void test(int M)
{
/* Original iterators. */
int i, j;
S3(2,1) ;
S1(3,1) ;
S1(4,1) ;
S4(4,2) ;
for (i=5;i<=M+1;i++) {
S1(i,1) ;
for (j=2;j<=floord(i-1,2);j++) {
S2(i,j) ;
}
if (i%2 == 0) {
S4(i,i/2) ;
}
}
for (i=M+2;i<=2*M-1;i++) {
for (j=i-M;j<=floord(i-1,2);j++) {
S2(i,j) ;
}
if (i%2 == 0) {
S4(i,i/2) ;
}
}
i = 2*M ;
S4(2*M,M) ;
}
void foo(int N)
{
int i;
int a[N];
#pragma scop
for (i = 0; i < floord(N, 2); ++i)
a[i] = i;
#pragma endscop
}
void test(int M)
{
/* Scattering iterators. */
int c1, c2, c3, c4;
/* Original iterators. */
int i, j, k, l;
if (M >= 3) {
for (c1=-1;c1<=min(2,floord(M+2,4));c1++) {
for (c2=max(ceild(2*c1-M+1,4),ceild(4*c1-M-2,4));c2<=min(0,floord(c1,2));c2++) {
for (c3=max(max(-4*c2-2,4*c2+3),4*c1-4*c2+1);c3<=min(min(min(M+3,-4*c2+9),4*c2+2*M),4*c1-4*c2+4);c3++) {
for (c4=max(3*c3-4*floord(c3+M+1,2)+6,4*c2-c3-4*floord(-c3+1,4)+2);c4<=min(min(4*c2+4,-c3+10),c3-2);c4+=4) {
if ((c2 <= floord(c4-1,4)) && (c2 >= ceild(c4-4,4))) {
S1(c1-c2,c2,(c3+c4-2)/4,(c3-c4)/2);
}
}
}
}
}
}
}
void test(int n)
{
/* Scattering iterators. */
int c1, c2;
/* Original iterators. */
int i, j;
for (c1=0;c1<=5*n;c1++) {
for (c2=max(c1-n,ceild(2*c1,3));c2<=min(c1,floord(2*c1+2*n,3));c2++) {
if (c2%2 == 0) {
i = (-2*c1+3*c2)/2 ;
j = c1-c2 ;
S1((-2*c1+3*c2)/2,c1-c2) ;
}
}
}
}
/* Generated from ../../../git/cloog/test/classen2.cloog by CLooG 0.14.0-271-gaa1e292 gmp bits in 0.14s. */
if ((M >= 2) && (N >= 3) && (outerProcTileScatter1 >= outerProcTileScatter2) && (5*outerProcTileScatter1 <= M+2*N-4) && (5*outerProcTileScatter1 <= 5*outerProcTileScatter2+N+2) && (outerProcTileScatter2 >= 0) && (5*outerProcTileScatter2 <= M+N-2) && (outerTimeTileScatter >= outerProcTileScatter1) && (outerTimeTileScatter <= 2*outerProcTileScatter1) && (outerTimeTileScatter <= outerProcTileScatter1+outerProcTileScatter2+1) && (5*outerTimeTileScatter <= 2*M+2*N-6) && (5*outerTimeTileScatter <= 5*outerProcTileScatter1+M+2) && (5*outerTimeTileScatter >= 10*outerProcTileScatter1-2*N-2) && (5*outerTimeTileScatter <= 5*outerProcTileScatter2+M+N) && (5*outerTimeTileScatter >= 10*outerProcTileScatter2-N-3) && (5*outerTimeTileScatter <= 10*outerProcTileScatter2+N+3) && (5*outerTimeTileScatter >= 5*outerProcTileScatter1+5*outerProcTileScatter2-N-4)) {
for (compScatter1=max(max(max(max(max(4,5*outerTimeTileScatter),5*outerProcTileScatter2+1),5*outerProcTileScatter1+5*outerProcTileScatter2-N),10*outerProcTileScatter1-2*N+2),10*outerProcTileScatter2-N+1);compScatter1<=min(min(min(min(min(5*outerTimeTileScatter+4,2*M+2*N-6),5*outerProcTileScatter1+M+2),5*outerProcTileScatter1+5*outerProcTileScatter2+5),5*outerProcTileScatter2+M+N),10*outerProcTileScatter2+N+3);compScatter1++) {
for (compScatter2=max(max(max(max(ceild(compScatter1+4,2),5*outerProcTileScatter1),5*outerProcTileScatter2+1),compScatter1-M+2),compScatter1-5*outerProcTileScatter2-1);compScatter2<=min(min(min(min(floord(compScatter1+2*N-2,2),compScatter1),5*outerProcTileScatter1+4),compScatter1-5*outerProcTileScatter2+N),5*outerProcTileScatter2+N+2);compScatter2++) {
for (compScatter3=max(max(5*outerProcTileScatter2,compScatter1-compScatter2+3),compScatter2-N+2);compScatter3<=min(min(compScatter2-1,5*outerProcTileScatter2+4),compScatter1-compScatter2+N);compScatter3++) {
S1(compScatter1-compScatter2+1,-compScatter1+compScatter2+compScatter3-2,compScatter2-compScatter3,compScatter1,compScatter2,compScatter3);
}
}
}
}
double test_1(){
double start_time = omp_get_wtime();
int read=0, write = 1;
// s is the number of non-pointy bit 2D slices of diamond tiling
// that is available for the current tile size.
int s = (tau/3) - 2;
// subset_s is an input parameter indicating how many of those
// slices we want to use in the repeated tiling pattern.
// subset_s should be less than s and greater than or equal to 2.
if (subset_s > s || subset_s<2) {
fprintf(stderr, "Error: need 2<=subset_s<=s\n");
exit(-1);
}
// Set lower and upper bounds for spatial dimensions.
// When did code gen have a non-inclusive upper bound.
// Ian's upper bound is inclusive.
int Li=1, Lj=1, Ui=upperBound+1, Uj=upperBound+1;
// Loop over the tiling pattern.
for (int toffset=0; toffset<T; toffset+=subset_s){
// Loop over phases of tiles within repeated tile pattern.
// This is like iterating over the A and B trapezoid tile types.
for (int c0 = -2; c0 <= 0; c0 += 1){
// Two loops over tiles within one phase.
// All of the tiles within one phase can be done in parallel.
// updates by Dave W, to the c1 and c2 loops, for OpenMP (from here to the end of the #if BOUNDING_BOX_FOR_PARALLEL_LOOPS
// hoist out min_c1 and max_c1, then use that to hoist a bounding box for c2
// initial version is just aiming for correct and parallel, without worrying about a loose boundingbox
int c1_lb =
max(
max(
floord(Lj + (tau/3) * c0 + (tau/3), tau),
c0 + floord(-2 * T + Lj - 1, tau) + 1),
floord(Lj + 1, tau)
); // end init block c1
int c1_ub =
min(
min(
floord(Uj + (tau/3) * c0 - ((tau/3)+2), tau) + 1,
floord(T + Uj - 1, tau)),
c0 + floord(Uj - 5, tau) + 2
); // end cond block c1
// The two expressions below are the same as in the previous version, except that
// in the c2_lb_min_expr, I have replaced c1 with:
// c1_min_value where it appears with a positive coefficient, and
// c1_max_value where it appears with a negative coefficient.
// and in the c2_ub_max_expr, the opposite (i.e., c1 becomes c1_max_value where positive)
// I will be embarrassed if I have done this wrong.
/// Note that I assume tau > 0
#define c2_lb_min_expr(c1_min_value, c1_max_value) \
max( \
max( \
max( \
max( \
max( \
max( \
c0 - 2 * c1_max_value + floord(-Ui + Lj + 1,tau), \
-c1_max_value + floord(-2 * Ui - Uj + tau * c0 + tau * c1_min_value - tau-3, tau*2)+1), \
c1_min_value + floord(-Ui - 2 * Uj + 3, tau)), \
floord(-Ui - Uj + 3, tau)), \
c0 - c1_max_value + floord(-Ui - (tau/3) * c0 + ((tau/3)+1), tau)), \
c0 - c1_max_value + floord(-T - Ui, tau) + 1), \
-c1_max_value + floord(-Ui + 4, tau) - 1 \
) /* end init block c2 */
#define c2_ub_max_expr(c1_min_value, c1_max_value) \
min( \
min( \
min( \
min( \
min( \
min( \
c0 - 2 * c1_min_value + floord(-Li + Uj - 2, tau) + 1, \
c0 - c1_min_value + floord(-Li - 2, tau) + 1), \
c0 - c1_min_value + floord(-Li - (tau/3) * c0 - ((tau/3)+1), tau) + 1), \
floord(T - Li - Lj, tau)), \
-c1_min_value + floord(2 * T - Li, tau)), \
c1_max_value + floord(-Li - 2 * Lj - 1, tau) + 1), \
-c1_min_value + floord(-2 * Li - Lj + tau * c0 + tau * c1_max_value + (tau-1), tau*2) \
) /* end cond block c2 */
#define c2_lb_expr(c1_value) c2_lb_min_expr(c1_value, c1_value)
#define c2_ub_expr(c1_value) c2_ub_max_expr(c1_value, c1_value)
#if BOUNDING_BOX_FOR_PARALLEL_LOOPS
int c2_box_lb = c2_lb_min_expr(c1_lb, c1_ub);
int c2_box_ub = c2_ub_max_expr(c1_lb, c1_ub);
#if PARALLEL
// don't need to mention c1...c5 below, since they're scoped inside the for loops
#pragma omp parallel for shared(start_time, s, Li, Lj, Ui, Uj, toffset, c0, c1_lb, c1_ub, c2_box_lb, c2_box_ub, ) private(read, write) collapse(2)
#endif
for (int c1 = c1_lb; c1 <= c1_ub; c1 += 1) {
for (int c2 = c2_box_lb; c2 <= c2_box_ub; c2 += 1) if (c2 >= c2_lb_expr(c1) && c2 <= c2_ub_expr(c1)) {
#else
for (int c1 = c1_lb; c1 <= c1_ub; c1 += 1) {
for (int c2 = c2_lb_expr(c1); c2 <= c2_ub_expr(c1); c2 += 1) {
#endif
//fprintf(stdout, "(%d,%d,%d)\n", c0,c1,c2);
// Loop over subset_s time steps within tiling pattern
// and within tile c0,c1,c2.
// Every time the pattern is repeated, toffset will be
// subset_s bigger.
// The real t value is c3+toffset. We are just using the
// tiling pattern from t=1 to t<=subset_s.
for (int c3 = 1; c3 <= min(T-toffset,subset_s); c3 += 1){
int t = c3+toffset;
// if t % 2 is 1, then read=0 and write=1
write = t & 1;
read = 1-write;
// x spatial dimension.
for (int c4 =
max(
max(
max(
-tau * c1 - tau * c2 + 2 * c3 - (2*tau-2),
-Uj - tau * c2 + c3 - (tau-2)),
tau * c0 - tau * c1 - tau * c2 - c3),
Li
); // end init block c4
c4 <=
min(
min(
min(
tau * c0 - tau * c1 - tau * c2 - c3 + (tau-1),
-tau * c1 - tau * c2 + 2 * c3),
-Lj - tau * c2 + c3),
Ui - 1
); // end cond block c4
c4 += 1){
// y spatial dimension.
