Horton

Re: refactor architecture-specific popcount code

Nathan Bossart <nathandbossart@gmail.com>

View thread View on postgr.es
From: Nathan Bossart <nathandbossart@gmail.com>
To: Greg Burd <greg@burd.me>
Cc: John Naylor <johncnaylorls@gmail.com>, Heikki Linnakangas <hlinnaka@iki.fi>, pgsql-hackers <pgsql-hackers@postgresql.org>
Date: 2026-02-03T20:44:10Z
Lists: pgsql-hackers

Attachments

Okay, here is an updated patch set with first drafts of the commit
messages.  I'm reasonably happy with these patches, but I'll admit my
justification for ripping out the 32-bit optimizations feels a bit flimsy.
I don't get the idea that we are all that concerned about things like
micro-regressions for popcount on 32-bit builds, but OTOH it isn't hard to
imagine someone objecting to these changes.

I ran the bms_num_members() benchmark on a couple of machines I had nearby:

    apple-m3 (neon)         intel-i5-13500T (sse4.2)
    words  HEAD   v8        words  HEAD   v8
        1    40   25            1    26   10
        2    57   51            2    37   29
        4    75   57            4    55   45
        8   105   56            8    88   51
       16   154   59           16   158   68
       32   265   73           32   296  102
       64   545  103           64   577  209
      128  1027  178          128  1212  423

I was going to run it on machines with SVE/AVX-512, but John already tested
the AVX-512 case [0], and I have no reason to believe that we'll see
regressions on machines with SVE.

[0] https://postgr.es/m/CANWCAZbWLX%3DEDd1Bq-8oGK2ZLVNR4m4BkGe%3D288t2V5oLcqeZA%40mail.gmail.com

-- 
nathan

Commits

  1. Make use of pg_popcount() in more places.

  2. Remove uses of popcount builtins.

  3. Remove some unnecessary optimizations in popcount code.

  4. Remove specialized word-length popcount implementations.

  5. Move x86-64-specific popcount code to pg_popcount_x86.c.

  6. Refactor some SIMD and popcount macros.

  7. Rename "fast" and "slow" popcount functions.

  8. Rename pg_popcount_avx512.c to pg_popcount_x86.c.

  9. Remove trailing zero words from Bitmapsets

  10. Don't use _BitScanForward64/_BitScanReverse64 on 32-bit MSVC builds