Ricing gentoo

[stefan11111]

I have seen multiple threads recently regarding optimization.
These are my COMMON_FLAGS right now:

[CaptainBlood]

Here's lto as much as possible... when it's worth it.

By worth I mean binary size shrink. Please note efficiency impact isn't considered there.

Most of the time lto shrinks, but not always.
Even within a single package: Some files may shrink when other may grow.

For this purpose, I have a script that builds a package twice and compare file sizes.
Classes of files considered here are:
*/*bin/*
/*.so
/*.pyc
I may be missing some, though...
Then I have to make decision when to activate lto in /etc/portage/package.env or not.

So I've opt for a per package setting, rather than global.
It could also be configured as global + negative /etc/portage/package.env to block it.

Some packages are or may become flto insensitive upon upstream decisions (source, or portage)
When package lto it blocked at gentoo/portage level, it should be easy to workaround it deriving its ebuild in local repository and modify it.

Besides,

I'm not sure -flto=4 is the regular way to activate...
Here is gcc by default, clang/llvm may be a # activation story.

rust/llvm has it own way to activate it.

Side note,
march / mtune should be considered if a binary package generated on a newer CPU is expected to be installed on a lower CPU.

The underlying general idea is that every optimization option should be well evaluated before deciding when and how to apply.

Thks 4 ur attention, interest & support.
_________________
USE="-* ..." in /etc/portage/make.conf here, i.e. a countermeasure to portage implicit braces, belt & diaper paradigm
LT: "I've been doing a passable imitation of the Fontana di Trevi, except my medium is mucus. Sooo much mucus. "

[stefan11111]

I do also have USE="lto" enabled globally.

[CaptainBlood]

Packages with USE=lto may indeed deserve some attention when activated along with lto global setting.

Thks 4 ur attention, interest & support
_________________
USE="-* ..." in /etc/portage/make.conf here, i.e. a countermeasure to portage implicit braces, belt & diaper paradigm
LT: "I've been doing a passable imitation of the Fontana di Trevi, except my medium is mucus. Sooo much mucus. "

[stefan11111]

Not many people interested in optimizing gentoo?
I also looked into these:
https://forums.gentoo.org/viewtopic.php?t=67777
https://github.com/InBetweenNames/gentooLTO

I will give it a try with these flags:
COMMON_FLAGS="-O3 -pipe -march=native -flto=4 -ftree-vectorize -ffast-math -funswitch-loops -fno-semantic-interposition -fipa-pta -fdevirtualize-at-ltrans"
LDFLAGS="-Wl,-O1 -Wl,--sort-common -s -Wl,--as-needed"
Will come back after an emerge -e @world.
_________________
My overlay: https://github.com/stefan11111/stefan_overlay
INSTALL_MASK="/etc/systemd /lib/systemd /usr/lib/systemd /usr/lib/modules-load.d *udev* /usr/lib/tmpfiles.d *tmpfiles* /var/lib/dbus /usr/bin/gdbus /lib/udev"

[stefan11111]

Results are in:
@world without palemoon builds in just under 6 hours.
The reason it's without palemoon is that palemoon fails to build.
The code has a #error to require me to define something to let me build it.
The fix is to add this to /etc/portage/env/www-client/palemoon:

[pjp]

[stefan11111]

[stefan11111]

Got it to build.
It took 46 minutes, 10 more that usual.
This bumps my @world emerge time to 6hours 30 minutes. Not too bad I'd say.
The assembler errors were a known issue:
https://repo.palemoon.org/MoonchildProductions/Pale-Moon/issues/1932

I also got some errors saying that libxul.so has 0 symbols.
Those were resolved after commenting out my LDFLAGS.

[Hu]

[stefan11111]

[Hu]

[stefan11111]

[Hu]

That could be, or it could be that this is their equivalent to Portage's I_KNOW_WHAT_I_AM_DOING. Their flag might be merely to disable a sanity check so that the developers can build with -ffast-math and see what is broken, as a necessary prerequisite to improving the code. If I were the developer, and I knew that -ffast-math broke the normal code, my preferences would be, in order:

• Fix the code to work with -ffast-math, if possible.
• Lock out the bad option, so that users cannot use it unknowingly.
• Detect the bad option and transparently switch to good code, even if it is less efficient. (PIC often did this on x86.)

Locking out the option, and requiring the user to manually switch to working code, and not naming the switch flag to make it obvious that this is what is happening, is not how I would approach this, when other options are available. Therefore, I am doubtful that the Palemoon developers took that approach.

[stefan11111]

[NichtDerHans]

Is your Gentoo faster then before?

[stefan11111]

[stefan11111]

Don't know how CFLAGS affect this, but at least it shows why minimalism is good.
I have qbittorrent open with 50 torrents summing up to about 3.8 TB with 6 MB/s traffic when I wrote this.
I have palemoon open with 60 tabs, which I made sure that they are all loaded into memory.
I also have kirc, vsftpd, sshd, utelnetd running.
The system uses about 2.6 GB ram and less that 10% cpu.
_________________
My overlay: https://github.com/stefan11111/stefan_overlay
INSTALL_MASK="/etc/systemd /lib/systemd /usr/lib/systemd /usr/lib/modules-load.d *udev* /usr/lib/tmpfiles.d *tmpfiles* /var/lib/dbus /usr/bin/gdbus /lib/udev"

[NichtDerHans]

Hmmm... I dont know a good benchmark for this case.

[pc_user4]

I know this thread hasn't had activity for a bit, but "ricing" / "rice" as a verb has racist context and I suggest to disavow such language within Gentoo. I understand the "Gentoo is Rice" meme page is several decades old and the world was a very different place in the early 2000's.

I am not the only one here who endorses the use of Gentoo in professional settings, and I would be very unhappy if a current or prospective employer investigating what Gentoo is all about saw such things or came across the "Gentoo is Rice" page.

We all use Gentoo because we enjoy full control over the system and squeezing every last bit of performance out of our systems. We would all be better off to refrain from using language which may be seen as culturally insensitive.

Kindly,
pc_user4

[pietinger]

I have the feeling this post is a counterattack on the thread creator. As a moderator, it is my job to point out early on that an escalation between two users is undesirable. If you want a discussion with the thread creator write him an email.

This is an international support forum in which it is only about a Linux system. No user here will be able to know all the specific rules of all the countries on this planet. That is why the reference to the subject is important.

I don't think there are people who don't realize what the subject is here. But the reference to the subject is crucial. Consider:

To rape a child is forbidden in every country of our planet. But a system can be raped very easily - just install windows.

[Zucca]

For me rice means a seed of a grass which is edible after boiling it a while.
I've seen posts on Lemmy network where people present their customized Unix/Linux desktop as "riced".

Also Wikipedia has few meanings for the word "Ricing". Of which one means tuning a car. Which I assume this topic is about, but instead of car it's Gentoo installation.

The little search I did, I didn't found any racist relation to this word. If there is one it is either "kept well underground" or is very new. Even a online dictionary for my language didn't find any translations.

Currently, I think it's up to the OP to decide if the title needs changing.
_________________
..: Zucca :..
Gentoo IRC channels reside on Libera.Chat.
--

[stefan11111]

[sMueggli]

[pietinger]