUENCY OF DATA SAVING = %8ld \n",NSAVE);
fprintf(six,"FREQUENCY TO WRITE RST FILE= %8ld \n",NRST);
fprintf(six,"SPHERICAL CUTOFF RADIUS = %8.4f ANGSTROM\n",CUTOFF);
fflush(six);
/* initialization routine; also reads displacements and
sets up random velocities*/
INITIA();
/*.....start molecular dynamic loop */
gl->tracktime = 0;
gl->intratime = 0;
gl->intertime = 0;
/* initialize Index to 1 so that the first created child gets
id 1, not 0 */
gl->Index = 1;
if (NSAVE > 0) /* not true for input decks provided */
fprintf(six,"COLLECTING X AND V DATA AT EVERY %4ld TIME STEPS \n",NSAVE);
/* spawn helper processes, each getting its unique process id */
CLOCK(gl->computestart);
CREATE(WorkStart, NumProcs);
/* macro to make main process wait for all others to finish */
WAIT_FOR_END(NumProcs);
CLOCK(gl->computeend);
printf("COMPUTESTART (after initialization) = %lu\n",gl->computestart);
printf("COMPUTEEND = %lu\n",gl->computeend);
printf("COMPUTETIME (after initialization) = %lu\n",gl->computeend-gl->computestart);
printf("Measured Time (2nd timestep onward) = %lu\n",gl->tracktime);
printf("Intramolecular time only (2nd timestep onward) = %lu\n",gl->intratime);
printf("Intermolecular time only (2nd timestep onward) = %lu\n",gl->intertime);
printf("Other time (2nd timestep onward) = %lu\n",gl->tracktime - gl->intratime - gl->intertime);
printf("\nExited Happily with XTT = %g (note: XTT value is garbage if NPRINT > NSTEP)\n", XTT);
MAIN_END;
} /* main.c */
int main(int argc, char *argv[])
{
long i;
long j;
long k;
long x_part;
long y_part;
long d_size;
long itemp;
long jtemp;
double procsqrt;
long temp = 0;
double min_total;
double max_total;
double avg_total;
double min_multi;
double max_multi;
double avg_multi;
double min_frac;
double max_frac;
double avg_frac;
long ch;
extern char *optarg;
unsigned long computeend;
unsigned long start;
CLOCK(start)
while ((ch = getopt(argc, argv, "n:p:e:r:t:soh")) != -1) {
switch(ch) {
case 'n': im = atoi(optarg);
if (log_2(im-2) == -1) {
printerr("Grid must be ((power of 2)+2) in each dimension\n");
exit(-1);
}
break;
case 'p': nprocs = atoi(optarg);
if (nprocs < 1) {
printerr("P must be >= 1\n");
exit(-1);
}
if (log_2(nprocs) == -1) {
printerr("P must be a power of 2\n");
exit(-1);
}
break;
case 'e': tolerance = atof(optarg); break;
case 'r': res = atof(optarg); break;
case 't': dtau = atof(optarg); break;
case 's': do_stats = !do_stats; break;
case 'o': do_output = !do_output; break;
case 'h': printf("Usage: OCEAN <options>\n\n");
printf("options:\n");
printf(" -nN : Simulate NxN ocean. N must be (power of 2)+2.\n");
printf(" -pP : P = number of processors. P must be power of 2.\n");
printf(" -eE : E = error tolerance for iterative relaxation.\n");
printf(" -rR : R = distance between grid points in meters.\n");
printf(" -tT : T = timestep in seconds.\n");
printf(" -s : Print timing statistics.\n");
printf(" -o : Print out relaxation residual values.\n");
printf(" -h : Print out command line options.\n\n");
printf("Default: OCEAN -n%1d -p%1d -e%1g -r%1g -t%1g\n",
DEFAULT_N,DEFAULT_P,DEFAULT_E,DEFAULT_R,DEFAULT_T);
exit(0);
break;
}
}
MAIN_INITENV(,60000000)
THREAD_INIT_FREE();
