BASKER_INLINE
void BaskerMatrix<Int,Entry,Exe_Space>::Finalize()
{
if(v_fill == BASKER_TRUE)
{
FREE_INT_1DARRAY(col_ptr);
FREE_INT_1DARRAY(row_idx);
FREE_ENTRY_1DARRAY(val);
v_fill = BASKER_FALSE;
}
if(w_fill == BASKER_TRUE)
{
FREE_INT_1DARRAY(iws);
FREE_ENTRY_1DARRAY(ews);
w_fill = BASKER_FALSE;
}
if(inc_lvl_flg == BASKER_TRUE)
{
FREE_INT_1DARRAY(inc_lvl);
inc_lvl_flg = BASKER_FALSE;
}
}//end finalize()
BASKER_INLINE
int Basker<Int, Entry, Exe_Space>::permute
(
ENTRY_1DARRAY vec,
INT_1DARRAY p,
Int n
)
{
ENTRY_1DARRAY temp;
MALLOC_ENTRY_1DARRAY(temp, n);
init_value(temp, n, (Entry) 0.0);
//Permute
for(Int i = 0; i < n; i++)
{
temp(i) = vec(p(i));
//temp(p(i)) = vec(i);
}
//Copy back
for(Int i = 0; i < n; i++)
{
vec(i) = temp(i);
}
FREE_ENTRY_1DARRAY(temp);
return 0;
}
BASKER_INLINE
int Basker<Int, Entry, Exe_Space>::permute
(
ENTRY_1DARRAY vec,
INT_1DARRAY p,
Int n, //n = size(vec) //n > m
Int m, //m = size(p)
Int start
)
{
if(m > n)
{
printf("ERROR permtue \n");
return BASKER_ERROR;
}
ENTRY_1DARRAY temp;
MALLOC_ENTRY_1DARRAY(temp, n);
init_value(temp, n, (Entry) 0.0);
//Permute
for(Int i = 0; i < n; i++)
{
temp(i) = vec(p(i));
//temp(p(i)) = vec(i);
}
//Copy back
for(Int i = 0; i < n; i++)
{
vec(i) = temp(i);
}
FREE_ENTRY_1DARRAY(temp);
return 0;
}
BASKER_INLINE
int Basker<Int,Entry,Exe_Space>::solve_interface
(
Entry *_x, //Solution (len = gn)
Entry *_y
)
{
//===== Move to view===========
ENTRY_1DARRAY x;
ENTRY_1DARRAY y;
MALLOC_ENTRY_1DARRAY(x, gn);
MALLOC_ENTRY_1DARRAY(y, gm);
for(Int i =0; i < gn; i++)
{
x(i) = (Entry) 0;
y(i) = (Entry) _y[i];
}
//printf("RHS: \n");
//printVec(y, gn);
//printf("\n");
//===== Permute
//printf("Permute RHS\n");
//==== Need to make this into one global perm
if(match_flag == BASKER_TRUE)
{
//printf("match order\n");
//printVec("match.txt", order_match_array, gn);
permute_inv(y,order_match_array, gn);
}
if(btf_flag == BASKER_TRUE)
{
//printf("btf order\n");
//printVec("btf.txt", order_btf_array, gn);
permute_inv(y,order_btf_array, gn);
//printVec("btf_amd.txt", order_c_csym_array, gn);
permute_inv(y,order_blk_amd_array, gn);
}
if(nd_flag == BASKER_TRUE)
{
//printf("ND order \n");
//printVec("nd.txt", part_tree.permtab, gn);
permute_inv(y,part_tree.permtab, gn);
}
if(amd_flag == BASKER_TRUE)
{
//printf("AMD order \n");
//printVec("amd.txt",order_csym_array, gn);
permute_inv(y,order_csym_array, gn);
}
//printVec("perm.txt" , gperm, gn);
permute_inv(y,gperm, gn);
solve_interface(x,y);
//Inverse perm
//Note: don't need to inverse a row only perm
if(btf_flag == BASKER_TRUE)
{
//printf("btf order\n");
//printVec(order_btf_array, gn);
permute(x,order_btf_array, gn);
}
if(nd_flag == BASKER_TRUE)
{
//printf("ND order \n");
//printVec(part_tree.permtab, gn);
permute(x,part_tree.permtab, gn);
}
if(amd_flag == BASKER_TRUE)
{
//printf("AMD order \n");
//printVec(order_csym_array, gn);
//FILE *fpamd;
//fpamd = fopen("amd.csv", "w");
//for(Int i = 0; i < gn; i++)
// {
// fprintf(fpamd, "%d \n",
// order_csym_array(i));
// }
//fclose(fpamd);
permute(x,order_csym_array, gn);
