Google's Propeller Is Beginning To Be Upstreamed For Spinning Faster Program Binaries

[phoronix]

Phoronix: Google's Propeller Is Beginning To Be Upstreamed For Spinning Faster Program Binaries

We have begun seeing the start of upstreaming on Google's Propeller Framework for offering post-link-time binary optimizations in the LLVM compiler stack to offer measurably faster (re)generated binaries...

http://www.phoronix.com/scan.php?page=news_item&px=Google-Propeller-Some-Upstream

[pyler]

Amazing work!

[tuxd3v]

Wouldn't this post-link optimizations cause regressions as security vulnerabilities?

[oiaohm]

Wouldn't this post-link optimizations cause regressions as security vulnerabilities?

Lot of ways no more dangerous than link time optimisations are in the first place. As in post-link and link time optimisation can both cause and cure regressions and security vulnerabilities to exactly the same degree. Yes you could have a do I fell lucky where link time optimisation creates a security fault then the post link optimisation stage fixes it.

Wonder if we will finally see someone do dynamic re-linking in binutils now.

[re:fi.64]

20% resource usage is quite impressive, dang.

[dragorth]

Is this only for linux, or will this work on all LLVM supported OSes targets? OSX is entirely compiled with LLVM, as are all its apps, I think google uses LLVM as much as it can with Android and ChromeOS, and Windows is a supported target. In fact, Firefox on all platforms now uses LLVM, so this could theoretically help them as well competing with Chrome, if this works across platforms.

That would be exciting, actually.

[carewolf]

20% resource usage is quite impressive, dang.

During building or during running? It could mean it gets the same 2-6% speed-up as BOLT but using fewer resources to get those better binaries.

[ms178]

Could someone explain to us laymen how this framework would be used? Would it be feasible for Linux distributions to use this?

[polarathene]

During building or during running?

To perform the optimization(if running meant runtime, not running the optimization).

Could someone explain to us laymen how this framewirk would be used? Would it be feasible for Linux distributions to use this?

Supposedly once upstreamed and that version of LLVM is available to you, it's just toggled via flag when compiling your software/packages(provided they use LLVM).

[flashmozzg]

To perform the optimization(if running meant runtime, not running the optimization).




Supposedly once upstreamed and that version of LLVM is available to you, it's just toggled via flag when compiling your software/packages(provided they use LLVM).

IRC, it depends on ELF and some arch-specific relocation so it's not that simple.