#include "BSprivate.h"

/*@ BSsetup_forward - Set up the communication structure for triangular
                      matrix solution

    Input Parameters:
.   A - a sparse matrix
.   procinfo - the usual processor stuff

    Returns:
    the communication structure for triangular matrix solution

 @*/
BScomm *BSsetup_forward(BSpar_mat *A, BSprocinfo *procinfo)
{
	BMcomp_msg *to_msg, *from_msg;
	BMphase *phase_ptr;
	BMmsg *msg;
	int	i, j, k;
	int	cl_ind, tcl_ind, cur_proc;
	int	count, size;
	int	*setup_data, *user_data;
	BScl_2_inode *clique2inode;
	BSnumbering *color2clique;
	BSinode *inodes;
	BScomm *comm_ptr;
	int	*data, data_len, ind, *perm;
	void (*map)(int,int *,int *,BSprocinfo *,BSmapping *);

	/* initialize communication structures */
	to_msg = BMcomp_init(COMP_MSG_BASE); CHKERRN(0);
	from_msg = BMcomp_init(COMP_MSG_BASE); CHKERRN(0);

	/* now go through and figure out everyone that we need stuff from */
	/* do it by color, where each color is a phase */
	color2clique = A->color2clique;
	clique2inode = A->clique2inode;
	inodes = A->inodes->list;
	for (i=0;i<color2clique->length-1;i++) {
		cl_ind = color2clique->numbers[i];
		while (cl_ind < color2clique->numbers[i+1]) {
			/* figure out the message size and destination */
			cur_proc = clique2inode->proc[cl_ind];
			tcl_ind = cl_ind;
			size = 0;
			while ((tcl_ind < color2clique->numbers[i+1]) && 
				(cur_proc == clique2inode->proc[tcl_ind])) {
				/* add in the lengths, nonlocals are negative */
				size += abs(clique2inode->d_mats[tcl_ind].size);
				tcl_ind++;
			}
			if (cur_proc != procinfo->my_id) {
				/* now we know the message size, allocate the message, etc */
				msg = BMcreate_msg(i,-1,MY_MPI_FLT,cur_proc); CHKERRN(0);
				BMadd_msg(from_msg,msg,procinfo); CHKERRN(0);
				MY_MALLOCN(user_data,(int *),sizeof(int)*2,1);
				user_data[0] = cl_ind;
				user_data[1] = tcl_ind-1;
				BMset_user_data(msg,user_data,2,BSfree_comm_data); CHKERRN(0);
				MY_MALLOCN(setup_data,(int *),sizeof(int)*size,1);
				count = 0;
				while (count < size) {
					for (j=clique2inode->inode_index[cl_ind];
						j<clique2inode->inode_index[cl_ind+1];j++) {
						for (k=0;k<inodes[j].num_cols;k++) {
							setup_data[count] = inodes[j].o_gcol_num[k];
							count++;
						}
					}
					cl_ind++;
				}
				BMset_setup_data(msg,setup_data,size,BSfree_comm_data);
				CHKERRN(0);
				BMset_msg_size(msg,size); CHKERRN(0);
			}
			cl_ind = tcl_ind;
		}

		/* now, let's work out what I need to send */
		phase_ptr = BMget_phase(from_msg,i); CHKERRN(0);
		count = BMfix_send(SETUP_FORWARD_MSG,COMP_MSG_BASE,MY_MPI_FLT,to_msg,
			phase_ptr,BSfree_comm_data,procinfo); CHKERRN(0);
		
		/* now free up the setup data */
		msg = NULL;
		while ((msg = BMnext_msg(phase_ptr,msg)) != NULL) {
			CHKERRN(0);
			BMfree_setup_data(msg);
			CHKERRN(0);
		}
		CHKERRN(0);
	}

	/* now, go over the messages that I have to send and translate the */
	/* user data */
	map = A->map->fglobal2local;
	perm = A->perm->perm;
	for (i=0;i<color2clique->length-1;i++) {
		phase_ptr = BMget_phase(to_msg,i); CHKERRN(0);
		msg = NULL;
		while ((msg = BMnext_msg(phase_ptr,msg)) != NULL) {
			CHKERRN(0);
			data = BMget_user(msg,&data_len); CHKERRN(0);
			BMset_msg_size(msg,data_len); CHKERRN(0);
			for (j=0;j<data_len;j++) {
				(*map)(1,&(data[j]),&ind,procinfo,A->map); CHKERRN(0);
				data[j] = perm[ind];
			}
		}
		CHKERRN(0);
	}

	/* organize the messages */
	BMalloc_msg(to_msg); CHKERRN(0);
	BMalloc_msg(from_msg); CHKERRN(0);

	MY_MALLOCN(comm_ptr,(BScomm *),sizeof(BScomm),1);
	comm_ptr->num_rhs = 1;
	comm_ptr->to_msg = to_msg;
	comm_ptr->from_msg = from_msg;
	return(comm_ptr);
}