jm = im;
printf("\n");
printf("Ocean simulation with W-cycle multigrid solver\n");
printf(" Processors : %1ld\n",nprocs);
printf(" Grid size : %1ld x %1ld\n",im,jm);
printf(" Grid resolution (meters) : %0.2f\n",res);
printf(" Time between relaxations (seconds) : %0.0f\n",dtau);
printf(" Error tolerance : %0.7g\n",tolerance);
printf("\n");
xprocs = 0;
yprocs = 0;
procsqrt = sqrt((double) nprocs);
j = (long) procsqrt;
while ((xprocs == 0) && (j > 0)) {
k = nprocs / j;
if (k * j == nprocs) {
if (k > j) {
xprocs = j;
yprocs = k;
} else {
xprocs = k;
yprocs = j;
}
}
j--;
}
if (xprocs == 0) {
printerr("Could not find factors for subblocking\n");
exit(-1);
}
minlevel = 0;
itemp = 1;
jtemp = 1;
numlev = 0;
minlevel = 0;
while (itemp < (im-2)) {
itemp = itemp*2;
jtemp = jtemp*2;
if ((itemp/yprocs > 1) && (jtemp/xprocs > 1)) {
numlev++;
}
}
if (numlev == 0) {
printerr("Must have at least 2 grid points per processor in each dimension\n");
exit(-1);
}
imx = (long *) G_MALLOC(numlev*sizeof(long));
jmx = (long *) G_MALLOC(numlev*sizeof(long));
lev_res = (double *) G_MALLOC(numlev*sizeof(double));
lev_tol = (double *) G_MALLOC(numlev*sizeof(double));
i_int_coeff = (double *) G_MALLOC(numlev*sizeof(double));
j_int_coeff = (double *) G_MALLOC(numlev*sizeof(double));
xpts_per_proc = (long *) G_MALLOC(numlev*sizeof(long));
ypts_per_proc = (long *) G_MALLOC(numlev*sizeof(long));
imx[numlev-1] = im;
jmx[numlev-1] = jm;
lev_res[numlev-1] = res;
lev_tol[numlev-1] = tolerance;
for (i=numlev-2;i>=0;i--) {
imx[i] = ((imx[i+1] - 2) / 2) + 2;
jmx[i] = ((jmx[i+1] - 2) / 2) + 2;
lev_res[i] = lev_res[i+1] * 2;
}
for (i=0;i<numlev;i++) {
xpts_per_proc[i] = (jmx[i]-2) / xprocs;
ypts_per_proc[i] = (imx[i]-2) / yprocs;
}
for (i=numlev-1;i>=0;i--) {
if ((xpts_per_proc[i] < 2) || (ypts_per_proc[i] < 2)) {
minlevel = i+1;
break;
}
}
for (i=0;i<numlev;i++) {
temp += imx[i];
}
temp = 0;
j = 0;
for (k=0;k<numlev;k++) {
for (i=0;i<imx[k];i++) {
j++;
temp += jmx[k];
}
}
d_size = nprocs*sizeof(double ***);
psi = (double ****) G_MALLOC(d_size);
psim = (double ****) G_MALLOC(d_size);
work1 = (double ****) G_MALLOC(d_size);
work4 = (double ****) G_MALLOC(d_size);
work5 = (double ****) G_MALLOC(d_size);
work7 = (double ****) G_MALLOC(d_size);
temparray = (double ****) G_MALLOC(d_size);
d_size = 2*sizeof(double **);
for (i=0;i<nprocs;i++) {
psi[i] = (double ***) G_MALLOC(d_size);
psim[i] = (double ***) G_MALLOC(d_size);
work1[i] = (double ***) G_MALLOC(d_size);
work4[i] = (double ***) G_MALLOC(d_size);
work5[i] = (double ***) G_MALLOC(d_size);
work7[i] = (double ***) G_MALLOC(d_size);
temparray[i] = (double ***) G_MALLOC(d_size);
}
d_size = nprocs*sizeof(double **);
psium = (double ***) G_MALLOC(d_size);
psilm = (double ***) G_MALLOC(d_size);
psib = (double ***) G_MALLOC(d_size);
ga = (double ***) G_MALLOC(d_size);
gb = (double ***) G_MALLOC(d_size);
work2 = (double ***) G_MALLOC(d_size);
work3 = (double ***) G_MALLOC(d_size);
work6 = (double ***) G_MALLOC(d_size);
tauz = (double ***) G_MALLOC(d_size);
oldga = (double ***) G_MALLOC(d_size);
oldgb = (double ***) G_MALLOC(d_size);
gp = (struct Global_Private *) G_MALLOC((nprocs+1)*sizeof(struct Global_Private));
