Wasmtime GitHub notifications bot (Sep 09 2021 at 16:15): Wasmtime GitHub notifications bot (Sep 09 2021 at 16:15):afonso360 commented on issue #3322:
It'd be a good idea add a
target aarch64 has_lsetoruntests/atomic-rmw.clifandruntests/atomic-rmw-2.clif, if QEMU supports LSE.
Wasmtime GitHub notifications bot (Sep 09 2021 at 17:35):akirilov-arm commented on issue #3322:
Currently we force QEMU to emulate a processor that lacks LSE support.
Wasmtime GitHub notifications bot (Sep 09 2021 at 17:43):cfallin commented on issue #3322:
I was curious if QEMU has any
-cpuoptions between A72 (the current) andmax(which we disabled to avoid pointer-auth issues on Linux, per #3183); the latest source appears to have a configuration option for the A64FX as well, which is ARMv8.2 (with LSE, without PAC). Maybe we could update the CI config and try that?
Wasmtime GitHub notifications bot (Sep 09 2021 at 17:45):cfallin commented on issue #3322:
(This was added 9 days ago so I expect we'd have to pull down a git checkout of qemu rather than build a release; but that's probably fine IMHO.)
Wasmtime GitHub notifications bot (Sep 10 2021 at 07:36):sparker-arm commented on issue #3322:
I was curious if QEMU has any
-cpuoptions between A72 (the current) andmax(which we disabled to avoid pointer-auth issues on Linux, per #3183); the latest source appears to have a configuration option for the A64FX as well, which is ARMv8.2 (with LSE, without PAC). Maybe we could update the CI config and try that?What might be a more scalable, though less pretty option, would be to use the max configuration and turn off the features that we don't want (only PAC?) @akirilov-arm is working on that feature currently so I imagine we'd need testing at some point in the near future. But I haven't ever had to disable features with QEMU, and briefly looking at the code, I'm wondering if all the features have a nice user facing option...
Wasmtime GitHub notifications bot (Sep 10 2021 at 08:55):afonso360 commented on issue #3322:
Reading this, it looks like we can just
-cpu max,pauth=off. My local qemu 6.0.0 build doesn't recognize this as a invalid feature, I'll try opening a PR to see if CI fails.
Wasmtime GitHub notifications bot (Sep 10 2021 at 08:57):afonso360 edited a comment on issue #3322:
Reading this, it looks like we can just
-cpu max,pauth=off. My local qemu 6.0.0 build recognizes this as a valid feature, I'll try opening a PR to see if CI fails.
Wasmtime GitHub notifications bot (Sep 15 2021 at 16:10):alexcrichton commented on issue #3322:
As a side note, this is somewhat unlikely to get used much in practice at least with Wasmtime itself due to this block which is gated on the off-by-default
stdsimdfeature (which requires nightly). If y'all are interseted in getting this supported in Wasmtime by default that'd need to be updated with a works-on-stable implementation. (I'm not sure of the status of stabilizing the required macro from libstd itself)
Wasmtime GitHub notifications bot (Sep 15 2021 at 16:28):akirilov-arm commented on issue #3322:
Yes, we are aware of that - my colleague @adamgemmell from Arm's Rust team has proposed stabilising the ISA extension test macro in issue rust-lang/rust#86941 independently of the
stdsimdfeature. We are going to update the block after the issue is closed, so that it just checks for the target architecture. In the meantime I think it is fine to continue with the code generation optimizations, and @afonso360's suggestion enables us to do some testing.
Wasmtime GitHub notifications bot (Sep 15 2021 at 16:31):alexcrichton commented on issue #3322:
Nice! And yeah of course I think this is fine to land (as well as any other improvements). Just because Wasmtime doesn't use something by default doesn't mean it's not useful!
Wasmtime GitHub notifications bot (Sep 22 2021 at 10:16):sparker-arm commented on issue #3322:
@cfallin, would you be able to look at this now the QEMU changes have landed?
Last updated: Dec 23 2024 at 13:07 UTC