int main(int argc, const char *argv[]) {
int **grid;
int n;
int i;
int j;
int element;
n = 4;
grid = (int**) malloc(sizeof(int*) * n);
element = 1;
for (i = 0; i < n; i++) {
grid[i] = (int*) malloc(sizeof(int) * n);
for (j = 0; j < n; j++) {
grid[i][j] = element;
element++;
}
}
print_grid(grid, n);
rotate_90 (grid, n);
printf("\n\n");
print_grid(grid, n);
return 0;
}
int main()
{
int **grid;
int i, a, b;
printf("First example:\n");
a = 5;
b = 20;
grid = alloc_grid(a, b);
print_grid(grid, a, b);
for (i = 0; i < a; i++)
free(grid[i]);
free(grid);
printf("Second example:\n");
grid = alloc_grid(5, 5);
print_grid(grid, 5, 5);
printf("\n");
grid[2][3] = 98;
grid[4][4] = 402;
print_grid(grid, 5, 5);
return (0);
}
/* read an image and generate a result */
int main(int argc, char **argv) {
int rows=0;
int cols=0;
int iterations=0;
int i;
int **grid = NULL;
int **neighbors = NULL;
if (argc<2 || !(grid=read_grid(stdin,&rows,&cols))
|| (iterations=atoi(argv[1]))<0) {
fprintf(stderr,"life usage: life iterations <inputfile\n");
exit(1);
}
#ifdef DEBUG
printf("input:\n");
print_grid(stdout,grid,rows,cols);
#endif /* DEBUG */
neighbors = make_grid(rows,cols);
for (i=0; i<iterations; i++) {
next(grid, neighbors, rows, cols);
#ifdef DEBUG
printf("next\n");
print_grid(stdout,grid,rows,cols);
#endif /* DEBUG */
}
print_grid(stdout,grid,rows,cols);
free_grid(grid,rows);
free_grid(neighbors,rows);
}
// main entry point for the program; initializes parameters, reads input file name argument if provided,
// declares two 2-dimensional arrays used as buffers to simulate the evolution of the cellular automaton
// grid, loads initial generation from input file into grid, runs automaton for 30 generations, prints
// results of evolution silently to file "output.txt" using a grid of 48 rows by 64 columns, (not including
// the additional frame which is printed at the perimeter of the grid), and prints error messages and exits
// if any part of this process fails;
int main( int argc, char ** argv )
{
char even_grid[ROWS][COLUMNS]; // for double-buffering the grid, used for even-numbered generations
char odd_grid[ROWS][COLUMNS]; // other buffer, used for odd-numbered generations
int g, r, c; // indices
char * input_filename; // "input.txt" if user does not provide a different file
char * output_filename = "output.txt"; // hardwired to always create or overwrite same output file
FILE * output_file_pointer; // used to print results to file as they are generated
for ( r = 0; r < ROWS; r++ ) { // initialize buffers so they are full of dead cells (spaces)
for ( c = 0; c < COLUMNS; c++ ) {
even_grid[r][c] = ' ';
odd_grid[r][c] = ' ';
}
}
// allow user to specify input file as command line argument to program
if ( 1 < argc ) { // if arguments are given, assume first is the name of input file
input_filename = argv[1];
} else { // no input filename provided, so use default name, "input.txt"
input_filename = "input.txt";
}
if ( !load_input( input_filename, even_grid ) ) {
fprintf( stderr, "usage: %s [ input_file ]\n", argv[0] );
fprintf( stderr, " quitting...\n" );
exit( EXIT_FAILURE );
}
if ( NULL == ( output_file_pointer = fopen( output_filename, "w" ) ) ) {
fprintf( stderr, "error: in 'main()', call to 'fopen(%s, 'w')' failed\n", output_filename );
perror( " " );
fprintf( stderr, "usage: %s [ input_file ]\n", argv[0] );
fprintf( stderr, " quitting...\n" );
exit( EXIT_FAILURE );
}
// run the 'Game of Life' on the provided input;
for ( g = 0; g < GENERATIONS; g++ ) { // iterate automaton for hardwired number of times;
if ( 0 == g % 2 ) { // modulo for double-buffering; even generations use even_grid;
print_grid( even_grid, g, output_file_pointer ); // once it's loaded or computed, print it;
generate_grid( even_grid, odd_grid ); // use even_grid to compute new odd_grid;
} else { // odd generations use odd grid
print_grid( odd_grid, g, output_file_pointer ); // once it's computed, print it;
generate_grid( odd_grid, even_grid ); // use odd_grid to compute new even_grid;
}
}
if ( EOF == fclose( output_file_pointer ) ) {
