Torvalds Is Unconvinced By LTO'ing A Linux Kernel

[milkylainen]

Smaller size benefits all, not just embedded.

I understand that 5% off is important in 4 MB flash storage. However in desktop apps it doesn't matter at all. Hard drives are now 1 TB (ssd) and 4 TB (3.5" hdd). You can also set up raid6 or zfs. So you get tens of terabytes and it's very cheap. You shouldn't bother with binary sizes. In fact there's plenty of room for more functionality in Firefox. Luckily they're working hard at implementing more new features with each release.

Saying that 5% size does not matter on desktop is a misunderstanding on how computers work.
While stability is a major concern as Linus points out, smaller size generally benefits everything.
Loading times, memory contention, cache usage etc.
Just reducing size while maintaining everything else will generate a speedup.
If it is measurable in comparison to general code behavior or not, that is another question.

[caligula]

Liska's thesis was linked in the thread. So soon LTO will allow the compiler to de-duplicate functions. They're also adding it to the linker, as each can catch different parts.

That saves 5% off Firefox binary size, because C++ generated a few thousand variants of a "increase reference count" function. Each with a different type of pointer, yet compiled to the exact same instructions.

I understand that 5% off is important in 4 MB flash storage. However in desktop apps it doesn't matter at all. Hard drives are now 1 TB (ssd) and 4 TB (3.5" hdd). You can also set up raid6 or zfs. So you get tens of terabytes and it's very cheap. You shouldn't bother with binary sizes. In fact there's plenty of room for more functionality in Firefox. Luckily they're working hard at implementing more new features with each release.

[curaga]

Liska's thesis was linked in the thread. So soon LTO will allow the compiler to de-duplicate functions. They're also adding it to the linker, as each can catch different parts.

That saves 5% off Firefox binary size, because C++ generated a few thousand variants of a "increase reference count" function. Each with a different type of pointer, yet compiled to the exact same instructions.

I....
I've toyed with it and have seen many weird errors, so I recompiled everything with bog standard "-march=native -O2"
LTO is interesting as an idea, but with this gcc and binutils... no, thanks. I'll wait for next round of gcc before I try again.

Actually -march=native is even more problematic than the LTO (depends on platform tho), for example, it generates sse instruction which make context switches slower since it takes a while to set them up, which is pretty bad, especially if you have realtime requirements.

LTO seems much safer, though any gain for desktop users is doubtful due to the fact that distribution kernels are super robust and have everything in modules, which is presumably least effective use-case for LTO (and yet, most widely used).

[gens]

Flash space, not the runtime footprint, is usually the problem in routers (which usually carry more RAM than Flash).

it fits now
more webgui ?

[combuster]

It looks like that Linus was so mad that he brought kernel.org down

Your privileges are revoked !

[caligula]

Flash space, not the runtime footprint, is usually the problem in routers (which usually carry more RAM than Flash).

Exactly. They can have 4 MB of Flash and 128-512 MB of RAM. Usually might be the case the whole file system is compressed and uncompressed to RAM while booting. Both kernel and root in the flash. The space requirements can be very strict.

[LinuxFaiT]

WTF

WTF is "Unconvined"? Is it Unconvinced because seriously spellcheck?

[Brane215]

I'm not sure I would want that thing in kernel just yet.

kernel is a thin layer of code anyway. There is not much potential for sizeable speed gains there, even if LTO would work much better than it does.

I've toyed with it and have seen many weird errors, so I recompiled everything with bog standard "-march=native -O2"

LTO is interesting as an idea, but with this gcc and binutils... no, thanks. I'll wait for next round of gcc before I try again.

And this goes even more for kernel. Having schisophrenic userland is bad enough, but stuffing that into kernel intself could be a disaster.

[Pseus]

thing is the code itself does not bring much to the footprint of a running kernel

Flash space, not the runtime footprint, is usually the problem in routers (which usually carry more RAM than Flash).