}
#ifdef BASKER_DEBUG_SOLVE_RHS
printf("\n\n");
printf("X: \n");
for(Int i = 0; i < gn; i++)
{
printf("%f, " , x(i));
}
printf("\n\n");
printf("RHS: \n");
for(Int i =0; i < gm; i++)
{
printf("%f, ", y(i));
}
printf("\n\n");
#endif
for(Int i = 0; i < gn; i++)
{
_x[i] = x(i);
}
#ifndef BASKER_KOKKOS
FREE_ENTRY_1DARRAY(x);
FREE_ENTRY_1DARRAY(y);
#endif
return 0;
}
BASKER_INLINE
int Basker<Int, Entry, Exe_Space>::permute_col
(
BASKER_MATRIX &M,
INT_1DARRAY col
)
{
if((M.ncol == 0)||(M.nnz == 0))
return 0;
Int n = M.ncol;
Int nnz = M.nnz;
//printf("Using n: %d nnz: %d \n", n, nnz);
INT_1DARRAY temp_p;
MALLOC_INT_1DARRAY(temp_p, n+1);
init_value(temp_p, n+1, (Int)0);
INT_1DARRAY temp_i;
MALLOC_INT_1DARRAY(temp_i, nnz);
init_value(temp_i, nnz, (Int)0);
ENTRY_1DARRAY temp_v;
MALLOC_ENTRY_1DARRAY(temp_v, nnz);
init_value(temp_v, nnz, (Entry)0.0);
//printf("done with init \n");
//Determine column ptr of output matrix
for(Int j = 0; j < n; j++)
{
Int i = col (j);
temp_p (i+1) = M.col_ptr (j+1) - M.col_ptr (j);
}
//Get ptrs from lengths
temp_p (0) = 0;
for(Int j = 0; j < n; j++)
{
temp_p (j+1) = temp_p (j+1) + temp_p (j);
}
//copy idxs
for(Int ii = 0; ii < n; ii++)
{
Int ko = temp_p (col (ii) );
for(Int k = M.col_ptr (ii); k < M.col_ptr (ii+1); k++)
{
temp_i (ko) = M.row_idx (k);
temp_v (ko) = M.val (k);
ko++;
}
}
//copy back int A
for(Int ii=0; ii < n+1; ii++)
{
M.col_ptr (ii) = temp_p (ii);
}
for(Int ii=0; ii < nnz; ii++)
{
M.row_idx (ii) = temp_i (ii);
M.val (ii) = temp_v (ii);
}
FREE_INT_1DARRAY(temp_p);
FREE_INT_1DARRAY(temp_i);
FREE_ENTRY_1DARRAY(temp_v);
return 0;
}//end permute_col(int)
BASKER_INLINE
void Basker<Int,Entry,Exe_Space>::Finalize()
{
//finalize all matrices
A.Finalize();
At.Finalize(); //??? is At even used
BTF_A.Finalize();
BTF_C.Finalize();
BTF_B.Finalize();
BTF_D.Finalize();
BTF_E.Finalize();
//finalize array of 2d matrics
FREE_MATRIX_VIEW_2DARRAY(AV, tree.nblks);
FREE_MATRIX_VIEW_2DARRAY(AL, tree.nblks);
FREE_MATRIX_2DARRAY(AVM, tree.nblks);
FREE_MATRIX_2DARRAY(ALM, tree.nblks);
FREE_MATRIX_2DARRAY(LL, tree.nblks);
FREE_MATRIX_2DARRAY(LU, tree.nblks);
FREE_INT_1DARRAY(LL_size);
FREE_INT_1DARRAY(LU_size);
//BTF structure
FREE_INT_1DARRAY(btf_tabs);
FREE_INT_1DARRAY(btf_blk_work);
FREE_INT_1DARRAY(btf_blk_nnz);
FREE_MATRIX_1DARRAY(LBTF);
FREE_MATRIX_1DARRAY(UBTF);
//Thread Array
FREE_THREAD_1DARRAY(thread_array);
basker_barrier.Finalize();
//S (Check on this)
FREE_INT_2DARRAY(S, tree.nblks);
//Permuations
FREE_INT_1DARRAY(gperm);
FREE_INT_1DARRAY(gpermi);
if(match_flag == BASKER_TRUE)
{
FREE_INT_1DARRAY(order_match_array);
match_flag = BASKER_FALSE;
}
if(btf_flag == BASKER_TRUE)
{
FREE_INT_1DARRAY(order_btf_array);
btf_flag = BASKER_FALSE;
}
if(nd_flag == BASKER_TRUE)
{
FREE_INT_1DARRAY(order_scotch_array);
nd_flag = BASKER_FALSE;
}
if(amd_flag == BASKER_TRUE)
{
FREE_INT_1DARRAY(order_csym_array);
amd_flag = BASKER_FALSE;
}
//NDE: Free workspace and permutation arrays
FREE_INT_1DARRAY(perm_comp_array);
FREE_INT_1DARRAY(perm_inv_comp_array);