for (i=0;i<nprocs;i++) {
gp[i].rel_num_x = (long *) G_MALLOC(numlev*sizeof(long));
gp[i].rel_num_y = (long *) G_MALLOC(numlev*sizeof(long));
gp[i].eist = (long *) G_MALLOC(numlev*sizeof(long));
gp[i].ejst = (long *) G_MALLOC(numlev*sizeof(long));
gp[i].oist = (long *) G_MALLOC(numlev*sizeof(long));
gp[i].ojst = (long *) G_MALLOC(numlev*sizeof(long));
gp[i].rlist = (long *) G_MALLOC(numlev*sizeof(long));
gp[i].rljst = (long *) G_MALLOC(numlev*sizeof(long));
gp[i].rlien = (long *) G_MALLOC(numlev*sizeof(long));
gp[i].rljen = (long *) G_MALLOC(numlev*sizeof(long));
gp[i].multi_time = 0;
gp[i].total_time = 0;
}
subblock();
x_part = (jm - 2)/xprocs + 2;
y_part = (im - 2)/yprocs + 2;
d_size = x_part*y_part*sizeof(double) + y_part*sizeof(double *);
global = (struct global_struct *) G_MALLOC(sizeof(struct global_struct));
for (i=0;i<nprocs;i++) {
psi[i][0] = (double **) G_MALLOC(d_size);
psi[i][1] = (double **) G_MALLOC(d_size);
psim[i][0] = (double **) G_MALLOC(d_size);
psim[i][1] = (double **) G_MALLOC(d_size);
psium[i] = (double **) G_MALLOC(d_size);
psilm[i] = (double **) G_MALLOC(d_size);
psib[i] = (double **) G_MALLOC(d_size);
ga[i] = (double **) G_MALLOC(d_size);
gb[i] = (double **) G_MALLOC(d_size);
work1[i][0] = (double **) G_MALLOC(d_size);
work1[i][1] = (double **) G_MALLOC(d_size);
work2[i] = (double **) G_MALLOC(d_size);
work3[i] = (double **) G_MALLOC(d_size);
work4[i][0] = (double **) G_MALLOC(d_size);
work4[i][1] = (double **) G_MALLOC(d_size);
work5[i][0] = (double **) G_MALLOC(d_size);
work5[i][1] = (double **) G_MALLOC(d_size);
work6[i] = (double **) G_MALLOC(d_size);
work7[i][0] = (double **) G_MALLOC(d_size);
work7[i][1] = (double **) G_MALLOC(d_size);
temparray[i][0] = (double **) G_MALLOC(d_size);
temparray[i][1] = (double **) G_MALLOC(d_size);
tauz[i] = (double **) G_MALLOC(d_size);
oldga[i] = (double **) G_MALLOC(d_size);
oldgb[i] = (double **) G_MALLOC(d_size);
}
f = (double *) G_MALLOC(im*sizeof(double));
multi = (struct multi_struct *) G_MALLOC(sizeof(struct multi_struct));
d_size = numlev*sizeof(double **);
if (numlev%2 == 1) { /* To make sure that the actual data
starts double word aligned, add an extra
pointer */
d_size += sizeof(double **);
}
for (i=0;i<numlev;i++) {
d_size += ((imx[i]-2)/yprocs+2)*((jmx[i]-2)/xprocs+2)*sizeof(double)+
((imx[i]-2)/yprocs+2)*sizeof(double *);
}
d_size *= nprocs;
if (nprocs%2 == 1) { /* To make sure that the actual data
starts double word aligned, add an extra
pointer */
d_size += sizeof(double ***);
}
d_size += nprocs*sizeof(double ***);
q_multi = (double ****) G_MALLOC(d_size);
rhs_multi = (double ****) G_MALLOC(d_size);
locks = (struct locks_struct *) G_MALLOC(sizeof(struct locks_struct));
bars = (struct bars_struct *) G_MALLOC(sizeof(struct bars_struct));
LOCKINIT(locks->idlock)
LOCKINIT(locks->psiailock)
LOCKINIT(locks->psibilock)
LOCKINIT(locks->donelock)
LOCKINIT(locks->error_lock)
LOCKINIT(locks->bar_lock)
#if defined(MULTIPLE_BARRIERS)
BARINIT(bars->iteration, nprocs)
BARINIT(bars->gsudn, nprocs)
BARINIT(bars->p_setup, nprocs)
BARINIT(bars->p_redph, nprocs)
BARINIT(bars->p_soln, nprocs)
BARINIT(bars->p_subph, nprocs)
BARINIT(bars->sl_prini, nprocs)
