Skip to content

Commit 337b6f5

Browse files
robertmhaasrobertmhaas
committed
Speed up in-memory tuplesorting.
Per recent work by Peter Geoghegan, it's significantly faster to tuplesort on a single sortkey if ApplySortComparator is inlined into quicksort rather reached via a function pointer. It's also faster in general to have a version of quicksort which is specialized for sorting SortTuple objects rather than objects of arbitrary size and type. This requires a couple of additional copies of the quicksort logic, which in this patch are generate using a Perl script. There might be some benefit in adding further specializations here too, but thus far it's not clear that those gains are worth their weight in code footprint.
1 parent ac9100f commit 337b6f5

File tree

8 files changed

+289
-36
lines changed

8 files changed

+289
-36
lines changed

‎src/backend/Makefile

Lines changed: 3 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -202,6 +202,7 @@ distprep:
202202
$(MAKE) -C replication repl_gram.c repl_scanner.c
203203
$(MAKE) -C utils fmgrtab.c fmgroids.h errcodes.h
204204
$(MAKE) -C utils/misc guc-file.c
205+
$(MAKE) -C utils/sort qsort_tuple.c
205206

206207

207208
##########################################################################
@@ -315,7 +316,8 @@ maintainer-clean: distclean
315316
utils/fmgroids.h \
316317
utils/fmgrtab.c \
317318
utils/errcodes.h \
318-
utils/misc/guc-file.c
319+
utils/misc/guc-file.c \
320+
utils/misc/qsort_tuple.c
319321

320322

321323
##########################################################################

‎src/backend/utils/sort/.gitignore

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1 @@
1+
/qsort_tuple.c

‎src/backend/utils/sort/Makefile

Lines changed: 8 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -14,4 +14,12 @@ include $(top_builddir)/src/Makefile.global
1414

1515
OBJS = logtape.o sortsupport.o tuplesort.o tuplestore.o
1616

17+
tuplesort.o: qsort_tuple.c
18+
19+
qsort_tuple.c: gen_qsort_tuple.pl
20+
$(PERL) $(srcdir)/gen_qsort_tuple.pl $< > $@
21+
1722
include $(top_srcdir)/src/backend/common.mk
23+
24+
maintainer-clean:
25+
rm -f qsort_tuple.c

