Changeset: 63affc95c402 for MonetDB
URL: https://dev.monetdb.org/hg/MonetDB/rev/63affc95c402
Modified Files:
sql/backends/monet5/rel_bin.c
sql/backends/monet5/sql_statement.c
sql/backends/monet5/sql_statement.h
Branch: pushcands
Log Message:
first attempts at pushing candidates into group and aggregates
diffs (truncated from 504 to 300 lines):
diff --git a/sql/backends/monet5/rel_bin.c b/sql/backends/monet5/rel_bin.c
--- a/sql/backends/monet5/rel_bin.c
+++ b/sql/backends/monet5/rel_bin.c
@@ -353,7 +353,7 @@ distinct_value_list(backend *be, list *v
s = stmt_append_bulk(be, stmt_temp(be, exp_subtype(vals->h->data)), l);
/* Probably faster to filter out the values directly in the underlying
list of atoms.
But for now use groupby to filter out duplicate values. */
- stmt* groupby = stmt_group(be, s, NULL, NULL, NULL, 1);
+ stmt* groupby = stmt_group(be, s, NULL, NULL, NULL, NULL, 1);
stmt* ext = stmt_result(be, groupby, 1);
return stmt_project(be, ext, s);
@@ -369,10 +369,13 @@ stmt_selectnonil( backend *be, stmt *col
}
static rel_bin_stmt *
-subrel_project( backend *be, rel_bin_stmt *s, list *refs, sql_rel *rel, bool
allow_single_cand)
+subrel_project( backend *be, rel_bin_stmt *s, list *refs, sql_rel *rel, bool
partial_push)
{
- if (!s || !s->cand || (s->cand && allow_single_cand && !s->pushed))
+ if (!s || !s->cand || (s->cand && partial_push && !s->pushed)) {
+ if (s->cand)
+ s->pushed = partial_push;
return s;
+ }
list *l = sa_list(be->mvc->sa);
stmt *cand = s->cand;
@@ -1374,7 +1377,7 @@ exp_bin(backend *be, sql_exp *e, rel_bin
as = exp_bin(be, at, left, right,
depth+1, 0, push);
if (as && as->nrcols <= 0 && left)
- as = stmt_const(be,
bin_find_smallest_column(be, left), NULL, as);
+ as = stmt_const(be,
bin_find_smallest_column(be, left), left->cand, as);
if (en == attr->h && !en->next &&
exp_aggr_is_count(e))
as = exp_count_no_nil_arg(e, right ?
right->ext : NULL, at, as);
/* insert single value into a column */
@@ -1392,7 +1395,7 @@ exp_bin(backend *be, sql_exp *e, rel_bin
stmt *ncnt = right ? right->cnt : NULL;
for (en = l->h; en; en = en->next) {
stmt *as = en->data;
- stmt *g = stmt_group(be, as, ngrp,
next, ncnt, 1);
+ stmt *g = stmt_group(be, as,
right->cand, ngrp, next, ncnt, 1);
ngrp = stmt_result(be, g, 0);
next = stmt_result(be, g, 1);
ncnt = stmt_result(be, g, 2);
@@ -1426,9 +1429,9 @@ exp_bin(backend *be, sql_exp *e, rel_bin
}
}
if (right)
- s = stmt_aggr(be, as, input_group, right->ext, a, 1,
need_no_nil(e) /* ignore nil*/, !zero_if_empty(e));
+ s = stmt_aggr(be, as, left->cand, input_group,
right->ext, a, 1, need_no_nil(e) /* ignore nil*/, !zero_if_empty(e));
else
- s = stmt_aggr(be, as, NULL, NULL, a, 1, need_no_nil(e)
/* ignore nil*/, !zero_if_empty(e));
+ s = stmt_aggr(be, as, left->cand, NULL, NULL, a, 1,
need_no_nil(e) /* ignore nil*/, !zero_if_empty(e));
if (find_prop(e->p, PROP_COUNT)) /* propagate count == 0 ipv
NULL in outer joins */
s->flag |= OUTER_ZERO;
} break;
@@ -3077,7 +3080,7 @@ rel2bin_distinct(backend *be, rel_bin_st
for (node *n = s->cols->h; n; n = n->next) {
stmt *t = n->data;
- g = stmt_group(be, column(be, t), grp, ext, cnt, !n->next);
+ g = stmt_group(be, column(be, t), NULL, grp, ext, cnt,
!n->next);
grp = stmt_result(be, g, 0);
ext = stmt_result(be, g, 1);
cnt = stmt_result(be, g, 2);
@@ -3116,7 +3119,7 @@ rel2bin_single(backend *be, rel_bin_stmt
const char *nme = column_name(sql->sa, t);
sql_subfunc *zero_or_one = sql_bind_func(sql, "sys",
"zero_or_one", tail_type(t), NULL, F_AGGR);
- t = stmt_aggr(be, t, NULL, NULL, zero_or_one, 1, 0, 1);
+ t = stmt_aggr(be, t, NULL, NULL, NULL, zero_or_one, 1, 0, 1);
t = stmt_alias(be, t, rnme, nme);
list_append(rl, t);
}
@@ -3220,13 +3223,13 @@ rel2bin_except(backend *be, sql_rel *rel
* number of rows.
*/
for (n = left->cols->h; n; n = n->next) {
- lg = stmt_group(be, column(be, n->data), lgrp, lext, lcnt,
!n->next);
+ lg = stmt_group(be, column(be, n->data), NULL, lgrp, lext,
lcnt, !n->next);
lgrp = stmt_result(be, lg, 0);
lext = stmt_result(be, lg, 1);
lcnt = stmt_result(be, lg, 2);
}
for (n = right->cols->h; n; n = n->next) {
- rg = stmt_group(be, column(be, n->data), rgrp, rext, rcnt,
!n->next);
+ rg = stmt_group(be, column(be, n->data), NULL, rgrp, rext,
rcnt, !n->next);
rgrp = stmt_result(be, rg, 0);
rext = stmt_result(be, rg, 1);
rcnt = stmt_result(be, rg, 2);
@@ -3330,13 +3333,13 @@ rel2bin_inter(backend *be, sql_rel *rel,
* number of rows.
*/
for (n = left->cols->h; n; n = n->next) {
- lg = stmt_group(be, column(be, n->data), lgrp, lext, lcnt,
!n->next);
+ lg = stmt_group(be, column(be, n->data), NULL, lgrp, lext,
lcnt, !n->next);
lgrp = stmt_result(be, lg, 0);
lext = stmt_result(be, lg, 1);
lcnt = stmt_result(be, lg, 2);
}
for (n = right->cols->h; n; n = n->next) {
- rg = stmt_group(be, column(be, n->data), rgrp, rext, rcnt,
!n->next);
+ rg = stmt_group(be, column(be, n->data), NULL, rgrp, rext,
rcnt, !n->next);
rgrp = stmt_result(be, rg, 0);
rext = stmt_result(be, rg, 1);
rcnt = stmt_result(be, rg, 2);
@@ -3508,10 +3511,8 @@ rel2bin_project(backend *be, sql_rel *re
}
if (sub && sub->cand && used < list_length(rel->exps)) {
psub->cand = sub->cand;
- if (used > 0) {
+ if (used > 0)
psub = subrel_project(be, psub, refs, rel, true);
- psub->pushed = true;
- }
} else if (sub && sub->cand && used == list_length(rel->exps)) {
sub = used_cands(sub);
psub = used_cands(psub);
@@ -3694,7 +3695,7 @@ rel2bin_groupby(backend *be, sql_rel *re
assert(rel->l); /* first construct the sub relation */
sub = subrel_bin(be, rel->l, refs);
- sub = subrel_project(be, sub, refs, rel->l, false);
+// sub = subrel_project(be, sub, refs, rel->l, false);
if (!sub)
return NULL;
@@ -3728,8 +3729,8 @@ rel2bin_groupby(backend *be, sql_rel *re
return NULL;
}
if (!gbcol->nrcols)
- gbcol = stmt_const(be,
bin_find_smallest_column(be, sub), NULL, gbcol);
- groupby = stmt_group(be, gbcol, grp, ext, cnt,
!en->next);
+ gbcol = stmt_const(be,
bin_find_smallest_column(be, sub), sub->cand, gbcol);
+ groupby = stmt_group(be, gbcol, sub->cand, grp, ext,
cnt, !en->next);
grp = stmt_result(be, groupby, 0);
ext = stmt_result(be, groupby, 1);
cnt = stmt_result(be, groupby, 2);
@@ -3960,7 +3961,7 @@ insert_check_ukey(backend *be, list *ins
s = releqjoin(be, lje, rje, NULL, 1 /* hash used */, 0,