FREE_INT_1DARRAY(perm_comp_iworkspace_array);
FREE_ENTRY_1DARRAY(perm_comp_fworkspace_array);
FREE_ENTRY_1DARRAY(x_view_ptr_copy);
FREE_ENTRY_1DARRAY(y_view_ptr_copy);
//Structures
part_tree.Finalize();
tree.Finalize();
stree.Finalize();
stats.Finalize();
}//end Finalize()
BASKER_INLINE
int Basker<Int,Entry,Exe_Space>::solve_interface
(
Entry *_x, //Solution (len = gn)
Entry *_y
)
{
//Need to modify to use global perm
INT_1DARRAY temp_array;
MALLOC_INT_1DARRAY(temp_array, gn);
//===== Move to view===========
ENTRY_1DARRAY x;
ENTRY_1DARRAY y;
MALLOC_ENTRY_1DARRAY(x, gn);
MALLOC_ENTRY_1DARRAY(y, gm);
for(Int i =0; i < gn; i++)
{
x(i) = (Entry) 0;
y(i) = (Entry) _y[i];
}
//printf("RHS: \n");
//printVec(y, gn);
//printf("\n");
//===== Permute
//printf("Permute RHS\n");
//==== Need to make this into one global perm
if(match_flag == BASKER_TRUE)
{
//printf("match order\n");
//printVec("match.txt", order_match_array, gn);
permute_inv(y,order_match_array, gn);
}
if(btf_flag == BASKER_TRUE)
{
//printf("btf order\n");
//printVec("btf.txt", order_btf_array, gn);
permute_inv(y,order_btf_array, gn);
//printVec("btf_amd.txt", order_c_csym_array, gn);
permute_inv(y,order_blk_amd_array, gn);
}
if(nd_flag == BASKER_TRUE)
{
//printf("ND order \n");
//printVec("nd.txt", part_tree.permtab, gn);
for(Int i = 0; i < BTF_A.ncol; ++i)
{
temp_array(i) = part_tree.permtab(i);
}
for(Int i = BTF_A.ncol; i < gn; ++i)
{
temp_array(i) = i;
}
//permute_inv(y,part_tree.permtab, gn);
permute_inv(y, temp_array,gn);
}
if(amd_flag == BASKER_TRUE)
{
//printf("AMD order \n");
//printVec("amd.txt",order_csym_array, gn);
for(Int i = 0; i < BTF_A.ncol; ++i)
{
temp_array(i) = order_csym_array(i);
}
for(Int i = BTF_A.ncol; i < gn; ++i)
{
temp_array(i) = i;
}
//permute_inv(y,order_csym_array, gn);
permute_inv(y,temp_array, gn);
}
//printVec("perm.txt" , gperm, gn);
permute_inv(y,gperm, gn);
solve_interface(x,y);
//Inverse perm
//Note: don't need to inverse a row only perm
if(btf_flag == BASKER_TRUE)
{
//printf("btf order\n");
//printVec(order_btf_array, gn);
permute(x,order_btf_array, gn);
}
if(nd_flag == BASKER_TRUE)
{
//printf("ND order \n");
//printVec(part_tree.permtab, gn);
for(Int i = 0; i < BTF_A.ncol; ++i)
{
temp_array(i) = part_tree.permtab(i);
}
for(Int i = BTF_A.ncol; i < gn; i++)
{
temp_array(i) = i;
}
//permute(x,part_tree.permtab, gn);
permute(x,temp_array, gn);
}
if(amd_flag == BASKER_TRUE)
{
//printf("AMD order \n");
//printVec(order_csym_array, gn);
for(Int i = 0; i < BTF_A.ncol; ++i)
{
temp_array(i) = order_csym_array(i);
}
for(Int i = BTF_A.ncol; i < gn; ++i)
{
temp_array(i) = order_csym_array(i);
}
//permute(x,order_csym_array, gn);
permute(x,temp_array,gn);
}
#ifdef BASKER_DEBUG_SOLVE_RHS
printf("\n\n");
printf("X: \n");
for(Int i = 0; i < gn; i++)
{
printf("%f, " , x(i));
}
printf("\n\n");
printf("RHS: \n");
for(Int i =0; i < gm; i++)
{
printf("%f, ", y(i));
}
printf("\n\n");
#endif
for(Int i = 0; i < gn; i++)
{
_x[i] = x(i);
}
FREE_ENTRY_1DARRAY(x);
FREE_ENTRY_1DARRAY(y);
FREE_INT_1DARRAY(temp_array);
return 0;
}