Chris Elrod

Closed by #50277. #48491 was the same issue.

```julia julia> @noinline foo(p) = unsafe_load(p,1) + unsafe_load(p,2) foo (generic function with 1 method) julia> function stacktest() x = Ref{NTuple{2,UInt8}}() GC.@preserve x begin p = Base.unsafe_convert(Ptr{UInt8}, x) unsafe_store!(p, 1, 1)...

I do think this would be useful. The problem I had in the other issue was resolved, and the result of misspecified analytical gradients/Jacobian which was used by AD. The...

Note that `@tturbo` and `@turbo` may really benefit from knowing the sizes at compile time. I had thought that this was part of the `ConvDims` type, but that appears not...

Oh, no, it is expected that LoopVectorization.jl does not support loops like that. LoopModels, the rewrite, will. But it will take some time before that is ready. I wrote about...

That's because you're on Julia master, which removed `Base.unsafe_convert(::Type{Ptr{T}}, x::Array{T})`. I guess I should've been calling `pointer` instead.

That message is `@warn maxlog=1`, so it should generally only display once. It means (as it says) `@turbo` is not actually being used, even on repeat runs -- it just...

I just observed this as well ``` $ cat /etc/os-release NAME="Clear Linux OS" VERSION=1 ID=clear-linux-os ID_LIKE=clear-linux-os VERSION_ID=39930 PRETTY_NAME="Clear Linux OS" ANSI_COLOR="1;35" HOME_URL="https://clearlinux.org" SUPPORT_URL="https://clearlinux.org" BUG_REPORT_URL="mailto:[email protected]" PRIVACY_POLICY_URL="http://www.intel.com/privacy" BUILD_ID=39930 $ lscpu Architecture: x86_64...

These arrays are too small for multithreading to be profitable. I'm surprised it ever did. We adjusted the threading thresholds, which may explain the difference. That is, try making the...

Interesting investigation! It may be that on 1.7, running `ThreadingUtilities.jl` on multiple threads at the same time could result in one `ccall`ing a not yet compiled function? I'm not sure...