for (int c5 =
max(
max(
tau * c1 - c3,
Lj),
-tau * c2 + c3 - c4 - (tau-1)
); // end init block c5
c5 <=
min(
min(
Uj - 1,
-tau * c2 + c3 - c4),
tau * c1 - c3 + (tau-1)
); // end cond block c5
c5 += 1){
//fprintf(stdout, "(%d,%d,%d,%d,%d,%d)\n", c0,c1,c2,c3,c4,c5);
stencil( read, write, c4, c5);
} // for c5
} // for c4
} // for c3
} // for c2
} // for c1
} // for c0
} // for toffset
double end_time = omp_get_wtime();
return (end_time - start_time);
}
int main( int argc, char* argv[] ){
setbuf(stdout, NULL); // set buffer to null, so prints ALWAYS print (for debug purposes mainly)
bool verify = false;
bool printtime = true;
// Command line parsing
char c;
while ((c = getopt (argc, argv, "nc:s:p:T:t:hv")) != -1){
switch( c ) {
case 'n':
printtime=false;
break;
case 'c': // problem size
cores = parseInt( optarg );
if( cores <= 0 ){
fprintf(stderr, "cores must be greater than 0: %d\n", cores);
exit(BAD_RUN_TIME_PARAMETERS);
}
break;
case 's': // subset
//globalSeed = parseInt( optarg );
subset_s = parseInt( optarg );
break;
case 'p': // problem size
problemSize = parseInt( optarg );
if( problemSize <= 0 ){
fprintf(stderr, "problemSize must be greater than 0: %d\n", problemSize);
exit(BAD_RUN_TIME_PARAMETERS);
}
break;
case 'T': // T (time steps)
T = parseInt( optarg );
if( T <= 0 ){
fprintf(stderr, "T must be greater than 0: %d\n", T);
exit(BAD_RUN_TIME_PARAMETERS);
}
break;
case 't': // tau
#if defined tau
fprintf(stderr, "don't use -t to set tau when you compiled with -Dtau=%d.\n", tau);
if (parseInt(optarg) != tau)
exit(BAD_COMPILE_TIME_PARAMETERS);
#else
tau = parseInt( optarg );
#endif
break;
case 'h': // help
printf("usage: %s\n-n \t dont print time \n-p <problem size> \t problem size in elements \n-T <time steps>\t number of time steps\n-c <cores>\tnumber of threads\n-s <subset_s>\t tile parameter\n-t <tau>\t tile parameter\n-h\tthis dialogue\n-v\tverify output\n", argv[0]);
exit(0);
case 'v': // verify;
verify = true;
break;
case '?':
if (optopt == 'p')
fprintf (stderr, "Option -%c requires positive int argument: problem size.\n", optopt);
else if (optopt == 'T')
fprintf (stderr, "Option -%c requires positive int argument: T.\n", optopt);
else if (optopt == 's')
fprintf (stderr, "Option -%c requires int argument: subset_s.\n", optopt);
else if (optopt == 'c')
fprintf (stderr, "Option -%c requires int argument: number of cores.\n", optopt);
else if (isprint (optopt))
fprintf (stderr, "Unknown option `-%c'.\n", optopt);
else
fprintf(stderr, "Unknown option character `\\x%x'.\n", optopt);
exit(0);
default:
exit(0);
}
}
if( !( tau % 3 == 0 && tau >= 15 ) ){
#if defined tau
fprintf(stderr, "tau must be a multiple of 3, and >= 15, but the program was compiled with -Dtau=%d, and thus can't run :-(\n", tau);
exit(BAD_COMPILE_TIME_PARAMETERS);
#else
fprintf(stderr, "tau must be a multiple of 3, and >= 15, but it's %d; re-run with a different -t value\n", tau);
exit(BAD_RUN_TIME_PARAMETERS);
#endif
}
init();
initSpace();
double time = test_1();
if( printtime ) {
printf( "Time: %f\n", time );
}
if( verify ){
verifyResult( true );
}
}
int main(void) {
int t, y, x, k;
double total_lattice_pts = (double)nY * (double)nX * (double)nTimesteps;
/* For timekeeping */
int ts_return = -1;
struct timeval start, end, result;
double tdiff = 0.0;
/* Compute values for global parameters */
omega = 1.0 / tau;
circle_R2 = circle_radius * circle_radius;
double rho_avg = (rho_in + rho_out) / 2.0;
printf(
"2D Flow Past Cylinder simulation with D2Q9 lattice:\n"
"\tscheme : 2-Grid, Fused, Pull\n"
"\tgrid size : %d x %d = %.2lf * 10^3 Cells\n"
"\tnTimeSteps : %d\n"
"\tomega : %.6lf\n",
nX, nY, nX * nY / 1.0e3, nTimesteps, omega);
/* Initialize all 9 PDFs for each point in the domain to 1.0 */
for (y = 0; y < nY + 2 + 4; y++) {
for (x = 0; x < nX + 2 + 2; x++) {
grid[0][y][x][0] = w1 * rho_avg;
grid[1][y][x][0] = w1 * rho_avg;
for (k = 1; k < 5; k++) {
grid[0][y][x][k] = w2 * rho_avg;
grid[1][y][x][k] = w2 * rho_avg;
}
for (k = 5; k < nK; k++) {
grid[0][y][x][k] = w3 * rho_avg;
grid[1][y][x][k] = w3 * rho_avg;
}
}
}
/* To satisfy PET */
short _nX = nX + 2;
short _nY = nY + 3;
int _nTimesteps = nTimesteps;
#ifdef TIME
gettimeofday(&start, 0);
#endif
int t1, t2, t3, t4, t5, t6;
int lb, ub, lbp, ubp, lb2, ub2;
register int lbv, ubv;
/* Start of CLooG code */
if ((_nTimesteps >= 1) && (_nX >= 3) && (_nY >= 4)) {
for (t1 = -1; t1 <= floord(5 * _nTimesteps + 3 * _nY - 8, 32); t1++) {
lbp = max(max(ceild(4 * t1, 5), ceild(16 * t1 - _nTimesteps + 1, 12)),
ceild(32 * t1 - _nTimesteps + 4, 32));
ubp = min(min(floord(4 * t1 + 4, 3), floord(_nTimesteps + _nY - 2, 8)),
floord(16 * t1 + _nY + 14, 20));
#pragma omp parallel for private(lbv, ubv, t3, t4, t5, t6)
for (t2 = lbp; t2 <= ubp; t2++) {
for (t3 = max(max(0, ceild(4 * t1 - 3 * t2 - 1, 2)),
ceild(8 * t2 - _nY - 4, 8));
t3 <= min(min(min(floord(_nTimesteps + _nX - 2, 8),
floord(8 * t2 + _nX + 3, 8)),
floord(16 * t1 - 12 * t2 + _nX + 18, 8)),
floord(32 * t1 - 32 * t2 + _nY + _nX + 29, 8));
t3++) {
for (t4 = max(
max(max(max(0, 16 * t1 - 12 * t2), 32 * t1 - 32 * t2 + 3),
8 * t2 - _nY + 1),
8 * t3 - _nX + 1);
t4 <= min(min(min(min(_nTimesteps - 1, 8 * t2 + 4), 8 * t3 + 5),
16 * t1 - 12 * t2 + 19),
32 * t1 - 32 * t2 + _nY + 30);
t4++) {
/* Hoisted loop conditional */
if (t4 % 2 == 0) {
for (t5 = max(max(8 * t2, t4 + 3),
-32 * t1 + 32 * t2 + 2 * t4 - 31);
t5 <= min(min(8 * t2 + 7, -32 * t1 + 32 * t2 + 2 * t4),
t4 + _nY - 1);
t5++) {
lbv = max(8 * t3, t4 + 2);
ubv = min(8 * t3 + 7, t4 + _nX - 1);
#pragma ivdep
#pragma vector always
for (t6 = lbv; t6 <= ubv; t6++) {
lbm_kernel(grid[0][(-t4 + t5)][(-t4 + t6)][0],
grid[0][(-t4 + t5) - 1][(-t4 + t6)][3],
grid[0][(-t4 + t5) + 1][(-t4 + t6)][4],
grid[0][(-t4 + t5)][(-t4 + t6) - 1][1],
grid[0][(-t4 + t5)][(-t4 + t6) + 1][2],
grid[0][(-t4 + t5) - 1][(-t4 + t6) - 1][5],
grid[0][(-t4 + t5) - 1][(-t4 + t6) + 1][6],
grid[0][(-t4 + t5) + 1][(-t4 + t6) - 1][7],
grid[0][(-t4 + t5) + 1][(-t4 + t6) + 1][8],
&grid[1][(-t4 + t5)][(-t4 + t6)][0],
&grid[1][(-t4 + t5)][(-t4 + t6)][3],
&grid[1][(-t4 + t5)][(-t4 + t6)][4],
&grid[1][(-t4 + t5)][(-t4 + t6)][1],
&grid[1][(-t4 + t5)][(-t4 + t6)][2],
&grid[1][(-t4 + t5)][(-t4 + t6)][5],
&grid[1][(-t4 + t5)][(-t4 + t6)][6],
&grid[1][(-t4 + t5)][(-t4 + t6)][7],
&grid[1][(-t4 + t5)][(-t4 + t6)][8], (t4),
((-t4 + t5)), ((-t4 + t6)));
;
}
}
} else {
for (t5 = max(max(8 * t2, t4 + 3),
-32 * t1 + 32 * t2 + 2 * t4 - 31);
t5 <= min(min(8 * t2 + 7, -32 * t1 + 32 * t2 + 2 * t4),
t4 + _nY - 1);
t5++) {
lbv = max(8 * t3, t4 + 2);
ubv = min(8 * t3 + 7, t4 + _nX - 1);
#pragma ivdep
#pragma vector always
for (t6 = lbv; t6 <= ubv; t6++) {
lbm_kernel(grid[1][(-t4 + t5)][(-t4 + t6)][0],
grid[1][(-t4 + t5) - 1][(-t4 + t6)][3],
grid[1][(-t4 + t5) + 1][(-t4 + t6)][4],
grid[1][(-t4 + t5)][(-t4 + t6) - 1][1],
grid[1][(-t4 + t5)][(-t4 + t6) + 1][2],
grid[1][(-t4 + t5) - 1][(-t4 + t6) - 1][5],
grid[1][(-t4 + t5) - 1][(-t4 + t6) + 1][6],
grid[1][(-t4 + t5) + 1][(-t4 + t6) - 1][7],
grid[1][(-t4 + t5) + 1][(-t4 + t6) + 1][8],
&grid[0][(-t4 + t5)][(-t4 + t6)][0],
&grid[0][(-t4 + t5)][(-t4 + t6)][3],
&grid[0][(-t4 + t5)][(-t4 + t6)][4],
&grid[0][(-t4 + t5)][(-t4 + t6)][1],
&grid[0][(-t4 + t5)][(-t4 + t6)][2],
&grid[0][(-t4 + t5)][(-t4 + t6)][5],
&grid[0][(-t4 + t5)][(-t4 + t6)][6],
&grid[0][(-t4 + t5)][(-t4 + t6)][7],
&grid[0][(-t4 + t5)][(-t4 + t6)][8], (t4),
((-t4 + t5)), ((-t4 + t6)));
;
}
}
}
/* end hoisted if */
}
}
}
}
}
/* End of CLooG code */
#ifdef TIME
gettimeofday(&end, 0);
ts_return = timeval_subtract(&result, &end, &start);
tdiff = (double)(result.tv_sec + result.tv_usec * 1.0e-6);
printf("\tTime taken : %7.5lfs\n", tdiff);
printf("\tMLUPS : %7.5lf\n", (total_lattice_pts / (1.0e6 * tdiff)));
#endif
#ifdef DEBUG
/* Dump rho, uX, uY for the entire domain to verify results */
dumpVelocities(t);
#endif
return 0;
}
if (m >= 8 * floord(m + 1, 8))
for (int c0 = 4 * floord(m + 1, 32); c0 <= n; c0 += 1)
s0(c0);
/* Generated from ./non_optimal/nul_complex1.cloog by CLooG 0.18.1-2-g43fc508 gmp bits in 0.00s. */
if (n >= 0) {
for (c1=0;c1<=5*n;c1++) {
for (c2=max(ceild(2*c1,3),c1-n);c2<=min(floord(2*c1+2*n,3),c1);c2++) {
if (c2%2 == 0) {
S1(((-2*c1+3*c2)/2),(c1-c2));
}
}
}
}
int main()
{
init_arrays();
double annot_t_start=0, annot_t_end=0, annot_t_total=0;
int annot_i;
for (annot_i=0; annot_i<REPS; annot_i++)
{
annot_t_start = rtclock();
#define ceild(n,d) ceil(((double)(n))/((double)(d)))
#define floord(n,d) floor(((double)(n))/((double)(d)))
#define max(x,y) ((x) > (y)? (x) : (y))
#define min(x,y) ((x) < (y)? (x) : (y))
#define S1(zT0,zT1,zT2,zT3,i,j) {B[i][j]=u2[i]*v2[j]+u1[i]*v1[j]+A[i][j];}
#define S2(zT0,zT1,zT2,zT3,i,j) {x[i]=beta*B[j][i]*y[j]+x[i];}
#define S3(i) {x[i]=z[i]+x[i];}