0);
s = stmt_result(be, s, 0);
}
- s = stmt_binop(be, stmt_aggr(be, s, NULL, NULL, cnt, 1, 0, 1),
stmt_atom_lng(be, 0), NULL, ne);
+ s = stmt_binop(be, stmt_aggr(be, s, NULL, NULL, NULL, cnt, 1,
0, 1), stmt_atom_lng(be, 0), NULL, ne);
/* 2nd stage: find out if inserted are unique */
if ((!idx_inserts && ins->nrcols) || (idx_inserts &&
idx_inserts->nrcols)) { /* insert columns not atoms */
@@ -3998,7 +3999,7 @@ insert_check_ukey(backend *be, list *ins
return NULL;
sum = sql_bind_func(sql, "sys", "not_unique",
tail_type(orderby_grp), NULL, F_AGGR);
- ssum = stmt_aggr(be, orderby_grp, NULL, NULL, sum, 1,
0, 1);
+ ssum = stmt_aggr(be, orderby_grp, NULL, NULL, NULL,
sum, 1, 0, 1);
/* combine results */
s = stmt_binop(be, s, ssum, NULL, or);
}
@@ -4022,11 +4023,11 @@ insert_check_ukey(backend *be, list *ins
s = stmt_join(be, s, h, 0, cmp_equal, 0, 0, false);
/* s should be empty */
s = stmt_result(be, s, 0);
- s = stmt_aggr(be, s, NULL, NULL, cnt, 1, 0, 1);
+ s = stmt_aggr(be, s, NULL, NULL, NULL, cnt, 1, 0, 1);
} else {
s = stmt_uselect(be, s, h, cmp_equal, NULL, 0, 0);
/* s should be empty */
- s = stmt_aggr(be, s, NULL, NULL, cnt, 1, 0, 1);
+ s = stmt_aggr(be, s, NULL, NULL, NULL, cnt, 1, 0, 1);
}
/* s should be empty */
s = stmt_binop(be, s, stmt_atom_lng(be, 0), NULL, ne);
@@ -4046,13 +4047,13 @@ insert_check_ukey(backend *be, list *ins
ins = stmt_project(be, nn, ins);
}
- g = stmt_group(be, ins, NULL, NULL, NULL, 1);
+ g = stmt_group(be, ins, NULL, NULL, NULL, NULL, 1);
ss = stmt_result(be, g, 2); /* use count */
/* (count(ss) <> sum(ss)) */
sum = sql_bind_func(sql, "sys", "sum", lng, NULL,
F_AGGR);
- ssum = stmt_aggr(be, ss, NULL, NULL, sum, 1, 0, 1);
+ ssum = stmt_aggr(be, ss, NULL, NULL, NULL, sum, 1, 0,
1);
ssum = sql_Nop_(be, "ifthenelse", sql_unop_(be,
"isnull", ssum), stmt_atom_lng(be, 0), ssum, NULL);
- count_sum = stmt_binop(be, check_types(be,
tail_type(ssum), stmt_aggr(be, ss, NULL, NULL, cnt, 1, 0, 1), type_equal),
ssum, NULL, ne);
+ count_sum = stmt_binop(be, check_types(be,
tail_type(ssum), stmt_aggr(be, ss, NULL, NULL, NULL, cnt, 1, 0, 1),
type_equal), ssum, NULL, ne);
/* combine results */
s = stmt_binop(be, s, count_sum, NULL, or);
@@ -4081,10 +4082,10 @@ insert_check_fkey(backend *be, list *ins
if (pin && list_length(pin->cols))
s = pin->cols->h->data;
if (s->key && s->nrcols == 0) {
- s = stmt_binop(be, stmt_aggr(be, idx_inserts, NULL, NULL, cnt,
1, 0, 1), stmt_atom_lng(be, 1), NULL, ne);
+ s = stmt_binop(be, stmt_aggr(be, idx_inserts, NULL, NULL, NULL,
