Suggestion Raised For Using PGO + LLVM BOLT To Optimize More Fedora Packages
([Fedora] 5 Hours Ago
PGO + LLVM Bolt Binaries)
- Reference: 0001499999
- News link: https://www.phoronix.com/news/Fedora-Idea-More-PGO-LLVM-BOLT
- Source link:
Outside of the likes of the Arch Linux based CachyOS and Intel's Clear Linux there aren't too many distributions that widely rely on aggressive compiler optimizations in the name of bettering the system performance. A suggestion was raised recently though for Fedora to use profile-guided optimizations (PGO) and post-link optimizations with the likes of LLVM BOLT for more packages, but at this stage it's not clear if such a shift in Fedora package optimizations will actually materialize.
LLVM's BOLT developed by Meta for optimizing the binary layout of compiled software has [1]shown to be able to deliver nice performance gains . Meta has shown over the past several years [2]the very nice performance benefits :
Using PGO with both LLVM Clang and GCC has also proven to be a valuable affair assuming you are able to collect accurate profiles of the application/software being compiled. But both approaches to relying on the compiler to provide faster, more optimized binaries places an additional burden on packagers and mandate further testing. It's rather uncommon seeing Linux distributions applying these compiler optimization techniques at scale.
There was a [3]Fedora discussion raised to begin using PGO and LLVM BOLT for more packages and now moved over to the [4]Fedora devel list .
So far not much commentary and hasn't been filed as any official Fedora change proposal. At this point it seems to be more staged as just an idea rather than the basis of an actual change proposal or commitment to engage in the work for carrying out more tuned packages. But in any event a worthy discussion that will hopefully be had about offering more optimized packages so that no performance is left behind.
[1] https://www.phoronix.com/news/LLVM-BOLT-Faster-Pango
[2] https://www.phoronix.com/news/LLVM-BOLT-Linux-Kernel-2024
[3] https://discussion.fedoraproject.org/t/expand-usage-of-profile-guided-optimization-pgo-and-llvm-bolt-across-fedora-packages/133724
[4] https://lists.fedoraproject.org/archives/list/devel@lists.fedoraproject.org/thread/TB7CTGA3BYMFAB36CGVSW5P6ICIPFM6V/
LLVM's BOLT developed by Meta for optimizing the binary layout of compiled software has [1]shown to be able to deliver nice performance gains . Meta has shown over the past several years [2]the very nice performance benefits :
Using PGO with both LLVM Clang and GCC has also proven to be a valuable affair assuming you are able to collect accurate profiles of the application/software being compiled. But both approaches to relying on the compiler to provide faster, more optimized binaries places an additional burden on packagers and mandate further testing. It's rather uncommon seeing Linux distributions applying these compiler optimization techniques at scale.
There was a [3]Fedora discussion raised to begin using PGO and LLVM BOLT for more packages and now moved over to the [4]Fedora devel list .
So far not much commentary and hasn't been filed as any official Fedora change proposal. At this point it seems to be more staged as just an idea rather than the basis of an actual change proposal or commitment to engage in the work for carrying out more tuned packages. But in any event a worthy discussion that will hopefully be had about offering more optimized packages so that no performance is left behind.
[1] https://www.phoronix.com/news/LLVM-BOLT-Faster-Pango
[2] https://www.phoronix.com/news/LLVM-BOLT-Linux-Kernel-2024
[3] https://discussion.fedoraproject.org/t/expand-usage-of-profile-guided-optimization-pgo-and-llvm-bolt-across-fedora-packages/133724
[4] https://lists.fedoraproject.org/archives/list/devel@lists.fedoraproject.org/thread/TB7CTGA3BYMFAB36CGVSW5P6ICIPFM6V/
edxposed