Changeset: 43e4c7e822a5 for MonetDB
URL: https://dev.monetdb.org/hg/MonetDB?cmd=changeset;node=43e4c7e822a5
Modified Files:
sql/backends/monet5/rel_weld.c
Branch: rel-weld
Log Message:
join: if the right side has 0 or 1 columns materialize it directly in the
dict/group merger
diffs (205 lines):
diff --git a/sql/backends/monet5/rel_weld.c b/sql/backends/monet5/rel_weld.c
--- a/sql/backends/monet5/rel_weld.c
+++ b/sql/backends/monet5/rel_weld.c
@@ -1432,102 +1432,128 @@ join_produce(backend *be, sql_rel *rel,
wprintf(wstate, "let v%dvecs = (", result_var);
wstate->num_parens++;
- /* {appender[{..}], appender[..], appender[..] ... } */
+ /* Build the key in probe_key */
len = 0;
- len += sprintf(new_builder + len, "{appender[");
if (list_length(rel->exps) > 1) {
- len += sprintf(new_builder + len, "{"); /* key is a struct */
+ len += sprintf(probe_key + len, "{"); /* key is a struct */
}
for (en = rel->exps->h; en; en = en->next) {
/* left cmp */
exp = ((sql_exp*)en->data)->l;
/* both have the same type */
int type = exp_subtype(exp)->type->localtype;
- len += sprintf(new_builder + len, "%s", getWeldType(type));
+ len += sprintf(probe_key + len, "%s", getWeldType(type));
if (en->next != NULL) {
- len += sprintf(new_builder + len, ", ");
+ len += sprintf(probe_key + len, ", ");
}
}
if (list_length(rel->exps) > 1) {
- len += sprintf(new_builder + len, "}"); /* key is a struct */
+ len += sprintf(probe_key + len, "}"); /* key is a struct */
}
- len += sprintf(new_builder + len, "]");
- /* Create a list of right columns that are not part of the join key */
- nulls_len = 0;
- nulls[0] = 0;
- for (en = right_cols->h, count = 1; en; en = en->next) {
- exp = en->data;
- int type = exp_subtype(exp)->type->localtype;
- if (type == TYPE_str) {
- type = TYPE_lng;
+ if (is_unique && list_length(right_cols) == 0) {
+ sprintf(new_builder, "dictmerger[%s, i64, +]", probe_key);
+ } else if (!is_unique && list_length(right_cols) == 1) {
+ sprintf(new_builder, "groupmerger[%s, ?]", probe_key);
+ } else {
+ /* {appender[{..}], appender[..], appender[..] ... } */
+ len = 0;
+ len += sprintf(new_builder + len, "{appender[%s]", probe_key);
+ /* Create a list of right columns that are not part of the join
key */
+ nulls_len = 0;
+ nulls[0] = 0;
+ for (en = right_cols->h, count = 1; en; en = en->next) {
+ exp = en->data;
+ int type = exp_subtype(exp)->type->localtype;
+ if (type == TYPE_str) {
+ type = TYPE_lng;
+ }
+ len += sprintf(new_builder + len, ", appender[%s]",
getWeldType(type));
+ nulls_len += sprintf(nulls + nulls_len, ",
merge(v%dvecs.$%d, %snil)", result_var, count, getWeldType(type));
+ count++;
}
- len += sprintf(new_builder + len, ", appender[%s]",
getWeldType(type));
- nulls_len += sprintf(nulls + nulls_len, ", merge(v%dvecs.$%d,
%snil)", result_var, count, getWeldType(type));
- count++;
+ len += sprintf(new_builder + len, "}"); /* complete the builder
*/
}
- len += sprintf(new_builder + len, "}"); /* complete the builder */
wstate->builder = new_builder;
right_produce(be, rel->r, wstate);
/* === Consume === */
- wprintf(wstate, "{merge(b%d.$0, ", wstate->num_loops); /* {key, value}
*/
/* Build the key */
+ len = 0;
if (list_length(right_cmp_cols) > 1) {
- wprintf(wstate, "{"); /* key is a struct */
+ len += sprintf(probe_key + len, "{"); /* key is a struct */
}
for (en = right_cmp_cols->h; en; en = en->next) {
- wprintf(wstate, "%s", (str)en->data);
+ len += sprintf(probe_key + len, "%s", (str)en->data);
if (en->next != NULL) {
- wprintf(wstate, ", ");
+ len += sprintf(probe_key + len, ", ");
}
}
if (list_length(right_cmp_cols) > 1) {
- wprintf(wstate, "}"); /* key is a struct */
- }
- wprintf(wstate, "), ");
- /* Append the values */
- for (en = right_cols->h, count = 0; en; en = en->next, count++) {
- exp = en->data;
- wprintf(wstate, "merge(b%d.$%d, %s", wstate->num_loops, count +
1, (str)list_fetch(right_cols_names, count));