fprintf( stderr, "error: in 'main()', call to 'fclose(%s)' failed\n", output_filename );
perror( " " );
fprintf( stderr, "usage: %s [ input_file ]\n", argv[0] );
fprintf( stderr, " quitting...\n" );
exit( EXIT_FAILURE );
}
exit( EXIT_SUCCESS );
}
int main()
{
int **grid;
grid = alloc_grid(5, 5);
print_grid(grid, 5, 5);
printf("\n");
grid[2][3] = 98;
grid[4][4] = 402;
print_grid(grid, 5, 5);
return (0);
}
gmx_bool print_forcefield(FILE *fp, real ener[], int natoms, rvec f[], rvec fshake[],
rvec x[], t_block *mols, real mass[], tensor pres)
{
real msf1;
if (ga)
{
msf1 = msf(natoms, f, fshake);
if (debug)
{
fprintf(fp, "Pressure: %12g, RMSF: %12g, Energy-Epot: %12g, cost: %12g\n",
ener[F_PRES], sqrt(msf1), ener[F_EPOT]/ff.nmol-ff.epot,
cost(pres, msf1, ener[F_EPOT]/ff.nmol));
}
if (print_ga(fp, ga, msf1, pres, scale, (ener[F_EPOT]/ff.nmol), range, ff.tol))
{
return TRUE;
}
fflush(fp);
}
else
{
print_grid(fp, ener, natoms, f, fshake, x, mols, mass, pres);
}
return FALSE;
}
void init(int policy, int prot)
{
allegro_init();
set_gfx_mode(GFX_AUTODETECT_WINDOWED, XMAX, YMAX, 0, 0);
clear_to_color(screen, BGC);
install_keyboard();
print_grid(policy, prot);
if (prot == PIP) ptask_init(policy, GLOBAL, PRIO_INHERITANCE);
else if (prot == PCP) ptask_init(policy, GLOBAL, PRIO_CEILING);
else {
allegro_exit();
ptask_syserror("pcp.c", "Wrong protocol");
}
if (prot == PIP) {
pmux_create_pi(&mxa);
pmux_create_pi(&muxA);
pmux_create_pi(&muxB);
}
else if (prot == PCP) {
pmux_create_pc(&mxa, prio[1]);
pmux_create_pc(&muxA, prio[1]);
pmux_create_pc(&muxB, prio[1]);
}
}
int find_best_move(grid_t grid)
{
int direction, best_direction = -1;
float best_score = 0.0f, score;
print_grid(grid);
std::printf("Current scores | real: %.0f, heuristic: %.3lf", score_real(grid), score_heuristic(grid));
std::cout << std::endl;
for (direction = 0; direction < 4; direction++) {
score = score_move(grid, direction);
if (score > best_score) {
best_score = score;
best_direction = direction;
} else if (score == best_score && best_score > 0.0f) {
// coin flip to break a tie
if (std::floor(rand() % 2) == 2) {
best_direction = direction;
}
}
}
if (best_direction != -1) {
std::printf("Move %s", GRID_MOVES[best_direction]);
std::cout << std::endl;
} else {
std::cout << "No move found" << std::endl;
}
return best_direction;
}
int main(int argc, char **argv)
{
char **grid;
if (error_check(argc, argv))
{
print_error();
return (0);
}
grid = set_grid_parameters(argv);
if (grid == NULL)
print_error();
sudoku(grid);
if (grid[9] == ERROR)
print_error();
else
{
grid[9] = UNIQUE;
if (sudoku(grid) && check_solution(grid))
print_grid(grid);
else
print_error();
}
free(grid);
return (0);
}
dir play_move(strategy s, grid g)
{
// If precedent move was right, we put all left again
/*if(((memory)(s->mem))->nbStages > 0 && (((((memory)(s->mem))->container)[((memory)(s->mem))->nbStages])->decision) == RIGHT && can_move(g, LEFT))
{
printf("test\n");
save_stage(s->mem, g, LEFT);
print_grid(g);
return LEFT;
}*/
if(((memory)(s->mem))->nbStages >=0)
{
//int temp = ((memory)(s->mem))->nbStages;
//printf("Test : [%d]%p\n", temp, (((memory)(s->mem))->container[5]));
printf("Base memory adress : %p.\n", (memory)(s->mem));
printf("Base container adress : %p.\n", ((memory)(s->mem))->container);
printf("Element at [%d] : %p.\n", ((memory)(s->mem))->nbStages, (((memory)(s->mem))->container)[(int)(((memory)(s->mem))->nbStages)]);
}
//printf("Test 2 : %d\n", ((memory)(s->mem))->nbStages);
if(can_move(g, DOWN))
{
save_stage(s->mem, g, DOWN);
print_grid(g);
return DOWN;
}
if(can_move(g, LEFT))
{
save_stage(s->mem, g, LEFT);
print_grid(g);
return LEFT;
}
if(can_move(g, RIGHT))
{
save_stage(s->mem, g, RIGHT);
print_grid(g);
return RIGHT;
}
if(can_move(g, UP))
{
save_stage(s->mem, g, UP);
print_grid(g);
return UP;
}
}
void handle_input() {
char command;
int x, y;
char input_buffer[20];
/* Lees het commando van de gebruiker in */
fgets(input_buffer, 20, stdin);
sscanf(input_buffer, "%c %i %i", &command, &x, &y);
switch (command) {
case 'F': // Flag
if (((x < 0) || (x > WIDTH) || (y < 0) || (y > HEIGHT))){ // Chack voor geldige input.