#define S4(i,j) {w[i]=alpha*B[i][j]*x[j]+w[i];}
int c1, c2, c3, c4, c5, c6, c7, c8, c9, c10;
register int lbv, ubv;
/* Generated from PLuTo-produced CLooG file by CLooG v0.14.1 64 bits in 0.05s. */
for (c2=0;c2<=floord(N-1,256);c2++) {
for (c3=0;c3<=floord(N-1,256);c3++) {
for (c4=max(0,8*c2);c4<=min(8*c2+7,floord(N-1,32));c4++) {
for (c5=max(8*c3,0);c5<=min(floord(N-1,32),8*c3+7);c5++) {
/*@ begin Loop(
transform UnrollJam(ufactor=32)
for (c6=max(32*c5,0);c6<=min(N-1,32*c5+31);c6++)
{
{
lbv=max(32*c4,0);
ubv=min(N-1,32*c4+31);
#pragma ivdep
#pragma vector always
for (c7=lbv; c7<=ubv; c7++) {
S1(c3,c2,c5,c4,c6,c7) ;
S2(c2,c3,c4,c5,c7,c6) ;
}
}
}
) @*/{
for (c6 = max(32 * c5, 0); c6 <= min(N - 1, 32 * c5 + 31) - 31; c6 = c6 + 32)
{
lbv=max(32*c4,0);
ubv=min(N-1,32*c4+31);
#pragma ivdep
#pragma vector always
for (c7=lbv; c7<=ubv; c7++) {
S1(c3, c2, c5, c4, c6, c7);
S2(c2, c3, c4, c5, c7, c6);
S1(c3, c2, c5, c4, (c6 + 1), c7);
S2(c2, c3, c4, c5, c7, (c6 + 1));
S1(c3, c2, c5, c4, (c6 + 2), c7);
S2(c2, c3, c4, c5, c7, (c6 + 2));
S1(c3, c2, c5, c4, (c6 + 3), c7);
S2(c2, c3, c4, c5, c7, (c6 + 3));
S1(c3, c2, c5, c4, (c6 + 4), c7);
S2(c2, c3, c4, c5, c7, (c6 + 4));
S1(c3, c2, c5, c4, (c6 + 5), c7);
S2(c2, c3, c4, c5, c7, (c6 + 5));
S1(c3, c2, c5, c4, (c6 + 6), c7);
S2(c2, c3, c4, c5, c7, (c6 + 6));
S1(c3, c2, c5, c4, (c6 + 7), c7);
S2(c2, c3, c4, c5, c7, (c6 + 7));
S1(c3, c2, c5, c4, (c6 + 8), c7);
S2(c2, c3, c4, c5, c7, (c6 + 8));
S1(c3, c2, c5, c4, (c6 + 9), c7);
S2(c2, c3, c4, c5, c7, (c6 + 9));
S1(c3, c2, c5, c4, (c6 + 10), c7);
S2(c2, c3, c4, c5, c7, (c6 + 10));
S1(c3, c2, c5, c4, (c6 + 11), c7);
S2(c2, c3, c4, c5, c7, (c6 + 11));
S1(c3, c2, c5, c4, (c6 + 12), c7);
S2(c2, c3, c4, c5, c7, (c6 + 12));
S1(c3, c2, c5, c4, (c6 + 13), c7);
S2(c2, c3, c4, c5, c7, (c6 + 13));
S1(c3, c2, c5, c4, (c6 + 14), c7);
S2(c2, c3, c4, c5, c7, (c6 + 14));
S1(c3, c2, c5, c4, (c6 + 15), c7);
S2(c2, c3, c4, c5, c7, (c6 + 15));
S1(c3, c2, c5, c4, (c6 + 16), c7);
S2(c2, c3, c4, c5, c7, (c6 + 16));
S1(c3, c2, c5, c4, (c6 + 17), c7);
S2(c2, c3, c4, c5, c7, (c6 + 17));
S1(c3, c2, c5, c4, (c6 + 18), c7);
S2(c2, c3, c4, c5, c7, (c6 + 18));
S1(c3, c2, c5, c4, (c6 + 19), c7);
S2(c2, c3, c4, c5, c7, (c6 + 19));
S1(c3, c2, c5, c4, (c6 + 20), c7);
S2(c2, c3, c4, c5, c7, (c6 + 20));
S1(c3, c2, c5, c4, (c6 + 21), c7);
S2(c2, c3, c4, c5, c7, (c6 + 21));
S1(c3, c2, c5, c4, (c6 + 22), c7);
S2(c2, c3, c4, c5, c7, (c6 + 22));
S1(c3, c2, c5, c4, (c6 + 23), c7);
S2(c2, c3, c4, c5, c7, (c6 + 23));
S1(c3, c2, c5, c4, (c6 + 24), c7);
S2(c2, c3, c4, c5, c7, (c6 + 24));
S1(c3, c2, c5, c4, (c6 + 25), c7);
S2(c2, c3, c4, c5, c7, (c6 + 25));
S1(c3, c2, c5, c4, (c6 + 26), c7);
S2(c2, c3, c4, c5, c7, (c6 + 26));
S1(c3, c2, c5, c4, (c6 + 27), c7);
S2(c2, c3, c4, c5, c7, (c6 + 27));
S1(c3, c2, c5, c4, (c6 + 28), c7);
S2(c2, c3, c4, c5, c7, (c6 + 28));
S1(c3, c2, c5, c4, (c6 + 29), c7);
S2(c2, c3, c4, c5, c7, (c6 + 29));
S1(c3, c2, c5, c4, (c6 + 30), c7);
S2(c2, c3, c4, c5, c7, (c6 + 30));
S1(c3, c2, c5, c4, (c6 + 31), c7);
S2(c2, c3, c4, c5, c7, (c6 + 31));
}
}
for (; c6 <= min(N - 1, 32 * c5 + 31); c6 = c6 + 1)
{
lbv=max(32*c4,0);
ubv=min(N-1,32*c4+31);
#pragma ivdep
#pragma vector always
for (c7=lbv; c7<=ubv; c7++) {
S1(c3, c2, c5, c4, c6, c7);
S2(c2, c3, c4, c5, c7, c6);
}
}
}
/*@ end @*/
}
}
}
}
for (c2=0;c2<=N-1;c2++) {
S3(c2) ;
}
for (c2=0;c2<=N-1;c2++) {
for (c3=0;c3<=N-1;c3++) {
S4(c2,c3) ;
}
}
/* End of CLooG code */
annot_t_end = rtclock();
annot_t_total += annot_t_end - annot_t_start;
}
annot_t_total = annot_t_total / REPS;
printf("%f\n", annot_t_total);
return ((int) w[0]);
}
for (int c0 = 4 * floord(m - 1, 12) + 4; c0 <= floord(n, 3); c0 += 4) s0(c0);
for (int c0 = 0; c0 <= 3; c0 += 1)
for (int c1 = max(0, 2 * c0 - 3); c1 <= min(c0 + c0 / 2 + 1, 3); c1 += 1)
for (int c2 = c0; c2 <= min(min(3, 3 * c1 + 2), 2 * c0 - c1 + 1); c2 += 1)
for (int c3 = max(max(max(c1 - (-c1 + 3) / 3, 0), c2 + floord(3 * c1 - c2 - 1, 6)), 2 * c0 - (c0 + c1 + 1) / 3 - 1); c3 <= min(c0 + 1, 3); c3 += 1)
for (int c4 = max(max(max(max(-200 * c1 + 400 * c3 - 199, 333 * c1 + c1 / 3), 333 * c2 + (c2 + 1) / 3), 667 * c0 - 333 * c1 - (c0 + c1 + 3) / 3 - 332), 250 * c3 + 1); c4 <= min(min(min(min(500 * c0 + 499, -200 * c1 + 400 * c3 + 400), 333 * c3 - (-c3 + 3) / 3 + 334), 1000), 333 * c2 - (-c2 + 3) / 3 + 333); c4 += 1)
for (int c5 = max(max(max(1000 * c3 - 2 * c4 + 2, 1000 * c0 - c4), 500 * c1 + (c4 + 1) / 2), c4); c5 <= min(min(min(1000 * c3 - 2 * c4 + 1001, 1000 * c0 - c4 + 999), 500 * c1 + (c4 + 1) / 2 + 499), 2 * c4 + 1); c5 += 1)
s0(c0, c1, c2, c3, c4, c5);
for (int c0 = 1; c0 < max((6 * M + 3 * N + 188) / 200 - 2, (N + 93) / 100 + 3 * ((2 * M + N - 4) / 200) - 1); c0 += 1)
for (int c1 = max(0, floord(-N + 100 * c0 + 106, 300)); c1 <= min((2 * M + N - 4) / 200 - 1, (c0 - 1) / 3); c1 += 1)
S2(c0 - c1, c1);
for (int c0 = 0; c0 <= 5 * n; c0 += 1)
for (int c1 = max(-((n + c0 + 1) % 2) - n + c0 + 1, 2 * floord(c0 - 1, 3) + 2); c1 <= min(n + c0 - (n + c0 + 2) / 3, c0); c1 += 2)
S1((-2 * c0 + 3 * c1) / 2, c0 - c1);
for (int c0 = 1; c0 <= min(4, floord(2 * m - 1, 17) + 1); c0 += 1)
for (int c1 = 1; c1 <= min(2, -2 * c0 + (2 * m + 3 * c0 - 4) / 10 + 3); c1 += 1)
for (int c2 = 0; c2 <= min(2, -c0 - c1 + (2 * m + 3 * c0 + 10 * c1 + 6) / 20 + 1); c2 += 1)
for (int c3 = 8 * c0 + (c0 + 1) / 2 - 8; c3 <= min(min(30, m - 5 * c1 - 10 * c2 + 5), 8 * c0 + c0 / 2); c3 += 1)
for (int c4 = 5 * c1 + 10 * c2 - 4; c4 <= min(5 * c1 + 10 * c2, m - c3 + 1); c4 += 1)
s0(c0, c1, c2, c3, c4, -9 * c0 + c3 + c0 / 2 + 9, -5 * c1 - 5 * c2 + c4 + 5);
int main()
{
int tx, ty, x, y;
int trial;
IN = (double **)malloc((N+2) * sizeof(double *));
for (int i=0; i<N+2; i++)
IN[i] = (double *)malloc((N) * sizeof(double));
BLURX = (double **)malloc((N) * sizeof(double *));
for (int i=0; i<N; i++)
BLURX[i] = (double *)malloc((N) * sizeof(double));
OUT = (double **)malloc((N) * sizeof(double *));
for (int i=0; i<N; i++)
OUT[i] = (double *)malloc((N) * sizeof(double));
init_array();
#ifdef PERFCTR
PERF_INIT;
#endif
IF_TIME(t_start = rtclock());
for (trial = 0; trial < 10 ; ++trial)
{
#pragma scop
for(ty = 0; ty <= floord((N-1),B); ++ty)
#pragma omp parallel for private(tx,x,y)
for(tx = 0; tx <= floord((N-1),B); ++tx){
for(x = 0; x <= B-1; ++x){
for(y = 0; y <= B-1; ++y)
/* P */ blurx(x,y)=in(x,y)+in((x+1),y)+in((x+2),y);
for(y = 0; y <= B-1; ++y)
if(ty*B+y>=2)
/* Q */ out(x,y)=blurx(x,y)+blurx(x,(y-1))+blurx(x,(y-2));
}
}
#pragma endscop
}
IF_TIME(t_end = rtclock());
IF_TIME(fprintf(stderr, "File:%s \t\t N=%d,T=%d \t Runtime=%0.6lfs\n", __FILE__, N, B, (t_end - t_start)/10));
#ifdef PERFCTR
PERF_EXIT;
#endif
#ifdef VERIFY
for(x = 0; x <= N-1; ++x)
for(y = 0; y <= N-1; ++y)
BLURX[x][y]=IN[x][y]+IN[x+1][y]+IN[x+2][y];
// Stage 2: vertical blur
for(x = 0; x <= N-1; ++x)
for(y = 2; y <= N-1; ++y)
{
if(OUT[x][y] != BLURX[x][y]+BLURX[x][y-1]+BLURX[x][y-2])
{
printf("Difference at (%d, %d) : %f versus %f\n", x, y, OUT[x][y], BLURX[x][y]+BLURX[x][y-1]+BLURX[x][y-2]);
break;
}
}
#endif
if (fopen(".test", "r")) {
// print_array();
}
return 0;
}
int main()
{
init_arrays();
double annot_t_start=0, annot_t_end=0, annot_t_total=0;
int annot_i;
for (annot_i=0; annot_i<REPS; annot_i++)
{
annot_t_start = rtclock();
int t, i, j, k, l, m, n,ii,jj;
#define S1(zT0,zT1,t,j) {ey[0][j]=t;}
#define S2(zT0,zT1,zT2,t,i,j) {ey[i][j]=ey[i][j]-((double)(1))/2*(hz[i][j]-hz[i-1][j]);}
#define S3(zT0,zT1,zT2,t,i,j) {ex[i][j]=ex[i][j]-((double)(1))/2*(hz[i][j]-hz[i][j-1]);}
#define S4(zT0,zT1,zT2,t,i,j) {hz[i][j]=hz[i][j]-((double)(7))/10*(ey[1+i][j]+ex[i][1+j]-ex[i][j]-ey[i][j]);}
int c1, c2, c3, c4, c5, c6, c7;
register int lbv, ubv;
for (c1=0;c1<=floord(tmax-1,32);c1++) {
for (c2=max(ceild(32*c1-31,32),0);c2<=min(floord(tmax+ny-1,32),floord(32*c1+ny+31,32));c2++) {
for (c3=max(max(max(max(ceild(32*c2-ny-30,32),0),ceild(64*c1-32*c2-61,32)),ceild(32*c1-31,32)),ceild(32*c1-992*c2-1891,992));c3<=min(min(floord(32*c2+nx+30,32),floord(tmax+nx-1,32)),floord(32*c1+nx+31,32));c3++) {
if ((c1 <= floord(32*c3-nx,32)) && (c2 <= floord(32*c3-nx+ny,32)) && (c3 >= ceild(nx,32))) {
for (c5=max(32*c3-nx+1,32*c2);c5<=min(32*c2+31,32*c3-nx+ny);c5++) {
S4(c1,-c1+c3,-c1+c2,32*c3-nx,nx-1,-32*c3+c5+nx-1) ;
}
}
if ((c1 <= floord(32*c2-ny,32)) && (c2 >= max(ceild(32*c3-nx+ny+1,32),ceild(ny,32)))) {
for (c6=max(32*c3,32*c2-ny+1);c6<=min(32*c2+nx-ny,32*c3+31);c6++) {
S4(c1,-c1+c3,-c1+c2,32*c2-ny,-32*c2+c6+ny-1,ny-1) ;
}
}
if (c1 == c3) {
for (c4=max(max(32*c2-ny+1,0),32*c3);c4<=min(min(32*c3+30,32*c2-ny+31),tmax-1);c4++) {
for (c5=32*c2;c5<=c4+ny-1;c5++) {
S1(c1,-c1+c2,c4,-c4+c5) ;
S3(c1,0,-c1+c2,c4,0,-c4+c5) ;
for (c6=c4+1;c6<=32*c3+31;c6++) {
S2(c1,0,-c1+c2,c4,-c4+c6,-c4+c5) ;
S3(c1,0,-c1+c2,c4,-c4+c6,-c4+c5) ;
S4(c1,0,-c1+c2,c4,-c4+c6-1,-c4+c5-1) ;