- if (exp_subtype(exp)->type->localtype == TYPE_str) {
- wprintf(wstate, "_stridx");
- }
- wprintf(wstate, ")");
- if (en->next != NULL) {
- wprintf(wstate, ", ");
- }
- }
- wprintf(wstate, "}");
- for (i = 0; i < wstate->num_parens; i++) {
- wprintf(wstate, ")");
- }
- wprintf(wstate, ";");
- /* Append the nulls at the end */
- wprintf(wstate, "let v%dvecs = {v%dvecs.$0 %s};", result_var,
result_var, nulls);
- /* Materialize the hashtable */
- wprintf(wstate, "let v%dkeys = result(v%dvecs.$0);", result_var,
result_var);
- for (en = right_cols->h, count = 0; en; en = en->next, count++) {
- wprintf(wstate, "let v%dvec%d = result(v%dvecs.$%d);",
result_var, count + 1, result_var, count + 1);
+ len += sprintf(probe_key + len, "}"); /* key is a struct */
}
- /* Build the hastable with indexes */
- if (is_unique) {
- wprintf(wstate,
- "let v%dtable = result("
- " for(v%dkeys, dictmerger[?, i64, +], |b, i, n|"
- " merge(b, {n, i})"
- " )"
- ");", result_var, result_var);
+ if (is_unique && list_length(right_cols) == 0) {
+ /* unique values and no right cols */
+ wprintf(wstate, "merge(b%d, {%s, 0L})", wstate->num_loops,
probe_key);
+ for (i = 0; i < wstate->num_parens; i++) {
+ wprintf(wstate, ")");
+ }
+ wprintf(wstate, ";");
+ wprintf(wstate, "let v%dtable = result(v%dvecs);", result_var,
result_var);
+ } else if (!is_unique && list_length(right_cols) == 1) {
+ /* groupmerger + a single right col */
+ wprintf(wstate, "merge(b%d, {%s, %s})", wstate->num_loops,
probe_key, (str)right_cols_names->h->data);
+ for (i = 0; i < wstate->num_parens; i++) {
+ wprintf(wstate, ")");
+ }
+ wprintf(wstate, ";");
+ wprintf(wstate, "let v%dtable = result(v%dvecs);", result_var,
result_var);
+ } else {
+ wprintf(wstate, "{merge(b%d.$0, %s), ", wstate->num_loops,
probe_key); /* {key, value} */
+ /* Append the values */
+ for (en = right_cols->h, count = 0; en; en = en->next, count++)
{
+ exp = en->data;
+ wprintf(wstate, "merge(b%d.$%d, %s", wstate->num_loops,
count + 1, (str)list_fetch(right_cols_names, count));
+ if (exp_subtype(exp)->type->localtype == TYPE_str) {
+ wprintf(wstate, "_stridx");
+ }
+ wprintf(wstate, ")");
+ if (en->next != NULL) {
+ wprintf(wstate, ", ");
+ }
+ }
+ wprintf(wstate, "}");
+ for (i = 0; i < wstate->num_parens; i++) {
+ wprintf(wstate, ")");
+ }
+ wprintf(wstate, ";");
+ /* Append the nulls at the end */
+ wprintf(wstate, "let v%dvecs = {v%dvecs.$0 %s};", result_var,
result_var, nulls);
+ /* Materialize the hashtable */
+ wprintf(wstate, "let v%dkeys = result(v%dvecs.$0);",
result_var, result_var);
+ for (en = right_cols->h, count = 0; en; en = en->next, count++)
{
+ wprintf(wstate, "let v%dvec%d = result(v%dvecs.$%d);",
result_var, count + 1, result_var, count + 1);
+ }
- } else {
- wprintf(wstate,
- "let v%dtable = result("
- " for(v%dkeys, groupmerger[?, i64], |b, i, n|"
- " merge(b, {n, i})"
- " )"
- ");", result_var, result_var);
+ /* Build the hastable with indexes */
+ if (is_unique) {
+ wprintf(wstate,
+ "let v%dtable = result("
+ " for(v%dkeys, dictmerger[?, i64, +], |b, i, n|"
+ " merge(b, {n, i})"
+ " )"
+ ");", result_var, result_var);
+
+ } else {
+ wprintf(wstate,
+ "let v%dtable = result("
+ " for(v%dkeys, groupmerger[?, i64], |b, i, n|"
+ " merge(b, {n, i})"
+ " )"
+ ");", result_var, result_var);
+ }
}
/* Resume the pipeline */
@@ -1574,6 +1600,8 @@ join_produce(backend *be, sql_rel *rel,
for (en = right_cols->h, count = 0; en; en = en->next, count++) {
if (is_unique) {
sprintf(value, "lookup(v%dvec%d, match)", result_var,
count + 1);
+ } else if (!is_unique && list_length(right_cols) == 1) {
+ sprintf(value, "n%d", wstate->num_loops);
} else {
sprintf(value, "lookup(v%dvec%d, n%d)", result_var,
count + 1, wstate->num_loops);
}
_______________________________________________
checkin-list mailing list
checkin-list@monetdb.org
https://www.monetdb.org/mailman/listinfo/checkin-list
