abrown opened PR #9185 from abrown:vtune-testing
to bytecodealliance:main
:
This checks that Wasmtime's
--profile=vtune
functionality continues to work. See commit messages for more details.
abrown updated PR #9185.
abrown updated PR #9185.
abrown updated PR #9185.
abrown updated PR #9185.
abrown updated PR #9185.
abrown submitted PR review.
abrown created PR review comment:
The reason for searching for this kind of string is due to error dumps like this one.
abrown updated PR #9185.
abrown updated PR #9185.
abrown has marked PR #9185 as ready for review.
abrown requested pchickey for a review on PR #9185.
abrown requested wasmtime-core-reviewers for a review on PR #9185.
abrown requested wasmtime-default-reviewers for a review on PR #9185.
abrown updated PR #9185.
abrown updated PR #9185.
One thing to note here: I've tried to only install VTune where needed (when we check wasmtime-cli) because it increases the build time by at least a minute. I know we've worked in the past to reduce build times so there's a trade-off to think about: faster builds or test more features?
alexcrichton submitted PR review.
alexcrichton created PR review comment:
To confirm, does this test fail if this flag is left out?
alexcrichton commented on PR #9185:
The integration path here seems like a reasonable starting point to me, thanks! I think it's fine to start here and see how it goes
abrown submitted PR review.
abrown created PR review comment:
No, one can always run vanilla Wasmtime in VTune and that should be valid whether this flag is present or not. What this flag does is inform VTune about the JIT-compiled code so it can display it appropriately in the UI. We've had problems in the past with incorrectly telling VTune about this code which _did_ result in errors and that is what this test is checking: in effect, this is a regression test to check that we correctly tell VTune about the JIT code.
The integration path here seems like a reasonable starting point to me, thanks! I think it's fine to start here and see how it goes
I worked on the
install-vtune-action
a bit more andv2
can now install VTune in 20 seconds using a ~700MB cache (see here). I haven't switched to that version for this PR because I'm not sure if we can afford caches that large; just saying it's available.
alexcrichton submitted PR review:
Ok sounds good to me :+1:. I think things run fast enough here we can eat the extra minute for vtune, but if that becomes an issue we can investigate caching in the future.
abrown updated PR #9185.
Not sure what is going on with CI here: with the latest
matrix.filter == 'linux-x64'
change, VTune no longer installs in the riscv64 and s390x runs, but somehow the check that runsvtune --version
is returning successfully (it should not!) and the full test attempts to run. Not sure what is going on: is thevtune --version
check incorrect somehow or is there some GitHub runner cached state causing issues here?
abrown edited a comment on PR #9185:
Not sure what is going on with CI: with the latest
matrix.filter == 'linux-x64'
change, VTune no longer installs in the riscv64 and s390x runs, but somehow the check that runsvtune --version
is returning successfully (it should not!) and the full test attempts to run. Is thevtune --version
check incorrect somehow or is there some GitHub runner cached state causing issues here?
abrown edited a comment on PR #9185:
Not sure what is going on with CI: with the latest
matrix.filter == 'linux-x64'
change, VTune no longer installs in the riscv64 and s390x runs, but somehow the check that runsvtune --version
is returning successfully (it should not!) and the full test attempts to run. Is thevtune --version
check incorrect somehow or is there some GitHub runner cached state causing issues here? [edit: let's try a rebase...]
abrown updated PR #9185.
alexcrichton commented on PR #9185:
Ah I think the issue is that this needs to be skipped in qemu. Otherwise it's running a non-x64 binary in vtune which isn't going to work. We don't have a great "am I in qemu" test other than
#[cfg(target_arch = "...")]
so running this test only on x64 should be fine
abrown updated PR #9185.
abrown merged PR #9185.
Last updated: Dec 23 2024 at 12:05 UTC