}
}
for (c6=c4+1;c6<=32*c3+31;c6++) {
S4(c1,0,-c1+c2,c4,-c4+c6-1,ny-1) ;
}
}
}
if (c1 == c3) {
for (c4=max(max(0,32*c3),32*c2-ny+32);c4<=min(min(tmax-1,32*c3+30),32*c2-1);c4++) {
for (c5=32*c2;c5<=32*c2+31;c5++) {
S1(c1,-c1+c2,c4,-c4+c5) ;
S3(c1,0,-c1+c2,c4,0,-c4+c5) ;
for (c6=c4+1;c6<=32*c3+31;c6++) {
S2(c1,0,-c1+c2,c4,-c4+c6,-c4+c5) ;
S3(c1,0,-c1+c2,c4,-c4+c6,-c4+c5) ;
S4(c1,0,-c1+c2,c4,-c4+c6-1,-c4+c5-1) ;
}
}
}
}
if (c1 == c3) {
for (c4=max(max(32*c2,0),32*c3);c4<=min(min(tmax-1,32*c3+30),32*c2+30);c4++) {
S1(c1,-c1+c2,c4,0) ;
for (c6=c4+1;c6<=32*c3+31;c6++) {
S2(c1,0,-c1+c2,c4,-c4+c6,0) ;
}
for (c5=c4+1;c5<=32*c2+31;c5++) {
S1(c1,-c1+c2,c4,-c4+c5) ;
S3(c1,0,-c1+c2,c4,0,-c4+c5) ;
for (c6=c4+1;c6<=32*c3+31;c6++) {
S2(c1,0,-c1+c2,c4,-c4+c6,-c4+c5) ;
S3(c1,0,-c1+c2,c4,-c4+c6,-c4+c5) ;
S4(c1,0,-c1+c2,c4,-c4+c6-1,-c4+c5-1) ;
}
}
}
}
for (c4=max(max(max(32*c1,0),32*c2-ny+1),32*c3-nx+1);c4<=min(min(min(32*c3-nx+31,32*c2-ny+31),32*c1+31),tmax-1);c4++) {
for (c5=32*c2;c5<=c4+ny-1;c5++) {
for (c6=32*c3;c6<=c4+nx-1;c6++) {
S2(c1,-c1+c3,-c1+c2,c4,-c4+c6,-c4+c5) ;
S3(c1,-c1+c3,-c1+c2,c4,-c4+c6,-c4+c5) ;
S4(c1,-c1+c3,-c1+c2,c4,-c4+c6-1,-c4+c5-1) ;
}
S4(c1,-c1+c3,-c1+c2,c4,nx-1,-c4+c5-1) ;
}
for (c6=32*c3;c6<=c4+nx;c6++) {
S4(c1,-c1+c3,-c1+c2,c4,-c4+c6-1,ny-1) ;
}
}
for (c4=max(max(max(32*c1,0),32*c3-nx+1),32*c2-ny+32);c4<=min(min(min(tmax-1,32*c1+31),32*c2-1),32*c3-nx+31);c4++) {
for (c5=32*c2;c5<=32*c2+31;c5++) {
for (c6=32*c3;c6<=c4+nx-1;c6++) {
S2(c1,-c1+c3,-c1+c2,c4,-c4+c6,-c4+c5) ;
S3(c1,-c1+c3,-c1+c2,c4,-c4+c6,-c4+c5) ;
S4(c1,-c1+c3,-c1+c2,c4,-c4+c6-1,-c4+c5-1) ;
}
S4(c1,-c1+c3,-c1+c2,c4,nx-1,-c4+c5-1) ;
}
}
for (c4=max(max(max(32*c3-nx+32,32*c1),0),32*c2-ny+1);c4<=min(min(min(32*c2-ny+31,32*c1+31),tmax-1),32*c3-1);c4++) {
for (c5=32*c2;c5<=c4+ny-1;c5++) {
for (c6=32*c3;c6<=32*c3+31;c6++) {
S2(c1,-c1+c3,-c1+c2,c4,-c4+c6,-c4+c5) ;
S3(c1,-c1+c3,-c1+c2,c4,-c4+c6,-c4+c5) ;
S4(c1,-c1+c3,-c1+c2,c4,-c4+c6-1,-c4+c5-1) ;
}
}
for (c6=32*c3;c6<=32*c3+31;c6++) {
S4(c1,-c1+c3,-c1+c2,c4,-c4+c6-1,ny-1) ;
}
}
for (c4=max(max(max(32*c2,32*c1),0),32*c3-nx+1);c4<=min(min(min(32*c2+30,tmax-1),32*c1+31),32*c3-nx+31);c4++) {
for (c6=32*c3;c6<=c4+nx-1;c6++) {
S2(c1,-c1+c3,-c1+c2,c4,-c4+c6,0) ;
}
for (c5=c4+1;c5<=32*c2+31;c5++) {
for (c6=32*c3;c6<=c4+nx-1;c6++) {
S2(c1,-c1+c3,-c1+c2,c4,-c4+c6,-c4+c5) ;
S3(c1,-c1+c3,-c1+c2,c4,-c4+c6,-c4+c5) ;
S4(c1,-c1+c3,-c1+c2,c4,-c4+c6-1,-c4+c5-1) ;
}
S4(c1,-c1+c3,-c1+c2,c4,nx-1,-c4+c5-1) ;
}
}
for (c4=max(max(max(32*c1,0),32*c3-nx+32),32*c2-ny+32);c4<=min(min(min(32*c3-1,tmax-1),32*c1+31),32*c2-1);c4++) {
/*@ begin Loop(
transform Composite(
tile = [('c5',T1,'ii'),('c6',T2,'jj')],
permut = [PERMUTS],
unrolljam = [('c5',U1),('c6',U2)],
vector = (VEC, ['ivdep','vector always'])
)
for (c5=32*c2;c5<=32*c2+31;c5++)
for (c6=32*c3;c6<=32*c3+31;c6++)
{
S2(c1,-c1+c3,-c1+c2,c4,-c4+c6,-c4+c5) ;
S3(c1,-c1+c3,-c1+c2,c4,-c4+c6,-c4+c5) ;
S4(c1,-c1+c3,-c1+c2,c4,-c4+c6-1,-c4+c5-1) ;
}
) @*/{
for (c6=32*c3; c6<=32*c3+28; c6=c6+4) {
register int cbv_1, cbv_2;
cbv_1=32*c2;
cbv_2=32*c2+31;
#pragma ivdep
#pragma vector always
for (c5=cbv_1; c5<=cbv_2; c5=c5+1) {
S2(c1,-c1+c3,-c1+c2,c4,-c4+c6,-c4+c5);
S2(c1,-c1+c3,-c1+c2,c4,-c4+c6+1,-c4+c5);
S2(c1,-c1+c3,-c1+c2,c4,-c4+c6+2,-c4+c5);
S2(c1,-c1+c3,-c1+c2,c4,-c4+c6+3,-c4+c5);
S3(c1,-c1+c3,-c1+c2,c4,-c4+c6,-c4+c5);
S3(c1,-c1+c3,-c1+c2,c4,-c4+c6+1,-c4+c5);
S3(c1,-c1+c3,-c1+c2,c4,-c4+c6+2,-c4+c5);
S3(c1,-c1+c3,-c1+c2,c4,-c4+c6+3,-c4+c5);
S4(c1,-c1+c3,-c1+c2,c4,-c4+c6-1,-c4+c5-1);
S4(c1,-c1+c3,-c1+c2,c4,-c4+c6,-c4+c5-1);
S4(c1,-c1+c3,-c1+c2,c4,-c4+c6+1,-c4+c5-1);
S4(c1,-c1+c3,-c1+c2,c4,-c4+c6+2,-c4+c5-1);
}
}
for (; c6<=32*c3+31; c6=c6+1) {
register int cbv_3, cbv_4;
cbv_3=32*c2;
cbv_4=32*c2+31;
#pragma ivdep
#pragma vector always
for (c5=cbv_3; c5<=cbv_4; c5=c5+1) {
S2(c1,-c1+c3,-c1+c2,c4,-c4+c6,-c4+c5);
S3(c1,-c1+c3,-c1+c2,c4,-c4+c6,-c4+c5);
S4(c1,-c1+c3,-c1+c2,c4,-c4+c6-1,-c4+c5-1);
}
}
}
/*@ end @*/
}
for (c4=max(max(max(32*c2,32*c3-nx+32),32*c1),0);c4<=min(min(min(32*c3-1,32*c2+30),tmax-1),32*c1+31);c4++) {
for (c6=32*c3;c6<=32*c3+31;c6++) {
S2(c1,-c1+c3,-c1+c2,c4,-c4+c6,0) ;
}
for (c5=c4+1;c5<=32*c2+31;c5++) {
for (c6=32*c3;c6<=32*c3+31;c6++) {
S2(c1,-c1+c3,-c1+c2,c4,-c4+c6,-c4+c5) ;
S3(c1,-c1+c3,-c1+c2,c4,-c4+c6,-c4+c5) ;
S4(c1,-c1+c3,-c1+c2,c4,-c4+c6-1,-c4+c5-1) ;
}
}
}
if ((c1 == c3) && (c2 <= min(floord(32*c3-1,32),floord(tmax-32,32)))) {
S1(c1,-c1+c2,32*c2+31,0) ;
for (c6=32*c2+32;c6<=32*c3+31;c6++) {
S2(c1,0,-c1+c2,32*c2+31,-32*c2+c6-31,0) ;
}
}
if ((-c1 == -c3) && (c1 >= ceild(32*c2-31,32)) && (c1 <= min(floord(tmax-32,32),floord(32*c2-1,32)))) {
S1(c1,-c1+c2,32*c1+31,0) ;
for (c5=32*c1+32;c5<=32*c2+31;c5++) {
S1(c1,-c1+c2,32*c1+31,-32*c1+c5-31) ;
S3(c1,0,-c1+c2,32*c1+31,0,-32*c1+c5-31) ;
}
}
if ((-c1 == -c3) && (c1 <= min(floord(tmax-32,32),c2-1))) {
for (c5=32*c2;c5<=min(32*c2+31,32*c1+ny+30);c5++) {
S1(c1,-c1+c2,32*c1+31,-32*c1+c5-31) ;
S3(c1,0,-c1+c2,32*c1+31,0,-32*c1+c5-31) ;
}
}
if ((-c1 == -c2) && (-c1 == -c3) && (c1 <= floord(tmax-32,32))) {
S1(c1,0,32*c1+31,0) ;
}
if ((c1 >= c2) && (c2 <= min(c3-1,floord(tmax-32,32)))) {
for (c6=32*c3;c6<=min(32*c2+nx+30,32*c3+31);c6++) {
S2(c1,-c1+c3,-c1+c2,32*c2+31,-32*c2+c6-31,0) ;
}
}
}
}
}
annot_t_end = rtclock();
annot_t_total += annot_t_end - annot_t_start;
}
annot_t_total = annot_t_total / REPS;
printf("%f\n", annot_t_total);
return 1;
}
for (int c0 = 0; c0 <= 3; c0 += 1)
for (int c1 = max(0, 2 * c0 - 3); c1 <= min(c0 + 1, 3); c1 += 1)
for (int c2 = c0; c2 <= min(min(3, 3 * c1 + 2), 2 * c0 - c1 + 1); c2 += 1)
for (int c3 = max(max(max(c2 - (c2 + 2) / 3, c2 + floord(3 * c1 - c2 - 1, 6)), c1 - (-c1 + 3) / 3), c0 - (-c2 + 3) / 3); c3 <= min(c0 + c0 / 2 + 1, 3); c3 += 1)
for (int c5 = max(max(max(max(c1 - (-c1 + 3) / 3, 0), 2 * c3 - 4), c3 - (c3 + 3) / 3), c2 - (c2 + 3) / 3); c5 <= min(min(-c2 + 2 * c3 - (c2 + 3) / 3 + 2, c1 + 1), c3); c5 += 1)
for (int c6 = max(max(max(max(max(250 * c3 + 1, 667 * c0 - 333 * c1 - (c0 + c1 + 3) / 3 - 332), -200 * c1 + 400 * c3 - 199), 333 * c1 + c1 / 3), 1000 * c0 - 500 * c5 - 501), 333 * c2 + (c2 + 1) / 3); c6 <= min(min(min(min(min(min(333 * c3 - (-c3 + 3) / 3 + 334, 1000), 333 * c2 - (-c2 + 3) / 3 + 333), 1000 * c0 - 500 * c5 + 997), 500 * c5 + 501), 500 * c0 + 499), -200 * c1 + 400 * c3 + 400); c6 += 1)
for (int c7 = max(max(max(max(c6, 500 * c1 + (c6 + 1) / 2), 1000 * c0 - c6), 500 * c5 + 2), 1000 * c3 - 2 * c6 + 2); c7 <= min(min(min(min(500 * c5 + 501, 2 * c6 + 1), 1000 * c3 - 2 * c6 + 1001), 1000 * c0 - c6 + 999), 500 * c1 + (c6 + 1) / 2 + 499); c7 += 1)
s0(c0, c1, c2, c3, c2 / 3, c5, c6, c7);
for (int c0 = floord(m, 4); c0 <= n; c0 += 1) s0(c0);
int main()
{
init_arrays();
double annot_t_start=0, annot_t_end=0, annot_t_total=0;
int annot_i;
for (annot_i=0; annot_i<REPS; annot_i++)
{
annot_t_start = rtclock();
/*@ begin PerfTuning (
def build
{
arg build_command = 'icc -O3 -openmp -I/usr/local/icc/include -lm';
}
def performance_counter
{
arg repetitions = 1;
}
def performance_params
{
# param T1_1[] = [1,16,32,64,128];
# param T1_2[] = [1,16,32,64,128];
# param T1_3[] = [1,16,32,64,128];
# param T2_1[] = [1,4,8,16,32];
# param T2_2[] = [1,4,8,16,32];
# param T2_3[] = [1,4,8,16,32];
param T1_1[] = [64];
param T1_2[] = [256];
param T1_3[] = [64];
param T2_1[] = [1];
param T2_2[] = [1];
param T2_3[] = [1];
constraint c1 = (T1_1*T2_1<=1024 and T1_1*T2_1<=1024 and T1_1*T2_1<=1024);
constraint c2 = ((T1_1 == T1_3) and (T2_1 == T2_3));
param U1[] = [1];
param U2[] = [1];
param U3[] = [7];
constraint c3 = (U1*U2*U3<=512);
param PERM[] = [
#[0,1,2],
#[0,2,1],
#[1,0,2],
#[1,2,0],
[2,0,1],
#[2,1,0],
];
param PAR[] = [True];
param SCREP[] = [False];
param IVEC[] = [True];
}
def search
{
arg algorithm = 'Exhaustive';
# arg algorithm = 'Simplex';
# arg time_limit = 5;
# arg total_runs = 1;
}
def input_params
{
param N[] = [1024];
}
def input_vars
{
arg decl_file = 'decl_code.h';
arg init_file = 'init_code.c';
}
) @*/
/**-- (Generated by Orio)
Best performance cost:
0.201184
Tuned for specific problem sizes:
N = 1024
Best performance parameters:
IVEC = True
PAR = True
PERM = [2, 0, 1]
SCREP = False
T1_1 = 64
T1_2 = 256
T1_3 = 64
T2_1 = 1
T2_2 = 1
T2_3 = 1
U1 = 1
U2 = 1
U3 = 7
--**/
register int i,j,k;
register int c1t, c2t, c3t, c4t, c5t, c6t, c7t, c8t, c9t, c10t, c11t, c12t;
register int newlb_c1, newlb_c2, newlb_c3, newlb_c4, newlb_c5, newlb_c6,
newlb_c7, newlb_c8, newlb_c9, newlb_c10, newlb_c11, newlb_c12;
register int newub_c1, newub_c2, newub_c3, newub_c4, newub_c5, newub_c6,
newub_c7, newub_c8, newub_c9, newub_c10, newub_c11, newub_c12;
/*@ begin PolySyn(
parallel = PAR;
tiles = [T1_1,T1_2,T1_3,T2_1,T2_2,T2_3];
permut = PERM;
unroll_factors = [U1,U2,U3];
scalar_replace = SCREP;
vectorize = IVEC;
profiling_code = 'lu_profiling.c';
