On Tue, Sep 15, 2020 at 3:09 AM Robert Berger <oecore.mailingl...@gmail.com> wrote: > > Hi Jon, > > That's not really a comment on the reorganization of compiler tunes, but > more like "Do they actually do something meaningful?" > > I posted here[1] some benchmarks and at least with the benchmarks I > tried on the chips I tried there is no obvious impact. > > i.mx6q: > > TUNE_FEATURES = "arm armv7a vfp thumb callconvention-hard" > TARGET_FPU = "hard" > > vs. > > TUNE_FEATURES = "arm vfp cortexa9 neon thumb callconvention-hard" > TARGET_FPU = "hard" > > > i.m8mm: > > TUNE_FEATURES = "aarch64 cortexa53 crc crypto" > TARGET_FPU = "" > > vs. > > TUNE_FEATURES = "aarch64 armv8a crc crypto" > TARGET_FPU = "" > > > [1] > https://yoctoproject.blogspot.com/2020/09/compiler-tunes-benchmarks-with-yocto.html > > Should we expect so see differences?
There are more things at play here than simply performance. But speaking of performance, there might not be much of a benefit between a generic ARMv8.0 and a ARMv8 based core (like A53), but I do expect to see a performance bump for a ARMv8.2 based core (like A76). The delta between the former is much smaller than the latter. The differences are not only performance. Tuning for A76 versus a more generic armv8a allows for security features like branch-protection to be enabled (as it isn't supported in older versions). You get these kind of things "by default" when tuning for the specific model. Thanks, Jon > If so can you suggest benchmarks which show those differences? > > > Regards, > > Robert
-=-=-=-=-=-=-=-=-=-=-=- Links: You receive all messages sent to this group. View/Reply Online (#142612): https://lists.openembedded.org/g/openembedded-core/message/142612 Mute This Topic: https://lists.openembedded.org/mt/76844469/21656 Group Owner: openembedded-core+ow...@lists.openembedded.org Unsubscribe: https://lists.openembedded.org/g/openembedded-core/unsub [arch...@mail-archive.com] -=-=-=-=-=-=-=-=-=-=-=-