printf("Invalid input.\n");
}
else {
flag_coordinate(x, y);
}
break;
case 'R': // Reveal
if (((x < 0) || (x > WIDTH) || (y < 0) || (y > HEIGHT))){ // Chack voor geldige input.
printf("Invalid input.\n");
}
else {
reveal_coordinate(x, y);
}
break;
case 'P': // Print
print_grid();
draw_grid(); // DEBUG
break;
case 'E': // Exit
update_stats();
deallocate_grid(WIDTH, HEIGHT);
exit(0); // Programma wordt afgesloten met code 0 = succes.
break;
case 'B': // Boem! Alle mijnen ontploffen
printf("BOEM! \n");
dead = 1;
print_grid();
break;
default:
/* De speler gaf een commando in dat niet begrepen werd door deze functie: probeer opnieuw. */
printf("Command '%c' not understood, please try again.\n", command);
handle_input();
}
}
int main(){
grid board = hard_sudoku();
print_grid(board);
solve(board, 0, 0);
return 0;
}
/* read an image and generate a result */
int main(int argc, char **argv) {
FILE *f;
int rows=0; int cols=0; int iterations=0;
int i;
if (argc>1 || !(grid=read_rle(stdin,&rows,&cols))) {
fprintf(stderr,"translate usage: translate <inputfile\n");
exit(1);
}
print_grid(stdout,grid,rows,cols);
free_grid(grid,rows);
}
int main(int argc, char **argv) {
TreeNode *root = new TreeNode(3);
root->left = new TreeNode(9);
root->left->left = new TreeNode(15);
root->right = new TreeNode(20);
root->right->right = new TreeNode(7);
Solution s;
print_grid(s.levelOrder(root));
return 0;
}
void preparation(t_map **map)
{
char **grid;
grid = NULL;
g_size = size_min_square(*map);
init_grid(&grid);
while (backtrack(grid, *map) != 0)
create_grid(&grid);
print_grid(grid);
delete_tab(&grid);
}
/*
* Initialize cells based on input file, otherwise all cells
* are DEAD.
*/
void init_grids (struct life_t * life)
{
FILE * fd;
int i,j;
if (life->infile != NULL) {
if ((fd = fopen(life->infile, "r")) == NULL) {
perror("Failed to open file for input");
exit(EXIT_FAILURE);
}
if (fscanf(fd, "%d %d\n", &life->nrows, &life->tcols) == EOF) {
printf("File must at least define grid dimensions!\nExiting.\n");
exit(EXIT_FAILURE);
}
}
// resize so each process is in charge of a vertical slice of the whole board
life->ubound = (((life->rank + 1) * life->tcols / life->size) - 1); // we want 1 col of (overlap?)
