Michael Kowalski

Thanks for explaining this Richard.

Hi @philipturner . I can get past your error above by removing some of the flags you set to install s4tf. I comment these flags: //%install-swiftpm-flags -c release -Xswiftc -Onone...

Success! I added `-rpath` flag. %install-swiftpm-flags -Xlinker "-rpath=/content/Library/tensorflow-2.4.0/usr/lib" The full colab is [here](https://colab.research.google.com/drive/1etOe1g0-mcIk62JkIOqniX9RrYvfoJbj?usp=sharing). Now this: import TensorFlow print(Device.default) let x = Tensor(0) print(x) print(x.device) let y = Tensor(0, on: .defaultXLA)...

> You're saying that if it's in debug mode, you don't have to restart the runtime to load the library? Maybe. I did not try restarting when compiled with `-c...

According to [this](https://github.com/modularml/mojo/issues/1245) it is done in main branch so it should be in nightly branch soon and released this month or next I suppose.

And if you are keen for this shuffle functionality the implementation is trivial. ``` @always_inline("nodebug") fn _shuffle_list[ output_size: Int, mask: StaticIntTuple[output_size] ](a: SIMD[_,_], b: SIMD[a.type, a.size]) -> SIMD[a.type, output_size]: #could...

Possible duplicate of modularml/mojo#1924.

I think the use of "instantiation" in the [`rebind`](https://docs.modular.com/mojo/roadmap.html#the-rebind-builtin) section of sharp edges refers to a related concept but in terms of the compile process. So those uses of `instantiation`...

Edit: originally I thought performance changed with v0.7.0 but that was on my end. Correct numbers are below. ~~I think for these tight loops run many times the difference between...

Sorry if my working example provided previously confused things. This is still broken when the deinterleave is nested as in my first example. Updated code: ```mojo from algorithm import vectorize...