Kevin Zhao

Any update on this? If not, I can take over this task now. We ( Linaro ) are using OpenStack deeply in a public cloud

@hasheddan Hi, I do not have time recently working on this. But will still pay attention on this topic.

for external/fmt, previously the commit b90895412f46e18e5b17efdea2c8f79e7d7504b3, build works well. But the master branch build failed.

/assign

Hi @rincebrain, thanks for the comment. The implementation just use 2 add pipeline concurrently, so that in each cycle, the Arm64 CPU can out of order to run the adding...

> > It's a little bit surprising that the superscalar implementations are winning on that many armv8 implementations. Does the NEON pipeline on these things only have 1 execution port...

Yes another reason is maybe @ryao referred, the test on hyperscalers is running on the VM, and the Neon better result I got is from a Baremetal machine. I will...

> Hi @rincebrain, I'm also interested in the SVE support for Raidz and Fletcher4. We have the AF64X hardware in the lab, and I plan to do the SVE support...

@ryao, thanks a lot! It looks that the GCC12 produced ASM looks quite nice to boost the neon performance. I will follow this to re-write the code.

> I would suggest staying with the 4 accumulator version here, but optimizing it based on the clang output that I provided earlier. If both PRs are merged and you...