Re: [PATCH] SVE popcount support

On Thu, Feb 06, 2025 at 10:33:35AM -0600, Nathan Bossart wrote:
> On Thu, Feb 06, 2025 at 08:44:35AM +0000, Chiranmoy.Bhattacharya@fujitsu.com wrote:
>>> Does this hand-rolled loop unrolling offer any particular advantage?  What
>>> do the numbers look like if we don't do this or if we process, say, 4
>>> vectors at a time?
>> 
>> The unrolled version performs better than the non-unrolled one, but
>> processing four vectors provides no additional benefit. The numbers
>> and code used are given below.
> 
> Hm.  Any idea why that is?  I wonder if the compiler isn't using as many
> SVE registers as it could for this.

I've also noticed that the latest patch doesn't compile on my M3 macOS
machine.  After a quick glance, I think the problem is that the
TRY_POPCNT_FAST macro is set, so it's trying to compile the assembly
versions.

../postgresql/src/port/pg_bitutils.c:230:41: error: invalid output constraint '=q' in asm
  230 | __asm__ __volatile__(" popcntl %1,%0\n":"=q"(res):"rm"(word):"cc");
      |                                         ^
../postgresql/src/port/pg_bitutils.c:247:41: error: invalid output constraint '=q' in asm
  247 | __asm__ __volatile__(" popcntq %1,%0\n":"=q"(res):"rm"(word):"cc");
      |                                         ^
2 errors generated.

-- 
nathan