Skip to content

Commit af025ee

Browse files
robertmhaasrobertmhaas
committed
Add combine functions for various floating-point aggregates.
This allows parallel aggregation to use them. It may seem surprising that we use float8_combine for both float4_accum and float8_accum transition functions, but that's because those functions differ only in the type of the non-transition-state argument. Haribabu Kommi, reviewed by David Rowley and Tomas Vondra
1 parent 38627f6 commit af025ee

File tree

5 files changed

+159
-46
lines changed

5 files changed

+159
-46
lines changed

‎doc/src/sgml/func.sgml‎

Lines changed: 20 additions & 20 deletions
Original file line numberDiff line numberDiff line change
@@ -12743,7 +12743,7 @@ NULL baz</literallayout>(3 rows)</entry>
1274312743
<type>double precision</type> for a floating-point argument,
1274412744
otherwise the same as the argument data type
1274512745
</entry>
12746-
<entry>All types apart from floating-point types</entry>
12746+
<entry>Yes</entry>
1274712747
<entry>the average (arithmetic mean) of all input values</entry>
1274812748
</row>
1274912749

@@ -13001,7 +13001,7 @@ NULL baz</literallayout>(3 rows)</entry>
1300113001
<type>bigint</type> arguments, otherwise the same as the
1300213002
argument data type
1300313003
</entry>
13004-
<entry>All types apart from floating-point types</entry>
13004+
<entry>Yes</entry>
1300513005
<entry>sum of <replaceable class="parameter">expression</replaceable> across all input values</entry>
1300613006
</row>
1300713007

@@ -13153,7 +13153,7 @@ SELECT xmlagg(x) FROM (SELECT x FROM test ORDER BY y DESC) AS tab;
1315313153
<entry>
1315413154
<type>double precision</type>
1315513155
</entry>
13156-
<entry>No</entry>
13156+
<entry>Yes</entry>
1315713157
<entry>correlation coefficient</entry>
1315813158
</row>
1315913159

@@ -13174,7 +13174,7 @@ SELECT xmlagg(x) FROM (SELECT x FROM test ORDER BY y DESC) AS tab;
1317413174
<entry>
1317513175
<type>double precision</type>
1317613176
</entry>
13177-
<entry>No</entry>
13177+
<entry>Yes</entry>
1317813178
<entry>population covariance</entry>
1317913179
</row>
1318013180

@@ -13195,7 +13195,7 @@ SELECT xmlagg(x) FROM (SELECT x FROM test ORDER BY y DESC) AS tab;
1319513195
<entry>
1319613196
<type>double precision</type>
1319713197
</entry>
13198-
<entry>No</entry>
13198+
<entry>Yes</entry>
1319913199
<entry>sample covariance</entry>
1320013200
</row>
1320113201

@@ -13212,7 +13212,7 @@ SELECT xmlagg(x) FROM (SELECT x FROM test ORDER BY y DESC) AS tab;
1321213212
<entry>
1321313213
<type>double precision</type>
1321413214
</entry>
13215-
<entry>No</entry>
13215+
<entry>Yes</entry>
1321613216
<entry>average of the independent variable
1321713217
(<literal>sum(<replaceable class="parameter">X</replaceable>)/<replaceable class="parameter">N</replaceable></literal>)</entry>
1321813218
</row>
@@ -13230,7 +13230,7 @@ SELECT xmlagg(x) FROM (SELECT x FROM test ORDER BY y DESC) AS tab;
1323013230
<entry>
1323113231
<type>double precision</type>
1323213232
</entry>
13233-
<entry>No</entry>
13233+
<entry>Yes</entry>
1323413234
<entry>average of the dependent variable
1323513235
(<literal>sum(<replaceable class="parameter">Y</replaceable>)/<replaceable class="parameter">N</replaceable></literal>)</entry>
1323613236
</row>
@@ -13248,7 +13248,7 @@ SELECT xmlagg(x) FROM (SELECT x FROM test ORDER BY y DESC) AS tab;
1324813248
<entry>
1324913249
<type>bigint</type>
1325013250
</entry>
13251-
<entry>No</entry>
13251+
<entry>Yes</entry>
1325213252
<entry>number of input rows in which both expressions are nonnull</entry>
1325313253
</row>
1325413254