compile_cmd = 'gcc';
compile_opts = '-lm';
) @*/
#include <math.h>
#include <assert.h>
#define ceild(n,d) ceil(((double)(n))/((double)(d)))
#define floord(n,d) floor(((double)(n))/((double)(d)))
#define max(x,y) ((x) > (y)? (x) : (y))
#define min(x,y) ((x) < (y)? (x) : (y))
int c1, c2, c3, c4, c5, c6, c7, c8, c9;
register int lb, ub, lb1, ub1, lb2, ub2;
/* Generated from PLuTo-produced CLooG file by CLooG v0.14.1 64 bits in 0.05s. */
for (c1=-1;c1<=floord(5*N-9,256);c1++) {
lb1=max(max(ceild(32*c1-127,160),ceild(64*c1-N+2,64)),0);
ub1=min(floord(64*c1+63,64),floord(N-1,256));
#pragma omp parallel for shared(c1,lb1,ub1) private(c2,c3,c4,c5,c6,c7,c8,c9)
for (c2=lb1; c2<=ub1; c2++) {
for (c3=max(ceild(32*c1-32*c2-1953,2016),ceild(32*c1-32*c2-31,32));c3<=floord(N-1,64);c3++) {
if (c1 == c2+c3) {
for (c7=max(64*c3,0);c7<=min(min(N-2,64*c3+62),256*c2+254);c7++) {
for (c8=max(c7+1,256*c2);c8<=min(N-1,256*c2+255);c8++) {
A[c7][c8]=A[c7][c8]/A[c7][c7] ;
for (c9=c7+1;c9<=min(N-1,64*c3+63);c9++) {
A[c9][c8]=A[c9][c8]-A[c9][c7]*A[c7][c8] ;
}
}
}
}
/*@ begin Loop(
transform Composite(
permut = [['c9', 'c7', 'c8']],
regtile = (['c7', 'c8', 'c9'],[1, 1, 7]),
scalarreplace = (False, 'double'),
vector = (True, ['ivdep','vector always']))
for (c7=max(0,64*c1-64*c2);c7<=min(min(256*c2+254,64*c1-64*c2+63),64*c3-1);c7++) {
for (c8=max(c7+1,256*c2);c8<=min(256*c2+255,N-1);c8++) {
for (c9=64*c3;c9<=min(N-1,64*c3+63);c9++) {
A[c9][c8]=A[c9][c8]-A[c9][c7]*A[c7][c8] ;
}
}
}
) @*/{
for (c9t=64*c3; c9t<=min(N-1,64*c3+63)-6; c9t=c9t+7) {
for (c7=max(0,64*c1-64*c2); c7<=min(min(256*c2+254,64*c1-64*c2+63),64*c3-1); c7++ ) {
register int cbv_1, cbv_2;
cbv_1=max(c7+1,256*c2);
cbv_2=min(256*c2+255,N-1);
#pragma ivdep
#pragma vector always
for (c8=cbv_1; c8<=cbv_2; c8++ ) {
A[c9t][c8]=A[c9t][c8]-A[c9t][c7]*A[c7][c8];
A[(c9t+1)][c8]=A[(c9t+1)][c8]-A[(c9t+1)][c7]*A[c7][c8];
A[(c9t+2)][c8]=A[(c9t+2)][c8]-A[(c9t+2)][c7]*A[c7][c8];
A[(c9t+3)][c8]=A[(c9t+3)][c8]-A[(c9t+3)][c7]*A[c7][c8];
A[(c9t+4)][c8]=A[(c9t+4)][c8]-A[(c9t+4)][c7]*A[c7][c8];
A[(c9t+5)][c8]=A[(c9t+5)][c8]-A[(c9t+5)][c7]*A[c7][c8];
A[(c9t+6)][c8]=A[(c9t+6)][c8]-A[(c9t+6)][c7]*A[c7][c8];
}
}
}
for (c9=c9t; c9<=min(N-1,64*c3+63); c9=c9+1) {
for (c7=max(0,64*c1-64*c2); c7<=min(min(256*c2+254,64*c1-64*c2+63),64*c3-1); c7++ ) {
register int cbv_3, cbv_4;
cbv_3=max(c7+1,256*c2);
cbv_4=min(256*c2+255,N-1);
#pragma ivdep
#pragma vector always
for (c8=cbv_3; c8<=cbv_4; c8++ ) {
A[c9][c8]=A[c9][c8]-A[c9][c7]*A[c7][c8];
}
}
}
}
/*@ end @*/
if ((-c1 == -c2-c3) && (c1 <= min(floord(320*c2+191,64),floord(64*c2+N-65,64)))) {
for (c8=max(256*c2,64*c1-64*c2+64);c8<=min(256*c2+255,N-1);c8++) {
A[64*c1-64*c2+63][c8]=A[64*c1-64*c2+63][c8]/A[64*c1-64*c2+63][64*c1-64*c2+63] ;
}
}
}
}
}
/* End of CLooG code */
/*@ end @*/
/*@ end @*/
annot_t_end = rtclock();
annot_t_total += annot_t_end - annot_t_start;
}
annot_t_total = annot_t_total / REPS;
#ifndef TEST
printf("%f\n", annot_t_total);
#else
{
int i, j;
for (i=0; i<N; i++) {
for (j=0; j<N; j++) {
if (j%100==0)
printf("\n");
printf("%f ",A[i][j]);
}
printf("\n");
}
}
#endif
return ((int) A[0][0]);
}
/* Generated from ../../../git/cloog/test/isl/jacobi-shared.cloog by CLooG 0.16.3-2-g5511bef gmp bits in 1.82s. */
if ((h0+1)%2 == 0) {
if ((16*floord(t0-1,16) >= -N+g1+t0+1) && (16*floord(N+15*g1+15*t0+15,16) >= 15*g1+15*t0+19) && (32*floord(t1-1,32) <= g2+t1-3) && (32*floord(t1-1,32) >= -N+g2+t1+1)) {
for (c0=max(-16*floord(t0-1,16)+t0,-16*floord(g1+t0-3,16)+t0);c0<=min(32,N-g1-1);c0+=16) {
c1 = -32*floord(t1-1,32)+t1;
if (c1 <= 32) {
S1(c0+g1-1,c1+g2-1);
}
}
}
}
for (int c0 = 0; c0 <= 5 * n; c0 += 1)
for (int c1 = max(-((5 * n - c0 + 1) % 2) - n + c0 + 1, 2 * floord(c0 - 1, 3) + 2); c1 <= min(c0, n + c0 - (n + c0 + 2) / 3); c1 += 2)
S1((3 * c1 / 2) - c0, c0 - c1);
int main()
{
init_arrays();
double annot_t_start=0, annot_t_end=0, annot_t_total=0;
int annot_i;
for (annot_i=0; annot_i<REPS; annot_i++)
{
annot_t_start = rtclock();
#include <math.h>
#include <assert.h>
#define ceild(n,d) ceil(((double)(n))/((double)(d)))
#define floord(n,d) floor(((double)(n))/((double)(d)))
#define max(x,y) ((x) > (y)? (x) : (y))
#define min(x,y) ((x) < (y)? (x) : (y))
#define S1(zT0,zT1,zT2,zT3,zT4,zT5,t,i,j) {A[i][j]=(A[1+i][1+j]+A[1+i][j]+A[1+i][j-1]+A[i][1+j]+A[i][j]+A[i][j-1]+A[i-1][1+j]+A[i-1][j]+A[i-1][j-1])/9;}
int c1, c2, c3, c4, c5, c6, c7, c8, c9;
register int lb, ub, lb1, ub1, lb2, ub2;
register int lbv, ubv;
omp_set_nested(1);
omp_set_num_threads(2);
/* Generated from PLuTo-produced CLooG file by CLooG v0.14.1 64 bits in 5.45s. */
for (c1=-2;c1<=floord(4*T+3*N-10,256);c1++) {
lb1=max(max(max(0,ceild(256*c1-2*T-N-251,512)),ceild(256*c1-3*T-2*N+7,256)),ceild(256*c1-N-761,1024));
ub1=min(min(min(floord(256*c1+2*N+505,1024),floord(256*c1+509,512)),floord(64*c1+127,64)),floord(T+N-3,256));
#pragma omp parallel for shared(c1,lb1,ub1) private(lb2,ub2,c2,c3,c4,c5,c6,c7,c8,c9)
for (c2=lb1; c2<=ub1; c2++) {
lb2=max(max(max(max(max(max(ceild(256*c1-256*c2-T+1,256),ceild(512*c1-512*c2-253,768)),0),ceild(512*c2-N-252,256)),ceild(128*c1-256*c2-127,128)),ceild(128*c2-127,128)),ceild(128*c1-127,256));
ub2=min(min(min(min(min(min(floord(256*c1-256*c2+255,256),floord(256*c1-512*c2+N+253,256)),floord(256*c2+T+N+252,256)),floord(T+N-3,128)),floord(256*c1+N+508,512)),floord(256*c1-256*c2+N+253,384)),floord(512*c2+N+507,256));
#pragma omp parallel for shared(c1,c2,lb1,ub1,lb2,ub2) private(c3,c4,c5,c6,c7,c8,c9)
for (c3=lb2; c3<=ub2; c3++) {
for (c4=max(max(max(max(0,ceild(-256*c2+256*c3-N-284,32)),8*c1-8*c2-8*c3),ceild(256*c2-N-29,32)),ceild(128*c3-N-29,32));c4<=min(min(min(min(8*c1-8*c2-8*c3+7,floord(256*c3+253,64)),floord(T-1,32)),floord(128*c2+127,16)),floord(-128*c2+128*c3+127,16));c4++) {
for (c5=max(max(max(max(max(8*c2,ceild(16*c4-15,16)),ceild(256*c3-T-N-28,32)),0),ceild(256*c3-32*c4-N-60,32)),ceild(256*c3-N-59,64));c5<=min(min(min(min(min(floord(32*c4+N+29,32),floord(128*c3+127,16)),8*c2+7),floord(128*c3-16*c4+127,16)),floord(T+N-3,32)),floord(256*c3+N+252,64));c5++) {
for (c6=max(max(max(max(max(ceild(64*c4-29,32),8*c3),ceild(16*c5-15,16)),ceild(16*c4+16*c5-15,16)),0),ceild(64*c5-N-28,32));c6<=min(min(min(min(min(8*c3+7,floord(T+N-3,16)),floord(32*c4+32*c5+N+60,32)),floord(32*c4+N+29,16)),floord(64*c5+N+59,32)),floord(32*c5+T+N+28,32));c6++) {
for (c7=max(max(max(max(0,32*c4),32*c5-N+2),16*c6-N+2),-32*c5+32*c6-N-29);c7<=min(min(min(min(-32*c5+32*c6+30,floord(32*c6+29,2)),T-1),32*c5+30),32*c4+31);c7++) {
/*@ begin Loop(
transform UnrollJam(ufactor=8)
for (c8=max(max(32*c5,c7+1),32*c6-c7-N+2);c8<=min(min(32*c6-c7+30,32*c5+31),c7+N-2);c8++)
transform Unroll(ufactor=8)
for (c9=max(c7+c8+1,32*c6);c9<=min(32*c6+31,c7+c8+N-2);c9++)
{
S1(c1-c2-c3,-c1+2*c2+c3,-c1+2*c3,c4,-c4+c5,-c4-c5+c6,c7,-c7+c8,-c7-c8+c9) ;
}
) @*/
for (c8=max(max(32*c5,c7+1),32*c6-c7-N+2);c8<=min(min(32*c6-c7+30,32*c5+31),c7+N-2);c8++) {
for (c9=max(c7+c8+1,32*c6);c9<=min(32*c6+31,c7+c8+N-2);c9++) {
S1(c1-c2-c3,-c1+2*c2+c3,-c1+2*c3,c4,-c4+c5,-c4-c5+c6,c7,-c7+c8,-c7-c8+c9) ;
}
}
/*@ end @*/
}
}
}
}
}
}
}
/* End of CLooG code */
annot_t_end = rtclock();
annot_t_total += annot_t_end - annot_t_start;
}
annot_t_total = annot_t_total / REPS;
printf("%f\n", annot_t_total);
return ((int) A[0][0]);
}
int main()
{
init_arrays();
double annot_t_start=0, annot_t_end=0, annot_t_total=0;
int annot_i;
omp_set_nested(1);
omp_set_num_threads(2);
for (annot_i=0; annot_i<REPS; annot_i++)
{
annot_t_start = rtclock();
register int i,j,k;
#define S1(zT0,zT1,zT2,zT3,k,j) {A[k][j]=A[k][j]/A[k][k];}
#define S2(zT0,zT1,zT2,zT3,zT4,zT5,k,i,j) {A[i][j]=A[i][j]-A[i][k]*A[k][j];}
int c1, c2, c3, c4, c5, c6, c7, c8, c9;
register int lb, ub, lb1, ub1, lb2, ub2;
register int lbv, ubv;
/* Generated from PLuTo-produced CLooG file by CLooG v0.14.1 64 bits in 2.21s. */
for (c1=-2;c1<=floord(3*N-4,256);c1++) {
lb1=max(max(0,ceild(256*c1-N-253,512)),ceild(256*c1-2*N+3,256));
ub1=min(floord(128*c1+255,128),floord(N-1,256));
#pragma omp parallel for shared(c1,lb1,ub1) private(lb2,ub2,c2,c3,c4,c5,c6,c7,c8,c9)
for (c2=lb1; c2<=ub1; c2++) {
lb2=max(max(max(ceild(256*c1-256*c2-N+2,256),ceild(128*c1-256*c2-127,128)),ceild(128*c1-128*c2-32385,32768)),ceild(128*c1-128*c2-127,256));
ub2=min(floord(N-1,256),floord(256*c1-256*c2+255,256));
#pragma omp parallel for shared(c1,c2,lb1,ub1,lb2,ub2) private(c3,c4,c5,c6,c7,c8,c9)
for (c3=lb2; c3<=ub2; c3++) {
for (c4=max(max(8*c1-8*c2-8*c3,0),8*c1-8*c2-1800*c3-1778);c4<=min(min(min(min(floord(3968*c3+3937,16),8*c1-8*c2-8*c3+7),floord(128*c2+127,16)),floord(N-2,32)),floord(128*c3+127,16));c4++) {
for (c5=max(max(ceild(16*c4-15,16),0),8*c2);c5<=min(floord(N-1,32),8*c2+7);c5++) {
