Re: Improving the heapgetpage function improves performance in common scenarios

On 2023/9/6 15:50, Quan Zongliang wrote:
>
>
> On 2023/9/5 18:46, John Naylor wrote:
>>
>> On Tue, Sep 5, 2023 at 4:27 PM Quan Zongliang <quanzongliang(at)yeah(dot)net
>> <mailto:quanzongliang(at)yeah(dot)net>> wrote:
>>
>>  > Here's how I test it
>>  >     EXPLAIN ANALYZE SELECT * FROM orders;
>>
>> Note that EXPLAIN ANALYZE has quite a bit of overhead, so it's not
>> good for these kinds of tests.
>>
>>  > I'll also try Andres Freund's test method next.
>>
>> Commit f691f5b80a85 from today removes another source of overhead in
>> this function, so I suggest testing against that, if you wish to test
>> again.
>>
> Test with the latest code of the master branch, see the attached results.
>
> If not optimized(--enable-debug CFLAGS='-O0'), there is a clear
> difference. When the compiler does the optimization, the performance is
> similar. I think the compiler does a good enough optimization with
> "pg_attribute_always_inline" and the last two constant parameters when
> calling heapgetpage_collect.
>
Add a note. The first execution time of an attachment is not calculated
in the average.

>
>> --
>> John Naylor
>> EDB: http://www.enterprisedb.com <http://www.enterprisedb.com>