life->lbound = life->rank * life->tcols / life->size;
life->ncols = (life->ubound - life->lbound) + 1;
printf("[Process %d] lower bound is %d upper bound is %d width is %d random seed is %d.\n", life->rank, life->lbound, life->ubound, life->ncols, life->randseed);
allocate_grids(life);
for (i = 0; i < life->nrows+2; i++) {
for (j = 0; j < life->ncols+2; j++) {
life->grid[i][j] = DEAD;
life->next_grid[i][j] = DEAD;
}
}
if (life->infile != NULL) {
while (fscanf(fd, "%d %d\n", &i, &j) != EOF) {
if (j <= life->ubound && j >= life->lbound){
fprintf(stderr, "[Process %d] %d %d -> %d %d.\n", life->rank, i, j, i, j - life->lbound);
life->grid[i+1][j - life->lbound + 1] = ALIVE;
life->next_grid[i+1][j- life->lbound + 1] = ALIVE;
}
}
fclose(fd);
} else {
randomize_grid(life, INIT_PROB);
}
if (life->print){
printf("[Host %d] printing initial slice.\n", life->rank);
print_grid (life);
}
}
int
main(int argc, char **argv)
{
char *err;
struct game g = { 0 };
long w, h, p;
if (argc != 4 && argc != 5) {
fprintf(stderr, "Usage: boxes height width playercount [filename]\n");
exit(1);
}
h = strtol(argv[1], &err, 10);
if (*err != '\0' || h < 2 || h > 999) {
fprintf(stderr, "Invalid grid dimensions\n");
exit(2);
}
w = strtol(argv[2], &err, 10);
if (*err != '\0' || w < 2 || w > 999) {
fprintf(stderr, "Invalid grid dimensions\n");
exit(2);
}
p = strtol(argv[3], &err, 10);
if (*err != '\0' || p < 2 || p > 100) {
fprintf(stderr, "Invalid player count\n");
exit(3);
}
g.width = w;
g.height = h;
g.num_players = p;
g.possible_closures = w * h;
allocate_empty_grid(&g);
if (argc == 5) {
read_grid_file(argv[4], &g);
}
while (1) {
print_grid(stdout, &g);
if (g.close_count == g.possible_closures) {
pick_winner(&g);
}
while (try_move(&g));
g.current_player = (g.current_player + 1) % g.num_players;
}
return 0;
}
main()
{
int t;
scanf("%d",&t);
while (t--)
{
read_grid();
init();
solve();
print_grid();
}
return 0;
}
/* remove random values from printed grid
* for solving by us apes
*/
static void remove_values(int* matrix)
{
puts("remove values...");
/* number of removed numbers */
int nr = DIFFICULTY;
int r;
// for each latin square, distribute evenly the number of cells
// to hide
for (int j=0; j<9; j++)
{
int result[9];
return_square(j, result);
}
// for 0 to 81, assign blanks to cells to present puzzle
for (int j=0; j<81; j++)
{
puzzle_matrix[j] = puzzle_solution[j];
if (nr>0)
{
/* TODO: This has to change, the distribution
* of removed values has to be spread across all
* latin squares, as it stands now, it will tend
* to only remove values from the beginning sequence
* of values. ie. the first few rows and first 3
* latin squares. Amateur.
*/
#ifdef JSW_RANDOM
r = jsw_rand() % 2;
#else
r = rand() % 2;
#endif
/* printf("%d\n", r); */
if (r)
{
puzzle_matrix[j] = 0;
--nr;
}
}
}
print_grid(puzzle_matrix);
}
void reveal_coordinate(int x, int y) {
/* Als een cel een mijn bevat, dan is de speler dood*/
if (get_cell(x, y)->is_mine == 1 )
{
printf("/+/+/+/ MINE EXPLODED /+/+/+/ \n");
printf("\n");
dead = 1;
print_grid();
printf("Spel wordt afgesloten...\n");
draw_grid();
update_stats();
show_stats();
sleep(2);
deallocate_grid(WIDTH, HEIGHT);
exit(0);
}
/* Als de cel geen mijn is, verander dan zijn staat van covered naar uncovered. */
else {
// Zet de status van het vakje op UNCOVERED.
get_cell(x, y)->state = UNCOVERED;
// Test of het vakje geen naburige mijnen heeft.
if (get_cell(x, y)->neighbouring_mines == 0)
{
// Als het 0 naburige vakjes heeft, moeten al zijn buren bekeken worden of zij een mijn in de buurt hebben.
for (int i = -1; i < 2; i++) // Begin bij de cel aan de linkse kant (vandaar de -1).
{
for (int j = -1; j < 2; j++) // Begin bij de cel aan de rechtse kant.
{
if ((x + i) < WIDTH && (x + i) >= 0 && (y + j) < HEIGHT && (y + j) >= 0) // Controle of een buurcel van de mijn zich niet buiten de randen van het veld bevindt.
{
if (get_cell(x + i, y + j)->state == COVERED)
{
reveal_coordinate(x + i, y + j); // Recursieve oproep op de naburige cellen.
