Expand JULIA_CPU_TARGET docs #58968
Expand JULIA_CPU_TARGET docs #58968
Conversation
IanButterworth
force-pushed
the
ib/cpu_targets_docs
branch
from
8d7b16f to
67fcbfb
Compare
There was a problem hiding this comment.
My understanding of https://docs.julialang.org/en/v1/devdocs/sysimg/#Specifying-multiple-system-image-targets is that clone_all should always be used for the second target, and base(1) for the following ones (if the ISA is strictly larger than the second one). Also, @gbaraldi mentioned on Slack that -xsaveopt and -rdrnd might be needed for the x86_64 targets to workaround some unspecified bugs, but I don't know more than what I'm reporting.
| # For heterogeneous environments, use generic to ensure broad compatibility | ||
| export JULIA_CPU_TARGET="generic" |
There was a problem hiding this comment.
I'm not sure we should recommend JULIA_CPU_TARGET="generic" for heterogeneous systems, stuff like JULIA_CPU_TARGET="generic;haswell" is better and provides the safe fallback of generic
There was a problem hiding this comment.
Shouldn't we recommend the one we actually use to ship Julia?
There was a problem hiding this comment.
True but that's different across platforms. I guess we could link to the build code?
There was a problem hiding this comment.
If we copy it to docs it's likely to get outdated because that's in a different repo
There was a problem hiding this comment.
Maybe we should have an option that copies the value that the sysimage was built with?
There was a problem hiding this comment.
That's proposed here #58970
Part of #55157
#55157 also talks about checks for less specific pkgimage vs. sysimage cpu targets not being in place. That's not addressed here.