tree.c

/*
 * Implementation of a distributed memory  kd-tree
 * The idea is to have a top level domain decomposition with a shallow shared
 * top level tree between computational nodes.
 *
 * Then each domain has a different set of points to work on separately
 * the top tree serves as a map to know later on in which processor ask for
 * neighbors
 */
#include "tree.h"
#include "heap.h"
#include "kdtreeV2.h"
#include "mpi.h"
#include <math.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <omp.h>
#include <sys/sysinfo.h>

//#define WRITE_NGBH
//#define WRITE_TOP_NODES
//#define WRITE_DENSITY
#define WRITE_CLUSTER_ASSIGN_H1

/* 
 * Maximum bytes to send with a single mpi send/recv, used 
 * while communicating results of ngbh search
 */

/* Maximum allowed is 4GB */
//#define MAX_MSG_SIZE 4294967296

/* Used slices of 10 mb ? Really good? Maybe at the cause of TID thing */
#define MAX_MSG_SIZE 1000000000 

#ifdef USE_FLOAT32
#define MPI_MY_FLOAT MPI_FLOAT
#else
#define MPI_MY_FLOAT MPI_DOUBLE
#endif

#define HERE printf("%d reached line %d\n", ctx -> mpi_rank, __LINE__);

#define I_AM_MASTER ctx->mpi_rank == 0

#define TOP_TREE_RCH 1
#define TOP_TREE_LCH 0
#define NO_CHILD -1

unsigned int data_dims;
const border_t border_null = {.density = -1.0, .error = 0, .idx = NOBORDER};
const sparse_border_t sparse_border_null = {.density = -1.0, .error = 0, .idx = NOBORDER, .i = NOBORDER, .j = NOBORDER};

#define PREALLOC_BORDERS 100


int cmp_float_t(const void* a, const void* b)
{
    float_t aa = *((float_t*)a);
    float_t bb = *((float_t*)b);
    return  (aa > bb) - (aa < bb);
}


float_t *read_data_file(global_context_t *ctx, const char *fname,
                        const int file_in_float32) 
{