Re: Popcount optimization using AVX512

On Tue, Oct 08, 2024 at 09:36:03PM -0500, Nathan Bossart wrote:
> On Wed, Jul 31, 2024 at 04:43:02PM -0500, Nathan Bossart wrote:
>> On Wed, Jul 31, 2024 at 01:52:54PM -0700, Andres Freund wrote:
>>> I think we'd be better off enabling architectural features on a per-function
>>> basis, roughly like this:
>>>
>>> [...]
>>>
>>> /* FIXME: Should be gated by configure check of -mavx512vpopcntdq -mavx512bw support */
>>> pg_enable_target("avx512vpopcntdq,avx512bw")
>>> uint64_t
>>> pg_popcount_avx512(const char *buf, int bytes)
>>> ...
>>
>> I remember wondering why the CRC-32C code wasn't already doing something
>> like this (old compiler versions? non-gcc-like compilers?), and I'm not
>> sure I ever discovered the reason, so out of an abundance of caution I used
>> the same approach for AVX-512. If we can convince ourselves that
>> __attribute__((target("..."))) is standard enough at this point, +1 for
>> moving to that.
>
> [...]
>
> So, at least for the CRC code, __attribute__((target("..."))) was probably
> not widely available enough yet when it was first added. Unfortunately,
> the ARMv8 CRC target support (without -march) is still pretty new, but it
> might be possible to switch the others to a per-function approach in v18.

Here is a first attempt at using __attribute__((target("..."))) for the
AVX-512 stuff. Besides allowing us to consolidate the code into a single
file, this simplifies the build file changes quite a bit.

--
nathan