for (c6=max(max(max(max(ceild(16*c4-465,496),ceild(8*c1-8*c2-16*c3-c4-217,223)),ceild(-8*c1+8*c2+16*c3+c4-217,225)),8*c3),ceild(16*c4-15,16));c6<=min(8*c3+7,floord(N-1,32));c6++) {
if ((c1 == c2+2*c3) && (c4 == c6)) {
for (c7=max(0,32*c6);c7<=min(min(32*c5+30,32*c6+30),N-2);c7++) {
for (c8=max(c7+1,32*c5);c8<=min(32*c5+31,N-1);c8++) {
if ((c1-c2)%2 == 0) {
S1((c1-c2)/2,c2,c4,c5,c7,c8) ;
}
for (c9=c7+1;c9<=min(32*c6+31,N-1);c9++) {
if ((c1-c2)%2 == 0) {
if ((c1-c2)%2 == 0) {
S2((c1-c2)/2,(c1-c2)/2,c2,c4,c4,c5,c7,c9,c8) ;
}
}
}
}
}
}
for (c7=max(32*c4,0);c7<=min(min(32*c6-1,32*c5+30),32*c4+31);c7++) {
/*@ begin Loop(
transform UnrollJam(ufactor=8)
for (c8=max(c7+1,32*c5);c8<=min(32*c5+31,N-1);c8++)
transform Unroll(ufactor=8)
for (c9=32*c6;c9<=min(N-1,32*c6+31);c9++)
{
S2(c1-c2-c3,c3,c2,c4,c6,c5,c7,c9,c8) ;
}
) @*/{
for (c8 = max(c7 + 1, 32 * c5); c8 <= min(32 * c5 + 31, N - 1) - 7; c8 = c8 + 8) {
for (c9 = 32 * c6; c9 <= min(N - 1, 32 * c6 + 31) - 7; c9 = c9 + 8) {
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, c9, c8);
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 1), c8);
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 2), c8);
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 3), c8);
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 4), c8);
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 5), c8);
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 6), c8);
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 7), c8);
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, c9, (c8 + 1));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 1), (c8 + 1));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 2), (c8 + 1));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 3), (c8 + 1));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 4), (c8 + 1));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 5), (c8 + 1));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 6), (c8 + 1));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 7), (c8 + 1));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, c9, (c8 + 2));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 1), (c8 + 2));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 2), (c8 + 2));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 3), (c8 + 2));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 4), (c8 + 2));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 5), (c8 + 2));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 6), (c8 + 2));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 7), (c8 + 2));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, c9, (c8 + 3));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 1), (c8 + 3));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 2), (c8 + 3));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 3), (c8 + 3));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 4), (c8 + 3));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 5), (c8 + 3));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 6), (c8 + 3));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 7), (c8 + 3));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, c9, (c8 + 4));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 1), (c8 + 4));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 2), (c8 + 4));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 3), (c8 + 4));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 4), (c8 + 4));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 5), (c8 + 4));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 6), (c8 + 4));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 7), (c8 + 4));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, c9, (c8 + 5));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 1), (c8 + 5));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 2), (c8 + 5));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 3), (c8 + 5));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 4), (c8 + 5));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 5), (c8 + 5));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 6), (c8 + 5));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 7), (c8 + 5));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, c9, (c8 + 6));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 1), (c8 + 6));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 2), (c8 + 6));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 3), (c8 + 6));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 4), (c8 + 6));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 5), (c8 + 6));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 6), (c8 + 6));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 7), (c8 + 6));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, c9, (c8 + 7));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 1), (c8 + 7));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 2), (c8 + 7));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 3), (c8 + 7));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 4), (c8 + 7));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 5), (c8 + 7));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 6), (c8 + 7));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 7), (c8 + 7));
}
for (; c9 <= min(N - 1, 32 * c6 + 31); c9 = c9 + 1) {
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, c9, c8);
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, c9, (c8 + 1));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, c9, (c8 + 2));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, c9, (c8 + 3));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, c9, (c8 + 4));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, c9, (c8 + 5));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, c9, (c8 + 6));
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, c9, (c8 + 7));
}
}
for (; c8 <= min(32 * c5 + 31, N - 1); c8 = c8 + 1) {
for (c9 = 32 * c6; c9 <= min(N - 1, 32 * c6 + 31) - 7; c9 = c9 + 8) {
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, c9, c8);
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 1), c8);
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 2), c8);
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 3), c8);
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 4), c8);
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 5), c8);
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 6), c8);
S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, (c9 + 7), c8);
}
for (; c9 <= min(N - 1, 32 * c6 + 31); c9 = c9 + 1) S2(c1 - c2 - c3, c3, c2, c4, c6, c5, c7, c9, c8);
}
}
/*@ end @*/
}
if ((c1 == c2+2*c3) && (-c4 == -c6) && (c4 <= min(floord(N-33,32),floord(32*c5-1,32)))) {
for (c8=max(32*c5,32*c4+32);c8<=min(N-1,32*c5+31);c8++) {
if ((c1-c2)%2 == 0) {
S1((c1-c2)/2,c2,c4,c5,32*c4+31,c8) ;
}
}
}
}
}
}
}
}
}
/* End of CLooG code */
annot_t_end = rtclock();
annot_t_total += annot_t_end - annot_t_start;
}
annot_t_total = annot_t_total / REPS;
printf("%f\n", annot_t_total);
return ((int) A[0][0]);
}
int main(void) {
int t = 0, z, y, x, k;
double total_lattice_pts =
(double)nZ * (double)nY * (double)nX * (double)nTimesteps;
/* For timekeeping */
int ts_return = -1;
struct timeval start, end, result;
double tdiff = 0.0;
/* Compute values for global parameters */
omega = 2.0 /
((6.0 * sqrt(uTopX * uTopX + uTopY * uTopY) * (nX - 0.5) / re) + 1.0);
printf(
"3D Lid Driven Cavity simulation with D3Q19 lattice:\n"
"\tscheme : 3-Grid, Fused, Pull\n"
"\tgrid size : %d x %d x %d = %.2lf * 10^3 Cells\n"
"\tnTimeSteps : %d\n"
"\tRe : %.2lf\n"
"\tuTopX : %.6lf\n"
"\tuTopY : %.6lf\n"
"\tomega : %.6lf\n",
nX, nY, nZ, nX * nY * nZ / 1.0e3, nTimesteps, re, uTopX, uTopY, omega);
/* Initialize all 19 PDFs for each point in the domain to w1, w2 or w3
* accordingly */
for (z = 0; z < nZ + 2 + 4; z++) {
for (y = 0; y < nY + 2 + 2; y++) {
for (x = 0; x < nX + 2 + 2; x++) {
grid[0][z][y][x][0] = w1;
grid[1][z][y][x][0] = w1;
for (k = 1; k < 7; k++) {
grid[0][z][y][x][k] = w2;
grid[1][z][y][x][k] = w2;
}
for (k = 7; k < nK; k++) {
grid[0][z][y][x][k] = w3;
grid[1][z][y][x][k] = w3;
}
}
}
}
/* To satisfy PET */
short _nX = nX + 3;
short _nY = nY + 3;
short _nZ = nZ + 4;
short _nTimesteps = nTimesteps;
#ifdef TIME
gettimeofday(&start, 0);
#endif
int t1, t2, t3, t4, t5, t6, t7, t8;
int lb, ub, lbp, ubp, lb2, ub2;
register int lbv, ubv;
/* Start of CLooG code */
if ((_nTimesteps >= 1) && (_nX >= 2) && (_nY >= 2) && (_nZ >= 3)) {
for (t1 = -1; t1 <= floord(_nTimesteps - 1, 8); t1++) {
lbp = max(ceild(t1, 2), ceild(16 * t1 - _nTimesteps + 3, 16));
ubp =
min(floord(_nTimesteps + _nZ - 2, 16), floord(8 * t1 + _nZ + 6, 16));
#pragma omp parallel for private(lbv, ubv, t3, t4, t5, t6, t7, t8)