@@ -13268,7 +13268,7 @@ SELECT xmlagg(x) FROM (SELECT x FROM test ORDER BY y DESC) AS tab;
1326813268
<entry>
1326913269
<type>double precision</type>
1327013270
</entry>
13271-
<entry>No</entry>
13271+
<entry>Yes</entry>
1327213272
<entry>y-intercept of the least-squares-fit linear equation
1327313273
determined by the (<replaceable
1327413274
class="parameter">X</replaceable>, <replaceable
@@ -13288,7 +13288,7 @@ SELECT xmlagg(x) FROM (SELECT x FROM test ORDER BY y DESC) AS tab;
1328813288
<entry>
1328913289
<type>double precision</type>
1329013290
</entry>
13291-
<entry>No</entry>
13291+
<entry>Yes</entry>
1329213292
<entry>square of the correlation coefficient</entry>
1329313293
</row>
1329413294

@@ -13308,7 +13308,7 @@ SELECT xmlagg(x) FROM (SELECT x FROM test ORDER BY y DESC) AS tab;
1330813308
<entry>
1330913309
<type>double precision</type>
1331013310
</entry>
13311-
<entry>No</entry>
13311+
<entry>Yes</entry>
1331213312
<entry>slope of the least-squares-fit linear equation determined
1331313313
by the (<replaceable class="parameter">X</replaceable>,
1331413314
<replaceable class="parameter">Y</replaceable>) pairs</entry>
@@ -13327,7 +13327,7 @@ SELECT xmlagg(x) FROM (SELECT x FROM test ORDER BY y DESC) AS tab;
1332713327
<entry>
1332813328
<type>double precision</type>
1332913329
</entry>
13330-
<entry>No</entry>
13330+
<entry>Yes</entry>
1333113331
<entry><literal>sum(<replaceable
1333213332
class="parameter">X</replaceable>^2) - sum(<replaceable
1333313333
class="parameter">X</replaceable>)^2/<replaceable
@@ -13348,7 +13348,7 @@ SELECT xmlagg(x) FROM (SELECT x FROM test ORDER BY y DESC) AS tab;
1334813348
<entry>
1334913349
<type>double precision</type>
1335013350
</entry>
13351-
<entry>No</entry>
13351+
<entry>Yes</entry>
1335213352
<entry><literal>sum(<replaceable
1335313353
class="parameter">X</replaceable>*<replaceable
1335413354
class="parameter">Y</replaceable>) - sum(<replaceable
@@ -13372,7 +13372,7 @@ SELECT xmlagg(x) FROM (SELECT x FROM test ORDER BY y DESC) AS tab;
1337213372
<entry>
1337313373
<type>double precision</type>
1337413374
</entry>
13375-
<entry>No</entry>
13375+
<entry>Yes</entry>
1337613376
<entry><literal>sum(<replaceable
1337713377
class="parameter">Y</replaceable>^2) - sum(<replaceable
1337813378
class="parameter">Y</replaceable>)^2/<replaceable
@@ -13399,7 +13399,7 @@ SELECT xmlagg(x) FROM (SELECT x FROM test ORDER BY y DESC) AS tab;
1339913399
<type>double precision</type> for floating-point arguments,
1340013400
otherwise <type>numeric</type>
1340113401
</entry>
13402-
<entry>All types apart from floating-point types</entry>
13402+
<entry>Yes</entry>
1340313403
<entry>historical alias for <function>stddev_samp</function></entry>
1340413404
</row>
1340513405

@@ -13423,7 +13423,7 @@ SELECT xmlagg(x) FROM (SELECT x FROM test ORDER BY y DESC) AS tab;
1342313423
<type>double precision</type> for floating-point arguments,
1342413424
otherwise <type>numeric</type>
1342513425
</entry>
13426-
<entry>All types apart from floating-point types</entry>
13426+
<entry>Yes</entry>
1342713427
<entry>population standard deviation of the input values</entry>
1342813428
</row>
1342913429

@@ -13447,7 +13447,7 @@ SELECT xmlagg(x) FROM (SELECT x FROM test ORDER BY y DESC) AS tab;
1344713447
<type>double precision</type> for floating-point arguments,
1344813448
otherwise <type>numeric</type>
1344913449
</entry>
13450-
<entry>All types apart from floating-point types</entry>
13450+
<entry>Yes</entry>
1345113451
<entry>sample standard deviation of the input values</entry>
1345213452
</row>
1345313453

@@ -13467,7 +13467,7 @@ SELECT xmlagg(x) FROM (SELECT x FROM test ORDER BY y DESC) AS tab;
1346713467
<type>double precision</type> for floating-point arguments,
1346813468
otherwise <type>numeric</type>
1346913469
</entry>
13470-
<entry>All types apart from floating-point types</entry>
13470+
<entry>Yes</entry>
1347113471
<entry>historical alias for <function>var_samp</function></entry>
1347213472
</row>
1347313473

@@ -13491,7 +13491,7 @@ SELECT xmlagg(x) FROM (SELECT x FROM test ORDER BY y DESC) AS tab;
1349113491
<type>double precision</type> for floating-point arguments,
1349213492
otherwise <type>numeric</type>
1349313493
</entry>
13494-
<entry>All types apart from floating-point types</entry>
13494+
<entry>Yes</entry>
1349513495
<entry>population variance of the input values (square of the population standard deviation)</entry>
1349613496
</row>
1349713497

@@ -13515,7 +13515,7 @@ SELECT xmlagg(x) FROM (SELECT x FROM test ORDER BY y DESC) AS tab;
1351513515
<type>double precision</type> for floating-point arguments,
1351613516
otherwise <type>numeric</type>
1351713517
</entry>
13518-
<entry>All types apart from floating-point types</entry>
13518+
<entry>Yes</entry>
1351913519
<entry>sample variance of the input values (square of the sample standard deviation)</entry>
1352013520
</row>
1352113521
</tbody>

‎src/backend/utils/adt/float.c‎

Lines changed: 107 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -2435,6 +2435,50 @@ check_float8_array(ArrayType *transarray, const char *caller, int n)
24352435
return (float8 *) ARR_DATA_PTR(transarray);
24362436
}
24372437

2438+
/*
2439+
* float8_combine
2440+
*
2441+
* An aggregate combine function used to combine two 3 fields
2442+
* aggregate transition data into a single transition data.
2443+
* This function is used only in two stage aggregation and
2444+
* shouldn't be called outside aggregate context.
2445+
*/
2446+
Datum
2447+
float8_combine(PG_FUNCTION_ARGS)
2448+
{
2449+
ArrayType *transarray1 = PG_GETARG_ARRAYTYPE_P(0);
2450+
ArrayType *transarray2 = PG_GETARG_ARRAYTYPE_P(1);
2451+
float8 *transvalues1;
2452+
float8 *transvalues2;
2453+
float8 N,
2454+
sumX,
2455+
sumX2;
2456+
2457+
if (!AggCheckCallContext(fcinfo, NULL))
2458+
elog(ERROR, "aggregate function called in non-aggregate context");
2459+
2460+
transvalues1 = check_float8_array(transarray1, "float8_combine", 3);
2461+
N = transvalues1[0];
2462+
sumX = transvalues1[1];
2463+
sumX2 = transvalues1[2];
2464+
2465+
transvalues2 = check_float8_array(transarray2, "float8_combine", 3);
2466+
2467+
N += transvalues2[0];
2468+
sumX += transvalues2[1];
2469+
CHECKFLOATVAL(sumX, isinf(transvalues1[1]) || isinf(transvalues2[1]),
2470+
true);
2471+
sumX2 += transvalues2[2];
2472+
CHECKFLOATVAL(sumX2, isinf(transvalues1[2]) || isinf(transvalues2[2]),
2473+
true);
2474+
2475+
transvalues1[0] = N;
2476+
transvalues1[1] = sumX;
2477+
transvalues1[2] = sumX2;
2478+
2479+
PG_RETURN_ARRAYTYPE_P(transarray1);
2480+
}
2481+
24382482
Datum
24392483
float8_accum(PG_FUNCTION_ARGS)
24402484
{
@@ -2762,6 +2806,69 @@ float8_regr_accum(PG_FUNCTION_ARGS)
27622806
}
27632807
}
27642808

2809+
/*
2810+
* float8_regr_combine
2811+
*
2812+
* An aggregate combine function used to combine two 6 fields
2813+
* aggregate transition data into a single transition data.
2814+
* This function is used only in two stage aggregation and
2815+
* shouldn't be called outside aggregate context.
2816+
*/
2817+
Datum
2818+
float8_regr_combine(PG_FUNCTION_ARGS)
2819+
{
2820+
ArrayType *transarray1 = PG_GETARG_ARRAYTYPE_P(0);
2821+
ArrayType *transarray2 = PG_GETARG_ARRAYTYPE_P(1);
2822+
float8 *transvalues1;
2823+
float8 *transvalues2;
2824+
float8 N,
2825+
sumX,
2826+
sumX2,
2827+
sumY,
2828+
sumY2,
2829+
sumXY;
2830+
2831+
if (!AggCheckCallContext(fcinfo, NULL))
2832+
elog(ERROR, "aggregate function called in non-aggregate context");
2833+
2834+
transvalues1 = check_float8_array(transarray1, "float8_regr_combine", 6);
2835+
N = transvalues1[0];
2836+
sumX = transvalues1[1];
2837+
sumX2 = transvalues1[2];
2838+
sumY = transvalues1[3];
2839+
sumY2 = transvalues1[4];
2840+
sumXY = transvalues1[5];
2841+
2842+
transvalues2 = check_float8_array(transarray2, "float8_regr_combine", 6);
2843+
2844+
N += transvalues2[0];
2845+
sumX += transvalues2[1];
2846+
CHECKFLOATVAL(sumX, isinf(transvalues1[1]) || isinf(transvalues2[1]),
2847+
true);
2848+
sumX2 += transvalues2[2];
2849+
CHECKFLOATVAL(sumX2, isinf(transvalues1[2]) || isinf(transvalues2[2]),
2850+
true);
2851+
sumY += transvalues2[3];
2852+
CHECKFLOATVAL(sumY, isinf(transvalues1[3]) || isinf(transvalues2[3]),
2853+
true);
2854+
sumY2 += transvalues2[4];
2855+
CHECKFLOATVAL(sumY2, isinf(transvalues1[4]) || isinf(transvalues2[4]),
2856+
true);
2857+
sumXY += transvalues2[5];
2858+
CHECKFLOATVAL(sumXY, isinf(transvalues1[5]) || isinf(transvalues2[5]),
2859+
true);
2860+
2861+
transvalues1[0] = N;
2862+
transvalues1[1] = sumX;
2863+
transvalues1[2] = sumX2;
2864+
transvalues1[3] = sumY;
2865+
transvalues1[4] = sumY2;
2866+
transvalues1[5] = sumXY;
2867+
2868+
PG_RETURN_ARRAYTYPE_P(transarray1);
2869+
}
2870+
2871+
27652872
Datum
27662873
float8_regr_sxx(PG_FUNCTION_ARGS)
27672874
{

0 commit comments

Comments
 (0)