Apply the "nodeAgg" optimization to more of the builtin transition

functions. This patch optimizes int2_sum(), int4_sum(), float4_accum()
and float8_accum() to avoid needing to copy the transition function's
state for each input tuple of the aggregate. In an extreme case
(e.g. SELECT sum(int2_col) FROM table where table has a single column),
it improves performance by about 20%. For more complex queries or tables
with wider rows, the relative performance improvement will not be as
significant.

1 files changed, 55 insertions, 19 deletions

diff --git a/src/backend/utils/adt/float.c b/src/backend/utils/adt/float.c
index e51babfebfb..c943ee2c71d 100644
--- a/src/backend/utils/adt/float.c
+++ b/src/backend/utils/adt/float.c
@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/utils/adt/float.c,v 1.113 2005/02/11 04:08:58 neilc Exp $
+ *	  $PostgreSQL: pgsql/src/backend/utils/adt/float.c,v 1.114 2005/04/06 23:56:07 neilc Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -1902,8 +1902,6 @@ float8_accum(PG_FUNCTION_ARGS)
 	float8		N,
 				sumX,
 				sumX2;
-	Datum		transdatums[3];
-	ArrayType  *result;
 
 	transvalues = check_float8_array(transarray, "float8_accum");
 	N = transvalues[0];
@@ -1914,15 +1912,35 @@ float8_accum(PG_FUNCTION_ARGS)
 	sumX += newval;
 	sumX2 += newval * newval;
 
-	transdatums[0] = Float8GetDatumFast(N);
-	transdatums[1] = Float8GetDatumFast(sumX);
-	transdatums[2] = Float8GetDatumFast(sumX2);
+	/*
+	 * If we're invoked by nodeAgg, we can cheat and modify our first
+	 * parameter in-place to reduce palloc overhead. Otherwise we
+	 * construct a new array with the updated transition data and
+	 * return it.
+	 */
+	if (fcinfo->context && IsA(fcinfo->context, AggState))
+	{
+		transvalues[0] = N;
+		transvalues[1] = sumX;
+		transvalues[2] = sumX2;
 
-	result = construct_array(transdatums, 3,
-							 FLOAT8OID,
-						 sizeof(float8), false /* float8 byval */ , 'd');
+		PG_RETURN_ARRAYTYPE_P(transarray);
+	}
+	else
+	{
+		Datum		transdatums[3];
+		ArrayType  *result;
+
+		transdatums[0] = Float8GetDatumFast(N);
+		transdatums[1] = Float8GetDatumFast(sumX);
+		transdatums[2] = Float8GetDatumFast(sumX2);
 
-	PG_RETURN_ARRAYTYPE_P(result);
+		result = construct_array(transdatums, 3,
+								 FLOAT8OID,
+								 sizeof(float8), false /* float8 byval */ , 'd');
+
+		PG_RETURN_ARRAYTYPE_P(result);
+	}
 }
 
 Datum
@@ -1935,8 +1953,6 @@ float4_accum(PG_FUNCTION_ARGS)
 				sumX,
 				sumX2,
 				newval;
-	Datum		transdatums[3];
-	ArrayType  *result;
 
 	transvalues = check_float8_array(transarray, "float4_accum");
 	N = transvalues[0];
@@ -1950,15 +1966,35 @@ float4_accum(PG_FUNCTION_ARGS)
 	sumX += newval;
 	sumX2 += newval * newval;
 
-	transdatums[0] = Float8GetDatumFast(N);
-	transdatums[1] = Float8GetDatumFast(sumX);
-	transdatums[2] = Float8GetDatumFast(sumX2);
+	/*
+	 * If we're invoked by nodeAgg, we can cheat and modify our first
+	 * parameter in-place to reduce palloc overhead. Otherwise we
+	 * construct a new array with the updated transition data and
+	 * return it.
+	 */
+	if (fcinfo->context && IsA(fcinfo->context, AggState))
+	{
+		transvalues[0] = N;
+		transvalues[1] = sumX;
+		transvalues[2] = sumX2;
 
-	result = construct_array(transdatums, 3,
-							 FLOAT8OID,
-						 sizeof(float8), false /* float8 byval */ , 'd');
+		PG_RETURN_ARRAYTYPE_P(transarray);
+	}
+	else
+	{
+		Datum		transdatums[3];
+		ArrayType  *result;
+
+		transdatums[0] = Float8GetDatumFast(N);
+		transdatums[1] = Float8GetDatumFast(sumX);
+		transdatums[2] = Float8GetDatumFast(sumX2);
 
-	PG_RETURN_ARRAYTYPE_P(result);
+		result = construct_array(transdatums, 3,
+								 FLOAT8OID,
+								 sizeof(float8), false /* float8 byval */ , 'd');
+
+		PG_RETURN_ARRAYTYPE_P(result);
+	}
 }
 
 Datum