cnt, 1, 0, 1), stmt_atom_lng(be, 1), NULL, ne);
} else {
/* releqjoin.count <> inserts[col1].count */
- s = stmt_binop(be, stmt_aggr(be, idx_inserts, NULL, NULL, cnt,
1, 0, 1), stmt_aggr(be, column(be, s), NULL, NULL, cnt, 1, 0, 1), NULL, ne);
+ s = stmt_binop(be, stmt_aggr(be, idx_inserts, NULL, NULL, NULL,
cnt, 1, 0, 1), stmt_aggr(be, column(be, s), NULL, NULL, NULL, cnt, 1, 0, 1),
NULL, ne);
}
/* s should be empty */
@@ -4193,7 +4194,7 @@ sql_insert_check_null(backend *be, sql_t
if (!(s->key && s->nrcols == 0)) {
s = stmt_selectnil(be, column(be, i));
- s = stmt_aggr(be, s, NULL, NULL, cnt, 1, 0, 1);
+ s = stmt_aggr(be, s, NULL, NULL, NULL, cnt, 1,
0, 1);
} else {
sql_subfunc *isnil = sql_bind_func(sql, "sys",
"isnull", &c->type, NULL, F_FUNC);
@@ -4270,7 +4271,7 @@ rel2bin_insert(backend *be, sql_rel *rel
if (ret->nrcols == 0) {
ret = stmt_atom_lng(be, 1);
} else {
- ret = stmt_aggr(be, ret, NULL, NULL, sql_bind_func(sql, "sys",
"count", sql_bind_localtype("void"), NULL, F_AGGR), 1, 0, 1);
+ ret = stmt_aggr(be, ret, NULL, NULL, NULL, sql_bind_func(sql,
"sys", "count", sql_bind_localtype("void"), NULL, F_AGGR), 1, 0, 1);
}
if (t->idxs) {
@@ -4405,7 +4406,7 @@ update_check_ukey(backend *be, stmt **up
}
s = releqjoin(be, lje, rje, NULL, 1 /* hash used */, 0,
0);
s = stmt_result(be, s, 0);
- s = stmt_binop(be, stmt_aggr(be, s, NULL, NULL, cnt, 1,
0, 1), stmt_atom_lng(be, 0), NULL, ne);
+ s = stmt_binop(be, stmt_aggr(be, s, NULL, NULL, NULL,
cnt, 1, 0, 1), stmt_atom_lng(be, 0), NULL, ne);
}
/* 2e stage: find out if the updated are unique */
@@ -4420,7 +4421,7 @@ update_check_ukey(backend *be, stmt **up
/* also take the hopefully unique hash keys, to reduce
(re)group costs */
if (k->idx && hash_index(k->idx->type)) {
- g = stmt_group(be, idx_updates, grp, ext, Cnt,
0);
+ g = stmt_group(be, idx_updates, NULL, grp, ext,
Cnt, 0);
grp = stmt_result(be, g, 0);
ext = stmt_result(be, g, 1);
Cnt = stmt_result(be, g, 2);
@@ -4463,7 +4464,7 @@ update_check_ukey(backend *be, stmt **up
}
/* apply group by on groups with Cnt > 1 */
- g = stmt_group(be, upd, grp, ext, Cnt,
!m->next);
+ g = stmt_group(be, upd, NULL, grp, ext, Cnt,
!m->next);
grp = stmt_result(be, g, 0);
ext = stmt_result(be, g, 1);
Cnt = stmt_result(be, g, 2);
@@ -4471,9 +4472,9 @@ update_check_ukey(backend *be, stmt **up
ss = Cnt; /* use count */
/* (count(ss) <> sum(ss)) */
sum = sql_bind_func(sql, "sys", "sum", lng, NULL,
F_AGGR);
- ssum = stmt_aggr(be, ss, NULL, NULL, sum, 1, 0, 1);