}
}
}
}
}
}
}
int main(void) {
int ret = 0, nc;
int c;
int key = 0;
int part = PARTITIONED; // PARTITIONED, GLOBAL
int sched = SCHED_FIFO;
get_data();
init();
ret = select_prot();
if (ret == -1) {
allegro_exit();
return 0;
}
print_grid(ret);
ptask_init(sched, part, ret);
t_start = ptask_gettime(MILLI);
set_sem_sezC(ret);
nc = ptask_getnumcores();
textprintf_ex(screen, font, 480, 10, 7, BGC, "(NumCores = %d)", nc);
int gen_id = ptask_create_prio(gen, 100, 30, NOW);
if (gen_id < 0) {
printf("Could not create task gen\n");
exit(-1);
}
while (key != KEY_ESC) {
if (keypressed()) {
c = readkey();
key = c >> 8;
}
}
pmux_destroy(&mx_sezNorm);
pmux_destroy(&mx_sezA);
pmux_destroy(&mx_sezB);
allegro_exit();
return 0;
}
/* randomly swap out values, by num_itr,
* number of iterations, of known solved puzzle;
* should result in a new (completed) random puzzle
*/
static int build_puzzle(int* puzzle, int* matrix)
{
int num_itr = 0, x = 0, y = 0;
puts("generating sudoku puzzle ...");
/* There is a stdlib call "rand"; this function is tuned primarily for speed
* and distribution, not for unpredictability. Almost all built-in random
* functions for various languages and frameworks use this function by
* default. There are also "cryptographic" random number generators that are
* much less predictable, but run much slower. These should be used in any sort
* of security-related application. - tylerl
* http://stackoverflow.com/questions/822323/how-to-generate-a-random-number-in-c
*/
#ifdef JSW_RANDOM
puts("using jsw_srand()");
jsw_seed(time(NULL));
num_itr = (jsw_rand() % 1000000);
#else
puts("using srand()");
srand(time(NULL));
num_itr = (rand() % 1000000);
#endif
printf("number iterations: %d\n", num_itr);
for (int i = 0; i<num_itr; i++)
{
#ifdef JSW_RANDOM
x = (jsw_rand() % 9);
y = (jsw_rand() % 9);
#else
x = (rand() % 9);
y = (rand() % 9);
#endif
x++;
y++;
/* printf("x = %d, y = %d\n", x, y); */
swap_numbers(puzzle, x, y);
}
print_grid(puzzle_solution);
remove_values(matrix);
return 0;
}
/*
* Deze functie wordt aangeroepen als de gebruiker een vlag wil plaatsten op het vakje met positie (x,y)
* in het veld.
*/
void flag_coordinate(int x, int y){
// UNFLAG //
if (get_cell(x, y)->state == FLAGGED) { // Is vakje gevlagd?
get_cell(x, y)->state = COVERED; // Zet status van vakje terug op covered.
nr_of_flags++; // Verhoog aantal vlaggen.
if (get_cell (x, y)->is_mine == 1){ // Als het onderliggend vakje een mijn was, dan wordt nr_of_flagged_mines verminderd.
nr_of_flagged_mines--;
}
}
// FLAG //
else if (nr_of_flags != 0){ // Als de vlaggen niet op zijn, kan de cel van status veranderen.
get_cell(x, y)->state = FLAGGED; // Verader de status van de cel naar FLAGGED (= 2)
nr_of_flags--; // Verminder het aantal vlaggen dat de speler ter beschikking nog heeft
if (get_cell(x, y)->is_mine == 1){
nr_of_flagged_mines++; // Verminder het aantal flagged mines.
}
if (nr_of_flagged_mines == NR_OF_MINES){
printf("/+/+/+/ YOU WON! /+/+/+/\n");
printf("\n");
print_grid();
draw_grid();
printf("Spel wordt afgesloten...\n");
update_stats();
show_stats();
sleep(2);
deallocate_grid(WIDTH, HEIGHT);
exit(0); // Programma wordt afgesloten met code 0 = succes.
}
}
// Het aantal vlaggen is op
else
{
printf("NO MORE FLAGS\n"); // Komt op het scherm al er geen vlggen meer zijn.
