Plan 9 from Bell Labs’s /usr/web/sources/contrib/fernan/nhc98/tests/nofib/real/fluid/Chl_decomp.hs
{- The second module of Choleski decomposition. Contains Choleski factorization function. XZ, 24/10/91 -} {- Modified to employ S_array. The decomposition is now coded in the column-Choleski feshion. It was in row-Choleski feshion. It might be benifitial if numerical evaluation is forced (access locality). XZ, 7/2/92 -} module Chl_decomp ( chl_factor ) where import Defs import S_Array -- not needed w/ proper module handling import Norm -- ditto import Asb_routs import Ix--1.3 infix 1 =: (=:) a b = (a,b) ----------------------------------------------------------- -- Choleski factorization: -- -- it adopts the matrix struct derived from Liu's so -- -- called generalized envelope method. Called at the -- -- data setup stage. -- ----------------------------------------------------------- cal_one_elem (j_off,v_off) (_,(j_s,v_s)) old_v = case v_s of [] -> old_v _ -> old_v - list_inner_prod (drop (max_j-j_s) v_s) (drop (max_j-j_off) v_off) where max_j = max j_s j_off cal_one_seg segs ((_,v_old),off@(j_off,v_off)) = zipWith (cal_one_elem off) segs v_old match_pairs a@((k1,l1):as) b@(b1@(k2,l2):bs) | k1<k2 = match_pairs as b | k2<k1 = match_pairs a bs | otherwise = (k1=:(l1,l2)):match_pairs as bs match_pairs _ _ = [] chl_factor :: S_array (S_array Float,S_array (Int,[Float])) -> S_array (S_array Float,S_array (Int,[Float])) chl_factor init_L = foldl f init_L (range (s_bounds init_L)) where f old_l j = old_l //^ [j=:step2 step1] where (l,u) = block_bnds block_bnds = s_bounds (fst this_block) this_block = (old_l!^j) bindTo x k = k x -- old Haskell 1.0 "let" step1 = foldl step1_f this_block -- previous_blocks... [ k=: ((snd (old_l!^k)) `bindTo` ( \ b -> filter (\ (i,_) -> (l<=i)&&(i<=u)) (s_assocs b) )) | k <- range (1,j-1) ] where step1_f (old_diag,old_off_diag) (k,segs_jk) = (new_diag,new_off_diag) where subs_segs ((i,pair@((j1,_),_)):rest) = (i=:(j1,cal_one_seg (drop (j1-l) segs_jk) pair)): subs_segs rest subs_segs _ = [] new_diag = s_accum (-) old_diag [ i =: list_inner_prod vs vs | (i,(_,vs@(_:_))) <- segs_jk ] new_off_diag = old_off_diag //^ ( subs_segs ( match_pairs (sparse_assocs old_off_diag) (sparse_assocs (snd (old_l!^k))) ) ) step2 (old_diag,old_off_diag) = ( s_listArray block_bnds (fst ass_res), old_off_diag //^ (tail (snd ass_res)) ) where ass_res = gen_assocs (sparse_assocs old_off_diag) ([sqrt (old_diag!^l)],[l=:(l+1,[])]) gen_assocs (old_line@(i,(j,vs)):rest) (t_diag,t_off_diag) = if i <= u then gen_assocs rest (t_diag++[new_diag],t_off_diag++[i=:(j,new_off_diag)]) else gen_assocs rest (t_diag,t_off_diag++[i=:(j,new_off_diag)]) where new_diag = sqrt ( (old_diag!^i) - list_inner_prod new_off_diag new_off_diag) new_off_diag = do_recur vs (drop (j-l) t_diag,drop (j-l) t_off_diag) [] do_recur (o_v:o_v_r) ((d:d_r),(off:off_r)) res = do_recur o_v_r (d_r,off_r) ( res ++ [ (cal_one_elem (j,res) off o_v)/d ] ) do_recur _ _ res = res gen_assocs _ res = res