Fedora's Firefox To Stick With GCC Over Clang, Beefed Up By LTO/PGO Optimizations

[phoronix]

Phoronix: Fedora's Firefox To Stick With GCC Over Clang, Beefed Up By LTO/PGO Optimizations

Last month Fedora developers were planning on building their Firefox package with Clang rather than GCC to follow the move by upstream Mozilla in transitioning their production builds from being built under GCC to LLVM Clang. But now Fedora has reversed course and will continue building with GCC though now benefiting also from PGO and LTO optimizations...

http://www.phoronix.com/scan.php?page=news_item&px=Fedora-Optimized-GCC-Firefox

[carewolf]

Cool. So we get better configured optimizations either way

[treba]

If they archive similar performance gains like the clang build that it huge. Weren't they talking about 20% in certain cases?

And hopefully they do the same for thunderbird, it needs it more badly IMO

[salsadoom]

Great, now do the same for every other package.

I mean, Clear Linux has shown us that LITERALLY NO (I mean, except Clear) Linux distros are making efforts for performance. Its sad that Intel had to come in and do the work that every distro should have been doing from the start. "Oh oh, some guy might still be running a Pentium 60, we can't possibly blah blah". Its a bunch of junk. If your hardware is that damn old nothing modern is going to run well regardless. Even so, there are distros that attempt maintain ancient systems anyway. Then clang shows up and people test it with sane options and lo and behold its faster than the stock crap they run gcc on. What's the damn point of even adding these features to gcc if no one bothers to use them.

Anyway, that was probably inflammatory but I feel like clear linux is shaming other distros into doing what they should be doing. Its just taking longer than it should.

[S.Pam]

Gentoo does compiler optimizations...

[kgonzales]

Great, now do the same for every other package.

I mean, Clear Linux has shown us that LITERALLY NO (I mean, except Clear) Linux distros are making efforts for performance. Its sad that Intel had to come in and do the work that every distro should have been doing from the start. "Oh oh, some guy might still be running a Pentium 60, we can't possibly blah blah". Its a bunch of junk. If your hardware is that damn old nothing modern is going to run well regardless. Even so, there are distros that attempt maintain ancient systems anyway. Then clang shows up and people test it with sane options and lo and behold its faster than the stock crap they run gcc on. What's the damn point of even adding these features to gcc if no one bothers to use them.

Anyway, that was probably inflammatory but I feel like clear linux is shaming other distros into doing what they should be doing. Its just taking longer than it should.

No they are not. They are showing what you can accomplish when you have no legacy users and no one depends on you for production.

[mlau]

No they are not. They are showing what you can accomplish when you have no legacy users and no one depends on you for production.

And when you only target very recent intel cpus.

EDIT: I don't blame them, since it clearly demonstrates what a modern distro for modern hardware can achieve. Gentoo is the same, just need to wait a lot longer for the same result.

[numacross]

And when you only target very recent intel cpus.

"2nd Generation, or later, Intel® Core™ processor family" is not very recent. Sandy Bridge is from 2011.

[skeevy420]

"2nd Generation, or later, Intel® Core™ processor family" is not very recent. Sandy Bridge is from 2011.

My system supports everything Clear Linux requires except for UEFI...so I'll never be able to use Clear Linux. It's a Westmere Xeon workstation. My solution is going to be using Gentoo, the LTO overlay, and keeping up with the various Solus and Clear tweaks.

Said this before and I'll say it again -- we need distros to start building multiple versions of binaries and make more use of march and mtune. For Intel we could do something like: march=generic mtune=westmere for gen 1, march=sandybridge mtune=broadwell for gen 2, march=skylake mtune=icelake-server for gen 3/current -- basically trying to separate processor families by similarly supported instruction sets. The gen 1 binaries would work on everything and the only "code bloat" is AES support for the earliest of processors that don't have AES.

[AsuMagic]

Great, now do the same for every other package.

I mean, Clear Linux has shown us that LITERALLY NO (I mean, except Clear) Linux distros are making efforts for performance. Its sad that Intel had to come in and do the work that every distro should have been doing from the start. "Oh oh, some guy might still be running a Pentium 60, we can't possibly blah blah". Its a bunch of junk. If your hardware is that damn old nothing modern is going to run well regardless. Even so, there are distros that attempt maintain ancient systems anyway. Then clang shows up and people test it with sane options and lo and behold its faster than the stock crap they run gcc on. What's the damn point of even adding these features to gcc if no one bothers to use them.

Anyway, that was probably inflammatory but I feel like clear linux is shaming other distros into doing what they should be doing. Its just taking longer than it should.

I still use a 2009 laptop which works very fine with a T9600. Clear Linux does not support its CPU.