BARINIT(bars->sl_psini, nprocs)
BARINIT(bars->sl_onetime, nprocs)
BARINIT(bars->sl_phase_1, nprocs)
BARINIT(bars->sl_phase_2, nprocs)
BARINIT(bars->sl_phase_3, nprocs)
BARINIT(bars->sl_phase_4, nprocs)
BARINIT(bars->sl_phase_5, nprocs)
BARINIT(bars->sl_phase_6, nprocs)
BARINIT(bars->sl_phase_7, nprocs)
BARINIT(bars->sl_phase_8, nprocs)
BARINIT(bars->sl_phase_9, nprocs)
BARINIT(bars->sl_phase_10, nprocs)
BARINIT(bars->error_barrier, nprocs)
#else
BARINIT(bars->barrier, nprocs)
#endif
link_all();
multi->err_multi = 0.0;
i_int_coeff[0] = 0.0;
j_int_coeff[0] = 0.0;
for (i=0;i<numlev;i++) {
i_int_coeff[i] = 1.0/(imx[i]-1);
j_int_coeff[i] = 1.0/(jmx[i]-1);
}
/* initialize constants and variables
id is a global shared variable that has fetch-and-add operations
performed on it by processes to obtain their pids. */
global->id = 0;
global->psibi = 0.0;
pi = atan(1.0);
pi = 4.*pi;
factjacob = -1./(12.*res*res);
factlap = 1./(res*res);
eig2 = -h*f0*f0/(h1*h3*gpr);
jmm1 = jm-1 ;
ysca = ((double) jmm1)*res ;
im = (imx[numlev-1]-2)/yprocs + 2;
jm = (jmx[numlev-1]-2)/xprocs + 2;
if (do_output) {
printf(" MULTIGRID OUTPUTS\n");
}
CREATE(slave, nprocs);
WAIT_FOR_END(nprocs);
CLOCK(computeend)
printf("\n");
printf(" PROCESS STATISTICS\n");
printf(" Total Multigrid Multigrid\n");
printf(" Proc Time Time Fraction\n");
printf(" 0 %15.0f %15.0f %10.3f\n", gp[0].total_time,gp[0].multi_time, gp[0].multi_time/gp[0].total_time);
if (do_stats) {
min_total = max_total = avg_total = gp[0].total_time;
min_multi = max_multi = avg_multi = gp[0].multi_time;
min_frac = max_frac = avg_frac = gp[0].multi_time/gp[0].total_time;
for (i=1;i<nprocs;i++) {
if (gp[i].total_time > max_total) {
max_total = gp[i].total_time;
}
if (gp[i].total_time < min_total) {
min_total = gp[i].total_time;
}
if (gp[i].multi_time > max_multi) {
max_multi = gp[i].multi_time;
}
if (gp[i].multi_time < min_multi) {
min_multi = gp[i].multi_time;
}
if (gp[i].multi_time/gp[i].total_time > max_frac) {
max_frac = gp[i].multi_time/gp[i].total_time;
}
if (gp[i].multi_time/gp[i].total_time < min_frac) {
min_frac = gp[i].multi_time/gp[i].total_time;
}
avg_total += gp[i].total_time;
avg_multi += gp[i].multi_time;
avg_frac += gp[i].multi_time/gp[i].total_time;
}
avg_total = avg_total / nprocs;
avg_multi = avg_multi / nprocs;
avg_frac = avg_frac / nprocs;
for (i=1;i<nprocs;i++) {
printf(" %3ld %15.0f %15.0f %10.3f\n", i,gp[i].total_time,gp[i].multi_time, gp[i].multi_time/gp[i].total_time);
}
printf(" Avg %15.0f %15.0f %10.3f\n", avg_total,avg_multi,avg_frac);
printf(" Min %15.0f %15.0f %10.3f\n", min_total,min_multi,min_frac);
printf(" Max %15.0f %15.0f %10.3f\n", max_total,max_multi,max_frac);
}
printf("\n");
global->starttime = start;
printf(" TIMING INFORMATION\n");
printf("Start time : %16lu\n", global->starttime);
printf("Initialization finish time : %16lu\n", global->trackstart);
printf("Overall finish time : %16lu\n", computeend);
printf("Total time with initialization : %16lu\n", computeend-global->starttime);
printf("Total time without initialization : %16lu\n", computeend-global->trackstart);
printf(" (excludes first timestep)\n");
printf("\n");
MAIN_END
}