}
}
int main()
{
initscr();
noecho();
// position
int* ptr_x;
int* ptr_y;
ptr_x = malloc(sizeof(int));
ptr_y = malloc(sizeof(int));
*ptr_x = *ptr_y = 0;
int moved = 1;
/// for (int i = 0; i < 10; i++ )
while(moved != 'q')
{
//moved = movement(ptr_x, ptr_y);
clear();
// printf("\e[2J\e[1;1H");
printw("Location is: (%d,%d)\n", *ptr_x, *ptr_y);
print_grid(ptr_x, ptr_y);
refresh();
moved = movement(ptr_x,ptr_y);
}
// grid array
//int grid[GRIDMAX][GRIDMAX];
free(ptr_x);
free(ptr_y);
endwin();
return 0;
}
/* For printing the Queens and Placing them in Grid */
void nqueens(int n)
{
int x[20];
int count=0;
int k=1;
x[k]=0;
while(k!=0)
{
x[k]=x[k]+1;
while((x[k]<=n)&&(!safetoplace(x,k)))
{
x[k]=x[k]+1;
}
if(x[k]<=n)
{
if(k==n)
{
count++;
printf("\n\tPlacement %d is : \n\n\n",count);
print_grid(n,x);
getch();
}
else
{
k++;
x[k]=0;
}
}
else
{
k--;
}
}
return;
}
int solve(grid board, int i, int j){
if(i==9){
printf("\n");
printf("Solved!\n\n");
print_grid(board);
return 1;
}
if(board.array[i][j]){
if (j == 8) {
if(solve(board, i+1, 0)){
return 1;
}
}else{
if(solve(board, i, j+1)){
return 1;
}
}
}
for (int possible = 1; possible < 10; possible++) {
if (validate(board, possible, i, j)) {
board.array[i][j] = possible;
if (j == 8) {
if(solve(board, i+1, 0)){
return 1;
}
}else{
if(solve(board, i, j+1)){
return 1;
}
}
}
}
board.array[i][j] = 0;
return 0;
}
gmx_bool pme_load_balance(pme_load_balancing_t pme_lb,
t_commrec *cr,
FILE *fp_err,
FILE *fp_log,
t_inputrec *ir,
t_state *state,
double cycles,
interaction_const_t *ic,
struct nonbonded_verlet_t *nbv,
struct gmx_pme_t ** pmedata,
gmx_int64_t step)
{
gmx_bool OK;
pme_setup_t *set;
double cycles_fast;
char buf[STRLEN], sbuf[22];
real rtab;
gmx_bool bUsesSimpleTables = TRUE;
if (pme_lb->stage == pme_lb->nstage)
{
return FALSE;
}
if (PAR(cr))
{
gmx_sumd(1, &cycles, cr);
cycles /= cr->nnodes;
}
set = &pme_lb->setup[pme_lb->cur];
set->count++;
rtab = ir->rlistlong + ir->tabext;
if (set->count % 2 == 1)
{
/* Skip the first cycle, because the first step after a switch
* is much slower due to allocation and/or caching effects.
*/
return TRUE;
}
sprintf(buf, "step %4s: ", gmx_step_str(step, sbuf));
print_grid(fp_err, fp_log, buf, "timed with", set, cycles);
if (set->count <= 2)
{
set->cycles = cycles;
}
else
{
if (cycles*PME_LB_ACCEL_TOL < set->cycles &&
pme_lb->stage == pme_lb->nstage - 1)
{
/* The performance went up a lot (due to e.g. DD load balancing).
* Add a stage, keep the minima, but rescan all setups.
*/
pme_lb->nstage++;
if (debug)
{
fprintf(debug, "The performance for grid %d %d %d went from %.3f to %.1f M-cycles, this is more than %f\n"
"Increased the number stages to %d"
" and ignoring the previous performance\n",
set->grid[XX], set->grid[YY], set->grid[ZZ],
cycles*1e-6, set->cycles*1e-6, PME_LB_ACCEL_TOL,
pme_lb->nstage);
}
}
set->cycles = min(set->cycles, cycles);
}
if (set->cycles < pme_lb->setup[pme_lb->fastest].cycles)
{
pme_lb->fastest = pme_lb->cur;
if (DOMAINDECOMP(cr))
{
/* We found a new fastest setting, ensure that with subsequent
* shorter cut-off's the dynamic load balancing does not make
* the use of the current cut-off impossible. This solution is
* a trade-off, as the PME load balancing and DD domain size
* load balancing can interact in complex ways.
* With the Verlet kernels, DD load imbalance will usually be
* mainly due to bonded interaction imbalance, which will often
* quickly push the domain boundaries beyond the limit for the
* optimal, PME load balanced, cut-off. But it could be that
* better overal performance can be obtained with a slightly
* shorter cut-off and better DD load balancing.
*/
change_dd_dlb_cutoff_limit(cr);
}
}
cycles_fast = pme_lb->setup[pme_lb->fastest].cycles;
/* Check in stage 0 if we should stop scanning grids.
* Stop when the time is more than SLOW_FAC longer than the fastest.
*/
if (pme_lb->stage == 0 && pme_lb->cur > 0 &&
cycles > pme_lb->setup[pme_lb->fastest].cycles*PME_LB_SLOW_FAC)
{
pme_lb->n = pme_lb->cur + 1;
/* Done with scanning, go to stage 1 */
switch_to_stage1(pme_lb);
}
if (pme_lb->stage == 0)
{
int gridsize_start;
gridsize_start = set->grid[XX]*set->grid[YY]*set->grid[ZZ];
do
{
if (pme_lb->cur+1 < pme_lb->n)
{
/* We had already generated the next setup */
OK = TRUE;
}
else
{
/* Find the next setup */
OK = pme_loadbal_increase_cutoff(pme_lb, ir->pme_order, cr->dd);
if (!OK)
{
pme_lb->elimited = epmelblimPMEGRID;
}
}
if (OK && ir->ePBC != epbcNONE)
{
OK = (sqr(pme_lb->setup[pme_lb->cur+1].rlistlong)
<= max_cutoff2(ir->ePBC, state->box));
if (!OK)
{
pme_lb->elimited = epmelblimBOX;
}
}
if (OK)
{
pme_lb->cur++;
if (DOMAINDECOMP(cr))
{
OK = change_dd_cutoff(cr, state, ir,
pme_lb->setup[pme_lb->cur].rlistlong);
if (!OK)
{
/* Failed: do not use this setup */
pme_lb->cur--;
pme_lb->elimited = epmelblimDD;
}
}
}
if (!OK)
{
/* We hit the upper limit for the cut-off,
* the setup should not go further than cur.
