Enzyme support for TimerOutputs.jl #1480
Labels
Comments
|
ah famous GC segfault. MWE defintely needed, or at least a reproducer |
|
@ChrisRackauckas can you post a reproducer/MWE here? |
|
I wasn't able to reproduce the GC segfault, but also am using a different computer for this. But I was able to show that Enzyme errors in the GC, so maybe that's just the newer behavior since I did get a version update between the time that was found and now. using TimerOutputs, Enzyme
const to = TimerOutput()
function f(x::Array{Float64}, y::Array{Float64})
@timeit to "a bit of time" y[1] = x[1] * x[1] + x[2] * x[1]
return nothing
end;
x = [2.0, 2.0]
bx = [0.0, 0.0]
y = [0.0]
by = [1.0];
Enzyme.autodiff(Reverse, f, Duplicated(x, bx), Duplicated(y, by));That gives: |
|
Looking at the message: it looks indeed like it might be the same thing, just Enzyme got better at not segfaulting. |
|
No that’s a different issue, I don’t need a MWE to fix it but add an issue
with the log ( it’s just marking that function as inactive).
…On Sun, Jun 16, 2024 at 9:25 AM Christopher Rackauckas < ***@***.***> wrote:
Looking at the message:
ijl_apply_generic at /cache/build/builder-amdci5-1/julialang/julia-release-1-dot-10/src/gf.c:3076
it looks indeed like it might be the same thing, just Enzyme got better at
not segfaulting.
—
Reply to this email directly, view it on GitHub
<#1480 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AAJTUXDKUHLQX7ZKBN4KXMLZHWG6NAVCNFSM6AAAAABIOJ4DM6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDCNZRGY2DKMZTGM>
.
You are receiving this because you commented.Message ID: <EnzymeAD/Enzyme.
***@***.***>
|
|
@ChrisRackauckas your micro test works on Enzyme main. What is the status of the bigger case? |
|
I think this solved it. I'll reopen if that test was done wrong. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Included is the GC segfault I get when a TimerOutputs.jl
@timeitcommand is within the Enzyme autodiff:message.txt
I think it's probably best to figure out what to label as inactive in that package as its timing data is inherently global I/O mess, though maybe fixing the GC stuff in that kind of situation is a better long-term solution.
I found this as part of a much larger push for Enzyme compatibility ODINN-SciML/ODINN.jl#151 so I don't have an MWE on hand, though I think if you take a standard autodiff test and add a
@timeit to begin ... endinside of it you should hit the same thing.The text was updated successfully, but these errors were encountered: