Christian Guinard

MWE (running from the Metal folder): ``` # Only run the failing test $ julia --project=test -e'using Pkg; Pkg.add(url="https://github.com/christiangnrd/GPUArrays.jl", rev="metal-mwe-hang")' # 1.8.5: Passes $ julia --project=test test/runtests.jl 'gpuarrays/broadcasting' # 1.8.5:...

The `metal-mwe-hang` branch has all broadcasting tests but the `x->getindex.(Ref(x), 1)` one commented out so it should pass or hang within a minute. I'm trying to reduce it but at...

I forgot about `--check-bounds=yes`. Here's a real MWE for you: ``` using Pkg; Pkg.activate(temp=true); Pkg.add(url="https://github.com/JuliaGPU/Metal.jl", rev="main") using Metal getindex.(Ref(MtlArray([0])), 1) ``` Pasting the above code into the REPL after starting...

Bumping GPUArrays seems to have fixed the hanging for me. If I understand correctly this gets around the issue by not calling the problematic code but the problem still exists?

I tried this on few different cpus and architectures, and I get some weird behaviours. Apple Silicon (no change) 1.10.2: ```julia-repl julia> T = Int16; v = ntuple(Returns(T(0)), 8); @b...

Maybe a check to see if a pointer is page-aligned and throw otherwise? And then for `unsafe_copyto!` it could fallback to the old implementation if the pointer is not page-aligned?

I believe that this code works in MacOS 14.4 but not MacOS 13.x. For some reason, on 14.4, `alloc_buffer_nobytes ` always works, even if the array is not page-aligned, while...

Apparently the array/buffer size also has to be divisible by 4096 to work in MacOS 13. https://developer.apple.com/library/archive/documentation/Performance/Conceptual/ManagingMemory/Articles/MemoryAlloc.html#//apple_ref/doc/uid/20001881-99117

All the changes you made seem like improvements to me.

If this gets approved, it might be worth adding a "See also:" reference to `unsafe_vector` in the `view` docstring.