static void init_setup_tab( void )
{
init_wg();
init_wgs();
init_wgt0();
init_wgpt0();
init_wgst0();
init_wgspt0();
init_wgf();
init_wgfs();
init_wgft0();
init_wgfpt0();
init_wgfst0();
init_wgfspt0();
init_t0();
init_f();
init_ft0();
init_g();
init_gs();
init_gt0();
init_gst0();
init_gf();
init_gfs();
init_gft0();
init_gfst0();
}
void ft_print_c(va_list *ap, t_format *f)
{
t_ullong arg;
arg = va_arg(*ap, t_ullong);
update_format(f);
get_result(arg, f);
init_f(f);
}
static void init_setup_tab( void )
{
init_wg(); /* pos + col */
init_wgt0(); /* pos + col + tex0 */
init_wgpt0(); /* pos + col + p-tex0 (?) */
init_t0(); /* tex0 */
init_g(); /* col */
init_gt0(); /* col + tex */
init_wgf();
init_wgft0();
init_wgfpt0();
init_f();
init_gf();
init_gft0();
}
static void init_setup_tab( void )
{
init_wg();
init_wgs();
init_wgt0();
init_wgt0t1();
init_wgpt0();
init_wgst0();
init_wgst0t1();
init_wgspt0();
init_wgf();
init_wgfs();
init_wgft0();
init_wgft0t1();
init_wgfpt0();
init_wgfst0();
init_wgfst0t1();
init_wgfspt0();
init_t0();
init_t0t1();
init_f();
init_ft0();
init_ft0t1();
init_g();
init_gs();
init_gt0();
init_gt0t1();
init_gst0();
init_gst0t1();
init_gf();
init_gfs();
init_gft0();
init_gft0t1();
init_gfst0();
init_gfst0t1();
/* Add proj texturing on t1 */
init_wgpt0t1();
init_wgspt0t1();
init_wgfpt0t1();
init_wgfspt0t1();
}
int main()
{
int enable_profiling = 0;
#ifdef DO_TIMING
enable_profiling = 1;
#endif
//print_platforms_devices();
cl_context ctx;
cl_command_queue queue;
create_context_on("NVIDIA", NULL, 0, &ctx, &queue, enable_profiling);
// --------------------------------------------------------------------------
// load kernels
// --------------------------------------------------------------------------
// read the cl file
char buf[100];
sprintf(buf, "mg-kernel-ver%d.cl", VERSION);
char *knl_text = read_file(buf);
//get work group dimensions and gflop info.
int wg_dims , wg_x, wg_y, wg_z, z_div, fetch_per_pt, flops_per_pt;
if (sscanf(knl_text, "// workgroup: (%d,%d,%d) z_div:%d fetch_per_pt:%d flops_per_pt:%d",
&wg_x, &wg_y, &wg_z, &z_div, &fetch_per_pt, &flops_per_pt) == 6)
{
wg_dims = 3;
}
else if (sscanf(knl_text, "// workgroup: (%d,%d) fetch_per_pt:%d flops_per_pt:%d",
&wg_x, &wg_y, &fetch_per_pt, &flops_per_pt) == 4)
{
wg_dims = 2;
wg_z = -1;
z_div = -1;
}
else
{
perror("reading workgroup spec");
abort();
}
#ifdef USE_DOUBLE
char *compile_opt = "-DFTYPE=double";
#else
char *compile_opt = "-DFTYPE=float";
#endif
// creation of the kernel
cl_kernel poisson_knl = kernel_from_string(ctx, knl_text, "fd_update", compile_opt);
free(knl_text); // my compiler complains about this one. OJO!!
// --------------------------------------------------------------------------
// set up grid
// --------------------------------------------------------------------------
const unsigned points = POINTS;
const ftype minus_bdry = -1, plus_bdry = 1;
// We're dividing into (points-1) intervals.
ftype dx = (plus_bdry-minus_bdry)/(points-1);
// --------------------------------------------------------------------------
// allocate and initialize CPU memory
// --------------------------------------------------------------------------
int use_alignment;
unsigned dim_other = points; //if order 2 then 1 point extra on each side
#ifdef USE_ALIGNMENT
// adjusts dimension so that the next row starts in a number divisible by 16
unsigned dim_x = ((dim_other + 15) / 16) * 16;
unsigned field_start = 0;
use_alignment = 1;
#else
unsigned dim_x = dim_other;
unsigned field_start = 0;// this one puts me right at the beginning
use_alignment = 0;
#endif
// --------Allocate forcing uexact, r and u vectors -------------------------
const size_t field_size = 0+dim_x*dim_x*dim_x; // extra large to fit the 2^n constrain in GPU
ftype *f = malloc(field_size*sizeof(ftype));
CHECK_SYS_ERROR(!f, "allocating f");
ftype *u = malloc (field_size*sizeof(ftype));
CHECK_SYS_ERROR(!u, "allocating u");
ftype *uexact = malloc (field_size*sizeof(ftype));
CHECK_SYS_ERROR(!uexact, "allocating uexact");
ftype *r = malloc(field_size * sizeof(ftype));
CHECK_SYS_ERROR(!r, "allocating residual r");
// --------------------------------------------------------------------------
// initialize
// --------------------------------------------------------------------------
// zero out (necessary to initialize everything bec. I measure norms)
for (size_t i = 0; i < field_size; ++i){
f[i] = 0;
u[i] = 0;
uexact[i] = 0;
r[i] = 0;
}
// set up the forcing field
init_f (points, f, dx, field_start, dim_x, dim_other, minus_bdry);
// Initialize u with initial boundary conditions
init_u ( points, u , minus_bdry, plus_bdry, dx, field_start, dim_x, dim_other);
// Initialize the exact solution
init_uexact(points, u, uexact, dx, field_size, field_start, dim_x, dim_other);
// --------------------------------------------------------------------------
// Setup the v-cycles
// --------------------------------------------------------------------------
unsigned n1, n2, n3, ncycles;
n1 = 50;
n2 = 60;
n3 = 1;
ncycles = 2;
ftype *sweeps = malloc (ncycles*sizeof(ftype));
ftype *rnorm = malloc (ncycles*sizeof(ftype));
ftype *enorm = malloc (ncycles*sizeof(ftype));
ftype rtol = 1.0e-05;
// Find the norm of the residual (choose your method)
sweeps[0] =0;
resid (r, f, u, dx, field_size, field_start, dim_x, dim_other);
rnorm[0] = norm( r , field_size) * dx;
U_error(u, uexact, r, field_size);
enorm[0] = norm( r, field_size ) * dx;
for(unsigned icycle = 1; icycle <= ncycles; icycle++){
mgv(f, u, dx, n1, n2, n3, field_size, points, use_alignment, dim_x, ctx, queue, poisson_knl, wg_dims , wg_x, wg_y, wg_z, z_div, fetch_per_pt, flops_per_pt); //update u through a v-cycle
sweeps[icycle] = sweeps[icycle -1] + (4 * (n1 + n2)/3);
resid (r, f, u, dx, field_size, field_start, dim_x, dim_other);
rnorm[icycle] = norm( r, field_size ) * dx;
U_error(u, uexact, r, field_size);
enorm[icycle] = norm( r, field_size ) * dx;
//cfacts = (rnorm(icycle)/rnorm(icycle - 1))^(1 / (n1 + n2)) not necessary
//disp something here if I want to.
//printf("norm of the cycle %f", enorm[icycle]);
if(rnorm[icycle] <= rtol * rnorm[0])
break;
}
#ifdef DO_TIMING
printf(" ftype:%d ver:%d align:%d pts:%d\tgflops:%.1f\tmcells:%.1f\tgbytes:%.1f [/sec]\tout_gflops:%.6f\n", (int) sizeof(ftype), VERSION, use_alignment, points, gflops_performed/seconds_taken, mcells_updated/seconds_taken, gbytes_accessed/seconds_taken, gflops_performed/tot_secs);
#endif
// --------------------------------------------------------------------------
// clean up
// --------------------------------------------------------------------------
CALL_CL_GUARDED(clReleaseKernel, (poisson_knl));
CALL_CL_GUARDED(clReleaseCommandQueue, (queue));
CALL_CL_GUARDED(clReleaseContext, (ctx));
}
void Lifter::init() {
init_f("Lifter");
}
void Tracer::init() {
init_f("Tracer");
}