Thanks to those implementing the new GC!

Thanks to those implementing the new GC!

[Alan Mackenzie]

To Gerd, Eli, Helmut, and anybody else working on the new GC:

Thanks!

The GC we have had over the last few decades is fine and works well, but
it consumes a large part of Emacs's run time (as can be seen on profiler
reports), and causes things expensive on computation to stutter (e.g.
scrolling).

I very much look forward to the new GC, which will be incremental and
take place on a different processor core from most of Emacs, given how
most (?all) modern PCs have several cores.  It will be good to use
these.

Gerd has already posted some impressive benchmarks.  I look forward to a
general speed increase when (hopefully not if) this new feature is
merged into master.

-- 
Alan Mackenzie (Nuremberg, Germany).

Re: Thanks to those implementing the new GC!

[Andrea Corallo]

Alan Mackenzie <acm@muc.de> writes:

> To Gerd, Eli, Helmut, and anybody else working on the new GC:

+1

Re: Thanks to those implementing the new GC!

[Gerd Möllmann]

Alan Mackenzie <acm@muc.de> writes:

> To Gerd, Eli, Helmut, and anybody else working on the new GC:
>
> Thanks!

Thanks Alan!

> Gerd has already posted some impressive benchmarks.  I look forward to a
> general speed increase when (hopefully not if) this new feature is
> merged into master.

I'd like to dampen expectations a bit. The annecdotal 2 build times I
once posted are just that, annecdotal, and actually quite the opposite
of what I presonally would expect. Let's rather wait and see what the
outcome in the end will be. If GC pause times are better that's all I
personally would want.

Re: Thanks to those implementing the new GC!

[Sam James]

Gerd Möllmann <gerd.moellmann@gmail.com> writes:

> Alan Mackenzie <acm@muc.de> writes:
>
>> To Gerd, Eli, Helmut, and anybody else working on the new GC:
>>
>> Thanks!
>
> Thanks Alan!
>
>> Gerd has already posted some impressive benchmarks.  I look forward to a
>> general speed increase when (hopefully not if) this new feature is
>> merged into master.
>
> I'd like to dampen expectations a bit. The annecdotal 2 build times I
> once posted are just that, annecdotal, and actually quite the opposite
> of what I presonally would expect. Let's rather wait and see what the
> outcome in the end will be. If GC pause times are better that's all I
> personally would want.

The effort being put in is worthy of applause, either way. Thank you.

Re: Thanks to those implementing the new GC!

[Richard Stallman]

[[[ To any NSA and FBI agents reading my email: please consider    ]]]
[[[ whether defending the US Constitution against all enemies,     ]]]
[[[ foreign or domestic, requires you to follow Snowden's example. ]]]

  > I very much look forward to the new GC, which will be incremental and
  > take place on a different processor core from most of Emacs, given how
  > most (?all) modern PCs have several cores.  It will be good to use
  > these.

This leads me to wonder what the effect of the new GC will be on a
single-processor machine.  Will Emacs run slower on those with
thw new GC?

-- 
Dr Richard Stallman (https://stallman.org)
Chief GNUisance of the GNU Project (https://gnu.org)
Founder, Free Software Foundation (https://fsf.org)
Internet Hall-of-Famer (https://internethalloffame.org)

Re: Thanks to those implementing the new GC!

[Lynn Winebarger]

[-- Attachment #1: Type: text/plain, Size: 808 bytes --]

On Wed, May 8, 2024, 7:24 PM Richard Stallman <rms@gnu.org> wrote:

>
>   > I very much look forward to the new GC, which will be incremental and
>   > take place on a different processor core from most of Emacs, given how
>   > most (?all) modern PCs have several cores.  It will be good to use
>   > these.
>
> This leads me to wonder what the effect of the new GC will be on a
> single-processor machine.  Will Emacs run slower on those with
> thw new GC?
>

That question is sort of ill-posed.  From Gerd's description, MPS is both
incremental and multi-threaded/concurrent.  I don't know if the dedicated
thread can be disabled, but even without it one would expect to get lower
pause times from incremental collection, possibly at the cost of higher
total collection times.

Lynn

[-- Attachment #2: Type: text/html, Size: 1391 bytes --]

Re: Thanks to those implementing the new GC!

[Björn Bidar]

Lynn Winebarger <owinebar@gmail.com> writes:

> On Wed, May 8, 2024, 7:24 PM Richard Stallman <rms@gnu.org> wrote:
>
>>
>>   > I very much look forward to the new GC, which will be incremental and
>>   > take place on a different processor core from most of Emacs, given how
>>   > most (?all) modern PCs have several cores.  It will be good to use
>>   > these.
>>
>> This leads me to wonder what the effect of the new GC will be on a
>> single-processor machine.  Will Emacs run slower on those with
>> thw new GC?
>>
>
> That question is sort of ill-posed.  From Gerd's description, MPS is both
> incremental and multi-threaded/concurrent.  I don't know if the dedicated
> thread can be disabled, but even without it one would expect to get lower
> pause times from incremental collection, possibly at the cost of higher
> total collection times.
>
> Lynn

Also I would think that if machines with more than one core profit while
those with only one core will be slightly slower is a good trade of
given that almost any machine from the last 20 years contains more than
one core.

In general Emacs shouldn't hold back to support very old legacy systems
on the cost of any recentish system from the last 10 years or less from
my point of view.

In case there's still the builtin gc..

Re: Thanks to those implementing the new GC!

[Richard Stallman]

[[[ To any NSA and FBI agents reading my email: please consider    ]]]
[[[ whether defending the US Constitution against all enemies,     ]]]
[[[ foreign or domestic, requires you to follow Snowden's example. ]]]

    I don't know if the dedicated
  > thread can be disabled, but even without it one would expect to get lower
  > pause times from incremental collection, possibly at the cost of higher
  > total collection times.

That isn't pbvious to me -- though it might be true.
Would people please benchmark this when they have a chance?


-- 
Dr Richard Stallman (https://stallman.org)
Chief GNUisance of the GNU Project (https://gnu.org)
Founder, Free Software Foundation (https://fsf.org)
Internet Hall-of-Famer (https://internethalloffame.org)

Re: Thanks to those implementing the new GC!

[Eli Zaretskii]

> From: Richard Stallman <rms@gnu.org>
> Cc: acm@muc.de, emacs-devel@gnu.org
> Date: Fri, 10 May 2024 19:12:50 -0400
> 
>     I don't know if the dedicated
>   > thread can be disabled, but even without it one would expect to get lower
>   > pause times from incremental collection, possibly at the cost of higher
>   > total collection times.
> 
> That isn't pbvious to me -- though it might be true.
> Would people please benchmark this when they have a chance?

The branch is not yet in a state that we can turn our attention to
performance aspects.  We are still working on adapting some basic
Emacs features to this new GC.

Re: Thanks to those implementing the new GC!

[Richard Stallman]

[[[ To any NSA and FBI agents reading my email: please consider    ]]]
[[[ whether defending the US Constitution against all enemies,     ]]]
[[[ foreign or domestic, requires you to follow Snowden's example. ]]]

  > > That isn't pbvious to me -- though it might be true.
  > > Would people please benchmark this when they have a chance?

  > The branch is not yet in a state that we can turn our attention to
  > performance aspects.  We are still working on adapting some basic
  > Emacs features to this new GC.

I understand.  The point is to make sure to benchmark this
becure proposing to merge in the code for the new GC.

-- 
Dr Richard Stallman (https://stallman.org)
Chief GNUisance of the GNU Project (https://gnu.org)
Founder, Free Software Foundation (https://fsf.org)
Internet Hall-of-Famer (https://internethalloffame.org)

Re: Thanks to those implementing the new GC!

[Andrea Corallo]

Richard Stallman <rms@gnu.org> writes:

> [[[ To any NSA and FBI agents reading my email: please consider    ]]]
> [[[ whether defending the US Constitution against all enemies,     ]]]
> [[[ foreign or domestic, requires you to follow Snowden's example. ]]]
>
>   > I very much look forward to the new GC, which will be incremental and
>   > take place on a different processor core from most of Emacs, given how
>   > most (?all) modern PCs have several cores.  It will be good to use
>   > these.
>
> This leads me to wonder what the effect of the new GC will be on a
> single-processor machine.  Will Emacs run slower on those with
> thw new GC?

Hi Richard,

I think we don't have ATM reliable performance figures for the new GC,
but even if it proves to be less efficient (more CPU cycles to execute
the same Lisp program), it might be still advantageous to have the GC
run in a parallel thread to avoid GC pauses.

Another point is that AFAIU we can't dump our classic GC anyway because
it is necessary for certain platforms we do support but ATM are not
supported by MPS.  So I guess our current GC will stay in any case at
least as a compile-time option.

Best Regards

  Andrea