Re: Add spin_delay() implementation for Arm in s_lock.h

Thanks for all the help Tom!

On 4/6/22, 6:07 PM, "Tom Lane" <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:

CAUTION: This email originated from outside of the organization. Do not click links or open attachments unless you can confirm the sender and know the content is safe.

"Blake, Geoff" <blakgeof(at)amazon(dot)com> writes:
> Hi Tom, Andres,
> Any additional feedback for this patch?

I did some more research and testing:

* Using a Mac with the M1 Pro chip (marginally beefier than the M1
I was testing on before), I think I can see some benefit in the
test case I proposed upthread. It's marginal though.

* On a Raspberry Pi 3B+, there's no outside-the-noise difference.

* ISB doesn't exist in pre-V7 ARM, so it seems prudent to restrict
the patch to ARM64. I doubt any flavor of ARM32 would be able to
benefit anyway. (Googling finds that MariaDB made this same
choice not long ago [1].)

So what we've got is that there seems to be benefit at high
core counts, and it at least doesn't hurt at lower ones.
That's good enough for me, so pushed.

regards, tom lane

[1] https://jira.mariadb.org/browse/MDEV-25807