From: Otto Moerbeek <otto@drijf.net>
Subject: Re: timing lld --threads for fun and profit
To: Landry Breuil <landry@openbsd.org>, robert@openbsd.org, tech@openbsd.org
Date: Sun, 10 Nov 2024 17:20:04 +0100

On Fri, Nov 08, 2024 at 01:01:50PM +0000, Stuart Henderson wrote:

> On 2024/11/08 13:26, Martin Pieuchot wrote:
> > On 08/11/24(Fri) 12:22, Landry Breuil wrote:
> > > [...] 
> > > someone(tm) should look into patching lld to avoid using more than
> > > MAX(ncpu,5) threads ? In the meantime, i'll probably fix the firefox
> > > ports to avoir using MAKE_JOBS for lld but cap it at 5.
> > 
> > Recent lld(1) include the following commits which limit the value to
> > 16 by default instead of the number of available CPUs.
> > 
> > See the following commits:
> > 
> > https://github.com/llvm/llvm-project/commit/a8788de1c3f3c8c3a591bd3aae2acee1b43b229a
> > https://github.com/llvm/llvm-project/commit/da68d2164efcc1f5e57f090e2ae2219056b120a0
> > 
> > Robert do you see the same with chromium?  Would it make sense to
> > backport these diff with a smaller value for OpenBSD?
> > 
> 
> Certainly helps for reorder_kernel.
> 
> $ sysctl hw.{model,ncpu,version}
> hw.model=12th Gen Intel(R) Core(TM) i5-1245U
> hw.ncpu=12
> hw.version=ThinkPad T14 Gen 3
> 
> # \time -l /usr/libexec/reorder_kernel
>         7.30 real         6.86 user         3.92 sys
>     879280  maximum resident set size
>          0  average shared memory size
>          0  average unshared data size
>          0  average unshared stack size
>     147812  minor page faults
>      79421  major page faults
>          0  swaps
>      11833  block input operations
>      15643  block output operations
>          1  messages sent
>          0  messages received
>         45  signals received
>      46512  voluntary context switches
>       7353  involuntary context switches
> # vi Makefile
> [...]
> $ grep ^LINKFL Makefile
> LINKFLAGS=	-T ld.script -X --warn-common -nopie -Wl,--threads=5
> LINKFLAGS+=	-S
> # \time -l /usr/libexec/reorder_kernel
>         0.41 real         0.26 user         0.09 sys
>     100920  maximum resident set size
>          0  average shared memory size
>          0  average unshared data size
>          0  average unshared stack size
>      17119  minor page faults
>         10  major page faults
>          0  swaps
>          5  block input operations
>         78  block output operations
>          1  messages sent
>          0  messages received
>         28  signals received
>        189  voluntary context switches
>          2  involuntary context switches
> 

There's more to this: even specyfying -Wl,--threads=1 gives me a big
speedup. Or even any value. Looking in the log I see:

ld: error: unknown argument '-Wl,--threads=1'

The correct flags is --threads=N

I dont see a lot of difference for the various flag values.

	-Otto