‎src/backend/utils/sort/gen_qsort_tuple.pl

Lines changed: 232 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,232 @@
1+
#!/usr/bin/perl -w
2+
3+
#
4+
# gen_qsort_tuple.pl
5+
#
6+
# This script generates specialized versions of the quicksort algorithm for
7+
# tuple sorting. The quicksort code is derived from the NetBSD code. The
8+
# code generated by this script runs significantly faster than vanilla qsort
9+
# when used to sort tuples. This speedup comes from a number of places.
10+
# The major effects are (1) inlining simple tuple comparators is much faster
11+
# than jumping through a function pointer and (2) swap and vecswap operations
12+
# specialized to the particular data type of interest (in this case, SortTuple)
13+
# are faster than the generic routines.
14+
#
15+
# Modifications from vanilla NetBSD source:
16+
# Add do ... while() macro fix
17+
# Remove __inline, _DIAGASSERTs, __P
18+
# Remove ill-considered "swap_cnt" switch to insertion sort,
19+
# in favor of a simple check for presorted input.
20+
# Instead of sorting arbitrary objects, we're always sorting SortTuples
21+
# Add CHECK_FOR_INTERRUPTS()
22+
#
23+
# CAUTION: if you change this file, see also qsort.c and qsort_arg.c
24+
#
25+
26+
use strict;
27+
28+
my $SUFFIX;
29+
my $EXTRAARGS;
30+
my $EXTRAPARAMS;
31+
my $CMPPARAMS;
32+
33+
emit_qsort_boilerplate();
34+
35+
$SUFFIX = 'tuple';
36+
$EXTRAARGS = ', SortTupleComparator cmp_tuple, Tuplesortstate *state';
37+
$EXTRAPARAMS = ', cmp_tuple, state';
38+
$CMPPARAMS = ', state';
39+
emit_qsort_implementation();
40+
41+
$SUFFIX = 'ssup';
42+
$EXTRAARGS = ', SortSupport ssup';
43+
$EXTRAPARAMS = ', ssup';
44+
$CMPPARAMS = ', ssup';
45+
print <<'EOM';
46+
#define cmp_ssup(a, b, ssup) \
47+
ApplySortComparator((a)->datum1, (a)->isnull1, \
48+
(b)->datum1, (b)->isnull1, ssup)
49+
EOM
50+
emit_qsort_implementation();
51+
52+
sub emit_qsort_boilerplate
53+
{
54+
print <<'EOM';
55+
/*
56+
* autogenerated by src/backend/utils/sort/gen_qsort_tuple.pl, do not edit
57+
* This file is included by tuplesort.c, rather than compiled separately.
58+
*/
59+
60+
/* $NetBSD: qsort.c,v 1.13 2003/08/07 16:43:42 agc Exp $ */
61+
62+
/*-
63+
* Copyright (c) 1992, 1993
64+
* The Regents of the University of California. All rights reserved.
65+
*
66+
* Redistribution and use in source and binary forms, with or without
67+
* modification, are permitted provided that the following conditions
68+
* are met:
69+
* 1. Redistributions of source code must retain the above copyright
70+
* notice, this list of conditions and the following disclaimer.
71+
* 2. Redistributions in binary form must reproduce the above copyright
72+
* notice, this list of conditions and the following disclaimer in the
73+
* documentation and/or other materials provided with the distribution.
74+
* 3. Neither the name of the University nor the names of its contributors
75+
* may be used to endorse or promote products derived from this software
76+
* without specific prior written permission.
77+
*
78+
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
79+
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
80+
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
81+
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
82+
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
83+
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
84+
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
85+
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
86+
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
87+
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
88+
* SUCH DAMAGE.
89+
*/
90+
91+
/*
92+
* Qsort routine based on J. L. Bentley and M. D. McIlroy,
93+
* "Engineering a sort function",
94+
* Software--Practice and Experience 23 (1993) 1249-1265.
95+
* We have modified their original by adding a check for already-sorted input,
96+
* which seems to be a win per discussions on pgsql-hackers around 2006-03-21.
97+
*/
98+
99+
static void
100+
swapfunc(SortTuple *a, SortTuple *b, size_t n)
101+
{
102+
do
103+
{
104+
SortTuple t = *a;
105+
*a++ = *b;
106+
*b++ = t;
107+
} while (--n > 0);
108+
}
109+
110+
#define swap(a, b) \
111+
do { \
112+
SortTuple t = *(a); \
113+
*(a) = *(b); \
114+
*(b) = t; \
115+
} while (0);
116+
117+
#define vecswap(a, b, n) if ((n) > 0) swapfunc((a), (b), (size_t)(n))
118+
EOM
119+
}
120+
121+
sub emit_qsort_implementation
122+
{
123+
print <<EOM;
124+
static SortTuple *
125+
med3_$SUFFIX(SortTuple *a, SortTuple *b, SortTuple *c$EXTRAARGS)
126+
{
127+
return cmp_$SUFFIX(a, b$CMPPARAMS) < 0 ?
128+
(cmp_$SUFFIX(b, c$CMPPARAMS) < 0 ? b :
129+
(cmp_$SUFFIX(a, c$CMPPARAMS) < 0 ? c : a))
130+
: (cmp_$SUFFIX(b, c$CMPPARAMS) > 0 ? b :
131+
(cmp_$SUFFIX(a, c$CMPPARAMS) < 0 ? a : c));
132+
}
133+
134+
static void
135+
qsort_$SUFFIX(SortTuple *a, size_t n$EXTRAARGS)
136+
{
137+
SortTuple *pa,
138+
*pb,
139+
*pc,
140+
*pd,
141+
*pl,
142+
*pm,
143+
*pn;
144+
int d,
145+
r,
146+
presorted;
147+
148+
loop:
149+
CHECK_FOR_INTERRUPTS();
150+
if (n < 7)
151+
{
152+
for (pm = a + 1; pm < a + n; pm++)
153+
for (pl = pm; pl > a && cmp_$SUFFIX(pl - 1, pl$CMPPARAMS) > 0; pl--)
154+
swap(pl, pl - 1);
155+
return;
156+
}
157+
presorted = 1;
158+
for (pm = a + 1; pm < a + n; pm++)
159+
{
160+
CHECK_FOR_INTERRUPTS();
161+
if (cmp_$SUFFIX(pm - 1, pm$CMPPARAMS) > 0)
162+
{
163+
presorted = 0;
164+
break;
165+
}
166+
}
167+
if (presorted)
168+
return;
169+
pm = a + (n / 2);
170+
if (n > 7)
171+
{
172+
pl = a;
173+
pn = a + (n - 1);
174+
if (n > 40)
175+
{
176+
d = (n / 8);
177+
pl = med3_$SUFFIX(pl, pl + d, pl + 2 * d$EXTRAPARAMS);
178+
pm = med3_$SUFFIX(pm - d, pm, pm + d$EXTRAPARAMS);
179+
pn = med3_$SUFFIX(pn - 2 * d, pn - d, pn$EXTRAPARAMS);
180+
}
181+
pm = med3_$SUFFIX(pl, pm, pn$EXTRAPARAMS);
182+
}
183+
swap(a, pm);
184+
pa = pb = a + 1;
185+
pc = pd = a + (n - 1);
186+
for (;;)
187+
{
188+
while (pb <= pc && (r = cmp_$SUFFIX(pb, a$CMPPARAMS)) <= 0)
189+
{
190+
CHECK_FOR_INTERRUPTS();
191+
if (r == 0)
192+
{
193+
swap(pa, pb);
194+
pa++;
195+
}
196+
pb++;
197+
}
198+
while (pb <= pc && (r = cmp_$SUFFIX(pc, a$CMPPARAMS)) >= 0)
199+
{
200+
CHECK_FOR_INTERRUPTS();
201+
if (r == 0)
202+
{
203+
swap(pc, pd);
204+
pd--;
205+
}
206+
pc--;
207+
}
208+
if (pb > pc)
209+
break;
210+
swap(pb, pc);
211+
pb++;
212+
pc--;
213+
}
214+
pn = a + n;
215+
r = Min(pa - a, pb - pa);
216+
vecswap(a, pb - r, r);
217+
r = Min(pd - pc, pn - pd - 1);
218+
vecswap(pb, pn - r, r);
219+
if ((r = pb - pa) > 1)
220+
qsort_$SUFFIX(a, r$EXTRAPARAMS);
221+
if ((r = pd - pc) > 1)
222+
{
223+
/* Iterate rather than recurse to save stack space */
224+
a = pn - r;
225+
n = r;
226+
goto loop;
227+
}
228+
/* qsort_$SUFFIX(pn - r, r$EXTRAPARAMS);*/
229+
}
230+
231+
EOM
232+
}

0 commit comments

Comments
 (0)
pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy