1
0
mirror of https://github.com/postgres/postgres.git synced 2025-06-11 20:28:21 +03:00

Improve performance of float overflow checks in btree_gist

The current code could do unnecessary calls to isinf() (two for the
argument values all the time while one could be sufficient in some
cases).  zero_is_valid was never used but the result value was still
checked on 0 in the first position of the check.

This is similar to 607f8ce.  btree_gist has just copy-pasted the code
doing those checks from the backend float4/8 code, as of the macro
CHECKFLOATVAL(), to do the work.

Author: Haiying Tang
Discussion: https://postgr.es/m/OS0PR01MB611358E3A7BC3C2F874AC36BFBF39@OS0PR01MB6113.jpnprd01.prod.outlook.com
This commit is contained in:
Michael Paquier
2021-08-19 10:42:44 +09:00
parent 2576dcfb76
commit 32cf7f7acc
3 changed files with 8 additions and 21 deletions

View File

@ -5,6 +5,7 @@
#include "btree_gist.h"
#include "btree_utils_num.h"
#include "utils/float.h"
typedef struct float4key
{
@ -98,7 +99,8 @@ float4_dist(PG_FUNCTION_ARGS)
float4 r;
r = a - b;
CHECKFLOATVAL(r, isinf(a) || isinf(b), true);
if (unlikely(isinf(r)) && !isinf(a) && !isinf(b))
float_overflow_error();
PG_RETURN_FLOAT4(Abs(r));
}