*/
pme_lb->n = pme_lb->cur + 1;
print_loadbal_limited(fp_err, fp_log, step, pme_lb);
/* Switch to the next stage */
switch_to_stage1(pme_lb);
}
}
while (OK &&
!(pme_lb->setup[pme_lb->cur].grid[XX]*
pme_lb->setup[pme_lb->cur].grid[YY]*
pme_lb->setup[pme_lb->cur].grid[ZZ] <
gridsize_start*PME_LB_GRID_SCALE_FAC
&&
pme_lb->setup[pme_lb->cur].grid_efficiency <
pme_lb->setup[pme_lb->cur-1].grid_efficiency*PME_LB_GRID_EFFICIENCY_REL_FAC));
}
if (pme_lb->stage > 0 && pme_lb->end == 1)
{
pme_lb->cur = 0;
pme_lb->stage = pme_lb->nstage;
}
else if (pme_lb->stage > 0 && pme_lb->end > 1)
{
/* If stage = nstage-1:
* scan over all setups, rerunning only those setups
* which are not much slower than the fastest
* else:
* use the next setup
*/
do
{
pme_lb->cur++;
if (pme_lb->cur == pme_lb->end)
{
pme_lb->stage++;
pme_lb->cur = pme_lb->start;
}
}
while (pme_lb->stage == pme_lb->nstage - 1 &&
pme_lb->setup[pme_lb->cur].count > 0 &&
pme_lb->setup[pme_lb->cur].cycles > cycles_fast*PME_LB_SLOW_FAC);
if (pme_lb->stage == pme_lb->nstage)
{
/* We are done optimizing, use the fastest setup we found */
pme_lb->cur = pme_lb->fastest;
}
}
if (DOMAINDECOMP(cr) && pme_lb->stage > 0)
{
OK = change_dd_cutoff(cr, state, ir, pme_lb->setup[pme_lb->cur].rlistlong);
if (!OK)
{
/* Failsafe solution */
if (pme_lb->cur > 1 && pme_lb->stage == pme_lb->nstage)
{
pme_lb->stage--;
}
pme_lb->fastest = 0;
pme_lb->start = 0;
pme_lb->end = pme_lb->cur;
pme_lb->cur = pme_lb->start;
pme_lb->elimited = epmelblimDD;
print_loadbal_limited(fp_err, fp_log, step, pme_lb);
}
}
/* Change the Coulomb cut-off and the PME grid */
set = &pme_lb->setup[pme_lb->cur];
ic->rcoulomb = set->rcut_coulomb;
ic->rlist = set->rlist;
ic->rlistlong = set->rlistlong;
ir->nstcalclr = set->nstcalclr;
ic->ewaldcoeff_q = set->ewaldcoeff_q;
/* TODO: centralize the code that sets the potentials shifts */
if (ic->coulomb_modifier == eintmodPOTSHIFT)
{
ic->sh_ewald = gmx_erfc(ic->ewaldcoeff_q*ic->rcoulomb);
}
if (EVDW_PME(ic->vdwtype))
{
/* We have PME for both Coulomb and VdW, set rvdw equal to rcoulomb */
ic->rvdw = set->rcut_coulomb;
ic->ewaldcoeff_lj = set->ewaldcoeff_lj;
if (ic->vdw_modifier == eintmodPOTSHIFT)
{
real crc2;
ic->dispersion_shift.cpot = -pow(ic->rvdw, -6.0);
ic->repulsion_shift.cpot = -pow(ic->rvdw, -12.0);
ic->sh_invrc6 = -ic->dispersion_shift.cpot;
crc2 = sqr(ic->ewaldcoeff_lj*ic->rvdw);
ic->sh_lj_ewald = (exp(-crc2)*(1 + crc2 + 0.5*crc2*crc2) - 1)*pow(ic->rvdw, -6.0);
}
}
bUsesSimpleTables = uses_simple_tables(ir->cutoff_scheme, nbv, 0);
nbnxn_gpu_pme_loadbal_update_param(nbv, ic);
/* With tMPI + GPUs some ranks may be sharing GPU(s) and therefore
* also sharing texture references. To keep the code simple, we don't
* treat texture references as shared resources, but this means that
* the coulomb_tab texture ref will get updated by multiple threads.