for (t2 = lbp; t2 <= ubp; t2++) {
for (t3 = max(max(0, ceild(t1 - 1, 2)), ceild(16 * t2 - _nZ - 13, 16));
t3 <= min(min(min(floord(_nTimesteps + _nY - 2, 16),
floord(8 * t1 + _nY + 14, 16)),
floord(16 * t2 + _nY + 12, 16)),
floord(16 * t1 - 16 * t2 + _nZ + _nY + 13, 16));
t3++) {
for (t4 = max(
max(max(0, ceild(t1 - 1, 2)), ceild(16 * t2 - _nZ - 13, 16)),
ceild(16 * t3 - _nY - 13, 16));
t4 <= min(min(min(min(floord(_nTimesteps + _nX - 2, 16),
floord(8 * t1 + _nX + 14, 16)),
floord(16 * t2 + _nX + 12, 16)),
floord(16 * t3 + _nX + 13, 16)),
floord(16 * t1 - 16 * t2 + _nZ + _nX + 13, 16));
t4++) {
for (t5 = max(max(max(max(max(0, 8 * t1), 16 * t1 - 16 * t2 + 2),
16 * t2 - _nZ + 1),
16 * t3 - _nY + 1),
16 * t4 - _nX + 1);
t5 <= min(min(min(min(min(_nTimesteps - 1, 8 * t1 + 15),
16 * t2 + 13),
16 * t3 + 14),
16 * t4 + 14),
16 * t1 - 16 * t2 + _nZ + 14);
t5++) {
/* Hoisted loop conditional */
if (t5 % 2 == 0) {
for (t6 = max(max(16 * t2, t5 + 2),
-16 * t1 + 16 * t2 + 2 * t5 - 15);
t6 <= min(min(16 * t2 + 15, -16 * t1 + 16 * t2 + 2 * t5),
t5 + _nZ - 1);
t6++) {
for (t7 = max(16 * t3, t5 + 1);
t7 <= min(16 * t3 + 15, t5 + _nY - 1); t7++) {
lbv = max(16 * t4, t5 + 1);
ubv = min(16 * t4 + 15, t5 + _nX - 1);
#pragma ivdep
#pragma vector always
for (t8 = lbv; t8 <= ubv; t8++) {
lbm_kernel(
grid[0][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][0],
grid[0][(-t5 + t6)][(-t5 + t7)][(-t5 + t8) - 1][1],
grid[0][(-t5 + t6)][(-t5 + t7)][(-t5 + t8) + 1][2],
grid[0][(-t5 + t6)][(-t5 + t7) - 1][(-t5 + t8)][3],
grid[0][(-t5 + t6)][(-t5 + t7) + 1][(-t5 + t8)][4],
grid[0][(-t5 + t6) - 1][(-t5 + t7)][(-t5 + t8)][5],
grid[0][(-t5 + t6) + 1][(-t5 + t7)][(-t5 + t8)][6],
grid[0][(-t5 + t6)][(-t5 + t7) - 1][(-t5 + t8) - 1]
[7],
grid[0][(-t5 + t6)][(-t5 + t7) - 1][(-t5 + t8) + 1]
[8],
grid[0][(-t5 + t6)][(-t5 + t7) + 1][(-t5 + t8) - 1]
[9],
grid[0][(-t5 + t6)][(-t5 + t7) + 1][(-t5 + t8) + 1]
[10],
grid[0][(-t5 + t6) - 1][(-t5 + t7)][(-t5 + t8) - 1]
[11],
grid[0][(-t5 + t6) - 1][(-t5 + t7)][(-t5 + t8) + 1]
[12],
grid[0][(-t5 + t6) + 1][(-t5 + t7)][(-t5 + t8) - 1]
[13],
grid[0][(-t5 + t6) + 1][(-t5 + t7)][(-t5 + t8) + 1]
[14],
grid[0][(-t5 + t6) - 1][(-t5 + t7) - 1][(-t5 + t8)]
[15],
grid[0][(-t5 + t6) - 1][(-t5 + t7) + 1][(-t5 + t8)]
[16],
grid[0][(-t5 + t6) + 1][(-t5 + t7) - 1][(-t5 + t8)]
[17],
grid[0][(-t5 + t6) + 1][(-t5 + t7) + 1][(-t5 + t8)]
[18],
&grid[1][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][0],
&grid[1][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][1],
&grid[1][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][2],
&grid[1][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][3],
&grid[1][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][4],
&grid[1][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][5],
&grid[1][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][6],
&grid[1][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][7],
&grid[1][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][8],
&grid[1][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][9],
&grid[1][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][10],
&grid[1][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][11],
&grid[1][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][12],
&grid[1][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][13],
&grid[1][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][14],
&grid[1][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][15],
&grid[1][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][16],
&grid[1][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][17],
&grid[1][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][18],
(t5), ((-t5 + t6)), ((-t5 + t7)), ((-t5 + t8)));
;
}
}
}
} else {
for (t6 = max(max(16 * t2, t5 + 2),
-16 * t1 + 16 * t2 + 2 * t5 - 15);
t6 <= min(min(16 * t2 + 15, -16 * t1 + 16 * t2 + 2 * t5),
t5 + _nZ - 1);
t6++) {
for (t7 = max(16 * t3, t5 + 1);
t7 <= min(16 * t3 + 15, t5 + _nY - 1); t7++) {
lbv = max(16 * t4, t5 + 1);
ubv = min(16 * t4 + 15, t5 + _nX - 1);
#pragma ivdep
#pragma vector always
for (t8 = lbv; t8 <= ubv; t8++) {
lbm_kernel(
grid[1][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][0],
grid[1][(-t5 + t6)][(-t5 + t7)][(-t5 + t8) - 1][1],
grid[1][(-t5 + t6)][(-t5 + t7)][(-t5 + t8) + 1][2],
grid[1][(-t5 + t6)][(-t5 + t7) - 1][(-t5 + t8)][3],
grid[1][(-t5 + t6)][(-t5 + t7) + 1][(-t5 + t8)][4],
grid[1][(-t5 + t6) - 1][(-t5 + t7)][(-t5 + t8)][5],
grid[1][(-t5 + t6) + 1][(-t5 + t7)][(-t5 + t8)][6],
grid[1][(-t5 + t6)][(-t5 + t7) - 1][(-t5 + t8) - 1]
[7],
grid[1][(-t5 + t6)][(-t5 + t7) - 1][(-t5 + t8) + 1]
[8],
grid[1][(-t5 + t6)][(-t5 + t7) + 1][(-t5 + t8) - 1]
[9],
grid[1][(-t5 + t6)][(-t5 + t7) + 1][(-t5 + t8) + 1]
[10],
grid[1][(-t5 + t6) - 1][(-t5 + t7)][(-t5 + t8) - 1]
[11],
grid[1][(-t5 + t6) - 1][(-t5 + t7)][(-t5 + t8) + 1]
[12],
grid[1][(-t5 + t6) + 1][(-t5 + t7)][(-t5 + t8) - 1]
[13],
grid[1][(-t5 + t6) + 1][(-t5 + t7)][(-t5 + t8) + 1]
[14],
grid[1][(-t5 + t6) - 1][(-t5 + t7) - 1][(-t5 + t8)]
[15],
grid[1][(-t5 + t6) - 1][(-t5 + t7) + 1][(-t5 + t8)]
[16],
grid[1][(-t5 + t6) + 1][(-t5 + t7) - 1][(-t5 + t8)]
[17],
grid[1][(-t5 + t6) + 1][(-t5 + t7) + 1][(-t5 + t8)]
[18],
&grid[0][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][0],
&grid[0][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][1],
&grid[0][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][2],
&grid[0][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][3],
&grid[0][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][4],
&grid[0][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][5],
&grid[0][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][6],
&grid[0][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][7],
&grid[0][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][8],
&grid[0][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][9],
&grid[0][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][10],
&grid[0][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][11],
&grid[0][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][12],
&grid[0][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][13],
&grid[0][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][14],
&grid[0][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][15],
&grid[0][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][16],
&grid[0][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][17],
&grid[0][(-t5 + t6)][(-t5 + t7)][(-t5 + t8)][18],
(t5), ((-t5 + t6)), ((-t5 + t7)), ((-t5 + t8)));
;
}
}
}
}
/* end hoisted if */
}
}
}
}
}
}
/* End of CLooG code */
#ifdef TIME
gettimeofday(&end, 0);
ts_return = timeval_subtract(&result, &end, &start);
tdiff = (double)(result.tv_sec + result.tv_usec * 1.0e-6);
printf("\tTime taken : %7.5lfm\n", tdiff / 60.0);
printf("\tMLUPS : %7.5lf\n", (total_lattice_pts / (1.0e6 * tdiff)));
#endif
#ifdef DEBUG
/* Dump rho, uX, uY for the entire domain to verify results */
dumpVelocities(t);
#endif
return 0;
}
#define S1(zT0,zT1,zT2,zT3,i,j) B[i][j]=A[i][j]+u1[i]*v1[j]+u2[i]*v2[j];
#define S2(zT0,zT1,zT2,zT3,i,j) x[i]=x[i]+beta*B[j][i]*y[j];
#define S3(zT0,zT1,zT2,zT3,i) x[i]=x[i]+z[i];
#define S4(zT0,zT1,zT2,zT3,i,j) w[i]=w[i]+alpha*B[i][j]*x[j];
int t0, t1, t2, t3, t4, t5, t6, t7;
register int lb, ub, lb1, ub1, lb2, ub2;
register int lbv, ubv;
/* Generated from PLUTO-produced CLooG file by CLooG v0.14.1 64 bits in 0.03s. */
lb1=0;
ub1=floord(N-1,8000);
#pragma omp parallel for shared(t0,lb1,ub1) private(t1,t2,t3,t4,t5,t6,t7)
for (t1=lb1; t1<=ub1; t1++) {
for (t2=0;t2<=floord(N-1,256);t2++) {
for (t3=max(20*t1,0);t3<=min(20*t1+19,floord(N-1,400));t3++) {
for (t4=max(0,16*t2);t4<=min(16*t2+15,floord(N-1,16));t4++) {
for (t5=max(0,16*t4);t5<=min(N-1,16*t4+15);t5++) {
{
lbv=max(0,400*t3); ubv=min(N-1,400*t3+399);
#pragma ivdep
#pragma vector always
for (t6=lbv; t6<=ubv; t6++) {
S1(t1,t2,t3,t4,t5,t6) ;
S2(t1,t2,t3,t4,t6,t5) ;
}
}
}
}
for (int c1 = 5; c1 <= 5 * M; c1 += 1) {
for (int c3 = max(2, floord(-M + c1, 4)); c3 < min(M, (c1 + 1) / 3 - 2); c3 += 1)
for (int c5 = max(1, -M - c3 + (M + c1) / 2 - 2); c5 < min(c3, -2 * c3 + (c1 + c3) / 2 - 2); c5 += 1)
S1(c1 - 2 * c3 - 2 * c5 - 5, c3, c5);
for (int c3 = max(1, floord(-M + c1, 4)); c3 < (c1 + 1) / 5; c3 += 1)
S2(c1 - 4 * c3 - 3, c3);
if (c1 % 5 == 0)
S4(c1 / 5);
for (int c3 = max(-3 * M - c1 + 3 * ((M + c1) / 2) + 1, -((c1 - 1) % 3) + 3); c3 < (c1 + 1) / 5; c3 += 3)
S3((c1 - 2 * c3 - 1) / 3, c3);
}
if (n % 2 == 0)
for (int c0 = (n / 2) + 2 * floord(-n - 1, 4) + 2; c0 <= 100; c0 += 2)
S(c0);
void test(int n)
{
/* Scattering iterators. */
int t1, t2, t3;
/* Original iterators. */
int i, j, k;
if (n >= 1) {
t1 = -n+1 ;
t2 = n+1 ;
for (t3=n+3;t3<=3*n+1;t3++) {
if ((t3+n+1)%2 == 0) {
