On 20/11/2026 09:14, David Geier wrote:
>> It's attached. I saw really good results on x86 but some regressions on
>> M5. That's why I started working on centralizing the case distinction
>> work scattered throughout all the macros in set_var_from_num() and have
>> all functions use that. This also makes the code a lot nicer. This
>> change isn't complete yet, so I cannot send it to you.
>>
>> I'll share the results on my box with you tomorrow. Right now, I don't
>> have access to my benchmark machine.
> 
> Attached are the results from my AMD Ryzen 7 9700X. You can see that
> there are lots of speedups but also lots of regressions. The regressions
> stem from the additional case distinction for if it's a packed /
> unpacked varlena that are repeatedly scattered throughout the code via
> the NUMERIC_ macros.
I wonder what your compiler is.
I test these numeric-related patches on Intel/AMD/ARM platforms with clang/gcc.
For GCC, I do recommend using -mstringop-strategy=libcall. It hides the
performance issue in GCC 12-15. For production code, I usually use a small,
non-committable hack that transforms memcpy(..,..,N) into a series of
memcpy(..,..,8).

-- 
regards, Andrei Lepikhov,
pgEdge


Reply via email to