When multiple calls to the same UDF are in a single statement, how many times will it be called?

Question

In the following t-sql statement, how many times will the dbo.FUNC function get called?

SELECT
    column1,
    column2,
    dbo.FUNC(column3) AS column3
FROM table1
WHERE dbo.FUNC(column3) >= 5
ORDER BY dbo.FUNC(column3) DESC

Will it called multiple separate times per row, or does the optimizer recognize that it is being used multiple times in a single statement, and only call it once?

How can I test this? I can't insert into a table inside of a function, so incrementing a counter wont work...

Answer 1

This isn't guaranteed.

You would need to check the execution plan to find out. Some examples.

CREATE FUNCTION dbo.FUNC1(@p1 int)
RETURNS int
AS
BEGIN
    RETURN @p1 + 1
END

GO

CREATE FUNCTION dbo.FUNC2(@p1 int)
RETURNS int
WITH SCHEMABINDING
AS
BEGIN
    RETURN @p1 + 1
END

GO
SELECT 
       OBJECTPROPERTYEX(OBJECT_ID('dbo.FUNC1'), 'IsDeterministic'),
       OBJECTPROPERTYEX(OBJECT_ID('dbo.FUNC2'), 'IsDeterministic') 
GO

FUNC2 is created WITH SCHEMABINDING and is treated as deterministic. FUNC1 isn't.

SELECT
    dbo.FUNC1(number) AS FUNC1,
    dbo.FUNC2(number) AS FUNC2
FROM master..spt_values
WHERE dbo.FUNC1(number) >= 5 AND dbo.FUNC2(number) >= 5
ORDER BY dbo.FUNC1(number), dbo.FUNC2(number)

Gives Plan

  |--Sort(ORDER BY:([Expr1003] ASC, [Expr1004] ASC))
       |--Compute Scalar(DEFINE:([Expr1003]=[test].[dbo].[FUNC1]([master].[dbo].[spt_values].[number])))
            |--Filter(WHERE:([test].[dbo].[FUNC1]([master].[dbo].[spt_values].[number])>=(5) AND [Expr1004]>=(5)))
                 |--Compute Scalar(DEFINE:([Expr1004]=[test].[dbo].[FUNC2]([master].[dbo].[spt_values].[number])))
                      |--Index Scan(OBJECT:([master].[dbo].[spt_values].[ix2_spt_values_nu_nc]))

FUNC1 is evaluated twice (once in the filter and once in a compute scalar outputting a calculated column used for both the projection and the ordering), FUNC2 is only evaluated once.

Rewriting as

SELECT
    FUNC1,
    FUNC2
FROM master..spt_values
CROSS APPLY (SELECT dbo.FUNC1(number), dbo.FUNC2(number)) C(FUNC1, FUNC2)
WHERE FUNC1 >= 5 AND FUNC2 >= 5
ORDER BY FUNC1, FUNC2

Changes the plan slightly and both are only evaluated once

  |--Sort(ORDER BY:([Expr1003] ASC, [Expr1004] ASC))
       |--Filter(WHERE:([Expr1003]>=(5)))
            |--Compute Scalar(DEFINE:([Expr1003]=[test].[dbo].[FUNC1]([master].[dbo].[spt_values].[number])))
                 |--Filter(WHERE:([Expr1004]>=(5)))
                      |--Compute Scalar(DEFINE:([Expr1004]=[test].[dbo].[FUNC2]([master].[dbo].[spt_values].[number])))
                           |--Index Scan(OBJECT:([master].[dbo].[spt_values].[ix2_spt_values_nu_nc]))

Now making a slight alteration to the query

SELECT
    FUNC1 + 10,
    FUNC2 + 10
FROM master..spt_values
CROSS APPLY (SELECT dbo.FUNC1(number), dbo.FUNC2(number)) C(FUNC1, FUNC2)
WHERE FUNC1 >= 5 AND FUNC2 >= 5
ORDER BY FUNC1, FUNC2

Gives the opposite of the original result in that FUNC2 is evaluated twice but FUNC1 only once.

  |--Compute Scalar(DEFINE:([Expr1005]=[Expr1003]+(10)))
       |--Sort(ORDER BY:([Expr1003] ASC, [Expr1004] ASC))
            |--Filter(WHERE:([Expr1003]>=(5)))
                 |--Compute Scalar(DEFINE:([Expr1003]=[test].[dbo].[FUNC1]([master].[dbo].[spt_values].[number])))
                      |--Filter(WHERE:([Expr1004]>=(5)))
                           |--Compute Scalar(DEFINE:([Expr1004]=[test].[dbo].[FUNC2]([master].[dbo].[spt_values].[number]), [Expr1006]=[test].[dbo].[FUNC2]([master].[dbo].[spt_values].[number])+(10)))
                                |--Index Scan(OBJECT:([master].[dbo].[spt_values].[ix2_spt_values_nu_nc]))

Answer 2

Firstly, it depends on if the function is deterministic.

Even then, that will only be used for multiple calls on a single row.

I believe your case would be optimized if the function is deterministic.