k = (t3-n-1)/2 ;
S1(1,n,(t3-n-1)/2) ;
}
}
}
if ((n >= 2) && (n <= 2)) {
t1 = -n+2 ;
for (t2=-n+4;t2<=3*n-2;t2++) {
for (t3=t2+2;t3<=t2+2*n;t3++) {
if ((t2+n)%2 == 0) {
i = (t2-n+2)/2 ;
j = (t2+n-2)/2 ;
if ((t3+n)%2 == 0) {
k = (-t2+t3)/2 ;
S1((t2-n+2)/2,(t2+n-2)/2,(-t2+t3)/2) ;
}
}
}
}
t2 = n+3 ;
for (t3=1;t3<=n;t3++) {
S2(1,n,t3) ;
}
}
if (n >= 3) {
t1 = -n+2 ;
for (t2=n;t2<=n+2;t2++) {
for (t3=t2+2;t3<=t2+2*n;t3++) {
if ((t2+n)%2 == 0) {
i = (t2-n+2)/2 ;
j = (t2+n-2)/2 ;
if ((t3+n)%2 == 0) {
k = (-t2+t3)/2 ;
S1((t2-n+2)/2,(t2+n-2)/2,(-t2+t3)/2) ;
}
}
}
}
t2 = n+3 ;
for (t3=1;t3<=n;t3++) {
S2(1,n,t3) ;
}
}
for (t1=ceild(-2*n+5,2);t1<=min(-n+6,-1);t1++) {
for (t2=-t1+2;t2<=-t1+4;t2++) {
for (t3=t2+2;t3<=t2+2*n;t3++) {
if ((t1+t2)%2 == 0) {
i = (t1+t2)/2 ;
j = (-t1+t2)/2 ;
if ((t1+t3)%2 == 0) {
k = (-t2+t3)/2 ;
S1((t1+t2)/2,(-t1+t2)/2,(-t2+t3)/2) ;
}
}
}
}
for (t2=-t1+5;t2<=t1+2*n;t2++) {
for (t3=1;t3<=n;t3++) {
if ((t1+t2+1)%2 == 0) {
i = (t1+t2-3)/2 ;
j = (-t1+t2-1)/2 ;
S2((t1+t2-3)/2,(-t1+t2-1)/2,t3) ;
}
}
for (t3=t2+2;t3<=t2+2*n;t3++) {
if ((t1+t2)%2 == 0) {
i = (t1+t2)/2 ;
j = (-t1+t2)/2 ;
if ((t1+t3)%2 == 0) {
k = (-t2+t3)/2 ;
S1((t1+t2)/2,(-t1+t2)/2,(-t2+t3)/2) ;
}
}
}
}
t2 = t1+2*n+1 ;
for (t3=1;t3<=n;t3++) {
i = t1+n-1 ;
S2(t1+n-1,n,t3) ;
}
}
if (n == 2) {
for (t3=5;t3<=7;t3++) {
if ((t3+1)%2 == 0) {
k = (t3-3)/2 ;
S1(2,1,(t3-3)/2) ;
}
}
for (t2=4;t2<=6;t2++) {
for (t3=1;t3<=2;t3++) {
if (t2%2 == 0) {
i = (t2-2)/2 ;
j = (t2-2)/2 ;
S2((t2-2)/2,(t2-2)/2,t3) ;
}
}
}
}
for (t1=-n+7;t1<=-1;t1++) {
for (t2=-t1+2;t2<=-t1+4;t2++) {
for (t3=t2+2;t3<=t2+2*n;t3++) {
if ((t1+t2)%2 == 0) {
i = (t1+t2)/2 ;
j = (-t1+t2)/2 ;
if ((t1+t3)%2 == 0) {
k = (-t2+t3)/2 ;
S1((t1+t2)/2,(-t1+t2)/2,(-t2+t3)/2) ;
}
}
}
}
for (t2=-t1+5;t2<=n-2;t2++) {
for (t3=1;t3<=t2+1;t3++) {
if ((t1+t2+1)%2 == 0) {
i = (t1+t2-3)/2 ;
j = (-t1+t2-1)/2 ;
S2((t1+t2-3)/2,(-t1+t2-1)/2,t3) ;
}
}
for (t3=t2+2;t3<=n;t3++) {
if ((t1+t2+1)%2 == 0) {
i = (t1+t2-3)/2 ;
j = (-t1+t2-1)/2 ;
S2((t1+t2-3)/2,(-t1+t2-1)/2,t3) ;
}
if ((t1+t2)%2 == 0) {
i = (t1+t2)/2 ;
j = (-t1+t2)/2 ;
if ((t1+t3)%2 == 0) {
k = (-t2+t3)/2 ;
S1((t1+t2)/2,(-t1+t2)/2,(-t2+t3)/2) ;
}
}
}
for (t3=n+1;t3<=t2+2*n;t3++) {
if ((t1+t2)%2 == 0) {
i = (t1+t2)/2 ;
j = (-t1+t2)/2 ;
if ((t1+t3)%2 == 0) {
k = (-t2+t3)/2 ;
S1((t1+t2)/2,(-t1+t2)/2,(-t2+t3)/2) ;
}
}
}
}
for (t2=n-1;t2<=t1+2*n;t2++) {
for (t3=1;t3<=n;t3++) {
if ((t1+t2+1)%2 == 0) {
i = (t1+t2-3)/2 ;
j = (-t1+t2-1)/2 ;
S2((t1+t2-3)/2,(-t1+t2-1)/2,t3) ;
}
}
for (t3=t2+2;t3<=t2+2*n;t3++) {
if ((t1+t2)%2 == 0) {
i = (t1+t2)/2 ;
j = (-t1+t2)/2 ;
if ((t1+t3)%2 == 0) {
k = (-t2+t3)/2 ;
S1((t1+t2)/2,(-t1+t2)/2,(-t2+t3)/2) ;
}
}
}
}
t2 = t1+2*n+1 ;
for (t3=1;t3<=n;t3++) {
i = t1+n-1 ;
S2(t1+n-1,n,t3) ;
}
}
if (n >= 3) {
for (t1=0;t1<=min(1,-n+6);t1++) {
for (t2=t1+2;t2<=-t1+4;t2++) {
for (t3=t2+2;t3<=t2+2*n;t3++) {
if ((t1+t2)%2 == 0) {
i = (t1+t2)/2 ;
j = (-t1+t2)/2 ;
if ((t1+t3)%2 == 0) {
k = (-t2+t3)/2 ;
S1((t1+t2)/2,(-t1+t2)/2,(-t2+t3)/2) ;
}
}
}
}
for (t2=-t1+5;t2<=-t1+2*n;t2++) {
for (t3=1;t3<=n;t3++) {
if ((t1+t2+1)%2 == 0) {
i = (t1+t2-3)/2 ;
j = (-t1+t2-1)/2 ;
S2((t1+t2-3)/2,(-t1+t2-1)/2,t3) ;
}
}
for (t3=t2+2;t3<=t2+2*n;t3++) {
if ((t1+t2)%2 == 0) {
i = (t1+t2)/2 ;
j = (-t1+t2)/2 ;
if ((t1+t3)%2 == 0) {
k = (-t2+t3)/2 ;
S1((t1+t2)/2,(-t1+t2)/2,(-t2+t3)/2) ;
}
}
}
}
for (t2=-t1+2*n+1;t2<=t1+2*n+1;t2++) {
for (t3=1;t3<=n;t3++) {
if ((t1+t2+1)%2 == 0) {
i = (t1+t2-3)/2 ;
j = (-t1+t2-1)/2 ;
S2((t1+t2-3)/2,(-t1+t2-1)/2,t3) ;
}
}
}
}
}
for (t1=max(-n+7,0);t1<=1;t1++) {
for (t2=t1+2;t2<=-t1+4;t2++) {
for (t3=t2+2;t3<=t2+2*n;t3++) {
if ((t1+t2)%2 == 0) {
i = (t1+t2)/2 ;
j = (-t1+t2)/2 ;
if ((t1+t3)%2 == 0) {
k = (-t2+t3)/2 ;
S1((t1+t2)/2,(-t1+t2)/2,(-t2+t3)/2) ;
}
}
}
}
for (t2=-t1+5;t2<=n-2;t2++) {
for (t3=1;t3<=t2+1;t3++) {
if ((t1+t2+1)%2 == 0) {
i = (t1+t2-3)/2 ;
j = (-t1+t2-1)/2 ;
S2((t1+t2-3)/2,(-t1+t2-1)/2,t3) ;
}
}
for (t3=t2+2;t3<=n;t3++) {
if ((t1+t2+1)%2 == 0) {
i = (t1+t2-3)/2 ;
j = (-t1+t2-1)/2 ;
S2((t1+t2-3)/2,(-t1+t2-1)/2,t3) ;
}
if ((t1+t2)%2 == 0) {
i = (t1+t2)/2 ;
j = (-t1+t2)/2 ;
if ((t1+t3)%2 == 0) {
k = (-t2+t3)/2 ;
S1((t1+t2)/2,(-t1+t2)/2,(-t2+t3)/2) ;
}
}
}
for (t3=n+1;t3<=t2+2*n;t3++) {
if ((t1+t2)%2 == 0) {
i = (t1+t2)/2 ;
j = (-t1+t2)/2 ;
if ((t1+t3)%2 == 0) {
k = (-t2+t3)/2 ;
S1((t1+t2)/2,(-t1+t2)/2,(-t2+t3)/2) ;
}
}
}
}
for (t2=n-1;t2<=-t1+2*n;t2++) {
for (t3=1;t3<=n;t3++) {
if ((t1+t2+1)%2 == 0) {
i = (t1+t2-3)/2 ;
j = (-t1+t2-1)/2 ;
S2((t1+t2-3)/2,(-t1+t2-1)/2,t3) ;
}
}
for (t3=t2+2;t3<=t2+2*n;t3++) {
if ((t1+t2)%2 == 0) {
i = (t1+t2)/2 ;
j = (-t1+t2)/2 ;
if ((t1+t3)%2 == 0) {
k = (-t2+t3)/2 ;
S1((t1+t2)/2,(-t1+t2)/2,(-t2+t3)/2) ;
}
}
}
}
for (t2=-t1+2*n+1;t2<=t1+2*n+1;t2++) {
for (t3=1;t3<=n;t3++) {
if ((t1+t2+1)%2 == 0) {
i = (t1+t2-3)/2 ;
j = (-t1+t2-1)/2 ;
S2((t1+t2-3)/2,(-t1+t2-1)/2,t3) ;
}
}
}
}
for (t1=2;t1<=n-5;t1++) {
t2 = t1+2 ;
for (t3=t1+4;t3<=t1+2*n+2;t3++) {
i = t1+1 ;
if ((t1+t3)%2 == 0) {
k = (-t1+t3-2)/2 ;
S1(t1+1,1,(-t1+t3-2)/2) ;
}
}
for (t2=t1+3;t2<=n-2;t2++) {
for (t3=1;t3<=t2+1;t3++) {
if ((t1+t2+1)%2 == 0) {
i = (t1+t2-3)/2 ;
j = (-t1+t2-1)/2 ;
S2((t1+t2-3)/2,(-t1+t2-1)/2,t3) ;
}
}
for (t3=t2+2;t3<=n;t3++) {
if ((t1+t2+1)%2 == 0) {
i = (t1+t2-3)/2 ;
j = (-t1+t2-1)/2 ;
S2((t1+t2-3)/2,(-t1+t2-1)/2,t3) ;
}
if ((t1+t2)%2 == 0) {
i = (t1+t2)/2 ;
j = (-t1+t2)/2 ;
if ((t1+t3)%2 == 0) {
k = (-t2+t3)/2 ;
S1((t1+t2)/2,(-t1+t2)/2,(-t2+t3)/2) ;
}
}
}
for (t3=n+1;t3<=t2+2*n;t3++) {
if ((t1+t2)%2 == 0) {
i = (t1+t2)/2 ;
j = (-t1+t2)/2 ;
if ((t1+t3)%2 == 0) {
k = (-t2+t3)/2 ;
S1((t1+t2)/2,(-t1+t2)/2,(-t2+t3)/2) ;
}
}
}
}
for (t2=n-1;t2<=-t1+2*n;t2++) {
for (t3=1;t3<=n;t3++) {
if ((t1+t2+1)%2 == 0) {
i = (t1+t2-3)/2 ;
j = (-t1+t2-1)/2 ;
S2((t1+t2-3)/2,(-t1+t2-1)/2,t3) ;
}
}
for (t3=t2+2;t3<=t2+2*n;t3++) {
if ((t1+t2)%2 == 0) {
i = (t1+t2)/2 ;
j = (-t1+t2)/2 ;
if ((t1+t3)%2 == 0) {
k = (-t2+t3)/2 ;
S1((t1+t2)/2,(-t1+t2)/2,(-t2+t3)/2) ;
}
}
}
}
for (t2=-t1+2*n+1;t2<=-t1+2*n+3;t2++) {
for (t3=1;t3<=n;t3++) {
if ((t1+t2+1)%2 == 0) {
i = (t1+t2-3)/2 ;
j = (-t1+t2-1)/2 ;
S2((t1+t2-3)/2,(-t1+t2-1)/2,t3) ;
}
}
}
}
for (t1=max(2,n-4);t1<=floord(2*n-3,2);t1++) {
t2 = t1+2 ;
for (t3=t1+4;t3<=t1+2*n+2;t3++) {
i = t1+1 ;
if ((t1+t3)%2 == 0) {
k = (-t1+t3-2)/2 ;
S1(t1+1,1,(-t1+t3-2)/2) ;
}
}
for (t2=t1+3;t2<=-t1+2*n;t2++) {
for (t3=1;t3<=n;t3++) {
if ((t1+t2+1)%2 == 0) {
i = (t1+t2-3)/2 ;
j = (-t1+t2-1)/2 ;
S2((t1+t2-3)/2,(-t1+t2-1)/2,t3) ;
}
}
for (t3=t2+2;t3<=t2+2*n;t3++) {
if ((t1+t2)%2 == 0) {
i = (t1+t2)/2 ;
j = (-t1+t2)/2 ;
if ((t1+t3)%2 == 0) {
k = (-t2+t3)/2 ;
S1((t1+t2)/2,(-t1+t2)/2,(-t2+t3)/2) ;
}
}
}
}
for (t2=-t1+2*n+1;t2<=-t1+2*n+3;t2++) {
for (t3=1;t3<=n;t3++) {
if ((t1+t2+1)%2 == 0) {
i = (t1+t2-3)/2 ;
j = (-t1+t2-1)/2 ;
S2((t1+t2-3)/2,(-t1+t2-1)/2,t3) ;
}
}
}
}
if (n >= 3) {
t1 = n-1 ;
t2 = n+1 ;
for (t3=n+3;t3<=3*n+1;t3++) {
if ((t3+n+1)%2 == 0) {
k = (t3-n-1)/2 ;
S1(n,1,(t3-n-1)/2) ;
}
}
for (t2=n+2;t2<=n+4;t2++) {
for (t3=1;t3<=n;t3++) {
if ((t2+n)%2 == 0) {
i = (t2+n-4)/2 ;
j = (t2-n)/2 ;
S2((t2+n-4)/2,(t2-n)/2,t3) ;
}
}
}
}
if (n >= 1) {
t2 = n+3 ;
for (t3=1;t3<=n;t3++) {
S2(n,1,t3) ;
}
}
}
for (int c0 = 0; c0 < m; c0 += 32)
for (int c1 = (n >= 32 && m >= c0 + 2) || (m == 1 && c0 == 0) ? 0 : 32 * n - 32 * floord(31 * n + 31, 32); c1 <= ((n <= -1 && c0 == 0) || (m == 1 && n >= 0 && c0 == 0) ? max(0, n - 1) : n); c1 += 32)
for (int c2 = c0; c2 <= (m >= 2 && c0 + 31 >= m && n >= c1 && c1 + 31 >= n ? 2 * m - 3 : (m >= 2 * c0 + 63 && c1 <= -32 && n >= c1 && c1 + 31 >= n) || (m >= c0 + 32 && 2 * c0 + 62 >= m && n >= c1 && c1 + 31 >= n) || (n >= 0 && c0 >= 32 && m >= 2 * c0 + 63 && c1 == n) || (m >= 63 && n >= 32 && c0 == 0 && c1 == n) ? 2 * c0 + 61 : m - 1); c2 += 32) {
if (n >= c1 + 32 && c1 >= 0 && 2 * c0 >= c2 + 32) {
for (int c4 = 0; c4 <= 31; c4 += 1)
for (int c5 = max(0, c0 - c2 + 1); c5 <= min(31, m - c2 - 1); c5 += 1)
S_27(c0, c2 + c5, c1 + c4);
} else if (c0 >= 32 && c1 >= 0 && c2 >= 2 * c0) {
for (int c4 = 0; c4 <= min(31, n - c1 - 1); c4 += 1)
for (int c5 = 0; c5 <= min(31, m - c2 - 1); c5 += 1)
S_27(c0, c2 + c5, c1 + c4);
} else if (c0 == 0 && c1 >= 0) {
for (int c4 = 0; c4 <= min(31, n - c1 - 1); c4 += 1)
for (int c5 = 0; c5 <= min(31, m - c2 - 1); c5 += 1) {
if (c1 == 0 && c4 == 0)
S_14(c2 + c5);
S_19(c1 + c4, c2 + c5);
if (c2 + c5 >= 1)
S_27(0, c2 + c5, c1 + c4);
}
}
if (c1 >= 0) {
for (int c3 = 1; c3 <= min(31, (c2 / 2) - c0); c3 += 1)
for (int c4 = 0; c4 <= min(31, n - c1 - 1); c4 += 1)
for (int c5 = 0; c5 <= min(31, m - c2 - 1); c5 += 1)
S_27(c0 + c3, c2 + c5, c1 + c4);
if (n >= c1 + 32) {
for (int c3 = max(1, (c2 / 2) - c0 + 1); c3 <= min(min(31, m - c0 - 2), -c0 + c2 + 30); c3 += 1)
for (int c4 = 0; c4 <= 31; c4 += 1)
for (int c5 = max(0, c0 - c2 + c3 + 1); c5 <= min(31, m - c2 - 1); c5 += 1)
S_27(c0 + c3, c2 + c5, c1 + c4);