* Hence, to ensure that the non-bonded kernels don't start before all
* texture binding operations are finished, we need to wait for all ranks
* to arrive here before continuing.
*
* Note that we could omit this barrier if GPUs are not shared (or
* texture objects are used), but as this is initialization code, there
* is not point in complicating things.
*/
#ifdef GMX_THREAD_MPI
if (PAR(cr) && use_GPU(nbv))
{
gmx_barrier(cr);
}
#endif /* GMX_THREAD_MPI */
/* Usually we won't need the simple tables with GPUs.
* But we do with hybrid acceleration and with free energy.
* To avoid bugs, we always re-initialize the simple tables here.
*/
init_interaction_const_tables(NULL, ic, bUsesSimpleTables, rtab);
if (cr->duty & DUTY_PME)
{
if (pme_lb->setup[pme_lb->cur].pmedata == NULL)
{
/* Generate a new PME data structure,
* copying part of the old pointers.
*/
gmx_pme_reinit(&set->pmedata,
cr, pme_lb->setup[0].pmedata, ir,
set->grid);
}
*pmedata = set->pmedata;
}
else
{
/* Tell our PME-only node to switch grid */
gmx_pme_send_switchgrid(cr, set->grid, set->ewaldcoeff_q, set->ewaldcoeff_lj);
}
if (debug)
{
print_grid(NULL, debug, "", "switched to", set, -1);
}
if (pme_lb->stage == pme_lb->nstage)
{
print_grid(fp_err, fp_log, "", "optimal", set, -1);
}
return TRUE;
}
/* Print the user statement of the host code to "p".
*
* The host code may contain original user statements, kernel launches,
* statements that copy data to/from the device and statements
* the initialize or clear the device.
* The original user statements and the kernel launches have
* an associated annotation, while the other statements do not.
* The latter are handled by print_device_node.
* The annotation on the user statements is called "user".
*
* In case of a kernel launch, print a block of statements that
* defines the grid and the block and then launches the kernel.
*/
static __isl_give isl_printer *print_host_user(__isl_take isl_printer *p,
__isl_take isl_ast_print_options *print_options,
__isl_keep isl_ast_node *node, void *user)
{
isl_id *id;
int is_user;
struct ppcg_kernel *kernel;
struct ppcg_kernel_stmt *stmt;
struct print_host_user_data *data;
isl_ast_print_options_free(print_options);
data = (struct print_host_user_data *) user;
id = isl_ast_node_get_annotation(node);
if (!id)
{
//p = isl_printer_print_str(p,"marker_NO_ID_CASE");
return print_device_node(p, node, data->prog);
}
is_user = !strcmp(isl_id_get_name(id), "user");
kernel = is_user ? NULL : isl_id_get_user(id);
stmt = is_user ? isl_id_get_user(id) : NULL;
isl_id_free(id);
if (is_user)
return ppcg_kernel_print_domain(p, stmt);
p = ppcg_start_block(p);
p = isl_printer_start_line(p);
p = isl_printer_print_str(p, "dim3 k");
p = isl_printer_print_int(p, kernel->id);
p = isl_printer_print_str(p, "_dimBlock");
print_reverse_list(isl_printer_get_file(p),
kernel->n_block, kernel->block_dim);
p = isl_printer_print_str(p, ";");
p = isl_printer_end_line(p);
p = print_grid(p, kernel);
p = isl_printer_start_line(p);
p = isl_printer_print_str(p, "kernel");
p = isl_printer_print_int(p, kernel->id);
p = isl_printer_print_str(p, " <<<k");
p = isl_printer_print_int(p, kernel->id);
p = isl_printer_print_str(p, "_dimGrid, k");
p = isl_printer_print_int(p, kernel->id);
p = isl_printer_print_str(p, "_dimBlock>>> (");
p = print_kernel_arguments(p, data->prog, kernel, 0);
p = isl_printer_print_str(p, ");");
p = isl_printer_end_line(p);
p = isl_printer_start_line(p);
p = isl_printer_print_str(p, "cudaCheckKernel();");
p = isl_printer_end_line(p);
p = ppcg_end_block(p);
p = isl_printer_start_line(p);
p = isl_printer_end_line(p);
p = copy_data_from_device_to_device(p,kernel);
printf("printing kernel");
print_kernel(data->prog, kernel, data->cuda);
printf("printing kernel done");
return p;
}
void show_grid() {
print_grid(gridP);
}
static void print_prompt(FILE *fp, const game_t *game)
{
print_grid(fp, game);
xprintf(fp, "Player %c> ", game->current_player);
}