+ ssum = stmt_aggr(be, ss, NULL, NULL, NULL, sum, 1, 0,
1);
ssum = sql_Nop_(be, "ifthenelse", sql_unop_(be,
"isnull", ssum), stmt_atom_lng(be, 0), ssum, NULL);
- count_sum = stmt_binop(be, stmt_aggr(be, ss, NULL,
NULL, cnt, 1, 0, 1), check_types(be, lng, ssum, type_equal), NULL, ne);
+ count_sum = stmt_binop(be, stmt_aggr(be, ss, NULL,
NULL, NULL, cnt, 1, 0, 1), check_types(be, lng, ssum, type_equal), NULL, ne);
/* combine results */
if (s)
@@ -4502,7 +4503,7 @@ update_check_ukey(backend *be, stmt **up
o = stmt_col(be, c->c, nu_tids, nu_tids->partition);
s = stmt_join(be, o, h, 0, cmp_equal, 0, 0, false);
s = stmt_result(be, s, 0);
- s = stmt_binop(be, stmt_aggr(be, s, NULL, NULL, cnt, 1,
0, 1), stmt_atom_lng(be, 0), NULL, ne);
+ s = stmt_binop(be, stmt_aggr(be, s, NULL, NULL, NULL,
cnt, 1, 0, 1), stmt_atom_lng(be, 0), NULL, ne);
}
/* 2e stage: find out if updated are unique */
@@ -4526,14 +4527,14 @@ update_check_ukey(backend *be, stmt **up
upd = stmt_project(be, nn, upd);
}
- g = stmt_group(be, upd, NULL, NULL, NULL, 1);
+ g = stmt_group(be, upd, NULL, NULL, NULL, NULL, 1);
ss = stmt_result(be, g, 2); /* use count */
/* (count(ss) <> sum(ss)) */
sum = sql_bind_func(sql, "sys", "sum", lng, NULL,
F_AGGR);
- ssum = stmt_aggr(be, ss, NULL, NULL, sum, 1, 0, 1);
+ ssum = stmt_aggr(be, ss, NULL, NULL, NULL, sum, 1, 0,
1);
ssum = sql_Nop_(be, "ifthenelse", sql_unop_(be,
"isnull", ssum), stmt_atom_lng(be, 0), ssum, NULL);
- count_sum = stmt_binop(be, check_types(be,
tail_type(ssum), stmt_aggr(be, ss, NULL, NULL, cnt, 1, 0, 1), type_equal),
ssum, NULL, ne);
+ count_sum = stmt_binop(be, check_types(be,
tail_type(ssum), stmt_aggr(be, ss, NULL, NULL, NULL, cnt, 1, 0, 1),
type_equal), ssum, NULL, ne);
/* combine results */
if (s)
@@ -4604,7 +4605,7 @@ update_check_fkey(backend *be, stmt **up
assert(0);
cur = stmt_col(be, c->c, dels, dels->partition);
}
- s = stmt_binop(be, stmt_aggr(be, idx_updates, NULL, NULL, cnt, 1, 0,
1), stmt_aggr(be, cur, NULL, NULL, cnt, 1, 0, 1), NULL, ne);
+ s = stmt_binop(be, stmt_aggr(be, idx_updates, NULL, NULL, NULL, cnt, 1,
0, 1), stmt_aggr(be, cur, NULL, NULL, NULL, cnt, 1, 0, 1), NULL, ne);
for (m = k->columns->h; m; m = m->next) {
sql_kc *c = m->data;
@@ -4627,12 +4628,12 @@ update_check_fkey(backend *be, stmt **up
}
}
if (null) {
- cntnulls = stmt_aggr(be, null, NULL, NULL, cnt, 1, 0, 1);
_______________________________________________
checkin-list mailing list
[email protected]
https://www.monetdb.org/mailman/listinfo/checkin-list
